sys/kern/kern_ras.c

   1 /*      $NetBSD: kern_ras.c,v 1.33 2008/06/09 11:49:40 ad Exp $ */
   2
   3 /*-
   4  * Copyright (c) 2002, 2006, 2007, 2008 The NetBSD Foundation, Inc.
   5  * All rights reserved.
   6  *
   7  * This code is derived from software contributed to The NetBSD Foundation
   8  * by Gregory McGarry, and by Andrew Doran.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  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 the
  17  *    documentation and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __KERNEL_RCSID(0, "$NetBSD: kern_ras.c,v 1.33 2008/06/09 11:49:40 ad Exp $");
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/kmem.h>
  39 #include <sys/proc.h>
  40 #include <sys/ras.h>
  41 #include <sys/sa.h>
  42 #include <sys/savar.h>
  43 #include <sys/xcall.h>
  44 #include <sys/syscallargs.h>
  45
  46 #include <uvm/uvm_extern.h>
  47
  48 #define MAX_RAS_PER_PROC        16
  49
  50 u_int ras_per_proc = MAX_RAS_PER_PROC;
  51
  52 #ifdef DEBUG
  53 int ras_debug = 0;
  54 #define DPRINTF(x)      if (ras_debug) printf x
  55 #else
  56 #define DPRINTF(x)      /* nothing */
  57 #endif
  58
  59 /*
  60  * Force all CPUs through cpu_switchto(), waiting until complete.
  61  * Context switching will drain the write buffer on the calling
  62  * CPU.
  63  */
  64 static void
  65 ras_sync(void)
  66 {
  67
  68         /* No need to sync if exiting or single threaded. */
  69         if (curproc->p_nlwps > 1 && ncpu > 1) {
  70 #ifdef NO_SOFTWARE_PATENTS
  71                 uint64_t where;
  72                 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
  73                 xc_wait(where);
  74 #else
  75                 /*
  76                  * Assumptions:
  77                  *
  78                  * o preemption is disabled by the thread in
  79                  *   ras_lookup().
  80                  * o proc::p_raslist is only inspected with
  81                  *   preemption disabled.
  82                  * o ras_lookup() plus loads reordered in advance
  83                  *   will take no longer than 1/8s to complete.
  84                  */
  85                 const int delta = hz >> 3;
  86                 int target = hardclock_ticks + delta;
  87                 do {
  88                         kpause("ras", false, delta, NULL);
  89                 } while (hardclock_ticks < target);
  90 #endif
  91         }
  92 }
  93
  94 /*
  95  * Check the specified address to see if it is within the
  96  * sequence.  If it is found, we return the restart address,
  97  * otherwise we return -1.  If we do perform a restart, we
  98  * mark the sequence as hit.
  99  *
 100  * No locking required: we disable preemption and ras_sync()
 101  * guarantees that individual entries are valid while we still
 102  * have visibility of them.
 103  */
 104 void *
 105 ras_lookup(struct proc *p, void *addr)
 106 {
 107         struct ras *rp;
 108         void *startaddr;
 109         lwp_t *l;
 110
 111         startaddr = (void *)-1;
 112         l = curlwp;
 113
 114         KPREEMPT_DISABLE(l);
 115         for (rp = p->p_raslist; rp != NULL; rp = rp->ras_next) {
 116                 if (addr > rp->ras_startaddr && addr < rp->ras_endaddr) {
 117                         startaddr = rp->ras_startaddr;
 118                         DPRINTF(("RAS hit: p=%p %p\n", p, addr));
 119                         break;
 120                 }
 121         }
 122         KPREEMPT_ENABLE(l);
 123
 124         return startaddr;
 125 }
 126
 127 /*
 128  * During a fork, we copy all of the sequences from parent p1 to
 129  * the child p2.
 130  *
 131  * No locking required as the parent must be paused.
 132  */
 133 int
 134 ras_fork(struct proc *p1, struct proc *p2)
 135 {
 136         struct ras *rp, *nrp;
 137
 138         for (rp = p1->p_raslist; rp != NULL; rp = rp->ras_next) {
 139                 nrp = kmem_alloc(sizeof(*nrp), KM_SLEEP);
 140                 nrp->ras_startaddr = rp->ras_startaddr;
 141                 nrp->ras_endaddr = rp->ras_endaddr;
 142                 nrp->ras_next = p2->p_raslist;
 143                 p2->p_raslist = nrp;
 144         }
 145
 146         DPRINTF(("ras_fork: p1=%p, p2=%p\n", p1, p2));
 147
 148         return 0;
 149 }
 150
 151 /*
 152  * Nuke all sequences for this process.
