Run a revision of the devstat interface:

Kernel:

Change statistics to use the *uptime() timescale (ie: relative to
boottime) rather than the UTC aligned timescale.  This makes the
device statistics code oblivious to clock steps.

Change timestamps to bintime format, they are cheaper.

Remove the "busy_count", and replace it with two counter fields:
"start_count" and "end_count", which are updated in the down and
up paths respectively.  This removes the locking constraint on
devstat.

Add a timestamp argument to devstat_start_transaction(), this will
normally be a timestamp set by the *_bio() function in bp->bio_t0.
Use this field to calculate duration of I/O operations.

Add two timestamp arguments to devstat_end_transaction(), one is
the current time, a NULL pointer means "take timestamp yourself",
the other is the timestamp of when this transaction started (see
above).

Change calculation of busy_time to operate on "the salami principle":
Only when we are idle, which we can determine by the start+end
counts being identical, do we update the "busy_from" field in the
down path.  In the up path we accumulate the timeslice in busy_time
and update busy_from.

Change the byte_* and num_* fields into two arrays: bytes[] and
operations[].

Userland:

Change the misleading "busy_time" name to be called "snap_time" and
make the time long double since that is what most users need anyway,
fill it using clock_gettime(CLOCK_MONOTONIC) to put it on the same
timescale as the kernel fields.

Change devstat_compute_etime() to operate on struct bintime.

Remove the version 2 legacy interface: the change to bintime makes
compatibility far too expensive.

Fix a bug in systat's "vm" page where boot relative busy times would
be bogus.

Bump __FreeBSD_version to 500107

Review & Collaboration by:	ken

12 files changed, 165 insertions, 265 deletions

diff --git a/lib/libdevstat/Makefile b/lib/libdevstat/Makefile
index 5d484bfaad6d..3c7454edfd8e 100644
--- a/lib/libdevstat/Makefile
+++ b/lib/libdevstat/Makefile
@@ -3,7 +3,8 @@
 MAINTAINER=ken@FreeBSD.ORG
 
 LIB=	devstat
-SHLIB_MAJOR= 3
+# Bump DEVSTAT_USER_API_VER in devstat.h every time this is incremented.
+SHLIB_MAJOR= 4
 SRCS=	devstat.c
 INCS=	devstat.h
 
diff --git a/lib/libdevstat/devstat.c b/lib/libdevstat/devstat.c
index 9c715d8afe2c..e5047be09a2d 100644
--- a/lib/libdevstat/devstat.c
+++ b/lib/libdevstat/devstat.c
@@ -315,6 +315,7 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 	int retval = 0;
 	struct devinfo *dinfo;
 	const char *func_name = "devstat_getdevs";
+	struct timespec ts;
 
 	dinfo = stats->dinfo;
 
@@ -327,8 +328,8 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 	oldnumdevs = dinfo->numdevs;
 	oldgeneration = dinfo->generation;
 
-	/* Get the current time when we get the stats */
-	gettimeofday(&stats->busy_time, NULL);
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	stats->snap_time = ts.tv_sec + ts.tv_nsec * 1e-9;
 
 	if (kd == NULL) {
 		/* If this is our first time through, mem_ptr will be null. */
@@ -338,7 +339,7 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 			 * error.  Don't bother setting the error string, since
 			 * getnumdevs() has already done that for us.
 			 */
-			if ((dinfo->numdevs = getnumdevs()) < 0)
+			if ((dinfo->numdevs = devstat_getnumdevs(NULL)) < 0)
 				return(-1);
 			
 			/*
@@ -371,9 +372,9 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 			if (errno == ENOMEM) {
 				/*
 				 * No need to set the error string here, 
-				 * getnumdevs() will do that if it fails.
+				 * devstat_getnumdevs() will do that if it fails.
 				 */
-				if ((dinfo->numdevs = getnumdevs()) < 0)
+				if ((dinfo->numdevs = devstat_getnumdevs(NULL)) < 0)
 					return(-1);
 
