Add a new column to the output of vmstat -z to indicate the number

of times the system was forced to sleep when requesting a new allocation.

Expand the debugger hook, db_show_uma, to display these results as well.

This has proven to be very useful in out of memory situations when
it is not known why systems have become sluggish or fail in odd ways.

Reviewed by:	rwatson alc
Approved by:	scottl (mentor) peter
Obtained from:	Yahoo Inc.

lib/libmemstat/memstat.c

@@ -188,6 +188,7 @@ _memstat_mt_reset_stats(struct memory_type *mtp)
 	mtp->mt_count = 0;
 	mtp->mt_free = 0;
 	mtp->mt_failures = 0;
+	mtp->mt_sleeps = 0;
 
 	mtp->mt_zonefree = 0;
 	mtp->mt_kegfree = 0;
@@ -304,6 +305,13 @@ memstat_get_failures(const struct memory_type *mtp)
 	return (mtp->mt_failures);
 }
 
+uint64_t
+memstat_get_sleeps(const struct memory_type *mtp)
+{
+
+	return (mtp->mt_sleeps);
+}
+
 void *
 memstat_get_caller_pointer(const struct memory_type *mtp, int index)
 {

lib/libmemstat/memstat.h

@@ -139,6 +139,7 @@ uint64_t	 memstat_get_bytes(const struct memory_type *mtp);
 uint64_t	 memstat_get_count(const struct memory_type *mtp);
 uint64_t	 memstat_get_free(const struct memory_type *mtp);
 uint64_t	 memstat_get_failures(const struct memory_type *mtp);
+uint64_t	 memstat_get_sleeps(const struct memory_type *mtp);
 void		*memstat_get_caller_pointer(const struct memory_type *mtp,
 		    int index);
 void		 memstat_set_caller_pointer(struct memory_type *mtp,

lib/libmemstat/memstat_internal.h

@@ -65,6 +65,7 @@ struct memory_type {
 	uint64_t	 mt_count;	/* Number of current allocations. */
 	uint64_t	 mt_free;	/* Number of cached free items. */
 	uint64_t	 mt_failures;	/* Number of allocation failures. */
+	uint64_t	 mt_sleeps;	/* Number of allocation sleeps. */
 
 	/*
 	 * Caller-owned memory.

lib/libmemstat/memstat_uma.c

@@ -208,6 +208,7 @@ retry:
 		mtp->mt_numallocs = uthp->uth_allocs;
 		mtp->mt_numfrees = uthp->uth_frees;
 		mtp->mt_failures = uthp->uth_fails;
+		mtp->mt_sleeps = uthp->uth_sleeps;
 
 		for (j = 0; j < maxcpus; j++) {
 			upsp = (struct uma_percpu_stat *)p;
@@ -402,6 +403,7 @@ memstat_kvm_uma(struct memory_type_list *list, void *kvm_handle)
 			mtp->mt_numallocs = uz.uz_allocs;
 			mtp->mt_numfrees = uz.uz_frees;
 			mtp->mt_failures = uz.uz_fails;
+			mtp->mt_sleeps = uz.uz_sleeps;
 			if (kz.uk_flags & UMA_ZFLAG_INTERNAL)
 				goto skip_percpu;
 			for (i = 0; i < mp_maxid + 1; i++) {

sys/vm/uma.h

@@ -600,7 +600,8 @@ struct uma_type_header {
 	u_int64_t	uth_allocs;	/* Zone: number of allocations. */
 	u_int64_t	uth_frees;	/* Zone: number of frees. */
 	u_int64_t	uth_fails;	/* Zone: number of alloc failures. */
-	u_int64_t	_uth_reserved1[3];	/* Reserved. */
+	u_int64_t	uth_sleeps;	/* Zone: number of alloc sleeps. */
+	u_int64_t	_uth_reserved1[2];	/* Reserved. */
 };
 
 struct uma_percpu_stat {

sys/vm/uma_core.c

@@ -1396,6 +1396,7 @@ zone_ctor(void *mem, int size, void *udata, int flags)
 	zone->uz_allocs = 0;
 	zone->uz_frees = 0;
 	zone->uz_fails = 0;
+	zone->uz_sleeps = 0;
 	zone->uz_fills = zone->uz_count = 0;
 	zone->uz_flags = 0;
 	keg = arg->keg;
@@ -2283,6 +2284,7 @@ zone_fetch_slab_multi(uma_zone_t zone, uma_keg_t last, int rflags)
 		 */
 		if (full && !empty) {
 			zone->uz_flags |= UMA_ZFLAG_FULL;
+			zone->uz_sleeps++;
 			msleep(zone, zone->uz_lock, PVM, "zonelimit", hz/100);
 			zone->uz_flags &= ~UMA_ZFLAG_FULL;
 			continue;
@@ -3094,13 +3096,13 @@ uma_print_zone(uma_zone_t zone)
  */
 static void
 uma_zone_sumstat(uma_zone_t z, int *cachefreep, u_int64_t *allocsp,
-    u_int64_t *freesp)
+    u_int64_t *freesp, u_int64_t *sleepsp)
 {
 	uma_cache_t cache;
-	u_int64_t allocs, frees;
+	u_int64_t allocs, frees, sleeps;
 	int cachefree, cpu;
 
