[PATCH] s390: common i/o layer

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

 - Make blacklist busid-aware. Add "all" keyword and ! operator to cio_ignore
   kernel parameter.
 - Add state change notify function for ccw devices (not mandatory) and
   introduce the "device disconnected" state.
 - Remove auto offline from remove function for ccw devices to be able to
   distinguish between user initiated offline and implicit offline due to
   device removal.
 - Store pointer to subchannel structure in the (hardware) subchannel intparm
   and remove the ioinfo array (hurray...). Remove intparm parameter of
   cio_start.
 - Use busid instead of subchannel number for debug output.
 - Use an opm mask to track which paths are logically online for a subchannel.
 - Pathgroup every device it was requested for, even single path devices.
 - Give i/o on a logically switched off path a grace period to complete, then
   kill the i/o to get the path offline.
 - Correctly initialize all spin_locks with spin_lock_init.
 - Handle status pending/busy while disabling subchannel.
 - Set busid already in cio_validate_subchannel.
 - Add s390_root_dev_{register,unregister} functions.
 - Do stcrw() inside a kernel thread. Add crw overflow handling.
 - Use subchannel lock directly instead of ccw device lock pointer in
   ccw_device_recognition to avoid accessing an already free structure.
 - Take/release ccw device lock in ccw_device_console_enable.
 - Don't wipe out the busid field in ccw_device_console_enable.
 - Call ccw_device_unregister() directly on a notoper event - delaying it via
   queue_work is harmful (subchannel may be removed before ccw_device).
 - Handle not opertional condition in ccw_device_cancel_halt_clear.
 - Correct status pending handling: don't collect pending status directly
   but wait for the interrupt to show up.
 - Enable subchannel when trying a steal lock operation.
 - Introduce doverify bit for delayed path verification.
 - Fix locking in __ccw_device_retry_loop/read_conf_data/read_dev/chars.
 - Make SPID retry mechanism more obvious.
 - qdio: check return code of ccw_device_{halt,clear} in qdio_cleanup. Don't
   try to wait for an interrupt we won't get.
 - qdio: fix shared indicators.
 - qdio: add code to handle i/o killed by cio with active queues.
 - qdio: don't do a shutdown on timeout in interrupt context.
 - Update cio documentation.

Documentation/s390/driver-model.txt

@@ -32,6 +32,9 @@ cutype:	    The control unit type / model.
 devtype:    The device type / model, if applicable.
 
 online:     An interface to set the device online and offline.
+	    In the special case of the device being disconnected (see the
+	    notify function under 1.2), piping 0 to online will focibly delete
+	    the device.
 
 The device drivers can add entries to export per-device data and interfaces.
 
@@ -82,6 +85,7 @@ struct ccw_driver {
 	int (*remove) (struct ccw_device *);
 	int (*set_online) (struct ccw_device *);
 	int (*set_offline) (struct ccw_device *);
+	int (*notify) (struct ccw_device *, int);
 	struct device_driver driver;
 	char *name;
 };
@@ -143,6 +147,28 @@ Parameters:   cdev       - the device to be deactivated. The common layer has
 			   verified that the device is online.
 
 
+notify: This function is called by the common I/O layer for some state changes
+	of the device.
+	Signalled to the driver are:
+	* In online state, device detached (CIO_GONE) or last path gone
+	  (CIO_NO_PATH). The driver must return !0 to keep the device; for
+	  return code 0, the device will be deleted as usual (also when no
+	  notify function is registerd). If the driver wants to keep the
+	  device, it is moved into disconnected state.
+	* In disconnected state, device operational again (CIO_OPER). The
+	  common I/O layer performs some sanity checks on device number and
+	  Device / CU to be reasonably sure if it is still the same device.
+	  If not, the old device is removed and a new one registered. By the
+	  return code of the notify function the device driver signals if it
+	  wants the device back: !0 for keeping, 0 to make the device being
+	  removed and re-registered.
+	
+int (*notify) (struct ccw_device *, int);
+
+Parameters:   cdev    - the device whose state changed.
+	      event   - the event that happened. This can be one of CIO_GONE,
+		        CIO_NO_PATH or CIO_OPER.
+
 The handler field of the struct ccw_device is meant to be set to the interrupt
 handler for the device. In order to accommodate drivers which use several 
 distinct handlers (e.g. multi subchannel devices), this is a member of ccw_device

drivers/s390/cio/blacklist.c

 /*
  *  drivers/s390/cio/blacklist.c
  *   S/390 common I/O routines -- blacklisting of specific devices
- *   $Revision: 1.24 $
+ *   $Revision: 1.27 $
  *
  *    Copyright (C) 1999-2002 IBM Deutschland Entwicklung GmbH,
  *			      IBM Corporation
@@ -62,51 +62,117 @@ blacklist_range (range_action action, unsigned int from, unsigned int to)
 }
 
 /*
- * function: blacklist_strtoul
- * Strip leading '0x' and interpret the values as Hex
+ * Function: blacklist_busid
+ * Get devno/busid from given string.
+ * Shamelessly grabbed from dasd_devmap.c.
  */
 static inline int
-blacklist_strtoul (const char *str, char **stra)
+blacklist_busid(char **str, int *id0, int *id1, int *devno)
 {
-	if (*str == '0') {
-		if (*(++str) == 'x')  /* strip leading zero */
-			str++;	      /* strip leading x */
+	int val, old_style;
+ 
+	/* check for leading '0x' */
+	old_style = 0;
+	if ((*str)[0] == '0' && (*str)[1] == 'x') {
+		*str += 2;
+		old_style = 1;
 	}
-	return simple_strtoul (str, stra, 16); /* interpret anything as hex */
+	if (!isxdigit((*str)[0]))	/* We require at least one hex digit */
+		return -EINVAL;
+	val = simple_strtoul(*str, str, 16);
+	if (old_style || (*str)[0] != '.') {
+		*id0 = *id1 = 0;
+		if (val < 0 || val > 0xffff)
+			return -EINVAL;
+		*devno = val;
+		return 0;
+	}
+	/* New style x.y.z busid */
+	if (val < 0 || val > 0xff)
+		return -EINVAL;
+	*id0 = val;
+	(*str)++;
+	if (!isxdigit((*str)[0]))	/* We require at least one hex digit */
+		return -EINVAL;
+	val = simple_strtoul(*str, str, 16);
+	if (val < 0 || val > 0xff || (*str)++[0] != '.')
+		return -EINVAL;
+	*id1 = val;
+	if (!isxdigit((*str)[0]))	/* We require at least one hex digit */
+		return -EINVAL;
+	val = simple_strtoul(*str, str, 16);
+	if (val < 0 || val > 0xffff)
+		return -EINVAL;
+	*devno = val;
+	return 0;
 }
 
 static inline int
 blacklist_parse_parameters (char *str, range_action action)
 {
-	unsigned int from, to;
+	unsigned int from, to, from_id0, to_id0, from_id1, to_id1;
+	char *sav;
 
+	sav = str;
 	while (*str != 0 && *str != '\n') {
-		if (!isxdigit(*str)) {
-			printk(KERN_WARNING "blacklist_setup: error parsing "
-			       "\"%s\"\n", str);
-			return 0;
+		range_action ra = action;
+		if (*str == '!') {
+			ra = !action;
+			++str;
 		}
 
-		from = blacklist_strtoul (str, &str);
-		to = (*str == '-') ? blacklist_strtoul (str+1, &str) : from;
-
+		/*
+		 * Since we have to parse the proc commands and the
+		 * kernel arguments we have to check four cases
+		 */
+		if (strncmp(str,"all,",4) == 0 || strcmp(str,"all") == 0 ||
+		    strncmp(str,"all\n",4) == 0 || strncmp(str,"all ",4) == 0) {
+			from = 0;
+			to = __MAX_SUBCHANNELS;
+			str += 3;
+		} else {
+			int rc;
+
+			rc = blacklist_busid(&str, &from_id0,
+					     &from_id1, &from);
+			if (rc)
+				goto out_err;
+			to = from;
+			to_id0 = from_id0;
+			to_id1 = from_id1;
+			if (*str == '-') {
+				str++;
+				rc = blacklist_busid(&str, &to_id0, &to_id1,
+						     &to);
+				if (rc)
+					goto out_err;
+			}
+			if ((from_id0 != to_id0) || (from_id1 != to_id1))
+				goto out_err;
+		}
+		/* FIXME: ignoring id0 and id1 here. */
 		pr_debug("blacklist_setup: adding range "
-			 "from 0x%04x to 0x%04x\n", from, to);
-		blacklist_range (action, from, to);
+			 "from 0.0.%04x to 0.0.%04x\n", from, to);
+		blacklist_range (ra, from, to);
 
 		if (*str == ',')
 			str++;
 	}
 	return 1;
+out_err:
+	printk(KERN_WARNING "blacklist_setup: error parsing \"%s\"\n", sav);
+	return 0;
 }
 
 /* Parsing the commandline for blacklist parameters, e.g. to blacklist
- * device IDs 0x1234, 0x1235 and 0x1236, you could use any of:
+ * bus ids 0.0.1234, 0.0.1235 and 0.0.1236, you could use any of:
  * - cio_ignore=1234-1236
  * - cio_ignore=0x1234-0x1235,1236
  * - cio_ignore=0x1234,1235-1236
  * - cio_ignore=1236 cio_ignore=1234-0x1236
  * - cio_ignore=1234 cio_ignore=1236 cio_ignore=0x1235
+ * - cio_ignore=0.0.1234-0.0.1236
+ * - cio_ignore=0.0.1234,0x1235,1236
  * - ...
  */
 static int __init
@@ -134,6 +200,7 @@ is_blacklisted (int devno)
 
 #ifdef CONFIG_PROC_FS
 
+extern void css_reiterate_subchannels(void);
 /*
  * Function: s390_redo_validation
  * Look for no longer blacklisted devices
@@ -141,12 +208,9 @@ is_blacklisted (int devno)
 static inline void
 s390_redo_validation (void)
 {
-	int irq;
-
 	CIO_TRACE_EVENT (0, "redoval");
 
-	for (irq = 0; irq <= highest_subchannel; irq++)
-		css_probe_device(irq);
+	css_reiterate_subchannels();
 }
 
 /*
@@ -157,10 +221,7 @@ static inline void
 blacklist_parse_proc_parameters (char *buf)
 {
 	if (strncmp (buf, "free ", 5) == 0) {
-		if (strstr (buf + 5, "all"))
-			blacklist_range (free, 0, __MAX_SUBCHANNELS);
-		else
-			blacklist_parse_parameters (buf + 5, free);
+		blacklist_parse_parameters (buf + 5, free);
 	} else if (strncmp (buf, "add ", 4) == 0) {
 		/* 
 		 * We don't need to check for known devices since
@@ -179,10 +240,11 @@ blacklist_parse_proc_parameters (char *buf)
 	s390_redo_validation ();
 }
 
+/* FIXME: These should be real bus ids and not home-grown ones! */
 static int cio_ignore_read (char *page, char **start, off_t off,
 			    int count, int *eof, void *data)
 {
-	const unsigned int entry_size = 14; /* "0xABCD-0xEFGH\n" */
+	const unsigned int entry_size = 18; /* "0.0.ABCD-0.0.EFGH\n" */
 	long devno;
 	int len;
 
@@ -192,12 +254,12 @@ static int cio_ignore_read (char *page, char **start, off_t off,
 	     devno <= __MAX_SUBCHANNELS && len + entry_size < count; devno++) {
 		if (!test_bit(devno, bl_dev))
 			continue;
-		len += sprintf(page + len, "0x%04lx", devno);
+		len += sprintf(page + len, "0.0.%04lx", devno);
 		if (test_bit(devno + 1, bl_dev)) { /* print range */
 			while (++devno < __MAX_SUBCHANNELS)
 				if (!test_bit(devno, bl_dev))
 					break;
-			len += sprintf(page + len, "-0x%04lx", --devno);
+			len += sprintf(page + len, "-0.0.%04lx", --devno);
 		}
 		len += sprintf(page + len, "\n");
 	}

drivers/s390/cio/ccwgroup.c

 /*
  *  drivers/s390/cio/ccwgroup.c
  *  bus driver for ccwgroup
- *   $Revision: 1.17 $
+ *   $Revision: 1.19 $
  *
  *    Copyright (C) 2002 IBM Deutschland Entwicklung GmbH,
  *                       IBM Corporation
@@ -337,7 +337,6 @@ ccwgroup_remove (struct device *dev)
 {
 	struct ccwgroup_device *gdev;
 	struct ccwgroup_driver *gdrv;
-	int ret;
 
 	gdev = to_ccwgroupdev(dev);
 	gdrv = to_ccwgroupdrv(dev->driver);
@@ -345,11 +344,10 @@ ccwgroup_remove (struct device *dev)
 	pr_debug("%s: device %s\n", __func__, gdev->dev.name);
 
 	device_remove_file(dev, &dev_attr_online);
-	ccwgroup_set_offline(gdev);
-
-	ret = (gdrv && gdrv->remove) ? gdrv->remove(gdev) : 0;
 
-	return ret;
+	if (gdrv && gdrv->remove)
+		gdrv->remove(gdev);
+	return 0;
 }
 
 int
@@ -418,20 +416,20 @@ __ccwgroup_get_gdev_by_cdev(struct ccw_device *cdev)
 	return gdev;
 }
 
-int
+void
 ccwgroup_remove_ccwdev(struct ccw_device *cdev)
 {
 	struct ccwgroup_device *gdev;
 
+	/* Ignore offlining errors, device is gone anyway. */
+	ccw_device_set_offline(cdev);
 	/* If one of its devices is gone, the whole group is done for. */
 	gdev = __ccwgroup_get_gdev_by_cdev(cdev);
 	if (gdev) {
-		ccwgroup_set_offline(gdev);
 		__ccwgroup_remove_symlinks(gdev);
 		device_unregister(&gdev->dev);
 		put_device(&gdev->dev);
 	}
-	return 0;
 }
 
 MODULE_LICENSE("GPL");

drivers/s390/cio/chsc.c

 /*
  *  drivers/s390/cio/chsc.c
  *   S/390 common I/O routines -- channel subsystem call
- *   $Revision: 1.78 $
+ *   $Revision: 1.92 $
  *
  *    Copyright (C) 1999-2002 IBM Deutschland Entwicklung GmbH,
  *			      IBM Corporation
@@ -17,40 +17,40 @@
 #include <linux/device.h>
 
 #include <asm/cio.h>
-#include <asm/ccwdev.h> // FIXME: layering violation, remove this
 
 #include "css.h"
 #include "cio.h"
 #include "cio_debug.h"
-#include "device.h" // FIXME: layering violation, remove this
 #include "ioasm.h"
 #include "chsc.h"
 
 #define CHPID_LONGS (256 / (8 * sizeof(long))) /* 256 chpids */
-static unsigned long chpids[CHPID_LONGS];
-static unsigned long chpids_logical[CHPID_LONGS] = {[0 ... CHPID_LONGS-1] -1};
-static unsigned long chpids_known[CHPID_LONGS];
-
 static struct channel_path *chps[NR_CHPIDS];
 
-static int cio_chsc_desc_avail;
-
 static int new_channel_path(int chpid, int status);
 
-static int
-set_chp_status(int chp, int status)
+static inline void
+set_chp_online(int chp, int onoff)
 {
-	if (chps[chp] == NULL)
-		return -EINVAL;
+	chps[chp]->state.online = onoff;
+}
 
-	chps[chp]->state = status;
-	return 0;
+static inline void
+set_chp_logically_online(int chp, int onoff)
+{
+	chps[chp]->state.logically_online = onoff;
 }
 
-static inline int
-chsc_chpid_logical (struct subchannel *sch, int chp)
+static int
+get_chp_status(int chp)
 {
-	return test_bit (sch->schib.pmcw.chpid[chp], chpids_logical);
+	int ret;
+
+	if (!chps[chp])
+		return 0;
+	ret = chps[chp]->state.online ? CHP_ONLINE : CHP_STANDBY;
+	return (chps[chp]->state.logically_online ?
+		ret : ret | CHP_LOGICALLY_OFFLINE);
 }
 
 void
@@ -58,24 +58,21 @@ chsc_validate_chpids(struct subchannel *sch)
 {
 	int mask, chp;
 
-	if (sch->lpm == 0)
-		return;
-
 	for (chp = 0; chp <= 7; chp++) {
 		mask = 0x80 >> chp;
-		if (sch->lpm & mask)
-			if (!chsc_chpid_logical(sch, chp))
-				/* disable using this path */
-				sch->lpm &= ~mask;
+		if (get_chp_status(sch->schib.pmcw.chpid[chp])
+		    &  CHP_LOGICALLY_OFFLINE)
+			/* disable using this path */
+			sch->opm &= ~mask;
 	}
 }
 
 /* FIXME: this is _always_ called for every subchannel. shouldn't we
  *	  process more than one at a time? */
 static int
-chsc_get_sch_desc_irq(int irq, void *page)
+chsc_get_sch_desc_irq(struct subchannel *sch, void *page)
 {
-	int ccode, chpid, j;
+	int ccode, j;
 
 	struct {
 		struct chsc_header request;
@@ -106,8 +103,8 @@ chsc_get_sch_desc_irq(int irq, void *page)
 		.code   = 0x0004,
 	};
 
-	ssd_area->f_sch = irq;
-	ssd_area->l_sch = irq;
+	ssd_area->f_sch = sch->irq;
+	ssd_area->l_sch = sch->irq;
 
 	ccode = chsc(ssd_area);
 	if (ccode > 0) {
@@ -120,19 +117,17 @@ chsc_get_sch_desc_irq(int irq, void *page)
 		break;
 	case 0x0002:
 		CIO_CRW_EVENT(2, "Invalid command!\n");
+		return -EINVAL;
 	case 0x0003:
 		CIO_CRW_EVENT(2, "Error in chsc request block!\n");
 		return -EINVAL;
-		break;
 	case 0x0004:
 		CIO_CRW_EVENT(2, "Model does not provide ssd\n");
 		return -EOPNOTSUPP;
-		break;
 	default:
 		CIO_CRW_EVENT(2, "Unknown CHSC response %d\n",
 			      ssd_area->response.code);
 		return -EIO;
-		break;
 	}
 
