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 #include <sys/mplock2.h>
  68
  69 static void umtx_sleep_page_action_cow(vm_page_t m, vm_page_action_t action);
  70
  71 /*
  72  * If the contents of the userland-supplied pointer matches the specified
  73  * value enter an interruptable sleep for up to <timeout> microseconds.
  74  * If the contents does not match then return immediately.
  75  *
  76  * Returns 0 if we slept and were woken up, -1 and EWOULDBLOCK if we slept
  77  * and timed out, and EBUSY if the contents of the pointer already does
  78  * not match the specified value.  A timeout of 0 indicates an unlimited sleep.
  79  * EINTR is returned if the call was interrupted by a signal (even if
  80  * the signal specifies that the system call should restart).
  81  *
  82  * This function interlocks against call to umtx_wakeup.  It does NOT interlock
  83  * against changes in *ptr.  However, it does not have to.  The standard use
  84  * of *ptr is to differentiate between an uncontested and a contested mutex
  85  * and call umtx_wakeup when releasing a contested mutex.  Therefore we can
  86  * safely race against changes in *ptr as long as we are properly interlocked
  87  * against the umtx_wakeup() call.
  88  *
  89  * The VM page associated with the mutex is held in an attempt to keep
  90  * the mutex's physical address consistent, allowing umtx_sleep() and
  91  * umtx_wakeup() to use the physical address as their rendezvous.  BUT
  92  * situations can arise where the physical address may change, particularly
  93  * if a threaded program fork()'s and the mutex's memory becomes
  94  * copy-on-write.  We register an event on the VM page to catch COWs.
  95  *
  96  * umtx_sleep { const int *ptr, int value, int timeout }
  97  *
  98  * MPALMOSTSAFE
  99  */
 100 int
 101 sys_umtx_sleep(struct umtx_sleep_args *uap)
 102 {
 103     int error = EBUSY;
 104     struct sf_buf *sf;
 105     struct vm_page_action action;
 106     vm_page_t m;
 107     void *waddr;
 108     int offset;
 109     int timeout;
 110
 111     if (uap->timeout < 0)
 112         return (EINVAL);
 113     if ((vm_offset_t)uap->ptr & (sizeof(int) - 1))
 114         return (EFAULT);
 115
 116     /*
 117      * When faulting in the page, force any COW pages to be resolved.
 118      * Otherwise the physical page we sleep on my not match the page
 119      * being woken up.
 120      */
 121     get_mplock();
 122     m = vm_fault_page_quick((vm_offset_t)uap->ptr, VM_PROT_READ|VM_PROT_WRITE, &error);
 123     if (m == NULL) {
 124         error = EFAULT;
 125         goto done;
 126     }
 127     sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
 128     offset = (vm_offset_t)uap->ptr & PAGE_MASK;
 129
 130     /*
 131      * The critical section is required to interlock the tsleep against
 132      * a wakeup from another cpu.  The lfence forces synchronization.
 133      */
 134     if (*(int *)(sf_buf_kva(sf) + offset) == uap->value) {
 135         if ((timeout = uap->timeout) != 0) {
 136             timeout = (timeout / 1000000) * hz +
 137                       ((timeout % 1000000) * hz + 999999) / 1000000;
 138         }
 139         waddr = (void *)((intptr_t)VM_PAGE_TO_PHYS(m) + offset);
 140         crit_enter();
 141         tsleep_interlock(waddr, PCATCH | PDOMAIN_UMTX);
 142         if (*(int *)(sf_buf_kva(sf) + offset) == uap->value) {
 143             vm_page_init_action(&action, umtx_sleep_page_action_cow, waddr);
 144             vm_page_register_action(m, &action, VMEVENT_COW);
 145             error = tsleep(waddr, PCATCH | PINTERLOCKED | PDOMAIN_UMTX,
 146                            "umtxsl", timeout);
 147             vm_page_unregister_action(m, &action);
 148         } else {
 149             error = EBUSY;
 150         }
 151         crit_exit();
 152         /* Always break out in case of signal, even if restartable */
 153         if (error == ERESTART)
 154                 error = EINTR;
 155     } else {
 156         error = EBUSY;
 157     }
 158
 159     sf_buf_free(sf);
 160     /*vm_page_dirty(m); we don't actually dirty the page */
 161     vm_page_unhold(m);
 162 done:
 163     rel_mplock();
 164     return(error);
 165 }
 166
 167 /*
 168  * If this page is being copied it may no longer represent the page
 169  * underlying our virtual address.  Wake up any umtx_sleep()'s
 170  * that were waiting on its physical address to force them to retry.
 171  */
 172 static void
 173 umtx_sleep_page_action_cow(vm_page_t m, vm_page_action_t action)
 174 {
 175     wakeup_domain(action->data, PDOMAIN_UMTX);
 176 }
 177
 178 /*
 179  * umtx_wakeup { const int *ptr, int count }
 180  *
 181  * Wakeup the specified number of processes held in umtx_sleep() on the
 182  * specified user address.  A count of 0 wakes up all waiting processes.
 183  *
 184  * XXX assumes that the physical address space does not exceed the virtual
 185  * address space.
 186  *
 187  * MPALMOSTSAFE
 188  */
 189 int
 190 sys_umtx_wakeup(struct umtx_wakeup_args *uap)
 191 {
 192     vm_page_t m;
 193     int offset;
 194     int error;
 195     void *waddr;
 196
 197     cpu_mfence();
 198     if ((vm_offset_t)uap->ptr & (sizeof(int) - 1))
 199         return (EFAULT);
 200     get_mplock();
 201     m = vm_fault_page_quick((vm_offset_t)uap->ptr, VM_PROT_READ, &error);
 202     if (m == NULL) {
 203         error = EFAULT;
 204         goto done;
 205     }
 206     offset = (vm_offset_t)uap->ptr & PAGE_MASK;
 207     waddr = (void *)((intptr_t)VM_PAGE_TO_PHYS(m) + offset);
 208
 209     if (uap->count == 1) {
 210         wakeup_domain_one(waddr, PDOMAIN_UMTX);
 211     } else {
 212         /* XXX wakes them all up for now */
 213         wakeup_domain(waddr, PDOMAIN_UMTX);
 214     }
 215     vm_page_unhold(m);
 216     error = 0;
 217 done:
 218     rel_mplock();
 219     return(error);
 220 }
 221