From: Matthew Dillon <dillon@apollo.backplane.com>
Date: Wed, 26 Oct 2011 21:48:10 +0000 (-0700)
Subject: kernel - Adjust pagezero/pagecopy assembly and re-enable VM_ALLOC_ZERO
X-Git-Tag: v3.0.0~814
X-Git-Url: https://gitweb.dragonflybsd.org/dragonfly.git/commitdiff_plain/8f901ed8c5e62fb529eb8944758cee9e70f0ea5b?hp=f2c5d4ab43eb3300236045f944aa0021db095c53

kernel - Adjust pagezero/pagecopy assembly and re-enable VM_ALLOC_ZERO

* Remove the movnti, sfence, and prefetch instructions from the pagezero(0
  and pagecopy() assembly.  They don't help and will eve hurt on some of
  the less powerful cpus.

* Re-enable the use of VM_ALLOC_ZERO.  There was no difference in test
  compile times w/ the concurrent buildkernel -j 48 NO_MODULES=TRUE test.
  It might help w/lower-load edge cases so keep it around.
---

diff --git a/sys/platform/pc64/x86_64/support.s b/sys/platform/pc64/x86_64/support.s
index 3c1b2e726b..ddd0a38ef2 100644
--- a/sys/platform/pc64/x86_64/support.s
+++ b/sys/platform/pc64/x86_64/support.s
@@ -67,13 +67,13 @@ ENTRY(pagezero)
 	subq	%rdx,%rdi
 	xorl	%eax,%eax
 1:
-	movnti	%rax,(%rdi,%rdx)
-	movnti	%rax,8(%rdi,%rdx)
-	movnti	%rax,16(%rdi,%rdx)
-	movnti	%rax,24(%rdi,%rdx)
+	movq	%rax,(%rdi,%rdx)	/* movnti */
+	movq	%rax,8(%rdi,%rdx)	/* movnti */
+	movq	%rax,16(%rdi,%rdx)	/* movnti */
+	movq	%rax,24(%rdi,%rdx)	/* movnti */
 	addq	$32,%rdx
 	jne	1b
-	sfence
+	/*sfence*/
 	ret
 
 ENTRY(bcmp)
@@ -172,21 +172,21 @@ ENTRY(pagecopy)
 	subq	%rax,%rdi
 	subq	%rax,%rsi
 1:
-	prefetchnta (%rdi,%rax)
-	addq	$64,%rax
-	jne	1b
+	/*prefetchnta (%rdi,%rax)*/
+	/*addq	$64,%rax*/
+	/*jne	1b*/
 2:
 	movq	(%rdi,%rdx),%rax
-	movnti	%rax,(%rsi,%rdx)
+	movq	%rax,(%rsi,%rdx)	/* movnti */
 	movq	8(%rdi,%rdx),%rax
-	movnti	%rax,8(%rsi,%rdx)
+	movq	%rax,8(%rsi,%rdx)	/* movnti */
 	movq	16(%rdi,%rdx),%rax
-	movnti	%rax,16(%rsi,%rdx)
+	movq	%rax,16(%rsi,%rdx)	/* movnti */
 	movq	24(%rdi,%rdx),%rax
-	movnti	%rax,24(%rsi,%rdx)
+	movq	%rax,24(%rsi,%rdx)	/* movnti */
 	addq	$32,%rdx
 	jne	2b
-	sfence
+	/*sfence*/
 	ret
 
 /* fillw(pat, base, cnt) */  
diff --git a/sys/vm/vm_fault.c b/sys/vm/vm_fault.c
index 522b8bfd26..bc12340468 100644
--- a/sys/vm/vm_fault.c
+++ b/sys/vm/vm_fault.c
@@ -1153,18 +1153,13 @@ vm_fault_object(struct faultstate *fs,
 
 			/*
 			 * Allocate a new page for this object/offset pair.
-			 *
-			 * XXX for now don't use the VM_ALLOC_ZERO flag
-			 *     because this will continuously cycle pages
-			 *     through the cpu caches.  Try to use a recently
-			 *     freed page.
 			 */
 			fs->m = NULL;
 			if (!vm_page_count_severe()) {
 				fs->m = vm_page_alloc(fs->object, pindex,
 				    ((fs->vp || fs->object->backing_object) ?
 					VM_ALLOC_NORMAL :
-					VM_ALLOC_NORMAL /*| VM_ALLOC_ZERO*/));
+					VM_ALLOC_NORMAL | VM_ALLOC_ZERO));
 			}
 			if (fs->m == NULL) {
 				vm_object_pip_wakeup(fs->first_object);
@@ -2240,14 +2235,9 @@ vm_prefault(pmap_t pmap, vm_offset_t addra, vm_map_entry_t entry, int prot)
 
 				/*
 				 * NOTE: Allocated from base object
-				 *
-				 * XXX for now don't use the VM_ALLOC_ZERO
-				 *     flag because this will continuously
-				 *     cycle pages through the cpu caches.
-				 *     Try to use a recently freed page.
 				 */
 				m = vm_page_alloc(object, index,
-					      VM_ALLOC_NORMAL /*| VM_ALLOC_ZERO*/);
+						  VM_ALLOC_NORMAL | VM_ALLOC_ZERO);
 
 				if ((m->flags & PG_ZERO) == 0) {
 					vm_page_zero_fill(m);