 	/*
@@ -151,7 +146,7 @@ chsc_get_sch_desc_irq(int irq, void *page)
 	 */
 	if (ssd_area->st > 3) { /* uhm, that looks strange... */
 		CIO_CRW_EVENT(0, "Strange subchannel type %d"
-			      " for sch %x\n", ssd_area->st, irq);
+			      " for sch %s\n", ssd_area->st, sch->dev.bus_id);
 		/*
 		 * There may have been a new subchannel type defined in the
 		 * time since this code was written; since we don't know which
@@ -160,14 +155,11 @@ chsc_get_sch_desc_irq(int irq, void *page)
 		return 0;
 	} else {
 		const char *type[4] = {"I/O", "chsc", "message", "ADM"};
-		CIO_CRW_EVENT(6, "ssd: sch %x is %s subchannel\n",
-			      irq, type[ssd_area->st]);
-		if (ioinfo[irq] == NULL)
-			/* FIXME: we should do device rec. here... */
-			return 0;
-
-		ioinfo[irq]->ssd_info.valid = 1;
-		ioinfo[irq]->ssd_info.type = ssd_area->st;
+		CIO_CRW_EVENT(6, "ssd: sch %s is %s subchannel\n",
+			      sch->dev.bus_id, type[ssd_area->st]);
+
+		sch->ssd_info.valid = 1;
+		sch->ssd_info.type = ssd_area->st;
 	}
 
 	if (ssd_area->st == 0 || ssd_area->st == 2) {
@@ -175,89 +167,73 @@ chsc_get_sch_desc_irq(int irq, void *page)
 			if (!((0x80 >> j) & ssd_area->path_mask &
 			      ssd_area->fla_valid_mask))
 				continue;
-			chpid = ssd_area->chpid[j];
-			if (chpid
-			    && (!test_and_set_bit (chpid, chpids_known))
-			    && (test_bit (chpid, chpids_logical)))
-				set_bit	 (chpid, chpids);
-
-			ioinfo[irq]->ssd_info.chpid[j] = chpid;
-			ioinfo[irq]->ssd_info.fla[j]   = ssd_area->fla[j];
+			sch->ssd_info.chpid[j] = ssd_area->chpid[j];
+			sch->ssd_info.fla[j]   = ssd_area->fla[j];
 		}
 	}
-
 	return 0;
 }
 
-static int
-chsc_get_sch_descriptions(void)
+int
+css_get_ssd_info(struct subchannel *sch)
 {
-	int irq;
-	int err;
+	int ret;
 	void *page;
 
-	CIO_TRACE_EVENT( 4, "gsdesc");
-
-	/*
-	 * get information about chpids and link addresses
-	 * by executing the chsc command 'store subchannel description'
-	 */
 	page = (void *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
 	if (!page)
 		return -ENOMEM;
-
-	err = 0;
-	for (irq = 0; irq <= highest_subchannel; irq++) {
-		/*
-		 * retrieve information for each sch
-		 */
-		err = chsc_get_sch_desc_irq(irq, page);
-		if (err) {
-			static int cio_chsc_err_msg;
-
-			if (!cio_chsc_err_msg) {
-				printk(KERN_ERR
-				       "chsc_get_sch_descriptions:"
-				       " Error %d while doing chsc; "
-				       "processing some machine checks may "
-				       "not work\n", err);
-				cio_chsc_err_msg = 1;
-			}
-			goto out;
+	spin_lock_irq(&sch->lock);
+	ret = chsc_get_sch_desc_irq(sch, page);
+	if (ret) {
+		static int cio_chsc_err_msg;
+		
+		if (!cio_chsc_err_msg) {
+			printk(KERN_ERR
+			       "chsc_get_sch_descriptions:"
+			       " Error %d while doing chsc; "
+			       "processing some machine checks may "
+			       "not work\n", ret);
+			cio_chsc_err_msg = 1;
 		}
-		clear_page(page);
 	}
-	cio_chsc_desc_avail = 1;
-out:
+	spin_unlock_irq(&sch->lock);
 	free_page((unsigned long)page);
-	return err;
+	if (!ret) {
+		int j, chpid;
+		/* Allocate channel path structures, if needed. */
+		for (j = 0; j < 8; j++) {
+			chpid = sch->ssd_info.chpid[j];
+			if (chpid && !get_chp_status(chpid))
+			    new_channel_path(chpid, CHP_ONLINE);
+		}
+	}
+	return ret;
 }
 
-__initcall(chsc_get_sch_descriptions);
-
-
-static inline void
-s390_subchannel_remove_chpid(struct subchannel *sch, __u8 chpid)
+static int
+s390_subchannel_remove_chpid(struct device *dev, void *data)
 {
 	int j;
 	int mask;
+	struct subchannel *sch;
+	__u8 *chpid;
 
+	sch = to_subchannel(dev);
+	chpid = data;
 	for (j = 0; j < 8; j++)
-		if (sch->schib.pmcw.chpid[j] == chpid)
+		if (sch->schib.pmcw.chpid[j] == *chpid)
 			break;
 	if (j >= 8)
-		return;
+		return 0;
 
 	mask = 0x80 >> j;
 	spin_lock(&sch->lock);
 
-	chsc_validate_chpids(sch);
-
 	stsch(sch->irq, &sch->schib);
-	if (sch->vpm == mask) {
-		dev_fsm_event(sch->dev.driver_data, DEV_EVENT_NOTOPER);
-		goto out_unlock;
-	}
+	if (sch->vpm == mask)
+		goto out_unreg;
+
 	if ((sch->schib.scsw.actl & (SCSW_ACTL_CLEAR_PEND |
 				     SCSW_ACTL_HALT_PEND |
 				     SCSW_ACTL_START_PEND |
@@ -265,83 +241,64 @@ s390_subchannel_remove_chpid(struct subchannel *sch, __u8 chpid)
 	    (sch->schib.pmcw.lpum == mask)) {
 		int cc = cio_cancel(sch);
 		
-		if (cc == -ENODEV) {
-			dev_fsm_event(sch->dev.driver_data, DEV_EVENT_NOTOPER);
-			goto out_unlock;
-		}
+		if (cc == -ENODEV)
+			goto out_unreg;
 
 		if (cc == -EINVAL) {
-			struct ccw_device *cdev;
-
 			cc = cio_clear(sch);
-			if (cc == -ENODEV) {
-				dev_fsm_event(sch->dev.driver_data,
-					      DEV_EVENT_NOTOPER);
-				goto out_unlock;
-			}
+			if (cc == -ENODEV)
+				goto out_unreg;
 			/* Call handler. */
-			cdev = sch->dev.driver_data;
-			cdev->private->state = DEV_STATE_CLEAR_VERIFY;
-			if (cdev->handler)
-				cdev->handler(cdev, cdev->private->intparm,
-					      ERR_PTR(-EIO));
+			if (sch->driver && sch->driver->termination)
+				sch->driver->termination(&sch->dev);
 			goto out_unlock;
 		}
 	} else if ((sch->schib.scsw.actl & SCSW_ACTL_DEVACT) &&
 		   (sch->schib.scsw.actl & SCSW_ACTL_SCHACT) &&
 		   (sch->schib.pmcw.lpum == mask)) {
-		struct ccw_device *cdev;
 		int cc;
 
 		cc = cio_clear(sch);
-		if (cc == -ENODEV) {
-			dev_fsm_event(sch->dev.driver_data, DEV_EVENT_NOTOPER);
-			goto out_unlock;
-		}
+		if (cc == -ENODEV)
+			goto out_unreg;
 		/* Call handler. */
-		cdev = sch->dev.driver_data;
-		cdev->private->state = DEV_STATE_CLEAR_VERIFY;
-		if (cdev->handler)
-			cdev->handler(cdev, cdev->private->intparm,
-				      ERR_PTR(-EIO));
+		if (sch->driver && sch->driver->termination)
+			sch->driver->termination(&sch->dev);
 		goto out_unlock;
 	}
 
 	/* trigger path verification. */
-	dev_fsm_event(sch->dev.driver_data, DEV_EVENT_VERIFY);
+	if (sch->driver && sch->driver->verify)
+		sch->driver->verify(&sch->dev);
 out_unlock:
 	spin_unlock(&sch->lock);
+	return 0;
+out_unreg:
+	spin_unlock(&sch->lock);
+	if (sch->driver && sch->driver->notify &&
+	    sch->driver->notify(&sch->dev, CIO_NO_PATH))
+		return 0;
+	device_unregister(&sch->dev);
+	sch->schib.pmcw.intparm = 0;
+	cio_modify(sch);
+	return 0;
 }
 
-/* FIXME: don't iterate all subchannels but use driver_for_each_dev */
 static inline void
 s390_set_chpid_offline( __u8 chpid)
 {
 	char dbf_txt[15];
-	struct subchannel *sch;
-	int irq;
 
 	sprintf(dbf_txt, "chpr%x", chpid);
 	CIO_TRACE_EVENT(2, dbf_txt);
 
-	clear_bit(chpid, chpids);
-	if (!test_and_clear_bit(chpid, chpids_known))
+	if (!get_chp_status(chpid))
 		return;	 /* we didn't know the chpid anyway */
 
-	set_chp_status(chpid, CHP_OFFLINE);
-
-#if 0
-	driver_for_each_dev(io_subchannel_driver, chpid, s390_subchannel_remove_chpid);
-#else
-
-	for (irq = 0; irq <= highest_subchannel; irq++) {
-		sch = ioinfo[irq];
-		if (sch == NULL)
-			continue;  /* we don't know the device anyway */
+	set_chp_online(chpid, 0);
 
-		s390_subchannel_remove_chpid(sch, chpid);
-	}
-#endif
+	bus_for_each_dev(&css_bus_type, NULL, &chpid,
+			 s390_subchannel_remove_chpid);
 }
 
 static int
@@ -353,7 +310,7 @@ s390_process_res_acc_sch(u8 chpid, __u16 fla, u32 fla_mask,
 	int ccode;
 	
 	/* Update our ssd_info */
-	if (chsc_get_sch_desc_irq(sch->irq, page))
+	if (chsc_get_sch_desc_irq(sch, page))
 		return 0;
 	
 	found = 0;
@@ -407,7 +364,7 @@ s390_process_res_acc (u8 chpid, __u16 fla, u32 fla_mask)
 	 * will we have to do.
 	 */
 
-	if (!test_bit(chpid, chpids_logical))
+	if (get_chp_status(chpid) & CHP_LOGICALLY_OFFLINE)
 		return; /* no need to do the rest */
 
 	page = (void *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
@@ -417,7 +374,7 @@ s390_process_res_acc (u8 chpid, __u16 fla, u32 fla_mask)
 	for (irq = 0; irq < __MAX_SUBCHANNELS; irq++) {
 		int chp_mask;
 
-		sch = ioinfo[irq];
+		sch = get_subchannel_by_schid(irq);
 		if (!sch) {
 			/*
 			 * We don't know the device yet, but since a path
@@ -450,17 +407,16 @@ s390_process_res_acc (u8 chpid, __u16 fla, u32 fla_mask)
 				continue;
 		}
 
-		sch->lpm = (sch->schib.pmcw.pim &
-			    sch->schib.pmcw.pam &
-			    sch->schib.pmcw.pom)
-			| chp_mask;
-
-		chsc_validate_chpids(sch);
+		sch->lpm = ((sch->schib.pmcw.pim &
+			     sch->schib.pmcw.pam &
+			     sch->schib.pmcw.pom)
+			    | chp_mask) & sch->opm;
 
-		dev_fsm_event(sch->dev.driver_data, DEV_EVENT_VERIFY);
+		if (sch->driver && sch->driver->verify)
+			sch->driver->verify(&sch->dev);
 
 		spin_unlock_irq(&sch->lock);
-
+		put_device(&sch->dev);
 		if (fla_mask != 0)
 			break;
 	}
@@ -497,8 +453,8 @@ __get_chpid_from_lir(void *data)
 	return (u16) (lir->indesc[0]&0x000000ff);
 }
 
-static void
-do_process_crw(void *ignore)
+void
+chsc_process_crw(void)
 {
 	int chpid;
 	struct {
@@ -608,7 +564,7 @@ do_process_crw(void *ignore)
 			if (chps[sei_area->rsid] == NULL)
 				new_channel_path(sei_area->rsid, CHP_ONLINE);
 			else
-				set_chp_status(sei_area->rsid, CHP_ONLINE);
+				set_chp_online(sei_area->rsid, 1);
 			
 			if ((sei_area->vf & 0x80) == 0) {
 				pr_debug("chpid: %x\n", sei_area->rsid);
@@ -639,14 +595,6 @@ do_process_crw(void *ignore)
 	free_page((unsigned long)sei_area);
 }
 
-void
-chsc_process_crw(void)
-{
-	static DECLARE_WORK(work, do_process_crw, 0);
-
-	queue_work(ccw_device_work, &work);
-}
-
 static void
 chp_add(int chpid)
 {
@@ -654,7 +602,7 @@ chp_add(int chpid)
 	int irq, ret;
 	char dbf_txt[15];
 
-	if (!test_bit(chpid, chpids_logical))
+	if (get_chp_status(chpid) & CHP_LOGICALLY_OFFLINE)
 		return; /* no need to do the rest */
 	
 	sprintf(dbf_txt, "cadd%x", chpid);
@@ -663,7 +611,7 @@ chp_add(int chpid)
 	for (irq = 0; irq < __MAX_SUBCHANNELS; irq++) {
 		int i;
 
-		sch = ioinfo[irq];
+		sch = get_subchannel_by_schid(irq);
 		if (!sch) {
 			ret = css_probe_device(irq);
 			if (ret == -ENXIO)
@@ -686,16 +634,16 @@ chp_add(int chpid)
 			spin_unlock(&sch->lock);
 			return;
 		}
-		sch->lpm = (sch->schib.pmcw.pim &
-			    sch->schib.pmcw.pam &
-			    sch->schib.pmcw.pom)
-			| 0x80 >> i;
-
-		chsc_validate_chpids(sch);
+		sch->lpm = ((sch->schib.pmcw.pim &
+			     sch->schib.pmcw.pam &
+			     sch->schib.pmcw.pom)
+			    | 0x80 >> i) & sch->opm;
 
-		dev_fsm_event(sch->dev.driver_data, DEV_EVENT_VERIFY);
+		if (sch->driver && sch->driver->verify)
+			sch->driver->verify(&sch->dev);
 
 		spin_unlock(&sch->lock);
+		put_device(&sch->dev);
 	}
 }
 
@@ -716,12 +664,80 @@ chp_process_crw(int chpid, int on)
 		if (chps[chpid] == NULL)
 			new_channel_path(chpid, CHP_ONLINE);
 		else
-			set_chp_status(chpid, CHP_ONLINE);
+			set_chp_online(chpid, 1);
 		/* Avoid the extra overhead in process_rec_acc. */
 		chp_add(chpid);
 	}
 }
 
+static inline void
+__check_for_io_and_kill(struct subchannel *sch, int index)
+{
+	int cc;
+
+	cc = stsch(sch->irq, &sch->schib);
+	if (cc)
+		return;
+	if (sch->schib.scsw.actl && sch->schib.pmcw.lpum == (0x80 >> index))
+		device_set_waiting(sch);
+}
+
+static inline void
+__s390_subchannel_vary_chpid(struct subchannel *sch, __u8 chpid, int on)
+{
+	int chp;
+
+	if (!sch->ssd_info.valid)
+		return;
+	
+	for (chp = 0; chp < 8; chp++) {
+		if (sch->ssd_info.chpid[chp] == chpid) {
+			if (on) {
+				sch->opm |= (0x80 >> chp);
+				sch->lpm |= (0x80 >> chp);
+			} else {
+				sch->opm &= ~(0x80 >> chp);
+				sch->lpm &= ~(0x80 >> chp);
+				/*
+				 * Give running I/O a grace period in which it
+				 * can successfully terminate, even using the
+				 * just varied off path. Then kill it.
+				 */
+				__check_for_io_and_kill(sch, chp);
+			}
+			if (sch->driver && sch->driver->verify)
+				sch->driver->verify(&sch->dev);
+			break;
+		}
+	}
+}
+
+static int
+s390_subchannel_vary_chpid_off(struct device *dev, void *data)
+{
+	struct subchannel *sch;
+	__u8 *chpid;
+
+	sch = to_subchannel(dev);
+	chpid = data;
+
+	__s390_subchannel_vary_chpid(sch, *chpid, 0);
+	return 0;
+}
+
+static int
+s390_subchannel_vary_chpid_on(struct device *dev, void *data)
+{
+	struct subchannel *sch;
+	__u8 *chpid;
+
+	sch = to_subchannel(dev);
+	chpid = data;
+
+	__s390_subchannel_vary_chpid(sch, *chpid, 1);
+	return 0;
+}
+
 /*
  * Function: s390_vary_chpid
  * Varies the specified chpid online or offline
@@ -730,66 +746,33 @@ static int
 s390_vary_chpid( __u8 chpid, int on)
 {
 	char dbf_text[15];
-	struct subchannel *sch;
-	int irq;
+	int status;
 
 	sprintf(dbf_text, on?"varyon%x":"varyoff%x", chpid);
 	CIO_TRACE_EVENT( 2, dbf_text);
 
-	chsc_get_sch_descriptions();
-	if (!cio_chsc_desc_avail) {
-		printk(KERN_ERR "Could not get chpid status, "
-		       "vary on/off not available\n");
-		return -EPERM;
-	}
-
-	if (!test_bit(chpid, chpids_known)) {
+	status = get_chp_status(chpid);
+	if (!status) {
 		printk(KERN_ERR "Can't vary unknown chpid %02X\n", chpid);
 		return -EINVAL;
 	}
 