 				dssize = (dinfo->numdevs * 
@@ -403,7 +404,7 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 		 * This is of course non-atomic, but since we are working
 		 * on a core dump, the generation is unlikely to change
 		 */
-		if ((dinfo->numdevs = getnumdevs()) == -1)
+		if ((dinfo->numdevs = devstat_getnumdevs(NULL)) == -1)
 			return(-1);
 		if ((dinfo->mem_ptr = get_devstat_kvm(kd)) == NULL)
 			return(-1);
@@ -430,8 +431,8 @@ devstat_getdevs(kvm_t *kd, struct statinfo *stats)
 	 * necessary.
 	 */
 	if (oldgeneration != dinfo->generation) {
-		if (getnumdevs() != dinfo->numdevs) {
-			if ((dinfo->numdevs = getnumdevs()) < 0)
+		if (devstat_getnumdevs(NULL) != dinfo->numdevs) {
+			if ((dinfo->numdevs = devstat_getnumdevs(NULL)) < 0)
 				return(-1);
 			dssize = (dinfo->numdevs * sizeof(struct devstat)) +
 				sizeof(long);
@@ -1142,21 +1143,21 @@ compute_stats(struct devstat *current, struct devstat *previous,
 	       DSM_NONE));
 }
 
+
+/* This is 1/2^64 */
+#define BINTIME_SCALE 5.42101086242752217003726400434970855712890625e-20
+
 long double
-devstat_compute_etime(struct timeval cur_time, struct timeval prev_time)
+devstat_compute_etime(struct bintime *cur_time, struct bintime *prev_time)
 {
-	struct timeval busy_time;
-	u_int64_t busy_usec;
 	long double etime;
 
-	timersub(&cur_time, &prev_time, &busy_time);
-
-        busy_usec = busy_time.tv_sec;  
-        busy_usec *= 1000000;          
-        busy_usec += busy_time.tv_usec;
-        etime = busy_usec;
-        etime /= 1000000;
-
+	etime = cur_time->sec;
+	etime += cur_time->frac * BINTIME_SCALE;
+	if (prev_time != NULL) {
+		etime -= prev_time->sec;
+		etime -= prev_time->frac * BINTIME_SCALE;
+	}
 	return(etime);
 }
 
@@ -1186,21 +1187,21 @@ devstat_compute_statistics(struct devstat *current, struct devstat *previous,
 		return(-1);
 	}
 
-	totalbytesread = current->bytes_read -
-			 ((previous) ? previous->bytes_read : 0);
-	totalbyteswrite = current->bytes_written -
-			    ((previous) ? previous->bytes_written : 0);
+	totalbytesread = current->bytes[DEVSTAT_READ] -
+			 ((previous) ? previous->bytes[DEVSTAT_READ] : 0);
+	totalbyteswrite = current->bytes[DEVSTAT_WRITE] -
+			    ((previous) ? previous->bytes[DEVSTAT_WRITE] : 0);
 
 	totalbytes = totalbytesread + totalbyteswrite;
 
-	totaltransfersread = current->num_reads -
-			     ((previous) ? previous->num_reads : 0);
+	totaltransfersread = current->operations[DEVSTAT_READ] -
+			     ((previous) ? previous->operations[DEVSTAT_READ] : 0);
 
-	totaltransferswrite = current->num_writes -
-			      ((previous) ? previous->num_writes : 0);
+	totaltransferswrite = current->operations[DEVSTAT_WRITE] -
+			      ((previous) ? previous->operations[DEVSTAT_WRITE] : 0);
 
-	totaltransfersother = current->num_other -
-			      ((previous) ? previous->num_other : 0);
+	totaltransfersother = current->operations[DEVSTAT_NO_DATA] -
+			      ((previous) ? previous->operations[DEVSTAT_NO_DATA] : 0);
 
 	totaltransfers = totaltransfersread + totaltransferswrite +
 			 totaltransfersother;
@@ -1496,7 +1497,7 @@ get_devstat_kvm(kvm_t *kd)
 	char *rv = NULL;
 	const char *func_name = "get_devstat_kvm";
 
