Move the uvm reserve enforcement from uvm_pagealloc to pmemrange.

More and more things are allocating outside of uvm_pagealloc these days making
it easy for something like the buffer cache to eat your last page with no
repercussions (other than a hung machine, of course).

ok ariane@ also ok ariane@ again after I spotted and fixed a possible underflow
problem in the calculation.

3 files changed, 45 insertions, 36 deletions

diff --git a/sys/uvm/uvm_extern.h b/sys/uvm/uvm_extern.h
index 72abe2a2540..72a9125340d 100644
--- a/sys/uvm/uvm_extern.h
+++ b/sys/uvm/uvm_extern.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_extern.h,v 1.101 2011/07/06 19:50:38 beck Exp $	*/
+/*	$OpenBSD: uvm_extern.h,v 1.102 2011/07/07 20:52:50 oga Exp $	*/
 /*	$NetBSD: uvm_extern.h,v 1.57 2001/03/09 01:02:12 chs Exp $	*/
 
 /*
@@ -222,6 +222,7 @@ typedef int		vm_prot_t;
 #define UVM_PLA_ZERO		0x0004	/* zero all pages before returning */
 #define UVM_PLA_TRYCONTIG	0x0008	/* try to allocate contig physmem */
 #define UVM_PLA_FAILOK		0x0010	/* caller can handle failure */
+#define UVM_PLA_USERESERVE	0x0020	/* ok to use reserve pages */
 
 /*
  * lockflags that control the locking behavior of various functions.
diff --git a/sys/uvm/uvm_page.c b/sys/uvm/uvm_page.c
index 48a62736256..318a0e52734 100644
--- a/sys/uvm/uvm_page.c
+++ b/sys/uvm/uvm_page.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_page.c,v 1.112 2011/07/06 19:50:38 beck Exp $	*/
+/*	$OpenBSD: uvm_page.c,v 1.113 2011/07/07 20:52:50 oga Exp $	*/
 /*	$NetBSD: uvm_page.c,v 1.44 2000/11/27 08:40:04 chs Exp $	*/
 
 /*
@@ -76,7 +76,6 @@
 #include <sys/sched.h>
 #include <sys/kernel.h>
 #include <sys/vnode.h>
-#include <sys/mount.h>
 #include <sys/proc.h>
 
 #include <uvm/uvm.h>
@@ -920,43 +919,19 @@ uvm_pagealloc(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
 	struct vm_page *pg;
 	struct pglist pgl;
 	int pmr_flags;
-	boolean_t use_reserve;
 
 	KASSERT(obj == NULL || anon == NULL);
 	KASSERT(off == trunc_page(off));
 
-	/*
-	 * check to see if we need to generate some free pages waking
-	 * the pagedaemon.
-	 */
-	if ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freemin ||
-	    ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freetarg &&
-	    (uvmexp.inactive + BUFPAGES_INACT) < uvmexp.inactarg))
-		wakeup(&uvm.pagedaemon);
-
-	/*
-	 * fail if any of these conditions is true:
-	 * [1]  there really are no free pages, or
-	 * [2]  only kernel "reserved" pages remain and
-	 *        the page isn't being allocated to a kernel object.
-	 * [3]  only pagedaemon "reserved" pages remain and
-	 *        the requestor isn't the pagedaemon.
-	 */
-
-	use_reserve = (flags & UVM_PGA_USERESERVE) ||
-		(obj && UVM_OBJ_IS_KERN_OBJECT(obj));
-	if ((uvmexp.free <= uvmexp.reserve_kernel && !use_reserve) ||
-	    (uvmexp.free <= uvmexp.reserve_pagedaemon &&
-	     !((curproc == uvm.pagedaemon_proc) ||
-	      (curproc == syncerproc))))
-		goto fail;
-
+	/* XXX these functions should share flags */
 	pmr_flags = UVM_PLA_NOWAIT;
 	if (flags & UVM_PGA_ZERO)
 		pmr_flags |= UVM_PLA_ZERO;
+	if (flags & UVM_PGA_USERESERVE)
+		pmr_flags |= UVM_PLA_USERESERVE;
 	TAILQ_INIT(&pgl);
 	if (uvm_pmr_getpages(1, 0, 0, 1, 0, 1, pmr_flags, &pgl) != 0)
-		goto fail;
+		return (NULL);
 
 	pg = TAILQ_FIRST(&pgl);
 	KASSERT(pg != NULL && TAILQ_NEXT(pg, pageq) == NULL);
@@ -967,10 +942,7 @@ uvm_pagealloc(struct uvm_object *obj, voff_t off, struct vm_anon *anon,
 	if (flags & UVM_PGA_ZERO)
 		atomic_clearbits_int(&pg->pg_flags, PG_CLEAN);
 