-	if (test_bit(chpid, chpids) == on) {
+	if ((on && !(status & CHP_LOGICALLY_OFFLINE)) ||
+	    (!on && (status & CHP_LOGICALLY_OFFLINE))) {
 		printk(KERN_ERR "chpid %x is "
 		       "already %sline\n", chpid, on ? "on" : "off");
 		return -EINVAL;
 	}
 
-	if (on) {
-		set_bit(chpid, chpids_logical);
-		set_bit(chpid, chpids);
-		set_chp_status(chpid, CHP_ONLINE);
-	} else {
-		clear_bit(chpid, chpids_logical);
-		clear_bit(chpid, chpids);
-		set_chp_status(chpid, CHP_LOGICALLY_OFFLINE);
-	}
+	set_chp_logically_online(chpid, on);
 
 	/*
 	 * Redo PathVerification on the devices the chpid connects to
 	 */
 
-	for (irq = 0; irq <= highest_subchannel; irq++) {
-		int chp;
-		/*
-		 * We don't need to adjust the lpm, as this will be done in
-		 * DevicePathVerification...
-		 */
-		sch = ioinfo[irq];
-		if (sch == NULL || sch->st || !sch->ssd_info.valid)
-			continue;
-
-		for (chp = 0; chp < 8; chp++) {
-			if (sch->ssd_info.chpid[chp] == chpid) {
-				if (on)
-					sch->lpm |= (0x80 >> chp);
-				else
-					sch->lpm &= ~(0x80 >> chp);
-				dev_fsm_event(sch->dev.driver_data,
-					      DEV_EVENT_VERIFY);
-				break;
-			}
-		}
-	}
+	bus_for_each_dev(&css_bus_type, NULL, &chpid, on ?
+			 s390_subchannel_vary_chpid_on :
+			 s390_subchannel_vary_chpid_off);
 	return 0;
 }
 
@@ -800,22 +783,16 @@ static ssize_t
 chp_status_show(struct device *dev, char *buf)
 {
 	struct channel_path *chp = container_of(dev, struct channel_path, dev);
+	int state;
 
 	if (!chp)
 		return 0;
-
-	switch(chp->state) {
-	case CHP_OFFLINE:
+	state = get_chp_status(chp->id);
+	if (state & CHP_STANDBY)
 		return sprintf(buf, "n/a\n");
-	case CHP_LOGICALLY_OFFLINE:
-		return sprintf(buf, "logically offline\n");
-	case CHP_STANDBY:
-		return sprintf(buf, "n/a\n");
-	case CHP_ONLINE:
-		return sprintf(buf, "online\n");
-	default:
-		return 0;
-	}
+	return (state & CHP_LOGICALLY_OFFLINE) ?
+		sprintf(buf, "logically offline\n") :
+		sprintf(buf, "online\n");
 }
 
 static ssize_t
@@ -847,6 +824,10 @@ static DEVICE_ATTR(status, 0644, chp_status_show, chp_status_write);
 static void
 chp_release(struct device *dev)
 {
+	struct channel_path *cp;
+	
+	cp = container_of(dev, struct channel_path, dev);
+	kfree(cp);
 }
 
 /*
@@ -868,7 +849,19 @@ new_channel_path(int chpid, int status)
 
 	/* fill in status, etc. */
 	chp->id = chpid;
-	chp->state = status;
+	switch (status) {
+	case CHP_STANDBY:
+		chp->state.online = 0;
+		chp->state.logically_online = 1;
+		break;
+	case CHP_LOGICALLY_OFFLINE:
+		chp->state.logically_online = 0;
+		chp->state.online = 1;
+		break;
+	case CHP_ONLINE:
+		chp->state.online = 1;
+		chp->state.logically_online = 1;
+	}
 	chp->dev = (struct device) {
 		.parent  = &css_bus_device,
 		.release = chp_release,
@@ -889,32 +882,3 @@ new_channel_path(int chpid, int status)
 	return ret;
 }
 
-static int __init
-register_channel_paths(void)
-{
-	int i;
-	int ret;
-
-	/* walk through the chpids arrays */
-	for (i = 0; i < NR_CHPIDS; i++) {
-		/* we are only interested in known chpids */
-		if (!test_bit(i, chpids_known))
-			continue;
-		if (!test_bit(i, chpids))
-			/* standby */
-			ret = new_channel_path(i, CHP_STANDBY);
-
-		else if (test_bit(i, chpids_logical))
-			/* online */
-			ret = new_channel_path(i, CHP_ONLINE);
-		else
-			/* logically offline */
-			ret = new_channel_path(i, CHP_LOGICALLY_OFFLINE);
-
-		if (ret)
-			return ret;
-	}
-	return 0;
-}
-
-module_init(register_channel_paths);

drivers/s390/cio/chsc.h

@@ -3,10 +3,9 @@
 
 #define NR_CHPIDS 256
 
-#define CHP_OFFLINE 0
-#define CHP_LOGICALLY_OFFLINE 1
-#define CHP_STANDBY 2
-#define CHP_ONLINE 3
+#define CHP_STANDBY 1
+#define CHP_LOGICALLY_OFFLINE 2
+#define CHP_ONLINE 4
 
 #define CHSC_SEI_ACC_CHPID        1
 #define CHSC_SEI_ACC_LINKADDR     2
@@ -19,7 +18,10 @@ struct chsc_header {
 
 struct channel_path {
 	int id;
-	int state;
+	struct {
+		unsigned int online:1;
+		unsigned int logically_online:1;
+	}__attribute__((packed)) state;
 	struct device dev;
 };
 

drivers/s390/cio/cio.c

 /*
  *  drivers/s390/cio/cio.c
  *   S/390 common I/O routines -- low level i/o calls
- *   $Revision: 1.105 $
+ *   $Revision: 1.114 $
  *
  *    Copyright (C) 1999-2002 IBM Deutschland Entwicklung GmbH,
  *			      IBM Corporation
@@ -147,7 +147,7 @@ cio_tpi(void)
 	if (tsch (tpi_info->irq, irb) != 0)
 		/* Not status pending or not operational. */
 		return 1;
-	sch = ioinfo[tpi_info->irq];
+	sch = (struct subchannel *)(unsigned long)tpi_info->intparm;
 	if (!sch)
 		return 1;
 	irq_enter ();
@@ -184,22 +184,22 @@ cio_start_handle_notoper(struct subchannel *sch, __u8 lpm)
 int
 cio_start (struct subchannel *sch,	/* subchannel structure */
 	   struct ccw1 * cpa,		/* logical channel prog addr */
-	   unsigned int intparm,	/* interruption parameter */
 	   __u8 lpm)			/* logical path mask */
 {
 	char dbf_txt[15];
 	int ccode;
 
-	sprintf (dbf_txt, "stIO%x", sch->irq);
-	CIO_TRACE_EVENT (4, dbf_txt);
+	CIO_TRACE_EVENT (4, "stIO");
+	CIO_TRACE_EVENT (4, sch->dev.bus_id);
 
-	sch->orb.intparm = intparm;
+	/* sch is always under 2G. */
+	sch->orb.intparm = (__u32)(unsigned long)sch;
 	sch->orb.fmt = 1;
 
 	sch->orb.pfch = sch->options.prefetch == 0;
 	sch->orb.spnd = sch->options.suspend;
 	sch->orb.ssic = sch->options.suspend && sch->options.inter;
-	sch->orb.lpm = (lpm != 0) ? lpm : sch->lpm;
+	sch->orb.lpm = (lpm != 0) ? (lpm & sch->opm) : sch->lpm;
 #ifdef CONFIG_ARCH_S390X
 	/*
 	 * for 64 bit we always support 64 bit IDAWs with 4k page size only
@@ -240,8 +240,8 @@ cio_resume (struct subchannel *sch)
 	char dbf_txt[15];
 	int ccode;
 
-	sprintf (dbf_txt, "resIO%x", sch->irq);
-	CIO_TRACE_EVENT (4, dbf_txt);
+	CIO_TRACE_EVENT (4, "resIO");
+	CIO_TRACE_EVENT (4, sch->dev.bus_id);
 
 	ccode = rsch (sch->irq);
 
@@ -277,8 +277,8 @@ cio_halt(struct subchannel *sch)
 	if (!sch)
 		return -ENODEV;
 
-	sprintf (dbf_txt, "haltIO%x", sch->irq);
-	CIO_TRACE_EVENT (2, dbf_txt);
+	CIO_TRACE_EVENT (2, "haltIO");
+	CIO_TRACE_EVENT (2, sch->dev.bus_id);
 
 	/*
 	 * Issue "Halt subchannel" and process condition code
@@ -312,8 +312,8 @@ cio_clear(struct subchannel *sch)
 	if (!sch)
 		return -ENODEV;
 
-	sprintf (dbf_txt, "clearIO%x", sch->irq);
-	CIO_TRACE_EVENT (2, dbf_txt);
+	CIO_TRACE_EVENT (2, "clearIO");
+	CIO_TRACE_EVENT (2, sch->dev.bus_id);
 
 	/*
 	 * Issue "Clear subchannel" and process condition code
@@ -348,8 +348,8 @@ cio_cancel (struct subchannel *sch)
 	if (!sch)
 		return -ENODEV;
 
-	sprintf (dbf_txt, "cancelIO%x", sch->irq);
-	CIO_TRACE_EVENT (2, dbf_txt);
+	CIO_TRACE_EVENT (2, "cancelIO");
+	CIO_TRACE_EVENT (2, sch->dev.bus_id);
 
 	ccode = xsch (sch->irq);
 
@@ -374,7 +374,7 @@ cio_cancel (struct subchannel *sch)
  * Function: cio_modify
  * Issues a "Modify Subchannel" on the specified subchannel
  */
-static int
+int
 cio_modify (struct subchannel *sch)
 {
 	int ccode, retry, ret;
@@ -411,8 +411,8 @@ cio_enable_subchannel (struct subchannel *sch, unsigned int isc)
 	int retry;
 	int ret;
 
-	sprintf (dbf_txt, "ensch%x", sch->irq);
-	CIO_TRACE_EVENT (2, dbf_txt);
+	CIO_TRACE_EVENT (2, "ensch");
+	CIO_TRACE_EVENT (2, sch->dev.bus_id);
 
 	ccode = stsch (sch->irq, &sch->schib);
 	if (ccode)
@@ -420,6 +420,7 @@ cio_enable_subchannel (struct subchannel *sch, unsigned int isc)
 
 	sch->schib.pmcw.ena = 1;
 	sch->schib.pmcw.isc = isc;
+	sch->schib.pmcw.intparm = (__u32)(unsigned long)sch;
 	for (retry = 5, ret = 0; retry > 0; retry--) {
 		ret = cio_modify(sch);
 		if (ret == -ENODEV)
@@ -457,8 +458,8 @@ cio_disable_subchannel (struct subchannel *sch)
 	int retry;
 	int ret;
 
-	sprintf (dbf_txt, "dissch%x", sch->irq);
-	CIO_TRACE_EVENT (2, dbf_txt);
+	CIO_TRACE_EVENT (2, "dissch");
+	CIO_TRACE_EVENT (2, sch->dev.bus_id);
 
 	ccode = stsch (sch->irq, &sch->schib);
 	if (ccode == 3)		/* Not operational. */
@@ -477,6 +478,13 @@ cio_disable_subchannel (struct subchannel *sch)
 		ret = cio_modify(sch);
 		if (ret == -ENODEV)
 			break;
+		if (ret == -EBUSY)
+			/*
+			 * The subchannel is busy or status pending.
+			 * We'll disable when the next interrupt was delivered
+			 * via the state machine.
+			 */
+			break;
 		if (ret == 0) {
 			stsch (sch->irq, &sch->schib);
 			if (!sch->schib.pmcw.ena)
@@ -512,6 +520,11 @@ cio_validate_subchannel (struct subchannel *sch, unsigned int irq)
 	/* Nuke all fields. */
 	memset(sch, 0, sizeof(struct subchannel));
 
+	spin_lock_init(&sch->lock);
+
+	/* Set a name for the subchannel */
+	snprintf (sch->dev.bus_id, BUS_ID_SIZE, "0.0.%04x", irq);
+
 	/*
 	 * The first subchannel that is not-operational (ccode==3)
 	 *  indicates that there aren't any more devices available.
@@ -551,12 +564,12 @@ cio_validate_subchannel (struct subchannel *sch, unsigned int irq)
 			      "at devno %04X\n", sch->schib.pmcw.dev);
 		return -ENODEV;
 	}
-
+	sch->opm = 0xff;
+	chsc_validate_chpids(sch);
 	sch->lpm = sch->schib.pmcw.pim &
 		sch->schib.pmcw.pam &
-		sch->schib.pmcw.pom;
-
-	chsc_validate_chpids(sch);
+		sch->schib.pmcw.pom &
+		sch->opm;
 
 	CIO_DEBUG(KERN_INFO, 0,
 		  "Detected device %04X on subchannel %s"
@@ -575,7 +588,6 @@ cio_validate_subchannel (struct subchannel *sch, unsigned int irq)
 	sch->schib.pmcw.isc = 3;	/* could be smth. else */
 	sch->schib.pmcw.csense = 1;	/* concurrent sense */
 	sch->schib.pmcw.ena = 0;
-	sch->schib.pmcw.intparm = sch->schib.pmcw.dev;
 	if ((sch->lpm & (sch->lpm - 1)) != 0)
 		sch->schib.pmcw.mp = 1;	/* multipath mode */
 	return 0;
@@ -612,7 +624,7 @@ do_IRQ (struct pt_regs *regs)
 			do_adapter_IO();
 			continue;
 		}
-		sch = ioinfo[tpi_info->irq];
+		sch = (struct subchannel *)(unsigned long)tpi_info->intparm;
 		if (sch)
 			spin_lock(&sch->lock);
 		/* Store interrupt response block to lowcore. */
@@ -733,16 +745,40 @@ cio_probe_console(void)
 	 */
 	ctl_set_bit(6, 24);
 	console_subchannel.schib.pmcw.isc = 7;
-	ioinfo[irq] = &console_subchannel;
+	console_subchannel.schib.pmcw.intparm =
+		(__u32)(unsigned long)&console_subchannel;
+	ret = cio_modify(&console_subchannel);
+	if (ret) {
+		console_subchannel_in_use = 0;
+		return ERR_PTR(ret);
+	}
 	return &console_subchannel;
 }
 
 void
 cio_release_console(void)
 {
-	ioinfo[console_subchannel.irq] = NULL;
+	console_subchannel.schib.pmcw.intparm = 0;
+	cio_modify(&console_subchannel);
 	ctl_clear_bit(6, 24);
 	console_subchannel_in_use = 0;
 }
 
+/* Bah... hack to catch console special sausages. */
+int
+cio_is_console(int irq)
+{
+	if (!console_subchannel_in_use)
+		return 0;
+	return (irq == console_subchannel.irq);
+}
+
+struct subchannel *
+cio_get_console_subchannel(void)
+{
+	if (!console_subchannel_in_use)
+		return 0;
+	return &console_subchannel;
+}
+
 #endif

drivers/s390/cio/cio.h

@@ -98,6 +98,7 @@ struct subchannel {
 
 	__u8 vpm;		/* verified path mask */
 	__u8 lpm;		/* logical path mask */
+	__u8 opm;               /* operational path mask */
 	struct schib schib;	/* subchannel information block */
 	struct orb orb;		/* operation request block */
 	struct ccw1 sense_ccw;	/* static ccw for sense command */
@@ -117,14 +118,16 @@ extern int cio_cancel (struct subchannel *);
 extern int cio_clear (struct subchannel *);
 extern int cio_resume (struct subchannel *);
 extern int cio_halt (struct subchannel *);
-extern int cio_start (struct subchannel *, struct ccw1 *, unsigned int, __u8);
+extern int cio_start (struct subchannel *, struct ccw1 *, __u8);
 extern int cio_cancel (struct subchannel *);
 extern int cio_set_options (struct subchannel *, int);
 extern int cio_get_options (struct subchannel *);
-
+extern int cio_modify (struct subchannel *);
 /* Use with care. */
 extern struct subchannel *cio_probe_console(void);
 extern void cio_release_console(void);
+extern int cio_is_console(int irq);
+extern struct subchannel *cio_get_console_subchannel(void);
 
 extern int cio_show_msg;
 

drivers/s390/cio/css.c

 /*
  *  drivers/s390/cio/css.c
  *  driver for channel subsystem
- *   $Revision: 1.49 $
+ *   $Revision: 1.65 $
  *
  *    Copyright (C) 2002 IBM Deutschland Entwicklung GmbH,
  *			 IBM Corporation
@@ -14,15 +14,11 @@
 #include <linux/slab.h>
 #include <linux/errno.h>
 
-#include <asm/ccwdev.h> // FIXME: layering violation, remove this
-
 #include "css.h"
 #include "cio.h"
 #include "cio_debug.h"
-#include "device.h" // FIXME: dito
 #include "ioasm.h"
 
-struct subchannel *ioinfo[__MAX_SUBCHANNELS];
 unsigned int highest_subchannel;
 int css_init_done = 0;
 
@@ -30,23 +26,19 @@ struct device css_bus_device = {
 	.bus_id = "css0",
 };
 
-static int
+static struct subchannel *
 css_alloc_subchannel(int irq)
 {
 	struct subchannel *sch;
 	int ret;
 