-	if ((num_devs = getnumdevs()) <= 0)
+	if ((num_devs = devstat_getnumdevs(kd)) <= 0)
 		return(NULL);
 	error = 0;
 	if (KREADNL(kd, X_DEVICE_STATQ, dhead) == -1)
@@ -1510,7 +1511,7 @@ get_devstat_kvm(kvm_t *kd)
 			 func_name);
 		return(NULL);
 	}
-	gen = getgeneration();
+	gen = devstat_getgeneration(kd);
 	memcpy(rv, &gen, sizeof(gen));
 	wp = sizeof(gen);
 	/*
@@ -1536,66 +1537,3 @@ get_devstat_kvm(kvm_t *kd)
 	}
 	return(rv);
 }
-
-/*
- * Compatability functions for libdevstat 2. These are deprecated and may
- * eventually be removed.
- */
-int
-getnumdevs(void)
-{
-	return(devstat_getnumdevs(NULL));
-}
-
-long
-getgeneration(void)
-{
-	return(devstat_getgeneration(NULL));
-}
-
-int
-getversion(void)
-{
-	return(devstat_getversion(NULL));
-}
-
-int
-checkversion(void)
-{
-	return(devstat_checkversion(NULL));
-}
-
-int
-getdevs(struct statinfo *stats)
-{
-	return(devstat_getdevs(NULL, stats));
-}
-
-int
-selectdevs(struct device_selection **dev_select, int *num_selected,
-	   int *num_selections, long *select_generation, 
-	   long current_generation, struct devstat *devices, int numdevs,
-	   struct devstat_match *matches, int num_matches,
-	   char **dev_selections, int num_dev_selections,
-	   devstat_select_mode select_mode, int maxshowdevs,
-	   int perf_select)
-{
-
-	return(devstat_selectdevs(dev_select, num_selected, num_selections,
-	       select_generation, current_generation, devices, numdevs,
-	       matches, num_matches, dev_selections, num_dev_selections,
-	       select_mode, maxshowdevs, perf_select));
-}
-
-int
-buildmatch(char *match_str, struct devstat_match **matches,
-	   int *num_matches)
-{
-	return(devstat_buildmatch(match_str, matches, num_matches));
-}
-
-long double
-compute_etime(struct timeval cur_time, struct timeval prev_time)
-{
-	return(devstat_compute_etime(cur_time, prev_time));
-}
diff --git a/lib/libdevstat/devstat.h b/lib/libdevstat/devstat.h
index c1ae355df6d8..09af2170ee2d 100644
--- a/lib/libdevstat/devstat.h
+++ b/lib/libdevstat/devstat.h
@@ -35,6 +35,16 @@
 
 #include <kvm.h>
 
+/*
+ * Bumped every time we change the userland API.  Hopefully this doesn't
+ * happen very often!  This should be bumped every time we have to
+ * increment SHLIB_MAJOR in the libdevstat Makefile (for non-backwards
+ * compatible API changes) and should also be bumped every time we make
+ * backwards-compatible API changes, so application writers have a way to 
+ * determine when a particular feature is available.
+ */
+#define	DEVSTAT_USER_API_VER	4
+
 #define DEVSTAT_ERRBUF_SIZE  2048 /* size of the devstat library error string */
 
 extern char devstat_errbuf[];
@@ -111,7 +121,7 @@ struct statinfo {
 	long		tk_nin;
 	long		tk_nout;
 	struct devinfo	*dinfo;
-	struct timeval	busy_time;
+	long double 	snap_time;
 };
 
 typedef enum {
@@ -122,31 +132,7 @@ typedef enum {
 } devstat_select_mode;
 
 __BEGIN_DECLS
-/* Legacy interfaces. */
-int getnumdevs(void);
-long getgeneration(void);
-int getversion(void);
-int checkversion(void);
-int getdevs(struct statinfo *stats);
-int selectdevs(struct device_selection **dev_select, int *num_selected,
-	       int *num_selections, long *select_generation, 
-	       long current_generation, struct devstat *devices, int numdevs,
-	       struct devstat_match *matches, int num_matches,
-	       char **dev_selections, int num_dev_selections,
-	       devstat_select_mode select_mode, int maxshowdevs,
-	       int perf_select);
-int buildmatch(char *match_str, struct devstat_match **matches,
-	       int *num_matches);
-int compute_stats(struct devstat *current, struct devstat *previous,
-		  long double etime, u_int64_t *total_bytes,
-		  u_int64_t *total_transfers, u_int64_t *total_blocks,
-		  long double *kb_per_transfer,
-		  long double *transfers_per_second, long double *mb_per_second,
-		  long double *blocks_per_second,
-		  long double *ms_per_transaction);
-long double compute_etime(struct timeval cur_time, struct timeval prev_time);
-
-/* New interfaces. */
+
 int devstat_getnumdevs(kvm_t *kd);
 long devstat_getgeneration(kvm_t *kd);
 int devstat_getversion(kvm_t *kd);
@@ -164,8 +150,8 @@ int devstat_buildmatch(char *match_str, struct devstat_match **matches,
 int devstat_compute_statistics(struct devstat *current,
 			       struct devstat *previous,
 			       long double etime, ...);
-long double devstat_compute_etime(struct timeval cur_time,
-				  struct timeval prev_time);
+long double devstat_compute_etime(struct bintime *cur_time,
+				  struct bintime *prev_time);
 __END_DECLS
 
