sys/kern/kern_umtx.c

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