-	if (ioinfo[irq])
-		/* There already is a struct subchannel for this irq. */
-		return -EBUSY;
-
 	sch = kmalloc (sizeof (*sch), GFP_KERNEL | GFP_DMA);
 	if (sch == NULL)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 	ret = cio_validate_subchannel (sch, irq);
 	if (ret < 0) {
 		kfree(sch);
-		return ret;
+		return ERR_PTR(ret);
 	}
 	if (irq > highest_subchannel)
 		highest_subchannel = irq;
@@ -54,27 +46,47 @@ css_alloc_subchannel(int irq)
 	if (sch->st != SUBCHANNEL_TYPE_IO) {
 		/* For now we ignore all non-io subchannels. */
 		kfree(sch);
-		return -EINVAL;
+		return ERR_PTR(-EINVAL);
 	}
 
-	ioinfo[irq] = sch;
-
-	return 0;
+	/* 
+	 * Set intparm to subchannel address.
+	 * This is fine even on 64bit since the subchannel is always located
+	 * under 2G.
+	 */
+	sch->schib.pmcw.intparm = (__u32)(unsigned long)sch;
+	ret = cio_modify(sch);
+	if (ret) {
+		kfree(sch);
+		return ERR_PTR(ret);
+	}
+	return sch;
 }
 
 static void
-css_free_subchannel(int irq)
+css_free_subchannel(struct subchannel *sch)
 {
-	struct subchannel *sch;
-
-	sch = ioinfo[irq];
 	if (sch) {
-		ioinfo[irq] = NULL;
+		/* Reset intparm to zeroes. */
+		sch->schib.pmcw.intparm = 0;
+		cio_modify(sch);
 		kfree(sch);
 	}
 	
 }
 
+static void
+css_subchannel_release(struct device *dev)
+{
+	struct subchannel *sch;
+
+	sch = to_subchannel(dev);
+	if (!cio_is_console(sch->irq))
+		kfree(sch);
+}
+
+extern int css_get_ssd_info(struct subchannel *sch);
+
 static int
 css_register_subchannel(struct subchannel *sch)
 {
@@ -83,15 +95,15 @@ css_register_subchannel(struct subchannel *sch)
 	/* Initialize the subchannel structure */
 	sch->dev.parent = &css_bus_device;
 	sch->dev.bus = &css_bus_type;
-
-	/* Set a name for the subchannel */
-	snprintf (sch->dev.bus_id, BUS_ID_SIZE, "0.0.%04x", sch->irq);
-
+	sch->dev.release = &css_subchannel_release;
+	
 	/* make it known to the system */
 	ret = device_register(&sch->dev);
 	if (ret)
 		printk (KERN_WARNING "%s: could not register %s\n",
 			__func__, sch->dev.bus_id);
+	else
+		css_get_ssd_info(sch);
 	return ret;
 }
 
@@ -99,28 +111,159 @@ int
 css_probe_device(int irq)
 {
 	int ret;
+	struct subchannel *sch;
 
-	ret = css_alloc_subchannel(irq);
+	sch = css_alloc_subchannel(irq);
+	if (IS_ERR(sch))
+		return PTR_ERR(sch);
+	ret = css_register_subchannel(sch);
 	if (ret)
-		return ret;
-	ret = css_register_subchannel(ioinfo[irq]);
-	if (ret)
-		css_free_subchannel(irq);
+		css_free_subchannel(sch);
+	return ret;
+}
+
+static struct subchannel *
+__get_subchannel_by_stsch(int irq)
+{
+	struct subchannel *sch;
+	int cc;
+	struct schib schib;
+
+	cc = stsch(irq, &schib);
+	if (cc)
+		return NULL;
+	sch = (struct subchannel *)(unsigned long)schib.pmcw.intparm;
+	if (!sch)
+		return NULL;
+	if (get_device(&sch->dev))
+		return sch;
+	return NULL;
+}
+
+struct subchannel *
+get_subchannel_by_schid(int irq)
+{
+	struct subchannel *sch;
+	struct list_head *entry;
+	struct device *dev;
+
+	if (!get_bus(&css_bus_type))
+		return NULL;
+
+	/* Try to get subchannel from pmcw first. */ 
+	sch = __get_subchannel_by_stsch(irq);
+	if (sch)
+		goto out;
+	if (!get_driver(&io_subchannel_driver.drv))
+		goto out;
+	down_read(&css_bus_type.subsys.rwsem);
+
+	list_for_each(entry, &io_subchannel_driver.drv.devices) {
+		dev = get_device(container_of(entry,
+					      struct device, driver_list));
+		if (!dev)
+			continue;
+		sch = to_subchannel(dev);
+		if (sch->irq == irq)
+			break;
+		put_device(dev);
+		sch = NULL;
+	}
+	up_read(&css_bus_type.subsys.rwsem);
+	put_driver(&io_subchannel_driver.drv);
+out:
+	put_bus(&css_bus_type);
+
+	return sch;
+}
+
+static inline int
+css_get_subchannel_status(struct subchannel *sch, int schid)
+{
+	struct schib schib;
+	int cc;
+
+	cc = stsch(schid, &schib);
+	if (cc)
+		return CIO_GONE;
+	if (!schib.pmcw.dnv)
+		return CIO_GONE;
+	if (sch && (schib.pmcw.dev != sch->schib.pmcw.dev))
+		return CIO_REVALIDATE;
+	return CIO_OPER;
+}
+	
+static inline int
+css_evaluate_subchannel(int irq)
+{
+	int event, ret, disc;
+	struct subchannel *sch;
+
+	sch = get_subchannel_by_schid(irq);
+	disc = sch ? device_is_disconnected(sch) : 0;
+	event = css_get_subchannel_status(sch, irq);
+	switch (event) {
+	case CIO_GONE:
+		if (!sch) {
+			/* Never used this subchannel. Ignore. */
+			ret = 0;
+			break;
+		}
+		if (sch->driver && sch->driver->notify &&
+		    sch->driver->notify(&sch->dev, CIO_GONE)) {
+			device_set_disconnected(sch);
+			ret = 0;
+			break;
+		}
+		/*
+		 * Unregister subchannel.
+		 * The device will be killed automatically.
+		 */
+		device_unregister(&sch->dev);
+		/* Reset intparm to zeroes. */
+		sch->schib.pmcw.intparm = 0;
+		cio_modify(sch);
+		put_device(&sch->dev);
+		ret = 0;
+		break;
+	case CIO_REVALIDATE:
+		/* 
+		 * Revalidation machine check. Sick.
+		 * We don't notify the driver since we have to throw the device
+		 * away in any case.
+		 */
+		device_unregister(&sch->dev);
+		/* Reset intparm to zeroes. */
+		sch->schib.pmcw.intparm = 0;
+		cio_modify(sch);
+		put_device(&sch->dev);
+		ret = css_probe_device(irq);
+		break;
+	case CIO_OPER:
+		if (disc)
+			/* Get device operational again. */
+			device_trigger_reprobe(sch);
+		ret = sch ? 0 : css_probe_device(irq);
+		break;
+	default:
+		BUG();
+		ret = 0;
+	}
 	return ret;
 }
 
 /*
  * Rescan for new devices. FIXME: This is slow.
+ * This function is called when we have lost CRWs due to overflows and we have
+ * to do subchannel housekeeping.
  */
-static void
-do_process_crw(void *ignore)
+void
+css_reiterate_subchannels(void)
 {
 	int irq, ret;
 
-	for (irq = 0; irq < __MAX_SUBCHANNELS; irq++) {
-		if (ioinfo[irq])
-			continue;
-		ret = css_probe_device(irq);
+	for (irq = 0; irq <= __MAX_SUBCHANNELS; irq++) {
+		ret = css_evaluate_subchannel(irq);
 		/* No more memory. It doesn't make sense to continue. No
 		 * panic because this can happen in midflight and just
 		 * because we can't use a new device is no reason to crash
@@ -135,39 +278,25 @@ do_process_crw(void *ignore)
 
 /*
  * Called from the machine check handler for subchannel report words.
- * Note: this is called disabled from the machine check handler itself.
  */
 void
 css_process_crw(int irq)
 {
-	static DECLARE_WORK(work, do_process_crw, 0);
-	struct subchannel *sch;
-	int ccode, devno;
-
 	CIO_CRW_EVENT(2, "source is subchannel %04X\n", irq);
 
-	sch = ioinfo[irq];
-	if (sch == NULL) {
-		queue_work(ccw_device_work, &work);
-		return;
-	}
-	if (!sch->dev.driver_data)
-		return;
-	devno = sch->schib.pmcw.dev;
-	/* FIXME: css_process_crw must not know about ccw_device */
-	dev_fsm_event(sch->dev.driver_data, DEV_EVENT_NOTOPER);
-	ccode = stsch(irq, &sch->schib);
-	if (!ccode)
-		if (devno != sch->schib.pmcw.dev)
-			queue_work(ccw_device_work, &work);
+	/* 
+	 * Since we are always presented with IPI in the CRW, we have to
+	 * use stsch() to find out if the subchannel in question has come
+	 * or gone.
+	 */
+	css_evaluate_subchannel(irq);
 }
 
 /*
  * some of the initialization has already been done from init_IRQ(),
  * here we do the rest now that the driver core is running.
- * Currently, this functions scans all the subchannel structures for
- * devices. The long term plan is to remove ioinfo[] and then the
- * struct subchannel's will be created during probing. 
+ * The struct subchannel's are created during probing (except for the
+ * static console subchannel).
  */
 static int __init
 init_channel_subsystem (void)
@@ -184,8 +313,16 @@ init_channel_subsystem (void)
 	ctl_set_bit(6, 28);
 
 	for (irq = 0; irq < __MAX_SUBCHANNELS; irq++) {
-		if (!ioinfo[irq]) {
-			ret = css_alloc_subchannel(irq);
+		struct subchannel *sch;
+
+		if (cio_is_console(irq))
+			sch = cio_get_console_subchannel();
+		else {
+			sch = css_alloc_subchannel(irq);
+			if (IS_ERR(sch))
+				ret = PTR_ERR(sch);
+			else
+				ret = 0;
 			if (ret == -ENOMEM)
 				panic("Out of memory in "
 				      "init_channel_subsystem\n");
@@ -202,7 +339,7 @@ init_channel_subsystem (void)
 		 * not working) so we do it now. This is true e.g. for the
 		 * console subchannel.
 		 */
-		css_register_subchannel(ioinfo[irq]);
+		css_register_subchannel(sch);
 	}
 	return 0;
 
@@ -236,6 +373,46 @@ struct bus_type css_bus_type = {
 
 subsys_initcall(init_channel_subsystem);
 
+/*
+ * Register root devices for some drivers. The release function must not be
+ * in the device drivers, so we do it here.
+ */
+static void
+s390_root_dev_release(struct device *dev)
+{
+	kfree(dev);
+}
+
+struct device *
+s390_root_dev_register(const char *name)
+{
+	struct device *dev;
+	int ret;
+
+	if (!strlen(name))
+		return ERR_PTR(-EINVAL);
+	dev = kmalloc(sizeof(struct device), GFP_KERNEL);
+	if (!dev)
+		return ERR_PTR(-ENOMEM);
+	memset(dev, 0, sizeof(struct device));
+	strncpy(dev->bus_id, name, min(strlen(name), (size_t)BUS_ID_SIZE));
+	dev->release = s390_root_dev_release;
+	ret = device_register(dev);
+	if (ret) {
+		kfree(dev);
+		return ERR_PTR(ret);
+	}
+	return dev;
+}
+
+void
+s390_root_dev_unregister(struct device *dev)
+{
+	if (dev)
+		device_unregister(dev);
+}
+
 MODULE_LICENSE("GPL");
 EXPORT_SYMBOL(css_bus_type);
-
+EXPORT_SYMBOL(s390_root_dev_register);
+EXPORT_SYMBOL(s390_root_dev_unregister);

drivers/s390/cio/css.h

@@ -79,6 +79,7 @@ struct ccw_device_private {
 		unsigned int esid:1;        /* Ext. SenseID supported by HW */
 		unsigned int dosense:1;	    /* delayed SENSE required */
 		unsigned int doverify:1;    /* delayed path verification */
+		unsigned int donotify:1;    /* call notify function */
 	} __attribute__((packed)) flags;
 	unsigned long intparm;	/* user interruption parameter */
 	struct qdio_irq *qdio_data;
@@ -100,22 +101,32 @@ struct css_driver {
 	unsigned int subchannel_type;
 	struct device_driver drv;
 	void (*irq)(struct device *);
+	int (*notify)(struct device *, int);
+	void (*verify)(struct device *);
+	void (*termination)(struct device *);
 };
 
 /*
  * all css_drivers have the css_bus_type
  */
 extern struct bus_type css_bus_type;
+extern struct css_driver io_subchannel_driver;
 
 int css_probe_device(int irq);
-
+extern struct subchannel * get_subchannel_by_schid(int irq);
 extern unsigned int highest_subchannel;
 extern int css_init_done;
 
 #define __MAX_SUBCHANNELS 65536
-extern struct subchannel *ioinfo[__MAX_SUBCHANNELS];
 
 extern struct bus_type css_bus_type;
 extern struct device css_bus_device;
 
+/* Some helper functions for disconnected state. */
+int device_is_disconnected(struct subchannel *);
+void device_set_disconnected(struct subchannel *);
+void device_trigger_reprobe(struct subchannel *);
+
+/* Helper function for vary on/off. */
+void device_set_waiting(struct subchannel *);
 #endif

drivers/s390/cio/device.c

 /*
  *  drivers/s390/cio/device.c
  *  bus driver for ccw devices
- *   $Revision: 1.70 $
+ *   $Revision: 1.85 $
  *
  *    Copyright (C) 2002 IBM Deutschland Entwicklung GmbH,
  *			 IBM Corporation
@@ -115,16 +115,24 @@ struct bus_type ccw_bus_type = {
 };
 
 static int io_subchannel_probe (struct device *);
+static int io_subchannel_remove (struct device *);
 void io_subchannel_irq (struct device *);
+static int io_subchannel_notify(struct device *, int);
+static void io_subchannel_verify(struct device *);
+static void io_subchannel_ioterm(struct device *);
 
-static struct css_driver io_subchannel_driver = {
+struct css_driver io_subchannel_driver = {
 	.subchannel_type = SUBCHANNEL_TYPE_IO,
 	.drv = {
 		.name = "io_subchannel",
 		.bus  = &css_bus_type,
 		.probe = &io_subchannel_probe,
+		.remove = &io_subchannel_remove,
 	},
 	.irq = io_subchannel_irq,
+	.notify = io_subchannel_notify,
+	.verify = io_subchannel_verify,
+	.termination = io_subchannel_ioterm,
 };
 
 struct workqueue_struct *ccw_device_work;
@@ -232,41 +240,58 @@ online_show (struct device *dev, char *buf)
 	return sprintf(buf, cdev->online ? "1\n" : "0\n");
 }
 
-void
+int
 ccw_device_set_offline(struct ccw_device *cdev)
 {
 	int ret;
 
 	if (!cdev)
-		return;
+		return -ENODEV;
 	if (!cdev->online || !cdev->drv)
-		return;
-
-	if (cdev->drv->set_offline)
-		if (cdev->drv->set_offline(cdev) != 0)
-			return;
+		return -EINVAL;
 
+	if (cdev->private->state == DEV_STATE_DISCONNECTED) {
+		struct subchannel *sch;
+		/* 
+		 * Forced offline in disconnected state means 
+		 * 'throw away device'.
+		 */
+		sch = to_subchannel(cdev->dev.parent);
+		device_unregister(&sch->dev);
+		/* Reset intparm to zeroes. */
+		sch->schib.pmcw.intparm = 0;
+		cio_modify(sch);
+		put_device(&sch->dev);
+		return 0;
+	}
+	if (cdev->drv->set_offline) {
+		ret = cdev->drv->set_offline(cdev);
+		if (ret != 0)
+			return ret;
+	}
 	cdev->online = 0;
 	spin_lock_irq(cdev->ccwlock);
 	ret = ccw_device_offline(cdev);
 	spin_unlock_irq(cdev->ccwlock);
 	if (ret == 0)
 		wait_event(cdev->private->wait_q, dev_fsm_final_state(cdev));
-	else
-		//FIXME: we can't fail!
+	else {
 		pr_debug("ccw_device_offline returned %d, device %s\n",
 			 ret, cdev->dev.bus_id);
+		cdev->online = 1;
+	}
+	return ret;
 }
 
-void
+int
 ccw_device_set_online(struct ccw_device *cdev)
 {
 	int ret;
 
 	if (!cdev)
-		return;
+		return -ENODEV;
 	if (cdev->online || !cdev->drv)
-		return;
+		return -EINVAL;
 
 	spin_lock_irq(cdev->ccwlock);
 	ret = ccw_device_online(cdev);
@@ -276,13 +301,13 @@ ccw_device_set_online(struct ccw_device *cdev)
 	else {
 		pr_debug("ccw_device_online returned %d, device %s\n",
 			 ret, cdev->dev.bus_id);
-		return;
+		return ret;
 	}
 	if (cdev->private->state != DEV_STATE_ONLINE)
-		return;
+		return -ENODEV;
 	if (!cdev->drv->set_online || cdev->drv->set_online(cdev) == 0) {
 		cdev->online = 1;
-		return;
+		return 0;
 	}
 	spin_lock_irq(cdev->ccwlock);
 	ret = ccw_device_offline(cdev);
@@ -292,6 +317,7 @@ ccw_device_set_online(struct ccw_device *cdev)
 	else 
 		pr_debug("ccw_device_offline returned %d, device %s\n",
 			 ret, cdev->dev.bus_id);
+	return (ret = 0) ? -ENODEV : ret;
 }
 
 static ssize_t
@@ -340,7 +366,7 @@ stlck_store(struct device *dev, const char *buf, size_t count)
 		  ccw_device_unbox_recog, (void *) cdev);
 	queue_work(ccw_device_work, &cdev->private->kick_work);
 	
-	return 0;
+	return count;
 }
 
 static DEVICE_ATTR(chpids, 0444, chpids_show, NULL);
@@ -430,7 +456,7 @@ ccw_device_add_stlck(void *data)
 