 #endif /* _DEVSTAT_H  */
diff --git a/libexec/rpc.rstatd/rstat_proc.c b/libexec/rpc.rstatd/rstat_proc.c
index bb04c16b75a2..440bdb6f588b 100644
--- a/libexec/rpc.rstatd/rstat_proc.c
+++ b/libexec/rpc.rstatd/rstat_proc.c
@@ -370,7 +370,7 @@ haveadisk(void)
 void
 updatexfers(int numdevs, int *devs)
 {
-	register int i, j, t;
+	register int i, j, k, t;
 	struct statinfo stats;
 	int num_devices = 0;
 	u_int64_t total_transfers;
@@ -400,9 +400,10 @@ updatexfers(int numdevs, int *devs)
 		      & DEVSTAT_TYPE_MASK) == DEVSTAT_TYPE_DIRECT)
 		 && ((stats.dinfo->devices[i].device_type
 		      & DEVSTAT_TYPE_PASS) == 0)) {
-			total_transfers = stats.dinfo->devices[i].num_reads +
-					  stats.dinfo->devices[i].num_writes +
-					  stats.dinfo->devices[i].num_other;
+			total_transfers = 0;
+			for (k = 0; k < DEVSTAT_N_TRANS_FLAGS; k++)
+				total_transfers +=
+				    stats.dinfo->devices[i].operations[k];
 			/*
 			 * XXX KDM If the total transfers for this device
 			 * are greater than the amount we can fit in a
diff --git a/sys/cam/cam_periph.c b/sys/cam/cam_periph.c
index 247ebb87a24e..09126e6bed38 100644
--- a/sys/cam/cam_periph.c
+++ b/sys/cam/cam_periph.c
@@ -866,7 +866,7 @@ cam_periph_runccb(union ccb *ccb,
 	 * this particular type of ccb, record the transaction start.
 	 */
 	if ((ds != NULL) && (ccb->ccb_h.func_code == XPT_SCSI_IO))
-		devstat_start_transaction(ds);
+		devstat_start_transaction(ds, NULL);
 
 	xpt_action(ccb);
  
@@ -896,7 +896,7 @@ cam_periph_runccb(union ccb *ccb,
 					CAM_DIR_NONE) ?  DEVSTAT_NO_DATA : 
 					(ccb->ccb_h.flags & CAM_DIR_OUT) ?
 					DEVSTAT_WRITE : 
-					DEVSTAT_READ);
+					DEVSTAT_READ, NULL, NULL);
 
 	return(error);
 }
diff --git a/sys/kern/subr_devstat.c b/sys/kern/subr_devstat.c
index 01322e020c8f..26fc06200d37 100644
--- a/sys/kern/subr_devstat.c
+++ b/sys/kern/subr_devstat.c
@@ -156,7 +156,7 @@ devstat_add_entry(struct devstat *ds, const char *dev_name,
 	ds->flags = flags;
 	ds->device_type = device_type;
 	ds->priority = priority;
-	getmicrotime(&ds->dev_creation_time);
+	binuptime(&ds->creation_time);
 }
 
 /*
@@ -183,9 +183,12 @@ devstat_remove_entry(struct devstat *ds)
 
