ia64/linux-2.6.18-xen.hg

view lib/semaphore-sleepers.c @ 897:329ea0ccb344

balloon: try harder to balloon up under memory pressure.

Currently if the balloon driver is unable to increase the guest's
reservation it assumes the failure was due to reaching its full
allocation, gives up on the ballooning operation and records the limit
it reached as the "hard limit". The driver will not try again until
the target is set again (even to the same value).

However it is possible that ballooning has in fact failed due to
memory pressure in the host and therefore it is desirable to keep
attempting to reach the target in case memory becomes available. The
most likely scenario is that some guests are ballooning down while
others are ballooning up and therefore there is temporary memory
pressure while things stabilise. You would not expect a well behaved
toolstack to ask a domain to balloon to more than its allocation nor
would you expect it to deliberately over-commit memory by setting
balloon targets which exceed the total host memory.

This patch drops the concept of a hard limit and causes the balloon
driver to retry increasing the reservation on a timer in the same
manner as when decreasing the reservation.

Also if we partially succeed in increasing the reservation
(i.e. receive less pages than we asked for) then we may as well keep
those pages rather than returning them to Xen.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Fri Jun 05 14:01:20 2009 +0100 (2009-06-05)
parents 831230e53067
children
line source
1 /*
2 * i386 and x86-64 semaphore implementation.
3 *
4 * (C) Copyright 1999 Linus Torvalds
5 *
6 * Portions Copyright 1999 Red Hat, Inc.
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * rw semaphores implemented November 1999 by Benjamin LaHaise <bcrl@kvack.org>
14 */
15 #include <linux/sched.h>
16 #include <linux/err.h>
17 #include <linux/init.h>
18 #include <asm/semaphore.h>
20 /*
21 * Semaphores are implemented using a two-way counter:
22 * The "count" variable is decremented for each process
23 * that tries to acquire the semaphore, while the "sleeping"
24 * variable is a count of such acquires.
25 *
26 * Notably, the inline "up()" and "down()" functions can
27 * efficiently test if they need to do any extra work (up
28 * needs to do something only if count was negative before
29 * the increment operation.
30 *
31 * "sleeping" and the contention routine ordering is protected
32 * by the spinlock in the semaphore's waitqueue head.
33 *
34 * Note that these functions are only called when there is
35 * contention on the lock, and as such all this is the
36 * "non-critical" part of the whole semaphore business. The
37 * critical part is the inline stuff in <asm/semaphore.h>
38 * where we want to avoid any extra jumps and calls.
39 */
41 /*
42 * Logic:
43 * - only on a boundary condition do we need to care. When we go
44 * from a negative count to a non-negative, we wake people up.
45 * - when we go from a non-negative count to a negative do we
46 * (a) synchronize with the "sleeper" count and (b) make sure
47 * that we're on the wakeup list before we synchronize so that
48 * we cannot lose wakeup events.
49 */
51 fastcall void __up(struct semaphore *sem)
52 {
53 wake_up(&sem->wait);
54 }
56 fastcall void __sched __down(struct semaphore * sem)
57 {
58 struct task_struct *tsk = current;
59 DECLARE_WAITQUEUE(wait, tsk);
60 unsigned long flags;
62 tsk->state = TASK_UNINTERRUPTIBLE;
63 spin_lock_irqsave(&sem->wait.lock, flags);
64 add_wait_queue_exclusive_locked(&sem->wait, &wait);
66 sem->sleepers++;
67 for (;;) {
68 int sleepers = sem->sleepers;
70 /*
71 * Add "everybody else" into it. They aren't
72 * playing, because we own the spinlock in
73 * the wait_queue_head.
74 */
75 if (!atomic_add_negative(sleepers - 1, &sem->count)) {
76 sem->sleepers = 0;
77 break;
78 }
79 sem->sleepers = 1; /* us - see -1 above */
80 spin_unlock_irqrestore(&sem->wait.lock, flags);
82 schedule();
84 spin_lock_irqsave(&sem->wait.lock, flags);
85 tsk->state = TASK_UNINTERRUPTIBLE;
86 }
87 remove_wait_queue_locked(&sem->wait, &wait);
88 wake_up_locked(&sem->wait);
89 spin_unlock_irqrestore(&sem->wait.lock, flags);
90 tsk->state = TASK_RUNNING;
91 }
93 fastcall int __sched __down_interruptible(struct semaphore * sem)
94 {
95 int retval = 0;
96 struct task_struct *tsk = current;
97 DECLARE_WAITQUEUE(wait, tsk);
98 unsigned long flags;
100 tsk->state = TASK_INTERRUPTIBLE;
101 spin_lock_irqsave(&sem->wait.lock, flags);
102 add_wait_queue_exclusive_locked(&sem->wait, &wait);
104 sem->sleepers++;
105 for (;;) {
106 int sleepers = sem->sleepers;
108 /*
109 * With signals pending, this turns into
110 * the trylock failure case - we won't be
111 * sleeping, and we* can't get the lock as
112 * it has contention. Just correct the count
113 * and exit.
114 */
115 if (signal_pending(current)) {
116 retval = -EINTR;
117 sem->sleepers = 0;
118 atomic_add(sleepers, &sem->count);
119 break;
120 }
122 /*
123 * Add "everybody else" into it. They aren't
124 * playing, because we own the spinlock in
125 * wait_queue_head. The "-1" is because we're
126 * still hoping to get the semaphore.
127 */
128 if (!atomic_add_negative(sleepers - 1, &sem->count)) {
129 sem->sleepers = 0;
130 break;
131 }
132 sem->sleepers = 1; /* us - see -1 above */
133 spin_unlock_irqrestore(&sem->wait.lock, flags);
135 schedule();
137 spin_lock_irqsave(&sem->wait.lock, flags);
138 tsk->state = TASK_INTERRUPTIBLE;
139 }
140 remove_wait_queue_locked(&sem->wait, &wait);
141 wake_up_locked(&sem->wait);
142 spin_unlock_irqrestore(&sem->wait.lock, flags);
144 tsk->state = TASK_RUNNING;
145 return retval;
146 }
148 /*
149 * Trylock failed - make sure we correct for
150 * having decremented the count.
151 *
152 * We could have done the trylock with a
153 * single "cmpxchg" without failure cases,
154 * but then it wouldn't work on a 386.
155 */
156 fastcall int __down_trylock(struct semaphore * sem)
157 {
158 int sleepers;
159 unsigned long flags;
161 spin_lock_irqsave(&sem->wait.lock, flags);
162 sleepers = sem->sleepers + 1;
163 sem->sleepers = 0;
165 /*
166 * Add "everybody else" and us into it. They aren't
167 * playing, because we own the spinlock in the
168 * wait_queue_head.
169 */
170 if (!atomic_add_negative(sleepers, &sem->count)) {
171 wake_up_locked(&sem->wait);
172 }
174 spin_unlock_irqrestore(&sem->wait.lock, flags);
175 return 1;
176 }