-	allocs = frees = 0;
+	allocs = frees = sleeps = 0;
 	cachefree = 0;
 	CPU_FOREACH(cpu) {
 		cache = &z->uz_cpu[cpu];
@@ -3113,12 +3115,15 @@ uma_zone_sumstat(uma_zone_t z, int *cachefreep, u_int64_t *allocsp,
 	}
 	allocs += z->uz_allocs;
 	frees += z->uz_frees;
+	sleeps += z->uz_sleeps;
 	if (cachefreep != NULL)
 		*cachefreep = cachefree;
 	if (allocsp != NULL)
 		*allocsp = allocs;
 	if (freesp != NULL)
 		*freesp = frees;
+	if (sleepsp != NULL)
+		*sleepsp = sleeps;
 }
 #endif /* DDB */
 
@@ -3226,6 +3231,7 @@ restart:
 			uth.uth_allocs = z->uz_allocs;
 			uth.uth_frees = z->uz_frees;
 			uth.uth_fails = z->uz_fails;
+			uth.uth_sleeps = z->uz_sleeps;
 			if (sbuf_bcat(&sbuf, &uth, sizeof(uth)) < 0) {
 				ZONE_UNLOCK(z);
 				mtx_unlock(&uma_mtx);
@@ -3277,32 +3283,33 @@ out:
 #ifdef DDB
 DB_SHOW_COMMAND(uma, db_show_uma)
 {
-	u_int64_t allocs, frees;
+	u_int64_t allocs, frees, sleeps;
 	uma_bucket_t bucket;
 	uma_keg_t kz;
 	uma_zone_t z;
 	int cachefree;
 
-	db_printf("%18s %8s %8s %8s %12s\n", "Zone", "Size", "Used", "Free",
-	    "Requests");
+	db_printf("%18s %8s %8s %8s %12s %8s\n", "Zone", "Size", "Used", "Free",
+	    "Requests", "Sleeps");
 	LIST_FOREACH(kz, &uma_kegs, uk_link) {
 		LIST_FOREACH(z, &kz->uk_zones, uz_link) {
 			if (kz->uk_flags & UMA_ZFLAG_INTERNAL) {
 				allocs = z->uz_allocs;
 				frees = z->uz_frees;
+				sleeps = z->uz_sleeps;
 				cachefree = 0;
 			} else
 				uma_zone_sumstat(z, &cachefree, &allocs,
-				    &frees);
+				    &frees, &sleeps);
 			if (!((z->uz_flags & UMA_ZONE_SECONDARY) &&
 			    (LIST_FIRST(&kz->uk_zones) != z)))
 				cachefree += kz->uk_free;
 			LIST_FOREACH(bucket, &z->uz_full_bucket, ub_link)
 				cachefree += bucket->ub_cnt;
-			db_printf("%18s %8ju %8jd %8d %12ju\n", z->uz_name,
+			db_printf("%18s %8ju %8jd %8d %12ju %8ju\n", z->uz_name,
 			    (uintmax_t)kz->uk_size,
 			    (intmax_t)(allocs - frees), cachefree,
-			    (uintmax_t)allocs);
+			    (uintmax_t)allocs, sleeps);
 		}
 	}
 }

sys/vm/uma_int.h

@@ -327,6 +327,7 @@ struct uma_zone {
 	u_int64_t	uz_allocs UMA_ALIGN; /* Total number of allocations */
 	u_int64_t	uz_frees;	/* Total number of frees */
 	u_int64_t	uz_fails;	/* Total number of alloc failures */
+	u_int64_t	uz_sleeps;	/* Total number of alloc sleeps */
 	uint16_t	uz_fills;	/* Outstanding bucket fills */
 	uint16_t	uz_count;	/* Highest value ub_ptr can have */
 

usr.bin/vmstat/vmstat.c

@@ -1294,16 +1294,17 @@ domemstat_zone(void)
 				    memstat_strerror(error));
 		}
 	}
-	printf("%-20s %8s  %8s  %8s  %8s  %8s  %8s\n\n", "ITEM", "SIZE",
-	    "LIMIT", "USED", "FREE", "REQUESTS", "FAILURES");
+	printf("%-20s %6s %6s %8s %8s %8s %4s %4s\n\n", "ITEM", "SIZE",
+	    "LIMIT", "USED", "FREE", "REQ", "FAIL", "SLEEP");
 	for (mtp = memstat_mtl_first(mtlp); mtp != NULL;
 	    mtp = memstat_mtl_next(mtp)) {
 		strlcpy(name, memstat_get_name(mtp), MEMTYPE_MAXNAME);
 		strcat(name, ":");
-		printf("%-20s %8llu, %8llu, %8llu, %8llu, %8llu, %8llu\n", name,
+		printf("%-20s %6llu, %6llu,%8llu,%8llu,%8llu,%4llu,%4llu\n",name,
 		    memstat_get_size(mtp), memstat_get_countlimit(mtp),
 		    memstat_get_count(mtp), memstat_get_free(mtp),
-		    memstat_get_numallocs(mtp), memstat_get_failures(mtp));
+		    memstat_get_numallocs(mtp), memstat_get_failures(mtp),
+		    memstat_get_sleeps(mtp));
 	}
 	memstat_mtl_free(mtlp);
 	printf("\n");