 /*
  * Record a transaction start.
+ *
+ * See comments for devstat_end_transaction().  Ordering is very important
+ * here.
  */
 void
-devstat_start_transaction(struct devstat *ds)
+devstat_start_transaction(struct devstat *ds, struct bintime *now)
 {
 	/* sanity check */
 	if (ds == NULL)
@@ -196,9 +199,21 @@ devstat_start_transaction(struct devstat *ds)
 	 * to busy.  The start time is really the start of the latest busy
 	 * period.
 	 */
-	if (ds->busy_count == 0)
-		getmicrouptime(&ds->start_time);
-	ds->busy_count++;
+	if (ds->start_count == ds->end_count) {
+		if (now != NULL)
+			ds->busy_from = *now;
+		else
+			binuptime(&ds->busy_from);
+	}
+	ds->start_count++;
+}
+
+void
+devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
+{
+
+	binuptime(&bp->bio_t0);
+	devstat_start_transaction(ds, &bp->bio_t0);
 }
 
 void
@@ -210,35 +225,35 @@ devstat_start_transaction_bio(struct devstat *ds, struct bio *bp)
 
 /*
  * Record the ending of a transaction, and incrment the various counters.
+ *
+ * Ordering in this function, and in devstat_start_transaction() is VERY
+ * important.  The idea here is to run without locks, so we are very
+ * careful to only modify some fields on the way "down" (i.e. at
+ * transaction start) and some fields on the way "up" (i.e. at transaction
+ * completion).  One exception is busy_from, which we only modify in
+ * devstat_start_transaction() when there are no outstanding transactions,
+ * and thus it can't be modified in devstat_end_transaction()
+ * simultaneously.
  */
 void
 devstat_end_transaction(struct devstat *ds, u_int32_t bytes, 
-			devstat_tag_type tag_type, devstat_trans_flags flags)
+			devstat_tag_type tag_type, devstat_trans_flags flags,
+			struct bintime *now, struct bintime *then)
 {
-	struct timeval busy_time;
+	struct bintime dt, lnow;
 
 	/* sanity check */
 	if (ds == NULL)
 		return;
 
-	getmicrouptime(&ds->last_comp_time);
-	ds->busy_count--;
+	if (now == NULL) {
+		now = &lnow;
+		binuptime(now);
+	}
 
-	/*
-	 * There might be some transactions (DEVSTAT_NO_DATA) that don't
-	 * transfer any data.
-	 */
-	if (flags == DEVSTAT_READ) {
-		ds->bytes_read += bytes;
-		ds->num_reads++;
-	} else if (flags == DEVSTAT_WRITE) {
-		ds->bytes_written += bytes;
-		ds->num_writes++;
-	} else if (flags == DEVSTAT_FREE) {
-		ds->bytes_freed += bytes;
-		ds->num_frees++;
-	} else
-		ds->num_other++;
+	/* Update byte and operations counts */
+	ds->bytes[flags] += bytes;
+	ds->operations[flags]++;
 
 	/*
 	 * Keep a count of the various tag types sent.
@@ -247,21 +262,20 @@ devstat_end_transaction(struct devstat *ds, u_int32_t bytes,
 	    tag_type != DEVSTAT_TAG_NONE)
 		ds->tag_types[tag_type]++;
 
-	/*
-	 * We only update the busy time when we go idle.  Otherwise, this
-	 * calculation would require many more clock cycles.
-	 */
-	if (ds->busy_count == 0) {
-		/* Calculate how long we were busy */
-		busy_time = ds->last_comp_time;
-		timevalsub(&busy_time, &ds->start_time);
-
-		/* Add our busy time to the total busy time. */
-		timevaladd(&ds->busy_time, &busy_time);
-	} else if (ds->busy_count < 0)
-		printf("devstat_end_transaction: HELP!! busy_count "
-		       "for %s%d is < 0 (%d)!\n", ds->device_name,
-		       ds->unit_number, ds->busy_count);
+	if (then != NULL) {
+		/* Update duration of operations */
+		dt = *now;
+		bintime_sub(&dt, then);
+		bintime_add(&ds->duration[flags], &dt);
+	}
+
+	/* Accumulate busy time */
+	dt = *now;
+	bintime_sub(&dt, &ds->busy_from);
+	bintime_add(&ds->busy_time, &dt);
+	ds->busy_from = *now;
+
+	ds->end_count++;
 }
 
 void
@@ -277,7 +291,7 @@ devstat_end_transaction_bio(struct devstat *ds, struct bio *bp)
 		flg = DEVSTAT_WRITE;
 
 	devstat_end_transaction(ds, bp->bio_bcount - bp->bio_resid,
-				DEVSTAT_TAG_SIMPLE, flg);
+				DEVSTAT_TAG_SIMPLE, flg, NULL, &bp->bio_t0);
 }
 
