kernel - Optimize the x86-64 lwbuf API
[dragonfly.git] / sys / kern / kern_umtx.c
CommitLineData
da5fb9ef 1/*
6a25f455
VS
2 * (MPSAFE)
3 *
4 * Copyright (c) 2003,2004,2010 The DragonFly Project. All rights reserved.
da5fb9ef
MD
5 *
6 * This code is derived from software contributed to The DragonFly Project
7 * by Matthew Dillon <dillon@backplane.com> and David Xu <davidxu@freebsd.org>
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 *
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
18 * distribution.
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
17cde63e 36 * $DragonFly: src/sys/kern/kern_umtx.c,v 1.9 2008/05/09 07:24:45 dillon Exp $
da5fb9ef
MD
37 */
38
39/*
40 * This module implements userland mutex helper functions. umtx_sleep()
41 * handling blocking and umtx_wakeup() handles wakeups. The sleep/wakeup
42 * functions operate on user addresses.
43 */
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/kernel.h>
48#include <sys/sysproto.h>
49#include <sys/sysunion.h>
50#include <sys/sysent.h>
51#include <sys/syscall.h>
52#include <sys/module.h>
53
5c5185ae
SG
54#include <cpu/lwbuf.h>
55
da5fb9ef
MD
56#include <vm/vm.h>
57#include <vm/vm_param.h>
58#include <sys/lock.h>
59#include <vm/pmap.h>
60#include <vm/vm_map.h>
61#include <vm/vm_object.h>
62#include <vm/vm_page.h>
63#include <vm/vm_pager.h>
64#include <vm/vm_pageout.h>
65#include <vm/vm_extern.h>
66#include <vm/vm_page.h>
67#include <vm/vm_kern.h>
68
10192bae
MD
69#include <vm/vm_page2.h>
70
71static void umtx_sleep_page_action_cow(vm_page_t m, vm_page_action_t action);
72
da5fb9ef
MD
73/*
74 * If the contents of the userland-supplied pointer matches the specified
75 * value enter an interruptable sleep for up to <timeout> microseconds.
76 * If the contents does not match then return immediately.
77 *
10192bae
MD
78 * Returns 0 if we slept and were woken up, -1 and EWOULDBLOCK if we slept
79 * and timed out, and EBUSY if the contents of the pointer already does
80 * not match the specified value. A timeout of 0 indicates an unlimited sleep.
02095b1c
MD
81 * EINTR is returned if the call was interrupted by a signal (even if
82 * the signal specifies that the system call should restart).
da5fb9ef
MD
83 *
84 * This function interlocks against call to umtx_wakeup. It does NOT interlock
85 * against changes in *ptr. However, it does not have to. The standard use
86 * of *ptr is to differentiate between an uncontested and a contested mutex
87 * and call umtx_wakeup when releasing a contested mutex. Therefore we can
88 * safely race against changes in *ptr as long as we are properly interlocked
89 * against the umtx_wakeup() call.
90 *
10192bae
MD
91 * The VM page associated with the mutex is held in an attempt to keep
92 * the mutex's physical address consistent, allowing umtx_sleep() and
93 * umtx_wakeup() to use the physical address as their rendezvous. BUT
94 * situations can arise where the physical address may change, particularly
95 * if a threaded program fork()'s and the mutex's memory becomes
96 * copy-on-write. We register an event on the VM page to catch COWs.
da5fb9ef
MD
97 *
98 * umtx_sleep { const int *ptr, int value, int timeout }
99 */
100int
753fd850 101sys_umtx_sleep(struct umtx_sleep_args *uap)
da5fb9ef 102{
7a683a24 103 struct lwbuf lwb_cache;
5c5185ae 104 struct lwbuf *lwb;
10192bae 105 struct vm_page_action action;
da5fb9ef
MD
106 vm_page_t m;
107 void *waddr;
06c5a8d6 108 int offset;
da5fb9ef 109 int timeout;
7a683a24 110 int error = EBUSY;
da5fb9ef 111
10192bae 112 if (uap->timeout < 0)
da5fb9ef 113 return (EINVAL);
06c5a8d6 114 if ((vm_offset_t)uap->ptr & (sizeof(int) - 1))
da5fb9ef 115 return (EFAULT);
10192bae
MD
116
117 /*
118 * When faulting in the page, force any COW pages to be resolved.
119 * Otherwise the physical page we sleep on my not match the page
120 * being woken up.
