Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 3 Feb 2019 18:43:20 +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: r343720 - head/sys/vm
Message-ID:  <201902031843.x13IhK0Y040606@repo.freebsd.org>

next in thread | raw e-mail | index | archive | help
Author: markj
Date: Sun Feb  3 18:43:20 2019
New Revision: 343720
URL: https://svnweb.freebsd.org/changeset/base/343720

Log:
  Allow vm_page_free_prep() to dequeue pages without the page lock.
  
  This is a step towards being able to free pages without the page
  lock held.  The approach is simply to add an implementation of
  vm_page_dequeue_deferred() which does not assert that the page
  lock is held.  Formally, the page lock is required to set
  PGA_DEQUEUE, but in the case of vm_page_free_prep() we get the
  same mutual exclusion for free by virtue of the fact that no
  other references to the page may exist.
  
  No functional change intended.
  
  Reviewed by:	kib (previous version)
  MFC after:	2 weeks
  Sponsored by:	Netflix
  Differential Revision:	https://reviews.freebsd.org/D19065

Modified:
  head/sys/vm/vm_page.c
  head/sys/vm/vm_page.h

Modified: head/sys/vm/vm_page.c
==============================================================================
--- head/sys/vm/vm_page.c	Sun Feb  3 18:38:58 2019	(r343719)
+++ head/sys/vm/vm_page.c	Sun Feb  3 18:43:20 2019	(r343720)
@@ -3175,7 +3175,11 @@ vm_pqbatch_submit_page(vm_page_t m, uint8_t queue)
 	struct vm_pagequeue *pq;
 	int domain;
 
-	vm_page_assert_locked(m);
+	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
+	    ("page %p is unmanaged", m));
+	KASSERT(mtx_owned(vm_page_lockptr(m)) ||
+	    (m->object == NULL && (m->aflags & PGA_DEQUEUE) != 0),
+	    ("missing synchronization for page %p", m));
 	KASSERT(queue < PQ_COUNT, ("invalid queue %d", queue));
 
 	domain = vm_phys_domain(m);
@@ -3197,8 +3201,9 @@ vm_pqbatch_submit_page(vm_page_t m, uint8_t queue)
 
 	/*
 	 * The page may have been logically dequeued before we acquired the
-	 * page queue lock.  In this case, the page lock prevents the page
-	 * from being logically enqueued elsewhere.
+	 * page queue lock.  In this case, since we either hold the page lock
+	 * or the page is being freed, a different thread cannot be concurrently
+	 * enqueuing the page.
 	 */
 	if (__predict_true(m->queue == queue))
 		vm_pqbatch_process_page(pq, m);
@@ -3290,6 +3295,30 @@ vm_page_dequeue_deferred(vm_page_t m)
 }
 
 /*
+ * A variant of vm_page_dequeue_deferred() that does not assert the page
+ * lock and is only to be called from vm_page_free_prep().  It is just an
+ * open-coded implementation of vm_page_dequeue_deferred().  Because the
+ * page is being freed, we can assume that nothing else is scheduling queue
+ * operations on this page, so we get for free the mutual exclusion that
+ * is otherwise provided by the page lock.
+ */
+static void
+vm_page_dequeue_deferred_free(vm_page_t m)
+{
+	uint8_t queue;
+
+	KASSERT(m->object == NULL, ("page %p has an object reference", m));
+
+	if ((m->aflags & PGA_DEQUEUE) != 0)
+		return;
+	atomic_thread_fence_acq();
+	if ((queue = m->queue) == PQ_NONE)
+		return;
+	vm_page_aflag_set(m, PGA_DEQUEUE);
+	vm_pqbatch_submit_page(m, queue);
+}
+
+/*
  *	vm_page_dequeue:
  *
  *	Remove the page from whichever page queue it's in, if any.
@@ -3474,7 +3503,7 @@ vm_page_free_prep(vm_page_t m)
 	 * dequeue.
 	 */
 	if ((m->oflags & VPO_UNMANAGED) == 0)
-		vm_page_dequeue_deferred(m);
+		vm_page_dequeue_deferred_free(m);
 
 	m->valid = 0;
 	vm_page_undirty(m);

Modified: head/sys/vm/vm_page.h
==============================================================================
--- head/sys/vm/vm_page.h	Sun Feb  3 18:38:58 2019	(r343719)
+++ head/sys/vm/vm_page.h	Sun Feb  3 18:43:20 2019	(r343720)
@@ -351,8 +351,10 @@ extern struct mtx_padalign pa_lock[];
  * queue, and cleared when the dequeue request is processed.  A page may
  * have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
  * is requested after the page is scheduled to be enqueued but before it is
- * actually inserted into the page queue.  The page lock must be held to set
- * this flag, and the queue lock for the page must be held to clear it.
+ * actually inserted into the page queue.  For allocated pages, the page lock
+ * must be held to set this flag, but it may be set by vm_page_free_prep()
+ * without the page lock held.  The page queue lock must be held to clear the
+ * PGA_DEQUEUE flag.
  *
  * PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
  * in its page queue.  The page lock must be held to set this flag, and the



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201902031843.x13IhK0Y040606>