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.

1 files changed, 37 insertions, 1 deletions

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.