 /*
@@ -427,3 +441,6 @@ devstat_free(struct devstat *dsp)
 		}
 	}
 }
+
+SYSCTL_INT(_debug_sizeof, OID_AUTO, devstat, CTLFLAG_RD,
+    0, sizeof(struct devstat), "sizeof(struct devstat)");
diff --git a/sys/sys/devicestat.h b/sys/sys/devicestat.h
index 872f3f89add5..c06df378768e 100644
--- a/sys/sys/devicestat.h
+++ b/sys/sys/devicestat.h
@@ -48,7 +48,7 @@
  * userland utilities to determine whether or not they are in sync with the
  * kernel.
  */
-#define DEVSTAT_VERSION	   5
+#define DEVSTAT_VERSION	   6
 
 /*
  * These flags specify which statistics features are supported or not
@@ -68,6 +68,7 @@ typedef enum {
 	DEVSTAT_WRITE	= 0x02,
 	DEVSTAT_FREE	= 0x03
 } devstat_trans_flags;
+#define DEVSTAT_N_TRANS_FLAGS	4
 
 typedef enum {
 	DEVSTAT_TAG_SIMPLE	= 0x00,
@@ -124,8 +125,15 @@ typedef enum {
 } devstat_type_flags;
 
 struct devstat {
-	int			sequence0;	     /* Update sequence# */
+	/* Internal house-keeping fields */
+	u_int			sequence0;	     /* Update sequence# */
 	int			allocated;	     /* Allocated entry */
+	u_int			start_count;	     /* started ops */
+	u_int			end_count;	     /* completed ops */
+	struct bintime		busy_from;	     /*
+						      * busy time unaccounted
+						      * for since this time
+						      */
 	STAILQ_ENTRY(devstat) 	dev_links;
 	u_int32_t		device_number;	     /*
 						      * Devstat device
@@ -133,44 +141,13 @@ struct devstat {
 						      */
 	char			device_name[DEVSTAT_NAME_LEN];
 	int			unit_number;
-	u_int64_t		bytes_read;	     /*
-						      * Total bytes read
-						      * from a device.
-						      */
-	u_int64_t		bytes_written;	     /*
-						      * Total bytes written
-						      * to a device.
-						      */
-	u_int64_t		bytes_freed;	     /*
-						      * Total bytes freed
-						      * from a device.
-						      */
-	u_int64_t		num_reads;	     /*
-						      * Total number of
-						      * read requests to 
-						      * the device.
-						      */
-	u_int64_t		num_writes;	     /*
-						      * Total number of
-						      * write requests to 
-						      * the device.
-						      */
-	u_int64_t		num_frees;	     /*
-						      * Total number of
-						      * free requests to 
-						      * the device.
-						      */
-	u_int64_t		num_other;	     /*
-						      * Total number of
-						      * transactions that 
-						      * don't read or write 
-						      * data.
-						      */
-	int32_t			busy_count;	     /*
-						      * Total number of 
-						      * transactions
-						      * outstanding for 
-						      * the device. 
+	u_int64_t		bytes[DEVSTAT_N_TRANS_FLAGS];
+	u_int64_t		operations[DEVSTAT_N_TRANS_FLAGS];
+	struct bintime		duration[DEVSTAT_N_TRANS_FLAGS];
+	struct bintime		busy_time;
+	struct bintime          creation_time;       /* 
+						      * Time the device was
+						      * created.
 						      */
 	u_int32_t		block_size;	     /* Block size, bytes */
 	u_int64_t		tag_types[3];	     /*
@@ -179,28 +156,6 @@ struct devstat {
 						      * and head of queue 
 						      * tags sent.
 						      */
-	struct timeval		dev_creation_time;   /* 
-						      * Time the device was
-						      * created.
-						      */
-	struct timeval 		busy_time;	     /*
-						      * Total amount of
-						      * time drive has spent 
-						      * processing requests.
-						      */
-	struct timeval		start_time;	     /*
-						      * The time when
-						      * busy_count was 
-						      * last == 0.  Or, the
-						      * start of the latest 
-						      * busy period.
-						      */
-	struct timeval		last_comp_time;	     /*
-						      * Last time a
-						      * transaction was 
-						      * completed.
-						      */
-
 	devstat_support_flags	flags;		     /*
 						      * Which statistics
 						      * are supported by a 
@@ -208,8 +163,11 @@ struct devstat {
 						      */
 	devstat_type_flags	device_type;	     /* Device type */
 	devstat_priority	priority;	     /* Controls list pos. */
-
-	int			sequence1;	     /* Update sequence# */
+	void			*id;		     /*
+						      * Identification for
+						      * GEOM nodes
+						      */
+	u_int			sequence1;	     /* Update sequence# */
 };
 
