6587140 page_retire()/page_trycapture should not try to retire non relocatable kernel pages

PSARC 2009/181 Page Retirement Kernel KStat Addition
6819876 pages slated for capture under x86 are being ignored

3 files changed, 43 insertions, 26 deletions

diff --git a/usr/src/uts/common/vm/page.h b/usr/src/uts/common/vm/page.h
index fe975a0900..da2879b687 100644
--- a/usr/src/uts/common/vm/page.h
+++ b/usr/src/uts/common/vm/page.h
@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -770,8 +770,9 @@ int	page_unretire_pp(page_t *, int);
 void	page_tryretire(page_t *);
 void	page_retire_mdboot();
 uint64_t	page_retire_pend_count(void);
-void	page_retire_incr_pend_count(void);
-void	page_retire_decr_pend_count(void);
+uint64_t	page_retire_pend_kas_count(void);
+void	page_retire_incr_pend_count(void *);
+void	page_retire_decr_pend_count(void *);
 void	page_clrtoxic(page_t *, uchar_t);
 void	page_settoxic(page_t *, uchar_t);
 
diff --git a/usr/src/uts/common/vm/page_retire.c b/usr/src/uts/common/vm/page_retire.c
index 0dbcd41245..fa454101f8 100644
--- a/usr/src/uts/common/vm/page_retire.c
+++ b/usr/src/uts/common/vm/page_retire.c
@@ -178,6 +178,7 @@ struct page_retire_kstat {
 	kstat_named_t	pr_enqueue_fail;
 	kstat_named_t	pr_dequeue_fail;
 	kstat_named_t	pr_pending;
+	kstat_named_t	pr_pending_kas;
 	kstat_named_t	pr_failed;
 	kstat_named_t	pr_failed_kernel;
 	kstat_named_t	pr_limit;
@@ -198,6 +199,7 @@ static struct page_retire_kstat page_retire_kstat = {
 	{ "pages_notenqueued", 		KSTAT_DATA_UINT64},
 	{ "pages_notdequeued", 		KSTAT_DATA_UINT64},
 	{ "pages_pending", 		KSTAT_DATA_UINT64},
+	{ "pages_pending_kas", 		KSTAT_DATA_UINT64},
 	{ "pages_deferred",		KSTAT_DATA_UINT64},
 	{ "pages_deferred_kernel",	KSTAT_DATA_UINT64},
 	{ "pages_limit",		KSTAT_DATA_UINT64},
@@ -222,6 +224,7 @@ static kstat_t  *page_retire_ksp = NULL;
 #define	PR_KSTAT_RETIRED_FMA	(page_retire_kstat.pr_fma.value.ui64)
 #define	PR_KSTAT_RETIRED_NOTUE	(PR_KSTAT_RETIRED_CE + PR_KSTAT_RETIRED_FMA)
 #define	PR_KSTAT_PENDING	(page_retire_kstat.pr_pending.value.ui64)
+#define	PR_KSTAT_PENDING_KAS	(page_retire_kstat.pr_pending_kas.value.ui64)
 #define	PR_KSTAT_EQFAIL		(page_retire_kstat.pr_enqueue_fail.value.ui64)
 #define	PR_KSTAT_DQFAIL		(page_retire_kstat.pr_dequeue_fail.value.ui64)
 
@@ -834,16 +837,30 @@ page_retire_pend_count(void)
 	return (PR_KSTAT_PENDING);
 }
 
+uint64_t
+page_retire_pend_kas_count(void)
+{
+	return (PR_KSTAT_PENDING_KAS);
+}
+
 void
-page_retire_incr_pend_count(void)
+page_retire_incr_pend_count(void *datap)
 {
 	PR_INCR_KSTAT(pr_pending);
+
+	if ((datap == &kvp) || (datap == &zvp)) {
+		PR_INCR_KSTAT(pr_pending_kas);
+	}
 }
 
 void
-page_retire_decr_pend_count(void)
+page_retire_decr_pend_count(void *datap)
 {
 	PR_DECR_KSTAT(pr_pending);
+
+	if ((datap == &kvp) || (datap == &zvp)) {
+		PR_DECR_KSTAT(pr_pending_kas);
+	}
 }
 