 153  */
 154 int
 155 ras_purgeall(void)
 156 {
 157         struct ras *rp, *nrp;
 158         proc_t *p;
 159
 160         p = curproc;
 161
 162         if (p->p_raslist == NULL)
 163                 return 0;
 164
 165         mutex_enter(&p->p_auxlock);
 166         if ((rp = p->p_raslist) != NULL) {
 167                 p->p_raslist = NULL;
 168                 ras_sync();
 169                 for(; rp != NULL; rp = nrp) {
 170                         nrp = rp->ras_next;
 171                         kmem_free(rp, sizeof(*rp));
 172                 }
 173         }
 174         mutex_exit(&p->p_auxlock);
 175
 176         return 0;
 177 }
 178
 179 #if defined(__HAVE_RAS)
 180
 181 /*
 182  * Install the new sequence.  If it already exists, return
 183  * an error.
 184  */
 185 static int
 186 ras_install(void *addr, size_t len)
 187 {
 188         struct ras *rp;
 189         struct ras *newrp;
 190         void *endaddr;
 191         int nras, error;
 192         proc_t *p;
 193
 194         endaddr = (char *)addr + len;
 195
 196         if (addr < (void *)VM_MIN_ADDRESS ||
 197             endaddr > (void *)VM_MAXUSER_ADDRESS)
 198                 return (EINVAL);
 199
 200         if (len <= 0)
 201                 return (EINVAL);
 202
 203         newrp = kmem_alloc(sizeof(*newrp), KM_SLEEP);
 204         newrp->ras_startaddr = addr;
 205         newrp->ras_endaddr = endaddr;
 206         error = 0;
 207         nras = 0;
 208         p = curproc;
 209
 210         mutex_enter(&p->p_auxlock);
 211         for (rp = p->p_raslist; rp != NULL; rp = rp->ras_next) {
 212                 if (++nras >= ras_per_proc) {
 213                         error = EINVAL;
 214                         break;
 215                 }
 216                 if (addr < rp->ras_endaddr && endaddr > rp->ras_startaddr) {
 217                         error = EEXIST;
 218                         break;
 219                 }
 220         }
 221         if (rp == NULL) {
 222                 newrp->ras_next = p->p_raslist;
 223                 p->p_raslist = newrp;
 224                 ras_sync();
 225                 mutex_exit(&p->p_auxlock);
 226         } else {
 227                 mutex_exit(&p->p_auxlock);
 228                 kmem_free(newrp, sizeof(*newrp));
 229         }
 230
 231         return error;
 232 }
 233
 234 /*
 235  * Nuke the specified sequence.  Both address and len must
 236  * match, otherwise we return an error.
 237  */
 238 static int
 239 ras_purge(void *addr, size_t len)
 240 {
 241         struct ras *rp, **link;
 242         void *endaddr;
 243         proc_t *p;
 244
 245         endaddr = (char *)addr + len;
 246         p = curproc;
 247
 248         mutex_enter(&p->p_auxlock);
 249         link = &p->p_raslist;
 250         for (rp = *link; rp != NULL; link = &rp->ras_next, rp = *link) {
 251                 if (addr == rp->ras_startaddr && endaddr == rp->ras_endaddr)
 252                         break;
 253         }
 254         if (rp != NULL) {
 255                 *link = rp->ras_next;
 256                 ras_sync();
 257                 mutex_exit(&p->p_auxlock);
 258                 kmem_free(rp, sizeof(*rp));
 259                 return 0;
 260         } else {
 261                 mutex_exit(&p->p_auxlock);
 262                 return ESRCH;
 263         }
 264 }
 265
 266 #endif /* defined(__HAVE_RAS) */
 267
 268 /*ARGSUSED*/
 269 int
 270 sys_rasctl(struct lwp *l, const struct sys_rasctl_args *uap, register_t *retval)
 271 {
 272
 273 #if defined(__HAVE_RAS)
 274         /* {
 275                 syscallarg(void *) addr;
 276                 syscallarg(size_t) len;
 277                 syscallarg(int) op;
 278         } */
 279         void *addr;
 280         size_t len;
 281         int op;
 282         int error;
 283
 284         /*
 285          * first, extract syscall args from the uap.
 286          */
 287
 288         addr = (void *)SCARG(uap, addr);
 289         len = (size_t)SCARG(uap, len);
 290         op = SCARG(uap, op);
 291
 292         DPRINTF(("sys_rasctl: p=%p addr=%p, len=%ld, op=0x%x\n",
 293             curproc, addr, (long)len, op));
 294
 295         switch (op) {
 296         case RAS_INSTALL:
 297                 error = ras_install(addr, len);
 298                 break;
 299         case RAS_PURGE:
 300                 error = ras_purge(addr, len);
 301                 break;
 302         case RAS_PURGE_ALL:
 303                 error = ras_purgeall();
 304                 break;
 305         default:
 306                 error = EINVAL;
 307                 break;
 308         }
 309
 310         return (error);
 311
 312 #else
 313
 314         return (EOPNOTSUPP);
 315
 316 #endif
 317
 318 }