svn commit: r268397 - in user/attilio/rm_vmobj_cache/sys: sys vm

[Attilio Rao]

Author: attilio
Date: Tue Jul  8 13:24:00 2014
New Revision: 268397
URL: http://svnweb.freebsd.org/changeset/base/268397

Log:
  - Fix vm_page_dispose() to also accept already enqueued pages and to
    correctly dequeue them.
    This aligns vm_page_dispose() with vm_page_activate() and
    vm_page_deactivate().
  - Rewrite vm_page_advise() to take advantage of the newly introduced
    disposed pagequeue.
    Pages advised with MADV_FREE will be infact moved to the high priority
    disposed pqueue, while pages advised with MADV_DONTNEED will be moved
    to the inactive queue but at the head, so that they will be more likely
    freed with higher priority.
    The only exception to this rule is with dirty pages that will be moved
    normally at the tail of the inactive queue, as the pagedaemon will
    try anyway to move them at the tail after pass == 1.

Modified:
  user/attilio/rm_vmobj_cache/sys/sys/pcpu.h
  user/attilio/rm_vmobj_cache/sys/vm/vm_page.c

Modified: user/attilio/rm_vmobj_cache/sys/sys/pcpu.h
==============================================================================
--- user/attilio/rm_vmobj_cache/sys/sys/pcpu.h	Tue Jul  8 13:16:03 2014	(r268396)
+++ user/attilio/rm_vmobj_cache/sys/sys/pcpu.h	Tue Jul  8 13:24:00 2014	(r268397)
@@ -162,7 +162,6 @@ struct pcpu {
 	long		pc_cp_time[CPUSTATES];	/* statclock ticks */
 	struct device	*pc_device;
 	void		*pc_netisr;		/* netisr SWI cookie */
-	int		pc_dnweight;		/* vm_page_dontneed() */
 	int		pc_domain;		/* Memory domain. */
 	struct rm_queue	pc_rm_queue;		/* rmlock list of trackers */
 	uintptr_t	pc_dynamic;		/* Dynamic per-cpu data area */

Modified: user/attilio/rm_vmobj_cache/sys/vm/vm_page.c
==============================================================================
--- user/attilio/rm_vmobj_cache/sys/vm/vm_page.c	Tue Jul  8 13:16:03 2014	(r268396)
+++ user/attilio/rm_vmobj_cache/sys/vm/vm_page.c	Tue Jul  8 13:24:00 2014	(r268397)
@@ -2449,12 +2449,15 @@ void
 vm_page_dispose(vm_page_t m)
 {
 	struct vm_pagequeue *pq;
+	int queue;
 
 	vm_page_lock_assert(m, MA_OWNED);
-	KASSERT(m->queue == PQ_NONE,
-	    ("vm_page_dispose: page %p already queued on %u queue", m,
-	    m->queue));
 
+	queue = m->queue;
+	if (queue == PQ_DISPOSED)
+		return;
+	if (queue != PQ_NONE)
+		vm_page_dequeue(m);
 	if (m->hold_count != 0)
 		panic("vm_page_dispose: page %p hold count %d",
 		    m, m->hold_count);
@@ -2628,23 +2631,17 @@ vm_page_cache(vm_page_t m)
 /*
  * vm_page_advise
  *
- *	Cache, deactivate, or do nothing as appropriate.  This routine
+ *	Dispose, deactivate, or do nothing as appropriate.  This routine
  *	is used by madvise().
  *
- *	Generally speaking we want to move the page into the cache so
- *	it gets reused quickly.  However, this can result in a silly syndrome
- *	due to the page recycling too quickly.  Small objects will not be
- *	fully cached.  On the other hand, if we move the page to the inactive
- *	queue we wind up with a problem whereby very large objects 
- *	unnecessarily blow away our inactive and cache queues.
- *
- *	The solution is to move the pages based on a fixed weighting.  We
- *	either leave them alone, deactivate them, or move them to the cache,
- *	where moving them to the cache has the highest weighting.
- *	By forcing some pages into other queues we eventually force the
- *	system to balance the queues, potentially recovering other unrelated
- *	space from active.  The idea is to not force this to happen too
- *	often.
+ *	For MADV_FREE the pages are moved directly to the higher priority
+ *	disposed pagequeue, for a quick reuse.
+ *	For MADV_DONTNEED the pages are moved directly at the head of the
+ *	inactive queue to boost their priority within the inactive queue.
+ *	The only exception to this last statement is in case of pages are
+ *	also dirty which are then moved to the tail of the inactive queue
+ *	as there are high chances they will be moved there anyway by
+ *	pagedaemon normal scanning.
  *
  *	The object and page must be locked.
  */
@@ -2673,17 +2670,10 @@ vm_page_advise(vm_page_t m, int advice)
 		m->act_count = 0;
 	} else if (advice != MADV_DONTNEED)
 		return;
-	dnw = PCPU_GET(dnweight);
-	PCPU_INC(dnweight);
 
-	/*
-	 * Occasionally leave the page alone.
-	 */
-	if ((dnw & 0x01F0) == 0 || m->queue == PQ_INACTIVE) {
-		if (m->act_count >= ACT_INIT)
-			--m->act_count;
-		return;
-	}
+	/* Set the dirty page bit if appropriate. */
+	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
+		vm_page_dirty(m);
 
 	/*
 	 * Clear any references to the page.  Otherwise, the page daemon will
@@ -2691,23 +2681,15 @@ vm_page_advise(vm_page_t m, int advice)
 	 */
 	vm_page_aflag_clear(m, PGA_REFERENCED);
 
-	if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
-		vm_page_dirty(m);
-
-	if (m->dirty || (dnw & 0x0070) == 0) {
-		/*
-		 * Deactivate the page 3 times out of 32.
-		 */
-		head = 0;
-	} else {
-		/*
-		 * Cache the page 28 times out of every 32.  Note that
-		 * the page is deactivated instead of cached, but placed
-		 * at the head of the queue instead of the tail.
-		 */
-		head = 1;
-	}
-	_vm_page_deactivate(m, head);
+	/*
+	 * For MADV_FREE put the pages into the disposed queue.
+	 * For MADV_DONTNEED, put the pages at the head of the inactive
+	 * queue if clean, otherwise normally at the tail.
+	 */
+	if (advice == MADV_FREE)
+		vm_page_dispose(m);
+	else
+		_vm_page_deactivate(m, !m->dirty);
 }
 
 /*