 /* this is a simple abstraction for device_register that sets the
  * correct bus type and adds the bus specific files */
-static int
+int
 ccw_device_register(struct ccw_device *cdev)
 {
 	struct device *dev = &cdev->dev;
@@ -448,16 +474,20 @@ ccw_device_register(struct ccw_device *cdev)
 }
 
 void
-ccw_device_unregister(void *data)
+ccw_device_do_unreg_rereg(void *data)
 {
 	struct device *dev;
 
 	dev = (struct device *)data;
-
 	device_remove_files(dev);
-	device_unregister(dev);
+	device_del(dev);
+	if (device_add(dev)) {
+		put_device(dev);
+		return;
+	}
+	if (device_add_files(dev))
+		device_unregister(dev);
 }
-	
 
 static void
 ccw_device_release(struct device *dev)
@@ -482,6 +512,10 @@ io_subchannel_register(void *data)
 	cdev = (struct ccw_device *) data;
 	sch = to_subchannel(cdev->dev.parent);
 
+	if (!list_empty(&sch->dev.children)) {
+		bus_rescan_devices(&ccw_bus_type);
+		goto out;
+	}
 	/* make it known to the system */
 	ret = ccw_device_register(cdev);
 	if (ret) {
@@ -495,8 +529,8 @@ io_subchannel_register(void *data)
 
 	ret = subchannel_add_files(cdev->dev.parent);
 	if (ret)
-		printk(KERN_WARNING "%s: could not add attributes to %04x\n",
-		       __func__, sch->irq);
+		printk(KERN_WARNING "%s: could not add attributes to %s\n",
+		       __func__, sch->dev.bus_id);
 	if (cdev->private->state == DEV_STATE_BOXED)
 		device_create_file(&cdev->dev, &dev_attr_steal_lock);
 out:
@@ -562,9 +596,9 @@ io_subchannel_recog(struct ccw_device *cdev, struct subchannel *sch)
 	atomic_inc(&ccw_device_init_count);
 
 	/* Start async. device sensing. */
-	spin_lock_irq(cdev->ccwlock);
+	spin_lock_irq(&sch->lock);
 	rc = ccw_device_recognition(cdev);
-	spin_unlock_irq(cdev->ccwlock);
+	spin_unlock_irq(&sch->lock);
 	if (rc) {
 		if (atomic_dec_and_test(&ccw_device_init_count))
 			wake_up(&ccw_device_init_wq);
@@ -637,6 +671,60 @@ io_subchannel_probe (struct device *pdev)
 	return 0;
 }
 
+static int
+io_subchannel_remove (struct device *dev)
+{
+	struct ccw_device *cdev;
+
+	if (!dev->driver_data)
+		return 0;
+	cdev = dev->driver_data;
+	/* Set ccw device to not operational and drop reference. */
+	dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
+	put_device(&cdev->dev);
+	dev->driver_data = NULL;
+	return 0;
+}
+
+static int
+io_subchannel_notify(struct device *dev, int event)
+{
+	struct ccw_device *cdev;
+
+	cdev = dev->driver_data;
+	if (!cdev)
+		return 0;
+	if (!cdev->drv)
+		return 0;
+	if (!cdev->online)
+		return 0;
+	return cdev->drv->notify ? cdev->drv->notify(cdev, event) : 0;
+}
+
+static void
+io_subchannel_verify(struct device *dev)
+{
+	struct ccw_device *cdev;
+
+	cdev = dev->driver_data;
+	if (cdev)
+		dev_fsm_event(cdev, DEV_EVENT_VERIFY);
+}
+
+static void
+io_subchannel_ioterm(struct device *dev)
+{
+	struct ccw_device *cdev;
+
+	cdev = dev->driver_data;
+	if (!cdev)
+		return;
+	cdev->private->state = DEV_STATE_CLEAR_VERIFY;
+	if (cdev->handler)
+		cdev->handler(cdev, cdev->private->intparm,
+			      ERR_PTR(-EIO));
+}
+
 #ifdef CONFIG_CCW_CONSOLE
 static struct ccw_device console_cdev;
 static struct ccw_device_private console_private;
@@ -652,25 +740,28 @@ ccw_device_console_enable (struct ccw_device *cdev, struct subchannel *sch)
 		.parent = &sch->dev,
 	};
 	/* Initialize the subchannel structure */
-	sch->dev = (struct device) {
-		.parent = &css_bus_device,
-		.bus	= &css_bus_type,
-	};
+	sch->dev.parent = &css_bus_device;
+	sch->dev.bus = &css_bus_type;
 
 	rc = io_subchannel_recog(cdev, sch);
 	if (rc)
 		return rc;
 
 	/* Now wait for the async. recognition to come to an end. */
+	spin_lock_irq(cdev->ccwlock);
 	while (!dev_fsm_final_state(cdev))
 		wait_cons_dev();
+	rc = -EIO;
 	if (cdev->private->state != DEV_STATE_OFFLINE)
-		return -EIO;
+		goto out_unlock;
 	ccw_device_online(cdev);
 	while (!dev_fsm_final_state(cdev))
 		wait_cons_dev();
 	if (cdev->private->state != DEV_STATE_ONLINE)
-		return -EIO;
+		goto out_unlock;
+	rc = 0;
+out_unlock:
+	spin_unlock_irq(cdev->ccwlock);
 	return 0;
 }
 
@@ -767,18 +858,28 @@ ccw_device_remove (struct device *dev)
 {
 	struct ccw_device *cdev = to_ccwdev(dev);
 	struct ccw_driver *cdrv = cdev->drv;
+	int ret;
 
-	/*
-	 * Set device offline, so device drivers don't end up with
-	 * doubled code.
-	 * This is safe because of the checks in ccw_device_set_offline.
-	 */
 	pr_debug("removing device %s, sch %d, devno %x\n",
-		 cdev->dev.name,
+		 cdev->dev.bus_id,
 		 cdev->private->irq,
 		 cdev->private->devno);
-	ccw_device_set_offline(cdev);
-	return cdrv->remove ? cdrv->remove(cdev) : 0;
+	if (cdrv->remove)
+		cdrv->remove(cdev);
+	if (cdev->online) {
+		cdev->online = 0;
+		spin_lock_irq(cdev->ccwlock);
+		ret = ccw_device_offline(cdev);
+		spin_unlock_irq(cdev->ccwlock);
+		if (ret == 0)
+			wait_event(cdev->private->wait_q,
+				   dev_fsm_final_state(cdev));
+		else
+			//FIXME: we can't fail!
+			pr_debug("ccw_device_offline returned %d, device %s\n",
+				 ret, cdev->dev.bus_id);
+	}
+	return 0;
 }
 
 int

drivers/s390/cio/device.h

@@ -17,6 +17,10 @@ enum dev_state {
 	/* states to wait for i/o completion before doing something */
 	DEV_STATE_CLEAR_VERIFY,
 	DEV_STATE_TIMEOUT_KILL,
+	DEV_STATE_WAIT4IO,
+	/* special states for devices gone not operational */
+	DEV_STATE_DISCONNECTED,
+	DEV_STATE_DISCONNECTED_SENSE_ID,
 	/* last element! */
 	NR_DEV_STATES
 };
@@ -63,7 +67,9 @@ extern struct workqueue_struct *ccw_device_work;
 
 void io_subchannel_recog_done(struct ccw_device *cdev);
 
-void ccw_device_unregister(void *);
+int ccw_device_register(struct ccw_device *);
+void ccw_device_do_unreg_rereg(void *);
+
 
 int ccw_device_recognition(struct ccw_device *);
 int ccw_device_online(struct ccw_device *);

drivers/s390/cio/device_fsm.c

@@ -19,8 +19,44 @@
 #include "cio_debug.h"
 #include "css.h"
 #include "device.h"
+#include "ioasm.h"
 #include "qdio.h"
 
+int
+device_is_disconnected(struct subchannel *sch)
+{
+	struct ccw_device *cdev;
+
+	if (!sch->dev.driver_data)
+		return 0;
+	cdev = sch->dev.driver_data;
+	return (cdev->private->state == DEV_STATE_DISCONNECTED ||
+		cdev->private->state == DEV_STATE_DISCONNECTED_SENSE_ID);
+}
+
+void
+device_set_disconnected(struct subchannel *sch)
+{
+	struct ccw_device *cdev;
+
+	if (!sch->dev.driver_data)
+		return;
+	cdev = sch->dev.driver_data;
+	cdev->private->state = DEV_STATE_DISCONNECTED;
+}
+
+void
+device_set_waiting(struct subchannel *sch)
+{
+	struct ccw_device *cdev;
+
+	if (!sch->dev.driver_data)
+		return;
+	cdev = sch->dev.driver_data;
+	ccw_device_set_timeout(cdev, 10*HZ);
+	cdev->private->state = DEV_STATE_WAIT4IO;
+}
+
 /*
  * Timeout function. It just triggers a DEV_EVENT_TIMEOUT.
  */
@@ -60,39 +96,76 @@ ccw_device_set_timeout(struct ccw_device *cdev, int expires)
  * Cancel running i/o. This is called repeatedly since halt/clear are
  * asynchronous operations. We do one try with cio_cancel, two tries
  * with cio_halt, 255 tries with cio_clear. If everythings fails panic.
+ * Returns 0 if device now idle, -ENODEV for device not operational and
+ * -EBUSY if an interrupt is expected (either from halt/clear or from a
+ * status pending).
  */
 static int
 ccw_device_cancel_halt_clear(struct ccw_device *cdev)
 {
 	struct subchannel *sch;
+	int ret;
 
 	sch = to_subchannel(cdev->dev.parent);
-	if (sch->schib.scsw.actl == 0)
+	ret = stsch(sch->irq, &sch->schib);
+	if (ret || !sch->schib.pmcw.dnv)
+		return -ENODEV; 
+	if (!sch->schib.pmcw.ena || sch->schib.scsw.actl == 0)
 		/* Not operational or no activity -> done. */
 		return 0;
 	/* Stage 1: cancel io. */
 	if (!(sch->schib.scsw.actl & SCSW_ACTL_HALT_PEND) &&
 	    !(sch->schib.scsw.actl & SCSW_ACTL_CLEAR_PEND)) {
-		if (cio_cancel (sch) == 0)
-			return 0;
+		ret = cio_cancel(sch);
+		if (ret != -EINVAL)
+			return ret;
 		/* cancel io unsuccessful. From now on it is asynchronous. */
 		cdev->private->iretry = 3;	/* 3 halt retries. */
 	}
 	if (!(sch->schib.scsw.actl & SCSW_ACTL_CLEAR_PEND)) {
 		/* Stage 2: halt io. */
-		while (cdev->private->iretry-- > 0)
-			if (cio_halt (sch) == 0)
-				return -EBUSY;
+		if (cdev->private->iretry) {
+			cdev->private->iretry--;
+			ret = cio_halt(sch);
+			return (ret == 0) ? -EBUSY : ret;
+		}
 		/* halt io unsuccessful. */
 		cdev->private->iretry = 255;	/* 255 clear retries. */
 	}
 	/* Stage 3: clear io. */
-	while (cdev->private->iretry-- > 0)
-		if (cio_clear (sch) == 0)
-			return -EBUSY;
+	if (cdev->private->iretry) {
+		cdev->private->iretry--;
+		ret = cio_clear (sch);
+		return (ret == 0) ? -EBUSY : ret;
+	}
 	panic("Can't stop i/o on subchannel.\n");
 }
 
+static void
+ccw_device_handle_oper(struct ccw_device *cdev)
+{
+	struct subchannel *sch;
+
+	sch = to_subchannel(cdev->dev.parent);
+	/*
+	 * Check if cu type and device type still match. If
+	 * not, it is certainly another device and we have to
+	 * de- and re-register.
+	 */
+	if (cdev->id.cu_type != cdev->private->senseid.cu_type ||
+	    cdev->id.cu_model != cdev->private->senseid.cu_model ||
+	    cdev->id.dev_type != cdev->private->senseid.dev_type ||
+	    cdev->id.dev_model != cdev->private->senseid.dev_model) {
+		PREPARE_WORK(&cdev->private->kick_work,
+			     ccw_device_do_unreg_rereg, (void *)&cdev->dev);
+		queue_work(ccw_device_work, &cdev->private->kick_work);
+		return;
+	}
+	cdev->private->flags.donotify = 1;
+	/* Get device online again. */
+	ccw_device_online(cdev);
+}
+
 /*
  * Stop device recognition.
  */
@@ -100,41 +173,54 @@ static void
 ccw_device_recog_done(struct ccw_device *cdev, int state)
 {
 	struct subchannel *sch;
+	int notify;
 
 	sch = to_subchannel(cdev->dev.parent);
 
 	ccw_device_set_timeout(cdev, 0);
 	cio_disable_subchannel(sch);
-	cdev->private->state = state;
-
+	if (cdev->private->state == DEV_STATE_DISCONNECTED_SENSE_ID) {
+		if (state == DEV_STATE_NOT_OPER) {
+			cdev->private->state = DEV_STATE_DISCONNECTED;
+			return;
+		}
+		/* Boxed devices don't need extra treatment. */
+	}
+	notify = 0;
 	switch (state) {
 	case DEV_STATE_NOT_OPER:
 		CIO_DEBUG(KERN_WARNING, 2,
-			  "SenseID : unknown device %04X on subchannel %04X\n",
-			  sch->schib.pmcw.dev, sch->irq);
+			  "SenseID : unknown device %s on subchannel %s\n",
+			  cdev->dev.bus_id, sch->dev.bus_id);
 		break;
 	case DEV_STATE_OFFLINE:
-		/* fill out sense information */
-		cdev->id = (struct ccw_device_id) {
-			.cu_type   = cdev->private->senseid.cu_type,
-			.cu_model  = cdev->private->senseid.cu_model,
-			.dev_type  = cdev->private->senseid.dev_type,
-			.dev_model = cdev->private->senseid.dev_model,
-		};
+		if (cdev->private->state == DEV_STATE_DISCONNECTED_SENSE_ID)
+			notify = 1;
+		else  /* fill out sense information */
+			cdev->id = (struct ccw_device_id) {
+				.cu_type   = cdev->private->senseid.cu_type,
+				.cu_model  = cdev->private->senseid.cu_model,
+				.dev_type  = cdev->private->senseid.dev_type,
+				.dev_model = cdev->private->senseid.dev_model,
+			};
 		/* Issue device info message. */
-		CIO_DEBUG(KERN_INFO, 2, "SenseID : device %04X reports: "
+		CIO_DEBUG(KERN_INFO, 2, "SenseID : device %s reports: "
 			  "CU  Type/Mod = %04X/%02X, Dev Type/Mod = "
-			  "%04X/%02X\n", sch->schib.pmcw.dev,
+			  "%04X/%02X\n", cdev->dev.bus_id,
 			  cdev->id.cu_type, cdev->id.cu_model,
 			  cdev->id.dev_type, cdev->id.dev_model);
 		break;
 	case DEV_STATE_BOXED:
 		CIO_DEBUG(KERN_WARNING, 2,
-			  "SenseID : boxed device %04X on subchannel %04X\n",
-			  sch->schib.pmcw.dev, sch->irq);
+			  "SenseID : boxed device %s on subchannel %s\n",
+			  cdev->dev.bus_id, sch->dev.bus_id);
 		break;
 	}
-	io_subchannel_recog_done(cdev);
+	cdev->private->state = state;
+	if (notify && state == DEV_STATE_OFFLINE)
+		ccw_device_handle_oper(cdev);
+	else
+		io_subchannel_recog_done(cdev);
 	if (state != DEV_STATE_NOT_OPER)
 		wake_up(&cdev->private->wait_q);
 }
@@ -158,6 +244,24 @@ ccw_device_sense_id_done(struct ccw_device *cdev, int err)
 	}
 }
 
+static void
+ccw_device_oper_notify(void *data)
+{
+	struct ccw_device *cdev;
+	struct subchannel *sch;
+	int ret;
+
+	cdev = (struct ccw_device *)data;
+	sch = to_subchannel(cdev->dev.parent);
+	ret = (sch->driver && sch->driver->notify) ?
+		sch->driver->notify(&sch->dev, CIO_OPER) : 0;
+	if (!ret)
+		/* Driver doesn't want device back. */
+		ccw_device_do_unreg_rereg((void *)&cdev->dev);
+	else
+		wake_up(&cdev->private->wait_q);
+}
+
 /*
  * Finished with online/offline processing.
  */
@@ -179,13 +283,19 @@ ccw_device_done(struct ccw_device *cdev, int state)
 
 	if (state == DEV_STATE_BOXED) {
 		CIO_DEBUG(KERN_WARNING, 2,
-			  "Boxed device %04X on subchannel %04X\n",
-			  sch->schib.pmcw.dev, sch->irq);
+			  "Boxed device %s on subchannel %s\n",
+			  cdev->dev.bus_id, sch->dev.bus_id);
 		INIT_WORK(&cdev->private->kick_work,
 			  ccw_device_add_stlck, (void *) cdev);
 		queue_work(ccw_device_work, &cdev->private->kick_work);
 	}
 
+	if (cdev->private->flags.donotify) {
+		cdev->private->flags.donotify = 0;
+		PREPARE_WORK(&cdev->private->kick_work, ccw_device_oper_notify,
+			     (void *)cdev);
+		queue_work(ccw_device_work, &cdev->private->kick_work);
+	}
 	wake_up(&cdev->private->wait_q);
 
 	if (css_init_done && state != DEV_STATE_ONLINE)
@@ -229,13 +339,15 @@ int
 ccw_device_recognition(struct ccw_device *cdev)
 {
 	struct subchannel *sch;
+	int ret;
 
 	if (cdev->private->state != DEV_STATE_NOT_OPER)
 		return -EINVAL;
 	sch = to_subchannel(cdev->dev.parent);
-	if (cio_enable_subchannel(sch, sch->schib.pmcw.isc) != 0)
+	ret = cio_enable_subchannel(sch, sch->schib.pmcw.isc);
+	if (ret != 0)
 		/* Couldn't enable the subchannel for i/o. Sick device. */
-		return -ENODEV;
+		return ret;
 