-	return(pg);
-
- fail:
-	return (NULL);
+	return (pg);
 }
 
 /*
diff --git a/sys/uvm/uvm_pmemrange.c b/sys/uvm/uvm_pmemrange.c
index 114709b6ac4..20b68bd8152 100644
--- a/sys/uvm/uvm_pmemrange.c
+++ b/sys/uvm/uvm_pmemrange.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_pmemrange.c,v 1.27 2011/07/06 19:50:38 beck Exp $	*/
+/*	$OpenBSD: uvm_pmemrange.c,v 1.28 2011/07/07 20:52:50 oga Exp $	*/
 
 /*
  * Copyright (c) 2009, 2010 Ariane van der Steldt <ariane@stack.nl>
@@ -20,6 +20,7 @@
 #include <sys/systm.h>
 #include <uvm/uvm.h>
 #include <sys/malloc.h>
+#include <sys/mount.h>		/* for BUFPAGES defines */
 #include <sys/proc.h>		/* XXX for atomic */
 #include <sys/kernel.h>
 
@@ -747,6 +748,7 @@ uvm_pmr_getpages(psize_t count, paddr_t start, paddr_t end, paddr_t align,
 	int	memtype;		/* Requested memtype. */
 	int	memtype_init;		/* Best memtype. */
 	int	desperate;		/* True if allocation failed. */
+	int	is_pdaemon;
 #ifdef DIAGNOSTIC
 	struct	vm_page *diag_prev;	/* Used during validation. */
 #endif /* DIAGNOSTIC */
@@ -762,6 +764,26 @@ uvm_pmr_getpages(psize_t count, paddr_t start, paddr_t end, paddr_t align,
 	    (boundary == 0 || maxseg * boundary >= count) &&
 	    TAILQ_EMPTY(result));
 
+	is_pdaemon = ((curproc == uvm.pagedaemon_proc) ||
+	      (curproc == syncerproc));
+
+	/*
+	 * All allocations by the pagedaemon automatically get access to
+	 * the kernel reserve of pages so swapping can catch up with memory
+	 * exhaustion
+	 */
+	if (is_pdaemon)
+		flags |= UVM_PLA_USERESERVE;
+
+	/*
+	 * check to see if we need to generate some free pages waking
+	 * the pagedaemon.
+	 */
+	if ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freemin ||
+	    ((uvmexp.free - BUFPAGES_DEFICIT) < uvmexp.freetarg &&
+	    (uvmexp.inactive + BUFPAGES_INACT) < uvmexp.inactarg))
+		wakeup(&uvm.pagedaemon);
+
 	/*
 	 * TRYCONTIG is a noop if you only want a single segment.
 	 * Remove it if that's the case: otherwise it'll deny the fast
@@ -836,6 +858,20 @@ retry:		/* Return point after sleeping. */
 	fcount = 0;
 	fnsegs = 0;
 
+	/*
+	 * fail if any of these conditions are true:
+	 * [1]  there really are no free pages, or
+	 * [2]  only kernel "reserved" pages remain and
+	 *        the we are not allowed to use them.
+	 * [3]  only pagedaemon "reserved" pages remain and
+	 *        the requestor isn't the pagedaemon.
+	 */
+	if (((uvmexp.free  < uvmexp.reserve_kernel + ptoa(count)) &&
+	    (flags & UVM_PLA_USERESERVE) == 0) ||
+	    ((uvmexp.free < uvmexp.reserve_pagedaemon + ptoa(count)) &&
+	    !is_pdaemon))
+		goto fail;
+
 retry_desperate:
 	/*
 	 * If we just want any page(s), go for the really fast option.