 STAILQ_HEAD(devstatlist, devstat);
@@ -217,18 +175,19 @@ STAILQ_HEAD(devstatlist, devstat);
 #ifdef _KERNEL
 struct bio;
 
-struct devstat *devstat_new_entry(const char *dev_name, 
-		       int unit_number, u_int32_t block_size,
-		       devstat_support_flags flags,
-		       devstat_type_flags device_type,
-		       devstat_priority priority);
+struct devstat *devstat_new_entry(const char *dev_name, int unit_number,
+				  u_int32_t block_size,
+				  devstat_support_flags flags,
+				  devstat_type_flags device_type,
+				  devstat_priority priority);
 
 void devstat_remove_entry(struct devstat *ds);
-void devstat_start_transaction(struct devstat *ds);
+void devstat_start_transaction(struct devstat *ds, struct bintime *now);
 void devstat_start_transaction_bio(struct devstat *ds, struct bio *bp);
 void devstat_end_transaction(struct devstat *ds, u_int32_t bytes, 
 			     devstat_tag_type tag_type,
-			     devstat_trans_flags flags);
+			     devstat_trans_flags flags,
+			     struct bintime *now, struct bintime *then);
 void devstat_end_transaction_bio(struct devstat *ds, struct bio *bp);
 #endif
 
diff --git a/sys/sys/param.h b/sys/sys/param.h
index f78e76e67c01..c737e834ce2f 100644
--- a/sys/sys/param.h
+++ b/sys/sys/param.h
@@ -55,7 +55,7 @@
  *	doc/en_US.ISO8859-1/books/porters-handbook/book.sgml
  */
 #undef __FreeBSD_version
-#define __FreeBSD_version 500106	/* Master, propagated to newvers */
+#define __FreeBSD_version 500107	/* Master, propagated to newvers */
 
 #ifndef NULL
 #define	NULL	0
diff --git a/usr.bin/systat/iostat.c b/usr.bin/systat/iostat.c
index ba39f3e6ea80..a54bb6d0ec11 100644
--- a/usr.bin/systat/iostat.c
+++ b/usr.bin/systat/iostat.c
@@ -148,8 +148,6 @@ fetchiostat()
 	last.dinfo = cur.dinfo;
 	cur.dinfo = tmp_dinfo;
          
-	last.busy_time = cur.busy_time;
-
 	/*
 	 * Here what we want to do is refresh our device stats.
 	 * getdevs() returns 1 when the device list has changed.
@@ -322,7 +320,7 @@ devstats(row, _col, dn)
 	
 	di = dev_select[dn].position;
 
-	busy_seconds = devstat_compute_etime(cur.busy_time, last.busy_time);
+	busy_seconds = cur.snap_time - last.snap_time;
 
 	if (devstat_compute_statistics(&cur.dinfo->devices[di],
 	    &last.dinfo->devices[di], busy_seconds,
diff --git a/usr.bin/systat/vmstat.c b/usr.bin/systat/vmstat.c
index 5fc198533aca..ff7a37b654eb 100644
--- a/usr.bin/systat/vmstat.c
+++ b/usr.bin/systat/vmstat.c
@@ -798,7 +798,7 @@ getinfo(ls)
 	last.dinfo = cur.dinfo;
 	cur.dinfo = tmp_dinfo;
 
-	last.busy_time = cur.busy_time;
+	last.snap_time = cur.snap_time;
 	switch (devstat_getdevs(NULL, &cur)) {
 	case -1:
 		errx(1, "%s", devstat_errbuf);
@@ -852,12 +852,19 @@ dinfo(dn, lc, now, then)
 
 	di = dev_select[dn].position;
 