 	/* After 60s the device recognition is considered to have failed. */
 	ccw_device_set_timeout(cdev, 60*HZ);
@@ -259,10 +371,19 @@ ccw_device_recognition(struct ccw_device *cdev)
 static void
 ccw_device_recog_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 {
-	if (ccw_device_cancel_halt_clear(cdev) == 0)
+	int ret;
+
+	ret = ccw_device_cancel_halt_clear(cdev);
+	switch (ret) {
+	case 0:
 		ccw_device_recog_done(cdev, DEV_STATE_BOXED);
-	else
+		break;
+	case -ENODEV:
+		ccw_device_recog_done(cdev, DEV_STATE_NOT_OPER);
+		break;
+	default:
 		ccw_device_set_timeout(cdev, 3*HZ);
+	}
 }
 
 
@@ -290,16 +411,19 @@ int
 ccw_device_online(struct ccw_device *cdev)
 {
 	struct subchannel *sch;
+	int ret;
 
 	if (cdev->private->state != DEV_STATE_OFFLINE)
 		return -EINVAL;
 	sch = to_subchannel(cdev->dev.parent);
 	if (css_init_done && !get_device(&cdev->dev))
 		return -ENODEV;
-	if (cio_enable_subchannel(sch, sch->schib.pmcw.isc) != 0) {
+	ret = cio_enable_subchannel(sch, sch->schib.pmcw.isc);
+	if (ret != 0) {
 		/* Couldn't enable the subchannel for i/o. Sick device. */
-		dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
-		return -ENODEV;
+		if (ret == -ENODEV)
+			dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
+		return ret;
 	}
 	/* Do we want to do path grouping? */
 	if (!cdev->private->options.pgroup) {
@@ -363,10 +487,19 @@ ccw_device_offline(struct ccw_device *cdev)
 static void
 ccw_device_onoff_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 {
-	if (ccw_device_cancel_halt_clear(cdev) == 0)
+	int ret;
+
+	ret = ccw_device_cancel_halt_clear(cdev);
+	switch (ret) {
+	case 0:
 		ccw_device_done(cdev, DEV_STATE_BOXED);
-	else
+		break;
+	case -ENODEV:
+		ccw_device_done(cdev, DEV_STATE_NOT_OPER);
+		break;
+	default:
 		ccw_device_set_timeout(cdev, 3*HZ);
+	}
 }
 
 /*
@@ -375,7 +508,10 @@ ccw_device_onoff_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 static void
 ccw_device_recog_notoper(struct ccw_device *cdev, enum dev_event dev_event)
 {
-	ccw_device_recog_done(cdev, DEV_STATE_NOT_OPER);
+	if (cdev->private->state == DEV_STATE_DISCONNECTED_SENSE_ID)
+		cdev->private->state = DEV_STATE_DISCONNECTED;
+	else
+		ccw_device_recog_done(cdev, DEV_STATE_NOT_OPER);
 }
 
 /*
@@ -385,9 +521,7 @@ static void
 ccw_device_offline_notoper(struct ccw_device *cdev, enum dev_event dev_event)
 {
 	cdev->private->state = DEV_STATE_NOT_OPER;
-	INIT_WORK(&cdev->private->kick_work,
-		  ccw_device_unregister, (void *) &cdev->dev);
-	queue_work(ccw_device_work, &cdev->private->kick_work);
+	device_unregister(&cdev->dev);
 	wake_up(&cdev->private->wait_q);
 }
 
@@ -406,9 +540,20 @@ ccw_device_online_notoper(struct ccw_device *cdev, enum dev_event dev_event)
 		// FIXME: not-oper indication to device driver ?
 		ccw_device_call_handler(cdev);
 	}
-	INIT_WORK(&cdev->private->kick_work,
-		  ccw_device_unregister, (void *) &cdev->dev);
-	queue_work(ccw_device_work, &cdev->private->kick_work);
+	device_unregister(&cdev->dev);
+	wake_up(&cdev->private->wait_q);
+}
+
+static void
+ccw_device_disconnected_notoper(struct ccw_device *cdev,
+				enum dev_event dev_event)
+{
+	struct subchannel *sch;
+
+	sch = to_subchannel(cdev->dev.parent);
+	cdev->private->state = DEV_STATE_NOT_OPER;
+	cio_disable_subchannel(sch);
+	device_unregister(&cdev->dev);
 	wake_up(&cdev->private->wait_q);
 }
 
@@ -478,13 +623,18 @@ ccw_device_irq(struct ccw_device *cdev, enum dev_event dev_event)
 static void
 ccw_device_online_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 {
+	int ret;
+
 	ccw_device_set_timeout(cdev, 0);
-	if (ccw_device_cancel_halt_clear(cdev) != 0) {
+	ret = ccw_device_cancel_halt_clear(cdev);
+	if (ret == -EBUSY) {
 		ccw_device_set_timeout(cdev, 3*HZ);
 		cdev->private->state = DEV_STATE_TIMEOUT_KILL;
 		return;
 	}
-	if (cdev->handler)
+	if (ret == -ENODEV)
+		dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
+	else if (cdev->handler)
 		cdev->handler(cdev, cdev->private->intparm,
 			      ERR_PTR(-ETIMEDOUT));
 }
@@ -552,10 +702,17 @@ ccw_device_killing_irq(struct ccw_device *cdev, enum dev_event dev_event)
 static void
 ccw_device_killing_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 {
-	if (ccw_device_cancel_halt_clear(cdev) != 0) {
+	int ret;
+
+	ret = ccw_device_cancel_halt_clear(cdev);
+	if (ret == -EBUSY) {
 		ccw_device_set_timeout(cdev, 3*HZ);
 		return;
 	}
+	if (ret == -ENODEV) {
+		dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
+		return;
+	}
 	//FIXME: Can we get here?
 	cdev->private->state = DEV_STATE_ONLINE;
 	if (cdev->handler)
@@ -563,6 +720,68 @@ ccw_device_killing_timeout(struct ccw_device *cdev, enum dev_event dev_event)
 			      ERR_PTR(-ETIMEDOUT));
 }
 
+static void
+ccw_device_wait4io_irq(struct ccw_device *cdev, enum dev_event dev_event)
+{
+	struct irb *irb;
+	struct subchannel *sch;
+
+	irb = (struct irb *) __LC_IRB;
+	/* Check for unsolicited interrupt. */
+	if (irb->scsw.stctl ==
+	    		(SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS)) {
+		if (cdev->handler)
+			cdev->handler (cdev, 0, irb);
+		return;
+	}
+	/*
+	 * Accumulate status and find out if a basic sense is needed.
+	 * This is fine since we have already adapted the lpm.
+	 */
+	ccw_device_accumulate_irb(cdev, irb);
+	if (cdev->private->flags.dosense) {
+		if (ccw_device_do_sense(cdev, irb) == 0) {
+			cdev->private->state = DEV_STATE_W4SENSE;
+		}
+		return;
+	}
+	/* Iff device is idle, reset timeout. */
+	sch = to_subchannel(cdev->dev.parent);
+	if (!stsch(sch->irq, &sch->schib))
+		if (sch->schib.scsw.actl == 0)
+			ccw_device_set_timeout(cdev, 0);
+	/* Call the handler. */
+	if (ccw_device_call_handler(cdev) && cdev->private->flags.doverify)
+		ccw_device_online_verify(cdev, 0);
+}
+
+static void
+ccw_device_wait4io_timeout(struct ccw_device *cdev, enum dev_event dev_event)
+{
+	int ret;
+
+	ccw_device_set_timeout(cdev, 0);
+	ret = ccw_device_cancel_halt_clear(cdev);
+	if (ret == -EBUSY) {
+		ccw_device_set_timeout(cdev, 3*HZ);
+		cdev->private->state = DEV_STATE_TIMEOUT_KILL;
+		return;
+	}
+	if (ret == -ENODEV)
+		dev_fsm_event(cdev, DEV_EVENT_NOTOPER);
+	else if (cdev->handler)
+		cdev->handler(cdev, cdev->private->intparm,
+			      ERR_PTR(-ETIMEDOUT));
+}
+
+static void
+ccw_device_wait4io_verify(struct ccw_device *cdev, enum dev_event dev_event)
+{
+	/* When the I/O has terminated, we have to start verification. */
+	if (cdev->private->options.pgroup)
+		cdev->private->flags.doverify = 1;
+}
+
 static void
 ccw_device_stlck_done(struct ccw_device *cdev, enum dev_event dev_event)
 {
@@ -586,6 +805,59 @@ ccw_device_stlck_done(struct ccw_device *cdev, enum dev_event dev_event)
 	wake_up(&cdev->private->wait_q);
 }
 
+static void
+ccw_device_start_id(struct ccw_device *cdev, enum dev_event dev_event)
+{
+	struct subchannel *sch;
+
+	sch = to_subchannel(cdev->dev.parent);
+	if (cio_enable_subchannel(sch, sch->schib.pmcw.isc) != 0)
+		/* Couldn't enable the subchannel for i/o. Sick device. */
+		return;
+
+	/* After 60s the device recognition is considered to have failed. */
+	ccw_device_set_timeout(cdev, 60*HZ);
+
+	cdev->private->state = DEV_STATE_DISCONNECTED_SENSE_ID;
+	ccw_device_sense_id_start(cdev);
+}
+
+void
+device_trigger_reprobe(struct subchannel *sch)
+{
+	struct ccw_device *cdev;
+	unsigned long flags;
+
+	if (!sch->dev.driver_data)
+		return;
+	cdev = sch->dev.driver_data;
+	if (cdev->private->state != DEV_STATE_DISCONNECTED)
+		return;
+	spin_lock_irqsave(&sch->lock, flags);
+	/* Re-set some bits in the pmcw that were lost. */
+	sch->schib.pmcw.isc = 3;
+	sch->schib.pmcw.csense = 1;
+	sch->schib.pmcw.ena = 0;
+	if ((sch->lpm & (sch->lpm - 1)) != 0)
+		sch->schib.pmcw.mp = 1;
+	sch->schib.pmcw.intparm = (__u32)(unsigned long)sch;
+	ccw_device_start_id(cdev, 0);
+	spin_unlock_irqrestore(&sch->lock, flags);
+}
+
+static void
+ccw_device_offline_irq(struct ccw_device *cdev, enum dev_event dev_event)
+{
+	struct subchannel *sch;
+
+	sch = to_subchannel(cdev->dev.parent);
+	/*
+	 * An interrupt in state offline means a previous disable was not
+	 * successful. Try again.
+	 */
+	cio_disable_subchannel(sch);
+}
+
 /*
  * No operation action. This is used e.g. to ignore a timeout event in
  * state offline.
@@ -630,7 +902,7 @@ fsm_func_t *dev_jumptable[NR_DEV_STATES][NR_DEV_EVENTS] = {
 	},
 	[DEV_STATE_OFFLINE] {
 		[DEV_EVENT_NOTOPER]	ccw_device_offline_notoper,
-		[DEV_EVENT_INTERRUPT]	ccw_device_bug,
+		[DEV_EVENT_INTERRUPT]	ccw_device_offline_irq,
 		[DEV_EVENT_TIMEOUT]	ccw_device_nop,
 		[DEV_EVENT_VERIFY]	ccw_device_nop,
 	},
@@ -677,6 +949,25 @@ fsm_func_t *dev_jumptable[NR_DEV_STATES][NR_DEV_EVENTS] = {
 		[DEV_EVENT_TIMEOUT]	ccw_device_killing_timeout,
 		[DEV_EVENT_VERIFY]	ccw_device_nop, //FIXME
 	},
+	[DEV_STATE_WAIT4IO] {
+		[DEV_EVENT_NOTOPER]	ccw_device_online_notoper,
+		[DEV_EVENT_INTERRUPT]	ccw_device_wait4io_irq,
+		[DEV_EVENT_TIMEOUT]	ccw_device_wait4io_timeout,
+		[DEV_EVENT_VERIFY]	ccw_device_wait4io_verify,
+	},
+	/* special states for devices gone not operational */
+	[DEV_STATE_DISCONNECTED] {
+		[DEV_EVENT_NOTOPER]	ccw_device_disconnected_notoper,
+		[DEV_EVENT_INTERRUPT]	ccw_device_start_id,
+		[DEV_EVENT_TIMEOUT]	ccw_device_bug,
+		[DEV_EVENT_VERIFY]	ccw_device_nop,
+	},
+	[DEV_STATE_DISCONNECTED_SENSE_ID] {
+		[DEV_EVENT_NOTOPER]	ccw_device_recog_notoper,
+		[DEV_EVENT_INTERRUPT]	ccw_device_sense_id_irq,
+		[DEV_EVENT_TIMEOUT]	ccw_device_recog_timeout,
+		[DEV_EVENT_VERIFY]	ccw_device_nop,
+	},
 };
 
 /*
@@ -686,13 +977,12 @@ fsm_func_t *dev_jumptable[NR_DEV_STATES][NR_DEV_EVENTS] = {
 void
 io_subchannel_irq (struct device *pdev)
 {
-	char dbf_txt[15];
 	struct ccw_device *cdev;
 
 	cdev = to_subchannel(pdev)->dev.driver_data;
 
-	sprintf (dbf_txt, "IRQ%04x", cdev->private->irq);
-	CIO_TRACE_EVENT (3, dbf_txt);
+	CIO_TRACE_EVENT (3, "IRQ");
+	CIO_TRACE_EVENT (3, pdev->bus_id);
 
 	dev_fsm_event(cdev, DEV_EVENT_INTERRUPT);
 }

drivers/s390/cio/device_id.c

@@ -22,6 +22,7 @@
 #include "cio_debug.h"
 #include "css.h"
 #include "device.h"
+#include "ioasm.h"
 
 /*
  * diag210 is used under VM to get information about a virtual device
@@ -195,15 +196,11 @@ __ccw_device_sense_id_start(struct ccw_device *cdev)
 	ret = -ENODEV;
 	while (cdev->private->imask != 0) {
 		if ((sch->lpm & cdev->private->imask) != 0 &&
-		    cdev->private->iretry-- > 0) {
-			/* 0x00E2C9C4 == ebcdic "SID" */
+		    cdev->private->iretry > 0) {
+			cdev->private->iretry--;
 			ret = cio_start (sch, cdev->private->iccws,
-					 0x00E2C9C4, cdev->private->imask);
+					 cdev->private->imask);
 			/* ret is 0, -EBUSY, -EACCES or -ENODEV */
-			if (ret == -EBUSY) {
-				udelay(100);
-				continue;
-			}
 			if (ret != -EACCES)
 				return ret;
 		}
@@ -223,7 +220,7 @@ ccw_device_sense_id_start(struct ccw_device *cdev)
 	cdev->private->imask = 0x80;
 	cdev->private->iretry = 5;
 	ret = __ccw_device_sense_id_start(cdev);
-	if (ret)
+	if (ret && ret != -EBUSY)
 		ccw_device_sense_id_done(cdev, ret);
 }
 
@@ -255,16 +252,16 @@ ccw_device_check_sense_id(struct ccw_device *cdev)
 		 * if the device doesn't support the SenseID
 		 *  command further retries wouldn't help ...
 		 */
-		CIO_MSG_EVENT(2, "SenseID : device %04X on Subchannel %04X "
+		CIO_MSG_EVENT(2, "SenseID : device %s on Subchannel %s "
 			      "reports cmd reject or intervention required\n",
-			      sch->schib.pmcw.dev, sch->irq);
+			      cdev->dev.bus_id, sch->dev.bus_id);
 		return -EOPNOTSUPP;
 	}
 	if (irb->esw.esw0.erw.cons) {
-		CIO_MSG_EVENT(2, "SenseID : UC on dev %04X, "
+		CIO_MSG_EVENT(2, "SenseID : UC on dev %s, "
 			      "lpum %02X, cnt %02d, sns :"
 			      " %02X%02X%02X%02X %02X%02X%02X%02X ...\n",
-			      sch->schib.pmcw.dev,
+			      cdev->dev.bus_id,
 			      irb->esw.esw0.sublog.lpum,
 			      irb->esw.esw0.erw.scnt,
 			      irb->ecw[0], irb->ecw[1],
@@ -274,15 +271,15 @@ ccw_device_check_sense_id(struct ccw_device *cdev)
 		return -EAGAIN;
 	}
 	if (irb->scsw.cc == 3) {
-		CIO_MSG_EVENT(2, "SenseID : path %02X for device %04X on "
-			      "subchannel %04X is 'not operational'\n",
-			      sch->orb.lpm, sch->schib.pmcw.dev, sch->irq);
+		CIO_MSG_EVENT(2, "SenseID : path %02X for device %s on "
+			      "subchannel %s is 'not operational'\n",
+			      sch->orb.lpm, cdev->dev.bus_id, sch->dev.bus_id);
 		return -EACCES;
 	}
 	/* Hmm, whatever happened, try again. */
-	CIO_MSG_EVENT(2, "SenseID : start_IO() for device %04X on "
-		      "subchannel %04X returns status %02X%02X\n",
-		      sch->schib.pmcw.dev, sch->irq,
+	CIO_MSG_EVENT(2, "SenseID : start_IO() for device %s on "
+		      "subchannel %s returns status %02X%02X\n",
+		      cdev->dev.bus_id, sch->dev.bus_id,
 		      irb->scsw.dstat, irb->scsw.cstat);
 	return -EAGAIN;
 }
@@ -299,11 +296,17 @@ ccw_device_sense_id_irq(struct ccw_device *cdev, enum dev_event dev_event)
 
