13833 want bhyve memory reservoir

13822 bhyve memory should exert memory pressure
13834 want extensible page_resv
13821 vmmctl ioctls should have more structure
Reviewed by: Andy Fiddaman <andy@omnios.org>
Reviewed by: Jason King <jason.brian.king@gmail.com>
Reviewed by: Dan Cross <cross@oxidecomputer.com>
Reviewed by: Hans Rosenfeld <rosenfeld@grumpf.hope-2000.org>
Reviewed by: Mike Zeller <mike.zeller@joyent.com>
Approved by: Dan McDonald <danmcd@joyent.com>

2 files changed, 63 insertions, 21 deletions

diff --git a/usr/src/uts/common/vm/page.h b/usr/src/uts/common/vm/page.h
index 8747b96acc..5b98acd24f 100644
--- a/usr/src/uts/common/vm/page.h
+++ b/usr/src/uts/common/vm/page.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2021 Oxide Computer Company
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -340,13 +341,13 @@ struct as;
  *
  * So, as a quick summary:
  *
- * 	pse_mutex[]'s protect the p_selock and p_cv fields.
+ *	pse_mutex[]'s protect the p_selock and p_cv fields.
  *
- * 	p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
+ *	p_selock protects the p_free, p_age, p_vnode, p_offset and p_hash,
  *
- * 	ph_mutex[]'s protect the page_hash[] array and its chains.
+ *	ph_mutex[]'s protect the page_hash[] array and its chains.
  *
- * 	vph_mutex[]'s protect the v_pages field and the vp page chains.
+ *	vph_mutex[]'s protect the v_pages field and the vp page chains.
  *
  *	First lock the page, then the hash chain, then the vnode chain.  When
  *	this is not possible `trylocks' must be used.  Sleeping while holding
@@ -762,6 +763,7 @@ void	page_lock_delete(page_t *);
 int	page_deleted(page_t *);
 int	page_pp_lock(page_t *, int, int);
 void	page_pp_unlock(page_t *, int, int);
+int	page_xresv(pgcnt_t, uint_t, int (*)(void));
 int	page_resv(pgcnt_t, uint_t);
 void	page_unresv(pgcnt_t);
 void	page_pp_useclaim(page_t *, page_t *, uint_t);
@@ -1078,7 +1080,7 @@ typedef struct kpm_hlk {
  * The state about how a kpm page is mapped and whether it is ready to go
  * is indicated by the following 1 byte kpm_spage structure. This byte is
  * split into two 4-bit parts - kp_mapped and kp_mapped_go.
- * 	- kp_mapped == 1	the page is mapped cacheable
+ *	- kp_mapped == 1	the page is mapped cacheable
  *	- kp_mapped == 2	the page is mapped non-cacheable
  *	- kp_mapped_go == 1	the mapping is ready to be dropped in
  *	- kp_mapped_go == 0	the mapping is not ready to be dropped in.
diff --git a/usr/src/uts/common/vm/vm_page.c b/usr/src/uts/common/vm/vm_page.c
index bcc6d05d47..89751b7b2c 100644
--- a/usr/src/uts/common/vm/vm_page.c
+++ b/usr/src/uts/common/vm/vm_page.c
@@ -23,6 +23,7 @@
  * Copyright (c) 2015, Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
  * Copyright (c) 2015, 2016 by Delphix. All rights reserved.
  * Copyright 2018 Joyent, Inc.
+ * Copyright 2021 Oxide Computer Company
  */
 
 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989  AT&T */
@@ -3919,29 +3920,68 @@ page_pp_unlock(
 }
 
 /*
- * This routine reserves availrmem for npages;
- *	flags: KM_NOSLEEP or KM_SLEEP
- *	returns 1 on success or 0 on failure
+ * This routine reserves availrmem for npages.
+ * It returns 1 on success or 0 on failure.
+ *
+ * flags: KM_NOSLEEP or KM_SLEEP
+ * cb_wait: called to induce delay when KM_SLEEP reservation requires kmem
+ *     reaping to potentially succeed.  If the callback returns 0, the
+ *     reservation attempts will cease to repeat and page_xresv() may
+ *     report a failure.  If cb_wait is NULL, the traditional delay(hz/2)
+ *     behavior will be used while waiting for a reap.
  */
 int
-page_resv(pgcnt_t npages, uint_t flags)
+page_xresv(pgcnt_t npages, uint_t flags, int (*cb_wait)(void))
 {
 	mutex_enter(&freemem_lock);
-	while (availrmem < tune.t_minarmem + npages) {
-		if (flags & KM_NOSLEEP) {
-			mutex_exit(&freemem_lock);
-			return (0);
-		}
+	if (availrmem >= tune.t_minarmem + npages) {
+		availrmem -= npages;
 		mutex_exit(&freemem_lock);
-		page_needfree(npages);
-		kmem_reap();
-		delay(hz >> 2);
-		page_needfree(-(spgcnt_t)npages);
-		mutex_enter(&freemem_lock);
+		return (1);
+	} else if ((flags & KM_NOSLEEP) != 0) {
+		mutex_exit(&freemem_lock);
+		return (0);
 	}
-	availrmem -= npages;
 	mutex_exit(&freemem_lock);
-	return (1);
+
+	/*
+	 * We signal memory pressure to the system by elevating 'needfree'.
+	 * Processes such as kmem reaping, pageout, and ZFS ARC shrinking can
+	 * then respond to said pressure by freeing pages.
+	 */
+	page_needfree(npages);
+	int nobail = 1;
+	do {
+		kmem_reap();
+		if (cb_wait == NULL) {
+			delay(hz >> 2);
+		} else {
+			nobail = cb_wait();
+		}
+
+		mutex_enter(&freemem_lock);
+		if (availrmem >= tune.t_minarmem + npages) {
+			availrmem -= npages;
+			mutex_exit(&freemem_lock);
+			page_needfree(-(spgcnt_t)npages);
+			return (1);
+		}
+		mutex_exit(&freemem_lock);
+	} while (nobail != 0);
+	page_needfree(-(spgcnt_t)npages);
+
+	return (0);
+}
+
+/*
+ * This routine reserves availrmem for npages;
+ *	flags: KM_NOSLEEP or KM_SLEEP
+ *	returns 1 on success or 0 on failure
+ */
+int
+page_resv(pgcnt_t npages, uint_t flags)
+{
+	return (page_xresv(npages, flags, NULL));
 }
 
 /*