121 */
6a25f455 122 lwkt_gettoken(&vm_token);
10192bae 123 m = vm_fault_page_quick((vm_offset_t)uap->ptr, VM_PROT_READ|VM_PROT_WRITE, &error);
3919ced0
MD
124 if (m == NULL) {
125 error = EFAULT;
126 goto done;
127 }
7a683a24 128 lwb = lwbuf_alloc(m, &lwb_cache);
06c5a8d6 129 offset = (vm_offset_t)uap->ptr & PAGE_MASK;
da5fb9ef 130
10192bae
MD
131 /*
132 * The critical section is required to interlock the tsleep against
133 * a wakeup from another cpu. The lfence forces synchronization.
134 */
5c5185ae 135 if (*(int *)(lwbuf_kva(lwb) + offset) == uap->value) {
10192bae
MD
136 if ((timeout = uap->timeout) != 0) {
137 timeout = (timeout / 1000000) * hz +
138 ((timeout % 1000000) * hz + 999999) / 1000000;
139 }
06c5a8d6 140 waddr = (void *)((intptr_t)VM_PAGE_TO_PHYS(m) + offset);
10192bae 141 crit_enter();
ae8e83e6 142 tsleep_interlock(waddr, PCATCH | PDOMAIN_UMTX);
5c5185ae 143 if (*(int *)(lwbuf_kva(lwb) + offset) == uap->value) {
906c754c
MD
144 vm_page_init_action(m, &action, umtx_sleep_page_action_cow, waddr);
145 vm_page_register_action(&action, VMEVENT_COW);
d9345d3a
MD
146 error = tsleep(waddr, PCATCH | PINTERLOCKED | PDOMAIN_UMTX,
147 "umtxsl", timeout);
906c754c 148 vm_page_unregister_action(&action);
10192bae
MD
149 } else {
150 error = EBUSY;
151 }
152 crit_exit();
02095b1c
MD
153 /* Always break out in case of signal, even if restartable */
154 if (error == ERESTART)
e392b468 155 error = EINTR;
da5fb9ef
MD
156 } else {
157 error = EBUSY;
158 }
06c5a8d6 159
5c5185ae 160 lwbuf_free(lwb);
17cde63e 161 /*vm_page_dirty(m); we don't actually dirty the page */
06c5a8d6 162 vm_page_unhold(m);
3919ced0 163done:
6a25f455 164 lwkt_reltoken(&vm_token);
da5fb9ef
MD
165 return(error);
166}
167
168/*
10192bae
MD
169 * If this page is being copied it may no longer represent the page
170 * underlying our virtual address. Wake up any umtx_sleep()'s
171 * that were waiting on its physical address to force them to retry.
172 */
173static void
174umtx_sleep_page_action_cow(vm_page_t m, vm_page_action_t action)
175{
c5724852 176 lwkt_gettoken(&vm_token);
10192bae 177 wakeup_domain(action->data, PDOMAIN_UMTX);
c5724852 178 lwkt_reltoken(&vm_token);
10192bae
MD
179}
180
181/*
da5fb9ef
MD
182 * umtx_wakeup { const int *ptr, int count }
183 *
184 * Wakeup the specified number of processes held in umtx_sleep() on the
185 * specified user address. A count of 0 wakes up all waiting processes.
186 *
187 * XXX assumes that the physical address space does not exceed the virtual
188 * address space.
189 */
190int
753fd850 191sys_umtx_wakeup(struct umtx_wakeup_args *uap)
da5fb9ef 192{
06c5a8d6
MD
193 vm_page_t m;
194 int offset;
195 int error;
da5fb9ef
MD
196 void *waddr;
197
35238fa5 198 cpu_mfence();
06c5a8d6 199 if ((vm_offset_t)uap->ptr & (sizeof(int) - 1))
da5fb9ef 200 return (EFAULT);
6a25f455 201 lwkt_gettoken(&vm_token);
06c5a8d6 202 m = vm_fault_page_quick((vm_offset_t)uap->ptr, VM_PROT_READ, &error);
3919ced0
MD
203 if (m == NULL) {
204 error = EFAULT;
205 goto done;
206 }
06c5a8d6
MD
207 offset = (vm_offset_t)uap->ptr & PAGE_MASK;
208 waddr = (void *)((intptr_t)VM_PAGE_TO_PHYS(m) + offset);
209
da5fb9ef
MD
210 if (uap->count == 1) {
211 wakeup_domain_one(waddr, PDOMAIN_UMTX);
212 } else {
213 /* XXX wakes them all up for now */
214 wakeup_domain(waddr, PDOMAIN_UMTX);
215 }
06c5a8d6 216 vm_page_unhold(m);
3919ced0
MD
217 error = 0;
218done:
6a25f455 219 lwkt_reltoken(&vm_token);
3919ced0 220 return(error);
da5fb9ef
MD
221}
222