 	sch = to_subchannel(cdev->dev.parent);
 	irb = (struct irb *) __LC_IRB;
-	/* Ignore unsolicited interrupts. */
+	/*
+	 * Unsolicited interrupts may pertain to an earlier status pending or
+	 * busy condition on the subchannel. Retry sense id.
+	 */
 	if (irb->scsw.stctl ==
-	    		(SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS))
+	    (SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS)) {
+		ret = __ccw_device_sense_id_start(cdev);
+		if (ret && ret != -EBUSY)
+			ccw_device_sense_id_done(cdev, ret);
 		return;
-
+	}
 	if (ccw_device_accumulate_and_sense(cdev, irb) != 0)
 		return;
 	ret = ccw_device_check_sense_id(cdev);

drivers/s390/cio/device_ops.c

@@ -49,6 +49,8 @@ ccw_device_clear(struct ccw_device *cdev, unsigned long intparm)
 
 	if (!cdev)
 		return -ENODEV;
+	if (cdev->private->state == DEV_STATE_NOT_OPER)
+		return -ENODEV;
 	if (cdev->private->state != DEV_STATE_ONLINE &&
 	    cdev->private->state != DEV_STATE_W4SENSE)
 		return -EINVAL;
@@ -73,6 +75,8 @@ ccw_device_start(struct ccw_device *cdev, struct ccw1 *cpa,
 	sch = to_subchannel(cdev->dev.parent);
 	if (!sch)
 		return -ENODEV;
+	if (cdev->private->state == DEV_STATE_NOT_OPER)
+		return -ENODEV;
 	if (cdev->private->state != DEV_STATE_ONLINE ||
 	    sch->schib.scsw.actl != 0 ||
 	    cdev->private->flags.doverify)
@@ -80,8 +84,7 @@ ccw_device_start(struct ccw_device *cdev, struct ccw1 *cpa,
 	ret = cio_set_options (sch, flags);
 	if (ret)
 		return ret;
-	/* 0xe4e2c5d9 == ebcdic "USER" */
-	ret = cio_start (sch, cpa, 0xe4e2c5d9, lpm);
+	ret = cio_start (sch, cpa, lpm);
 	if (ret == 0)
 		cdev->private->intparm = intparm;
 	return ret;
@@ -111,6 +114,8 @@ ccw_device_halt(struct ccw_device *cdev, unsigned long intparm)
 
 	if (!cdev)
 		return -ENODEV;
+	if (cdev->private->state == DEV_STATE_NOT_OPER)
+		return -ENODEV;
 	if (cdev->private->state != DEV_STATE_ONLINE &&
 	    cdev->private->state != DEV_STATE_W4SENSE)
 		return -EINVAL;
@@ -133,6 +138,8 @@ ccw_device_resume(struct ccw_device *cdev)
 	sch = to_subchannel(cdev->dev.parent);
 	if (!sch)
 		return -ENODEV;
+	if (cdev->private->state == DEV_STATE_NOT_OPER)
+		return -ENODEV;
 	if (cdev->private->state != DEV_STATE_ONLINE ||
 	    !(sch->schib.scsw.actl & SCSW_ACTL_SUSPENDED))
 		return -EINVAL;
@@ -251,7 +258,7 @@ __ccw_device_retry_loop(struct ccw_device *cdev, struct ccw1 *ccw, long magic)
 
 	sch = to_subchannel(cdev->dev.parent);
 	do {
-		ret = cio_start (sch, ccw, magic, 0);
+		ret = cio_start (sch, ccw, 0);
 		if ((ret == -EBUSY) || (ret == -EACCES)) {
 			/* Try again later. */
 			spin_unlock_irq(&sch->lock);
@@ -302,7 +309,6 @@ int
 read_dev_chars (struct ccw_device *cdev, void **buffer, int length)
 {
 	void (*handler)(struct ccw_device *, unsigned long, struct irb *);
-	char dbf_txt[15];
 	struct subchannel *sch;
 	int ret;
 	struct ccw1 *rdc_ccw;
@@ -313,8 +319,8 @@ read_dev_chars (struct ccw_device *cdev, void **buffer, int length)
 		return -EINVAL;
 	sch = to_subchannel(cdev->dev.parent);
 
-	sprintf (dbf_txt, "rddevch%x", sch->irq);
-	CIO_TRACE_EVENT (4, dbf_txt);
+	CIO_TRACE_EVENT (4, "rddevch");
+	CIO_TRACE_EVENT (4, sch->dev.bus_id);
 
 	rdc_ccw = kmalloc(sizeof(struct ccw1), GFP_KERNEL | GFP_DMA);
 	if (!rdc_ccw)
@@ -359,7 +365,6 @@ int
 read_conf_data (struct ccw_device *cdev, void **buffer, int *length)
 {
 	void (*handler)(struct ccw_device *, unsigned long, struct irb *);
-	char dbf_txt[15];
 	struct subchannel *sch;
 	struct ciw *ciw;
 	char *rcd_buf;
@@ -372,8 +377,8 @@ read_conf_data (struct ccw_device *cdev, void **buffer, int *length)
 		return -EINVAL;
 	sch = to_subchannel(cdev->dev.parent);
 
-	sprintf (dbf_txt, "rdconf%x", sch->irq);
-	CIO_TRACE_EVENT (4, dbf_txt);
+	CIO_TRACE_EVENT (4, "rdconf");
+	CIO_TRACE_EVENT (4, sch->dev.bus_id);
 
 	/*
 	 * scan for RCD command in extended SenseID data
@@ -449,21 +454,25 @@ ccw_device_stlck(struct ccw_device *cdev)
 	CIO_TRACE_EVENT(2, "stl lock");
 	CIO_TRACE_EVENT(2, cdev->dev.bus_id);
 
+	spin_lock_irqsave(&sch->lock, flags);
+	ret = cio_enable_subchannel(sch, 3);
+	if (ret)
+		goto out_unlock;
 	/* Setup ccw. This cmd code seems not to be in use elsewhere. */
 	cdev->private->iccws[0].cmd_code = CCW_CMD_STLCK;
 	cdev->private->iccws[0].cda = (__u32) __pa(buf);
 	cdev->private->iccws[0].count = 32;
 	cdev->private->iccws[0].flags = CCW_FLAG_SLI;
 
-	spin_lock_irqsave(&sch->lock, flags);
-	ret = cio_start(sch, cdev->private->iccws, 0xE2D3C3D2, 0);
+	ret = cio_start(sch, cdev->private->iccws, 0);
+	if (ret) {
+		cio_disable_subchannel(sch); //FIXME: return code?
+		goto out_unlock;
+	}
 	spin_unlock_irqrestore(&sch->lock, flags);
-	if (ret)
-		return ret;
-
-	wait_event(cdev->private->wait_q, sch->schib.scsw.actl == 0);
+	wait_event(cdev->private->wait_q, cdev->private->irb.scsw.actl == 0);
 	spin_lock_irqsave(&sch->lock, flags);
-
+	cio_disable_subchannel(sch); //FIXME: return code?
 	if ((cdev->private->irb.scsw.dstat !=
 	     (DEV_STAT_CHN_END|DEV_STAT_DEV_END)) ||
 	    (cdev->private->irb.scsw.cstat != 0))
@@ -471,8 +480,8 @@ ccw_device_stlck(struct ccw_device *cdev)
 
 	/* Clear irb. */
 	memset(&cdev->private->irb, 0, sizeof(struct irb));
+out_unlock:
 	spin_unlock_irqrestore(&sch->lock, flags);
-
 	return ret;
 }
 

drivers/s390/cio/device_pgid.c

@@ -48,25 +48,17 @@ __ccw_device_sense_pgid_start(struct ccw_device *cdev)
 	ret = -ENODEV;
 	while (cdev->private->imask != 0) {
 		/* Try every path multiple times. */
-		if (cdev->private->iretry-- > 0) {
-			/* 0xe2d5c9c4 == ebcdic "SNID" */
+		if (cdev->private->iretry > 0) {
+			cdev->private->iretry--;
 			ret = cio_start (sch, cdev->private->iccws, 
-					 0xE2D5C9C4, cdev->private->imask);
+					 cdev->private->imask);
 			/* ret is 0, -EBUSY, -EACCES or -ENODEV */
-			if (ret == -EBUSY) {
-				CIO_MSG_EVENT(2, 
-					      "SNID - device %04X, start_io() "
-					      "reports rc : %d, retrying ...\n",
-					      sch->schib.pmcw.dev, ret);
-				udelay(100);
-				continue;
-			}
 			if (ret != -EACCES)
 				return ret;
-			CIO_MSG_EVENT(2, "SNID - Device %04X on Subchannel "
-				      "%04X, lpm %02X, became 'not "
+			CIO_MSG_EVENT(2, "SNID - Device %s on Subchannel "
+				      "%s, lpm %02X, became 'not "
 				      "operational'\n",
-				      sch->schib.pmcw.dev, sch->irq,
+				      cdev->dev.bus_id, sch->dev.bus_id,
 				      cdev->private->imask);
 
 		}
@@ -86,7 +78,7 @@ ccw_device_sense_pgid_start(struct ccw_device *cdev)
 	cdev->private->iretry = 5;
 	memset (&cdev->private->pgid, 0, sizeof (struct pgid));
 	ret = __ccw_device_sense_pgid_start(cdev);
-	if (ret)
+	if (ret && ret != -EBUSY)
 		ccw_device_sense_pgid_done(cdev, ret);
 }
 
@@ -113,10 +105,10 @@ __ccw_device_check_sense_pgid(struct ccw_device *cdev)
 		return -EOPNOTSUPP;
 	}
 	if (irb->esw.esw0.erw.cons) {
-		CIO_MSG_EVENT(2, "SNID - device %04X, unit check, "
+		CIO_MSG_EVENT(2, "SNID - device %s, unit check, "
 			      "lpum %02X, cnt %02d, sns : "
 			      "%02X%02X%02X%02X %02X%02X%02X%02X ...\n",
-			      sch->schib.pmcw.dev,
+			      cdev->dev.bus_id,
 			      irb->esw.esw0.sublog.lpum,
 			      irb->esw.esw0.erw.scnt,
 			      irb->ecw[0], irb->ecw[1],
@@ -126,16 +118,15 @@ __ccw_device_check_sense_pgid(struct ccw_device *cdev)
 		return -EAGAIN;
 	}
 	if (irb->scsw.cc == 3) {
-		CIO_MSG_EVENT(2, "SNID - Device %04X on Subchannel "
-			      "%04X, lpm %02X, became 'not "
-			      "operational'\n",
-			      sch->schib.pmcw.dev, sch->irq, sch->orb.lpm);
+		CIO_MSG_EVENT(2, "SNID - Device %s on Subchannel "
+			      "%s, lpm %02X, became 'not operational'\n",
+			      cdev->dev.bus_id, sch->dev.bus_id, sch->orb.lpm);
 		return -EACCES;
 	}
 	if (cdev->private->pgid.inf.ps.state2 == SNID_STATE2_RESVD_ELSE) {
-		CIO_MSG_EVENT(2, "SNID - Device %04X on Subchannel %04X "
+		CIO_MSG_EVENT(2, "SNID - Device %s on Subchannel %s "
 			      "is reserved by someone else\n",
-			      sch->schib.pmcw.dev, sch->irq);
+			      cdev->dev.bus_id, sch->dev.bus_id);
 		return -EUSERS;
 	}
 	return 0;
@@ -150,34 +141,30 @@ ccw_device_sense_pgid_irq(struct ccw_device *cdev, enum dev_event dev_event)
 	struct subchannel *sch;
 	struct irb *irb;
 	int ret;
-	int opm;
-	int i;
 
 	irb = (struct irb *) __LC_IRB;
-	/* Ignore unsolicited interrupts. */
+	/*
+	 * Unsolicited interrupts may pertain to an earlier status pending or
+	 * busy condition on the subchannel. Retry sense pgid.
+	 */
 	if (irb->scsw.stctl ==
-	    		(SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS))
+	    (SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS)) {
+		ret = __ccw_device_sense_pgid_start(cdev);
+		if (ret && ret != -EBUSY)
+			ccw_device_sense_pgid_done(cdev, ret);
 		return;
+	}
 	if (ccw_device_accumulate_and_sense(cdev, irb) != 0)
 		return;
 	sch = to_subchannel(cdev->dev.parent);
 	switch (__ccw_device_check_sense_pgid(cdev)) {
 	/* 0, -ETIME, -EOPNOTSUPP, -EAGAIN, -EACCES or -EUSERS */
 	case 0:			/* Sense Path Group ID successful. */
-		opm = sch->schib.pmcw.pim &
-			sch->schib.pmcw.pam &
-			sch->schib.pmcw.pom;
-		for (i=0;i<8;i++) {
-			if (opm == (0x80 << i)) {
-				/* Don't group single path devices. */
-				cdev->private->options.pgroup = 0;
-				break;
-			}
-		}
 		if (cdev->private->pgid.inf.ps.state1 == SNID_STATE1_RESET)
 			memcpy(&cdev->private->pgid, &global_pgid,
 			       sizeof(struct pgid));
-		/* fall through. */
+		ccw_device_sense_pgid_done(cdev, 0);
+		break;
 	case -EOPNOTSUPP:	/* Sense Path Group ID not supported */
 		ccw_device_sense_pgid_done(cdev, -EOPNOTSUPP);
 		break;
@@ -235,26 +222,20 @@ __ccw_device_do_pgid(struct ccw_device *cdev, __u8 func)
 
 	/* Try multiple times. */
 	ret = -ENODEV;
-	while (cdev->private->iretry-- > 0) {
-		/* 0xE2D7C9C4 == ebcdic "SPID" */
+	if (cdev->private->iretry > 0) {
+		cdev->private->iretry--;
 		ret = cio_start (sch, cdev->private->iccws,
-				 0xE2D7C9C4, cdev->private->imask);
+				 cdev->private->imask);
 		/* ret is 0, -EBUSY, -EACCES or -ENODEV */
-		if (ret == -EACCES)
-			break;
-		if (ret != -EBUSY)
+		if (ret != -EACCES)
 			return ret;
-		udelay(100);
-		continue;
 	}
 	/* PGID command failed on this path. Switch it off. */
 	sch->lpm &= ~cdev->private->imask;
 	sch->vpm &= ~cdev->private->imask;
-	CIO_MSG_EVENT(2, "SPID - Device %04X on Subchannel "
-		      "%04X, lpm %02X, became 'not "
-		      "operational'\n",
-		      sch->schib.pmcw.dev, sch->irq,
-		      cdev->private->imask);
+	CIO_MSG_EVENT(2, "SPID - Device %s on Subchannel "
+		      "%s, lpm %02X, became 'not operational'\n",
+		      cdev->dev.bus_id, sch->dev.bus_id, cdev->private->imask);
 	return ret;
 }
 
@@ -276,9 +257,9 @@ __ccw_device_check_pgid(struct ccw_device *cdev)
 		if (irb->ecw[0] & SNS0_CMD_REJECT)
 			return -EOPNOTSUPP;
 		/* Hmm, whatever happened, try again. */
-		CIO_MSG_EVENT(2, "SPID - device %04X, unit check, cnt %02d, "
+		CIO_MSG_EVENT(2, "SPID - device %s, unit check, cnt %02d, "
 			      "sns : %02X%02X%02X%02X %02X%02X%02X%02X ...\n",
-			      sch->schib.pmcw.dev, irb->esw.esw0.erw.scnt,
+			      cdev->dev.bus_id, irb->esw.esw0.erw.scnt,
 			      irb->ecw[0], irb->ecw[1],
 			      irb->ecw[2], irb->ecw[3],
 			      irb->ecw[4], irb->ecw[5],
@@ -286,10 +267,9 @@ __ccw_device_check_pgid(struct ccw_device *cdev)
 		return -EAGAIN;
 	}
 	if (irb->scsw.cc == 3) {
-		CIO_MSG_EVENT(2, "SPID - Device %04X on Subchannel "
-			      "%04X, lpm %02X, became 'not "
-			      "operational'\n",
-			      sch->schib.pmcw.dev, sch->irq,
+		CIO_MSG_EVENT(2, "SPID - Device %s on Subchannel "
+			      "%s, lpm %02X, became 'not operational'\n",
+			      cdev->dev.bus_id, sch->dev.bus_id,
 			      cdev->private->imask);
 		return -EACCES;
 	}
@@ -313,7 +293,7 @@ __ccw_device_verify_start(struct ccw_device *cdev)
 		func = (sch->vpm & imask) ?
 			SPID_FUNC_RESIGN : SPID_FUNC_ESTABLISH;
 		ret = __ccw_device_do_pgid(cdev, func);
-		if (ret == 0)
+		if (ret == 0 || ret == -EBUSY)
 			return;
 		cdev->private->iretry = 5;
 	}
@@ -330,10 +310,15 @@ ccw_device_verify_irq(struct ccw_device *cdev, enum dev_event dev_event)
 	struct irb *irb;
 
 	irb = (struct irb *) __LC_IRB;
-	/* Ignore unsolicited interrupts. */
+	/*
+	 * Unsolicited interrupts may pertain to an earlier status pending or
+	 * busy condition on the subchannel. Restart path verification.
+	 */
 	if (irb->scsw.stctl ==
-	    		(SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS))
+	    (SCSW_STCTL_STATUS_PEND | SCSW_STCTL_ALERT_STATUS)) {
+		__ccw_device_verify_start(cdev);
 		return;
+	}
 	if (ccw_device_accumulate_and_sense(cdev, irb) != 0)
 		return;
 	sch = to_subchannel(cdev->dev.parent);

drivers/s390/cio/device_status.c

@@ -62,8 +62,8 @@ ccw_device_path_notoper(struct ccw_device *cdev)
 	sch = to_subchannel(cdev->dev.parent);
 	stsch (sch->irq, &sch->schib);
 