-	elapsed_time = devstat_compute_etime(now->busy_time,
-	    then ? then->busy_time : now->dinfo->devices[di].dev_creation_time);
-
-	device_busy =  devstat_compute_etime(now->dinfo->devices[di].busy_time,
-	    then ? then->dinfo->devices[di].busy_time :
-	    now->dinfo->devices[di].dev_creation_time);
+	if (then != NULL) {
+		/* Calculate relative to previous sample */
+		elapsed_time = now->snap_time - then->snap_time;
+		device_busy = devstat_compute_etime(
+		    &now->dinfo->devices[di].busy_time,
+		    &then->dinfo->devices[di].busy_time);
+	} else {
+		/* Calculate relative to device creation */
+	        elapsed_time = now->snap_time - devstat_compute_etime(
+		    &now->dinfo->devices[di].creation_time, NULL);
+		device_busy = devstat_compute_etime(
+		    &now->dinfo->devices[di].busy_time, NULL);
+	}
 
 	if (devstat_compute_statistics(&now->dinfo->devices[di], then ?
 	    &then->dinfo->devices[di] : NULL, elapsed_time,
@@ -866,11 +873,6 @@ dinfo(dn, lc, now, then)
 	    DSM_NONE) != 0)
 		errx(1, "%s", devstat_errbuf);
 
-	if ((device_busy == 0) && (transfers_per_second > 5))
-		/* the device has been 100% busy, fake it because
-		 * as long as the device is 100% busy the busy_time
-		 * field in the devstat struct is not updated */
-		device_busy = elapsed_time;
 	if (device_busy > elapsed_time)
 		/* this normally happens after one or more periods
 		 * where the device has been 100% busy, correct it */
diff --git a/usr.bin/vmstat/vmstat.c b/usr.bin/vmstat/vmstat.c
index fb99b2cd2a1f..88c1d217cb39 100644
--- a/usr.bin/vmstat/vmstat.c
+++ b/usr.bin/vmstat/vmstat.c
@@ -428,7 +428,7 @@ dovmstat(interval, reps)
 		tmp_dinfo = last.dinfo;
 		last.dinfo = cur.dinfo;
 		cur.dinfo = tmp_dinfo;
-		last.busy_time = cur.busy_time;
+		last.snap_time = cur.snap_time;
 
 		/*
 		 * Here what we want to do is refresh our device stats.
@@ -680,7 +680,7 @@ devstats()
 		last.cp_time[state] = tmp;
 	}
 
-	busy_seconds = devstat_compute_etime(cur.busy_time, last.busy_time);
+	busy_seconds = cur.snap_time - last.snap_time;
 
 	for (dn = 0; dn < num_devices; dn++) {
 		int di;
diff --git a/usr.sbin/iostat/iostat.c b/usr.sbin/iostat/iostat.c
index 3370df283de8..724756ebe32e 100644
--- a/usr.sbin/iostat/iostat.c
+++ b/usr.sbin/iostat/iostat.c
@@ -407,12 +407,10 @@ main(int argc, char **argv)
 	cur.tk_nin = 0;
 
 	/*
-	 * Set the busy time to the system boot time, so the stats are
-	 * calculated since system boot.
+	 * Set the snap time to the system boot time (ie: zero), so the 
+	 * stats are calculated since system boot.
 	 */
-	if (readvar(kd, "kern.boottime", X_BOOTTIME, &cur.busy_time,
-		    sizeof(cur.busy_time)) != 0)
-		exit(1);
+	cur.snap_time = 0;
 
 	/*
 	 * If the user stops the program (control-Z) and then resumes it,
@@ -452,7 +450,7 @@ main(int argc, char **argv)
 		last.dinfo = cur.dinfo;
 		cur.dinfo = tmp_dinfo;
 
-		last.busy_time = cur.busy_time;
+		last.snap_time = cur.snap_time;
 
 		/*
 		 * Here what we want to do is refresh our device stats.
@@ -538,7 +536,7 @@ main(int argc, char **argv)
 			last.tk_nout = tmp;
 		}
 
-		etime = devstat_compute_etime(cur.busy_time, last.busy_time);
+		etime = cur.snap_time - last.snap_time;
 
 		if (etime == 0.0)
 			etime = 1.0;