 /*
@@ -1042,7 +1059,7 @@ page_retire(uint64_t pa, uchar_t reason)
 
 	if (MTBF(pr_calls, pr_mtbf)) {
 		page_settoxic(pp, reason);
-		if (page_trycapture(pp, 0, CAPTURE_RETIRE, NULL) == 0) {
+		if (page_trycapture(pp, 0, CAPTURE_RETIRE, pp->p_vnode) == 0) {
 			PR_DEBUG(prd_prlocked);
 		} else {
 			PR_DEBUG(prd_prnotlocked);
diff --git a/usr/src/uts/common/vm/vm_page.c b/usr/src/uts/common/vm/vm_page.c
index 836881f6d2..f5a36ae2cf 100644
--- a/usr/src/uts/common/vm/vm_page.c
+++ b/usr/src/uts/common/vm/vm_page.c
@@ -6451,7 +6451,7 @@ page_capture_add_hash(page_t *pp, uint_t szc, uint_t flags, void *datap)
 		page_capture_hash[index].lists[0].next = bp1;
 		page_capture_hash[index].num_pages++;
 		if (flags & CAPTURE_RETIRE) {
-			page_retire_incr_pend_count();
+			page_retire_incr_pend_count(datap);
 		}
 		mutex_exit(&page_capture_hash[index].pchh_mutex);
 		rw_exit(&pc_cb[cb_index].cb_rwlock);
@@ -6477,7 +6477,8 @@ page_capture_add_hash(page_t *pp, uint_t szc, uint_t flags, void *datap)
 			if (bp2->pp == pp) {
 				if (flags & CAPTURE_RETIRE) {
 					if (!(bp2->flags & CAPTURE_RETIRE)) {
-						page_retire_incr_pend_count();
+						page_retire_incr_pend_count(
+						    datap);
 						bp2->flags = flags;
 						bp2->expires = bp1->expires;
 						bp2->datap = datap;
@@ -6829,7 +6830,7 @@ page_capture_take_action(page_t *pp, uint_t flags, void *datap)
 	if (ret >= 0) {
 		if (found) {
 			if (bp1->flags & CAPTURE_RETIRE) {
-				page_retire_decr_pend_count();
+				page_retire_decr_pend_count(datap);
 			}
 			kmem_free(bp1, sizeof (*bp1));
 		}
@@ -7278,28 +7279,16 @@ page_capture_handle_outstanding(void)
 {
 	int ntry;
 
-	if (!page_retire_pend_count()) {
-		/*
-		 * Do we really want to be this aggressive
-		 * for things other than page_retire?
-		 * Maybe have a counter for each callback
-		 * type to guide how aggressive we should
-		 * be here.  Thus if there's at least one
-		 * page for page_retire we go ahead and reap
-		 * like this.
-		 */
-		kmem_reap();
-		seg_preap();
-		page_capture_async();
-	} else if (hat_supported(HAT_DYNAMIC_ISM_UNMAP, (void *)0)) {
+	/* Reap pages before attempting capture pages */
+	kmem_reap();
+
+	if ((page_retire_pend_count() > page_retire_pend_kas_count()) &&
+	    hat_supported(HAT_DYNAMIC_ISM_UNMAP, (void *)0)) {
 		/*
 		 * Note: Purging only for platforms that support
 		 * ISM hat_pageunload() - mainly SPARC. On x86/x64
 		 * platforms ISM pages SE_SHARED locked until destroyed.
-		 * There are pages pending retirement, so
-		 * we reap prior to attempting to capture.
 		 */
-		kmem_reap();
 
 		/* disable and purge seg_pcache */
 		(void) seg_p_disable();
@@ -7317,7 +7306,17 @@ page_capture_handle_outstanding(void)
 		}
 		/* reenable seg_pcache */
 		seg_p_enable();
+
+		/* completed what can be done.  break out */
+		return;
 	}
+
+	/*
+	 * For kernel pages and/or unsupported HAT_DYNAMIC_ISM_UNMAP, reap
+	 * and then attempt to capture.
+	 */
+	seg_preap();
+	page_capture_async();
 }
 
 /*