-	CIO_MSG_EVENT(0, "%s(%04X) - path(s) %02x are "
-		      "not operational \n", __FUNCTION__, sch->irq,
+	CIO_MSG_EVENT(0, "%s(%s) - path(s) %02x are "
+		      "not operational \n", __FUNCTION__, sch->dev.bus_id,
 		      sch->schib.pmcw.pnom);
 
 	sch->lpm &= ~sch->schib.pmcw.pnom;
@@ -321,8 +321,7 @@ ccw_device_do_sense(struct ccw_device *cdev, struct irb *irb)
 	sch->sense_ccw.count = SENSE_MAX_COUNT;
 	sch->sense_ccw.flags = CCW_FLAG_SLI;
 
-	/* 0xe2C5D5E2 == "SENS" in ebcdic */
-	return cio_start (sch, &sch->sense_ccw, 0xE2C5D5E2, 0xff);
+	return cio_start (sch, &sch->sense_ccw, 0xff);
 }
 
 /*

drivers/s390/cio/qdio.c

@@ -56,7 +56,7 @@
 #include "ioasm.h"
 #include "chsc.h"
 
-#define VERSION_QDIO_C "$Revision: 1.62 $"
+#define VERSION_QDIO_C "$Revision: 1.67 $"
 
 /****************** MODULE PARAMETER VARIABLES ********************/
 MODULE_AUTHOR("Utz Bacher <utz.bacher@de.ibm.com>");
@@ -80,6 +80,7 @@ static int hydra_thinints;
 static int indicator_used[INDICATORS_PER_CACHELINE];
 static __u32 * volatile indicators;
 static __u32 volatile spare_indicator;
+static atomic_t spare_indicator_usecount;
 
 static debug_info_t *qdio_dbf_setup;
 static debug_info_t *qdio_dbf_sbal;
@@ -238,7 +239,7 @@ qdio_get_indicator(void)
 			indicator_used[i]=1;
 			return indicators+i;
 		}
-
+	atomic_inc(&spare_indicator_usecount);
 	return (__u32 * volatile) &spare_indicator;
 }
 
@@ -252,6 +253,8 @@ qdio_put_indicator(__u32 *addr)
 		i=addr-indicators;
 		indicator_used[i]=0;
 	}
+	if (addr == &spare_indicator)
+		atomic_dec(&spare_indicator_usecount);
 }
 
 static inline volatile void 
@@ -622,7 +625,8 @@ qdio_outbound_processing(struct qdio_q *q)
 	if (q->is_iqdio_q) {
 		/* 
 		 * for asynchronous queues, we better check, if the fill
-		 * level is too high 
+		 * level is too high. for synchronous queues, the fill
+		 * level will never be that high. 
 		 */
 		if (atomic_read(&q->number_of_buffers_used)>
 		    IQDIO_FILL_LEVEL_TO_POLL)
@@ -920,7 +924,7 @@ qdio_kick_inbound_handler(struct qdio_q *q)
 }
 
 static inline void
-tiqdio_inbound_processing(struct qdio_q *q)
+__tiqdio_inbound_processing(struct qdio_q *q, int spare_ind_was_set)
 {
 	struct qdio_irq *irq_ptr;
 	struct qdio_q *oq;
@@ -954,10 +958,21 @@ tiqdio_inbound_processing(struct qdio_q *q)
 		goto out;
 	}
 
-	if (!(*(q->dev_st_chg_ind)))
-		goto out;
+	/* 
+	 * we reset spare_ind_was_set, when the queue does not use the
+	 * spare indicator
+	 */
+	if (spare_ind_was_set)
+		spare_ind_was_set = (q->dev_st_chg_ind == &spare_indicator);
 
-	tiqdio_clear_summary_bit((__u32*)q->dev_st_chg_ind);
+	if (!(*(q->dev_st_chg_ind)) && !spare_ind_was_set)
+		goto out;
+	/*
+	 * q->dev_st_chg_ind is the indicator, be it shared or not.
+	 * only clear it, if indicator is non-shared
+	 */
+	if (!spare_ind_was_set)
+		tiqdio_clear_summary_bit((__u32*)q->dev_st_chg_ind);
 
 	if (q->hydra_gives_outbound_pcis) {
 		if (!q->siga_sync_done_on_thinints) {
@@ -1003,6 +1018,12 @@ tiqdio_inbound_processing(struct qdio_q *q)
 	qdio_release_q(q);
 }
 
+static void
+tiqdio_inbound_processing(struct qdio_q *q)
+{
+	__tiqdio_inbound_processing(q, atomic_read(&spare_indicator_usecount));
+}
+
 static void
 qdio_inbound_processing(struct qdio_q *q)
 {
@@ -1106,6 +1127,7 @@ static inline void
 tiqdio_inbound_checks(void)
 {
 	struct qdio_q *q;
+	int spare_ind_was_set=0;
 #ifdef QDIO_USE_PROCESSING_STATE
 	int q_laps=0;
 #endif /* QDIO_USE_PROCESSING_STATE */
@@ -1117,11 +1139,17 @@ tiqdio_inbound_checks(void)
 again:
 #endif /* QDIO_USE_PROCESSING_STATE */
 
+	/* when the spare indicator is used and set, save that and clear it */
+	if ((atomic_read(&spare_indicator_usecount)) && spare_indicator) {
+		spare_ind_was_set = 1;
+		tiqdio_clear_summary_bit((__u32*)&spare_indicator);
+	}
+
 	q=(struct qdio_q*)tiq_list;
 	do {
 		if (!q)
 			break;
-		tiqdio_inbound_processing(q);
+		__tiqdio_inbound_processing(q, spare_ind_was_set);
 		q=(struct qdio_q*)q->list_next;
 	} while (q!=(struct qdio_q*)tiq_list);
 
@@ -1582,6 +1610,16 @@ qdio_handle_activate_check(struct ccw_device *cdev, unsigned long intparm,
 
 }
 
+static void
+qdio_call_shutdown(void *data)
+{
+	struct ccw_device *cdev;
+
+	cdev = (struct ccw_device *)data;
+	qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
+	put_device(&cdev->dev);
+}
+
 static void
 qdio_timeout_handler(struct ccw_device *cdev)
 {
@@ -1608,6 +1646,20 @@ qdio_timeout_handler(struct ccw_device *cdev)
 				irq_ptr->irq);
 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
 		break;
+	case QDIO_IRQ_STATE_ESTABLISHED:
+	case QDIO_IRQ_STATE_ACTIVE:
+		/* I/O has been terminated by common I/O layer. */
+		QDIO_PRINT_INFO("Queues on irq %04x killed by cio.\n",
+				irq_ptr->irq);
+		QDIO_DBF_TEXT2(1, trace, "cio:term");
+		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
+		if (get_device(&cdev->dev)) {
+			/* Can't call shutdown from interrupt context. */
+			PREPARE_WORK(&cdev->private->kick_work,
+				     qdio_call_shutdown, (void *)cdev);
+			queue_work(ccw_device_work, &cdev->private->kick_work);
+		}
+		break;
 	default:
 		BUG();
 	}
@@ -2098,7 +2150,7 @@ qdio_shutdown(struct ccw_device *cdev, int how)
 						 !atomic_read(&irq_ptr->
 							      input_qs[i]->
 							      use_count),
-						 QDIO_NO_USE_COUNT_TIMEOUT*HZ);
+						 QDIO_NO_USE_COUNT_TIMEOUT);
 		if (atomic_read(&irq_ptr->input_qs[i]->use_count))
 			/*
 			 * FIXME:
@@ -2116,7 +2168,7 @@ qdio_shutdown(struct ccw_device *cdev, int how)
 						 !atomic_read(&irq_ptr->
 							      output_qs[i]->
 							      use_count),
-						 QDIO_NO_USE_COUNT_TIMEOUT*HZ);
+						 QDIO_NO_USE_COUNT_TIMEOUT);
 		if (atomic_read(&irq_ptr->output_qs[i]->use_count))
 			/*
 			 * FIXME:
@@ -2134,23 +2186,34 @@ qdio_shutdown(struct ccw_device *cdev, int how)
 	/* cleanup subchannel */
 	spin_lock_irqsave(get_ccwdev_lock(cdev),flags);
 	if (how&QDIO_FLAG_CLEANUP_USING_CLEAR) {
-		ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
+		result = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
 		timeout=QDIO_CLEANUP_CLEAR_TIMEOUT;
 	} else if (how&QDIO_FLAG_CLEANUP_USING_HALT) {
-		ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
+		result = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
 		timeout=QDIO_CLEANUP_HALT_TIMEOUT;
 	} else { /* default behaviour */
-		ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
+		result = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
 		timeout=QDIO_CLEANUP_HALT_TIMEOUT;
 	}
-	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
-	ccw_device_set_timeout(cdev, timeout);
-	spin_unlock_irqrestore(get_ccwdev_lock(cdev),flags);
-
-	wait_event(cdev->private->wait_q,
-		   irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
-		   irq_ptr->state == QDIO_IRQ_STATE_ERR);
-
+	if (result == -ENODEV) {
+		/* No need to wait for device no longer present. */
+		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
+		result = 0; /* No error. */
+		spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+	} else if (result == 0) {
+		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
+		ccw_device_set_timeout(cdev, timeout);
+		spin_unlock_irqrestore(get_ccwdev_lock(cdev),flags);
+
+		wait_event(cdev->private->wait_q,
+			   irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
+			   irq_ptr->state == QDIO_IRQ_STATE_ERR);
+	} else {
+		QDIO_PRINT_INFO("ccw_device_{halt,clear} returned %d for "
+				"device %s\n", result, cdev->dev.bus_id);
+		spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
+		goto out;
+	}
 	if (irq_ptr->is_thinint_irq) {
 		qdio_put_indicator((__u32*)irq_ptr->dev_st_chg_ind);
 		tiqdio_set_subchannel_ind(irq_ptr,1); 
@@ -2796,7 +2859,7 @@ qdio_activate(struct ccw_device *cdev, int flags)
 					  QDIO_IRQ_STATE_STOPPED) ||
 					  (irq_ptr->state ==
 					   QDIO_IRQ_STATE_ERR)),
-					 (QDIO_ACTIVATE_TIMEOUT>>10)*HZ);
+					 QDIO_ACTIVATE_TIMEOUT);
 
 	switch (irq_ptr->state) {
 	case QDIO_IRQ_STATE_STOPPED:

drivers/s390/cio/qdio.h

 #ifndef _CIO_QDIO_H
 #define _CIO_QDIO_H
 
-#define VERSION_CIO_QDIO_H "$Revision: 1.20 $"
+#define VERSION_CIO_QDIO_H "$Revision: 1.22 $"
 
 //#define QDIO_DBF_LIKE_HELL
 
@@ -56,14 +56,14 @@
 #define QDIO_STATS_CLASSES 2
 #define QDIO_STATS_COUNT_NEEDED 2*/
 
-#define QDIO_NO_USE_COUNT_TIMEOUT 1000 /* wait for 1 sec on each q before
-					  exiting without having use_count
-					  of the queue to 0 */
+#define QDIO_NO_USE_COUNT_TIMEOUT (1*HZ) /* wait for 1 sec on each q before
+					    exiting without having use_count
+					    of the queue to 0 */
 
-#define QDIO_ESTABLISH_TIMEOUT 1000
-#define QDIO_ACTIVATE_TIMEOUT 5
-#define QDIO_CLEANUP_CLEAR_TIMEOUT 20000
-#define QDIO_CLEANUP_HALT_TIMEOUT 10000
+#define QDIO_ESTABLISH_TIMEOUT (1*HZ)
+#define QDIO_ACTIVATE_TIMEOUT ((5*HZ)>>10)
+#define QDIO_CLEANUP_CLEAR_TIMEOUT (20*HZ)
+#define QDIO_CLEANUP_HALT_TIMEOUT (10*HZ)
 
 enum qdio_irq_states {
 	QDIO_IRQ_STATE_INACTIVE,

drivers/s390/s390mach.c

@@ -19,9 +19,12 @@
 #define DBG printk
 // #define DBG(args,...) do {} while (0);
 
+static struct semaphore s_sem;
+
 extern void css_process_crw(int);
 extern void chsc_process_crw(void);
 extern void chp_process_crw(int, int);
+extern void css_reiterate_subchannels(void);
 
 static void
 s390_handle_damage(char *msg)
@@ -38,12 +41,19 @@ s390_handle_damage(char *msg)
  *
  * Note : we currently process CRWs for io and chsc subchannels only
  */
-static void
-s390_collect_crw_info(void)
+static int
+s390_collect_crw_info(void *param)
 {
 	struct crw crw;
 	int ccode;
+	struct semaphore *sem;
 
+	sem = (struct semaphore *)param;
+	/* Set a nice name. */
+	daemonize("kmcheck");
+
+repeat:
+	down_interruptible(sem);
 	while (1) {
 		ccode = stcrw(&crw);
 		if (ccode != 0)
@@ -52,6 +62,12 @@ s390_collect_crw_info(void)
 		    "chn=%d, rsc=%X, anc=%d, erc=%X, rsid=%X\n",
 		    crw.slct, crw.oflw, crw.chn, crw.rsc, crw.anc,
 		    crw.erc, crw.rsid);
+		/* Check for overflows. */
+		if (crw.oflw) {
+			pr_debug("%s: crw overflow detected!\n", __FUNCTION__);
+			css_reiterate_subchannels();
+			continue;
+		}
 		switch (crw.rsc) {
 		case CRW_RSC_SCH:
 			pr_debug("source is subchannel %04X\n", crw.rsid);
@@ -86,6 +102,8 @@ s390_collect_crw_info(void)
 			break;
 		}
 	}
+	goto repeat;
+	return 0;
 }
 
 /*
@@ -122,7 +140,7 @@ s390_do_machine_check(void)
 				   "check\n");
 
 	if (mci->cp)		/* channel report word pending */
-		s390_collect_crw_info();
+		up(&s_sem);
 
 #ifdef CONFIG_MACHCHK_WARNING
 /*
@@ -154,6 +172,7 @@ s390_do_machine_check(void)
 static int
 machine_check_init(void)
 {
+	init_MUTEX_LOCKED( &s_sem );
 	ctl_clear_bit(14, 25);	/* disable damage MCH */
 	ctl_set_bit(14, 26);	/* enable degradation MCH */
 	ctl_set_bit(14, 27);	/* enable system recovery MCH */
@@ -176,6 +195,7 @@ arch_initcall(machine_check_init);
 static int __init
 machine_check_crw_init (void)
 {
+	kernel_thread(s390_collect_crw_info, &s_sem, CLONE_FS|CLONE_FILES);
 	ctl_set_bit(14, 28);	/* enable channel report MCH */
 	return 0;
 }

include/asm-s390/ccwdev.h

@@ -89,10 +89,11 @@ struct ccw_driver {
 	struct module *owner;		/* for automatic MOD_INC_USE_COUNT   */
 	struct ccw_device_id *ids;	/* probe driver with these devs      */
 	int (*probe) (struct ccw_device *); /* ask driver to probe dev 	     */
-	int (*remove) (struct ccw_device *);
+	void (*remove) (struct ccw_device *);
 					/* device is no longer available     */
 	int (*set_online) (struct ccw_device *);
 	int (*set_offline) (struct ccw_device *);
+	int (*notify) (struct ccw_device *, int);
 	struct device_driver driver;	/* higher level structure, don't init
 					   this from your driver	     */
 	char *name;
@@ -149,8 +150,8 @@ extern int ccw_device_clear(struct ccw_device *, unsigned long);
 extern int read_dev_chars(struct ccw_device *cdev, void **buffer, int length);
 extern int read_conf_data(struct ccw_device *cdev, void **buffer, int *length);
 
-extern void ccw_device_set_online(struct ccw_device *cdev);
-extern void ccw_device_set_offline(struct ccw_device *cdev);
+extern int ccw_device_set_online(struct ccw_device *cdev);
+extern int ccw_device_set_offline(struct ccw_device *cdev);
 
 
 extern struct ciw *ccw_device_get_ciw(struct ccw_device *, __u32 cmd);
@@ -167,4 +168,7 @@ extern struct ccw_device *ccw_device_probe_console(void);
 extern int _ccw_device_get_device_number(struct ccw_device *);
 extern int _ccw_device_get_subchannel_number(struct ccw_device *);
 
+extern struct device *s390_root_dev_register(const char *);
+extern void s390_root_dev_unregister(struct device *);
+
 #endif /* _S390_CCWDEV_H_ */

include/asm-s390/ccwgroup.h

@@ -21,8 +21,7 @@ struct ccwgroup_driver {
 	unsigned long driver_id;
 
 	int (*probe) (struct ccwgroup_device *);
-	int (*remove) (struct ccwgroup_device *);
-	int (*release) (struct ccwgroup_driver *);
+	void (*remove) (struct ccwgroup_device *);
 	int (*set_online) (struct ccwgroup_device *);
 	int (*set_offline) (struct ccwgroup_device *);
 
@@ -37,7 +36,7 @@ extern int ccwgroup_create (struct device *root,
 			    int argc, char *argv[]);
 
 extern int ccwgroup_probe_ccwdev(struct ccw_device *cdev);
-extern int ccwgroup_remove_ccwdev(struct ccw_device *cdev);
+extern void ccwgroup_remove_ccwdev(struct ccw_device *cdev);
 
 #define to_ccwgroupdev(x) container_of((x), struct ccwgroup_device, dev)
 #define to_ccwgroupdrv(x) container_of((x), struct ccwgroup_driver, driver)

include/asm-s390/cio.h

@@ -247,6 +247,14 @@ struct ciw {
 #define DOIO_DENY_PREFETCH	 0x0002 /* don't allow for CCW prefetch */
 #define DOIO_SUPPRESS_INTER	 0x0004 /* suppress intermediate inter. */
 					/* ... for suspended CCWs */
+/* Device or subchannel gone. */
+#define CIO_GONE       0x0001
+/* No path to device. */
+#define CIO_NO_PATH    0x0002
+/* Device has appeared. */
+#define CIO_OPER       0x0004
+/* Sick revalidation of device. */
+#define CIO_REVALIDATE 0x0008
 
 struct diag210 {
 	__u16 vrdcdvno : 16;   /* device number (input) */