Skip site navigation (1)Skip section navigation (2) 
Date:      Sat, 8 Sep 2018 21:49:44 +0000 (UTC)
From:      Mark Johnston <markj@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r338536 - head/sys/vm
Message-ID:  <201809082149.w88Lniko022447@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help 
Author: markj
Date: Sat Sep  8 21:49:43 2018
New Revision: 338536
URL: https://svnweb.freebsd.org/changeset/base/338536

Log:
  Relax an assertion in vm_pqbatch_process_page().
  
  While executing vm_pqbatch_process_page(m), m->queue may change to
  PQ_NONE if the page daemon is concurrently freeing the page.  In this
  case m's queue state flags must be clear, so vm_pqbatch_process_page()
  will be a no-op, but the race could cause spurious assertion failures.
  Correct the assertion which assumed that m->queue's value does not
  change while the page queue lock is held.
  
  Reviewed by:	alc, kib
  Reported and tested by:	pho
  Approved by:	re (gjb)
  Sponsored by:	The FreeBSD Foundation
  Differential Revision:	https://reviews.freebsd.org/D17027

Modified:
  head/sys/vm/vm_page.c

Modified: head/sys/vm/vm_page.c
==============================================================================
--- head/sys/vm/vm_page.c	Sat Sep  8 19:17:22 2018	(r338535)
+++ head/sys/vm/vm_page.c	Sat Sep  8 21:49:43 2018	(r338536)
@@ -3107,28 +3107,35 @@ static inline void
 vm_pqbatch_process_page(struct vm_pagequeue *pq, vm_page_t m)
 {
 	struct vm_domain *vmd;
-	uint8_t aflags;
+	uint8_t qflags;
 
 	CRITICAL_ASSERT(curthread);
 	vm_pagequeue_assert_locked(pq);
-	KASSERT(pq == vm_page_pagequeue(m),
-	    ("page %p doesn't belong to %p", m, pq));
 
-	aflags = m->aflags;
-	if ((aflags & PGA_DEQUEUE) != 0) {
-		if (__predict_true((aflags & PGA_ENQUEUED) != 0)) {
+	/*
+	 * The page daemon is allowed to set m->queue = PQ_NONE without
+	 * the page queue lock held.  In this case it is about to free the page,
+	 * which must not have any queue state.
+	 */
+	qflags = atomic_load_8(&m->aflags) & PGA_QUEUE_STATE_MASK;
+	KASSERT(pq == vm_page_pagequeue(m) || qflags == 0,
+	    ("page %p doesn't belong to queue %p but has queue state %#x",
+	    m, pq, qflags));
+
+	if ((qflags & PGA_DEQUEUE) != 0) {
+		if (__predict_true((qflags & PGA_ENQUEUED) != 0)) {
 			TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
 			vm_pagequeue_cnt_dec(pq);
 		}
 		vm_page_dequeue_complete(m);
-	} else if ((aflags & (PGA_REQUEUE | PGA_REQUEUE_HEAD)) != 0) {
-		if ((aflags & PGA_ENQUEUED) != 0)
+	} else if ((qflags & (PGA_REQUEUE | PGA_REQUEUE_HEAD)) != 0) {
+		if ((qflags & PGA_ENQUEUED) != 0)
 			TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
 		else {
 			vm_pagequeue_cnt_inc(pq);
 			vm_page_aflag_set(m, PGA_ENQUEUED);
 		}
-		if ((aflags & PGA_REQUEUE_HEAD) != 0) {
+		if ((qflags & PGA_REQUEUE_HEAD) != 0) {
 			KASSERT(m->queue == PQ_INACTIVE,
 			    ("head enqueue not supported for page %p", m));
 			vmd = vm_pagequeue_domain(m);

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201809082149.w88Lniko022447>