sys/kern/kern_usched.c

   1 /*
   2  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Sergey Glushchenko <deen@smz.com.ua>
   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  */
  35
  36 #include <sys/errno.h>
  37 #include <sys/globaldata.h>             /* curthread */
  38 #include <sys/proc.h>
  39 #include <sys/priv.h>
  40 #include <sys/sysproto.h>               /* struct usched_set_args */
  41 #include <sys/systm.h>                  /* strcmp() */
  42 #include <sys/usched.h>
  43
  44 #include <sys/mplock2.h>
  45
  46 #include <machine/cpumask.h>
  47 #include <machine/smp.h>
  48
  49 static TAILQ_HEAD(, usched) usched_list = TAILQ_HEAD_INITIALIZER(usched_list);
  50
  51 cpumask_t usched_mastermask = CPUMASK_INITIALIZER_ALLONES;
  52
  53 /*
  54  * Called from very low level boot code, i386/i386/machdep.c/init386().
  55  * We cannot do anything fancy.  no malloc's, no nothing other then
  56  * static initialization.
  57  */
  58 struct usched *
  59 usched_init(void)
  60 {
  61         const char *defsched;
  62
  63         defsched = kgetenv("kern.user_scheduler");
  64
  65         /*
  66          * Add various userland schedulers to the system.
  67          */
  68         usched_ctl(&usched_bsd4, USCH_ADD);
  69         usched_ctl(&usched_dfly, USCH_ADD);
  70         usched_ctl(&usched_dummy, USCH_ADD);
  71         if (defsched == NULL )
  72                 return(&usched_dfly);
  73         if (strcmp(defsched, "bsd4") == 0)
  74                 return(&usched_bsd4);
  75         if (strcmp(defsched, "dfly") == 0)
  76                 return(&usched_dfly);
  77         kprintf("WARNING: Running dummy userland scheduler\n");
  78         return(&usched_dummy);
  79 }
  80
  81 /*
  82  * USCHED_CTL
  83  *
  84  * SYNOPSIS:
  85  *      Add/remove usched to/from list.
  86  *
  87  * ARGUMENTS:
  88  *      usched - pointer to target scheduler
  89  *      action - addition or removal ?
  90  *
  91  * RETURN VALUES:
  92  *      0 - success
  93  *      EINVAL - error
  94  */
  95 int
  96 usched_ctl(struct usched *usched, int action)
  97 {
  98         struct usched *item;    /* temporaly for TAILQ processing */
  99         int error = 0;
 100
 101         switch(action) {
 102         case USCH_ADD:
 103                 /*
 104                  * Make sure it isn't already on the list
 105                  */
 106 #ifdef INVARIANTS
 107                 TAILQ_FOREACH(item, &usched_list, entry) {
 108                         KKASSERT(item != usched);
 109                 }
 110 #endif
 111                 /*
 112                  * Optional callback to the scheduler before we officially
 113                  * add it to the list.
 114                  */
 115                 if (usched->usched_register)
 116                         usched->usched_register();
 117                 TAILQ_INSERT_TAIL(&usched_list, usched, entry);
 118                 break;
 119         case USCH_REM:
 120                 /*
 121                  * Do not allow the default scheduler to be removed
 122                  */
 123                 if (strcmp(usched->name, "bsd4") == 0) {
 124                         error = EINVAL;
 125                         break;
 126                 }
 127                 TAILQ_FOREACH(item, &usched_list, entry) {
 128                         if (item == usched)
 129                                 break;
 130                 }
 131                 if (item) {
 132                         if (item->usched_unregister)
 133                                 item->usched_unregister();
 134                         TAILQ_REMOVE(&usched_list, item, entry);
 135                 } else {
 136                         error = EINVAL;
 137                 }
 138                 break;
 139         default:
 140                 error = EINVAL;
 141                 break;
 142         }
 143         return (error);
 144 }
 145
 146 /*
 147  * Called from the scheduler clock on each cpu independently at the
 148  * common scheduling rate.  If th scheduler clock interrupted a running
 149  * lwp the lp will be non-NULL.
 150  */
 151 void
 152 usched_schedulerclock(struct lwp *lp, sysclock_t periodic, sysclock_t time)
 153 {
 154         struct usched *item;
 155
 156         TAILQ_FOREACH(item, &usched_list, entry) {
 157                 if (lp && lp->lwp_proc->p_usched == item)
 158                         item->schedulerclock(lp, periodic, time);
 159                 else
 160                         item->schedulerclock(NULL, periodic, time);
 161         }
 162 }
 163
 164 /*
 165  * USCHED_SET(syscall)
 166  *
 167  * SYNOPSIS:
 168  *      Setting up a proc's usched.
 169  *
 170  * ARGUMENTS:
 171  *      pid     -
 172  *      cmd     -
 173  *      data    -
 174  *      bytes   -
 175  * RETURN VALUES:
 176  *      0 - success
 177  *      EFBIG  - error (invalid cpu#)
 178  *      EPERM  - error (failed to delete cpu#)
 179  *      EINVAL - error (other reasons)
 180  *
 181  * MPALMOSTSAFE
 182  */
 183 int
 184 sys_usched_set(struct usched_set_args *uap)
 185 {
 186         struct proc *p = curthread->td_proc;
 187         struct usched *item;    /* temporaly for TAILQ processing */
 188         int error;
 189         char buffer[NAME_LENGTH];
 190         cpumask_t mask;
 191         struct lwp *lp;
 192         int cpuid;
 193
 194         if (uap->pid != 0 && uap->pid != curthread->td_proc->p_pid)
 195                 return (EINVAL);
 196
 197         lp = curthread->td_lwp;
 198         get_mplock();
 199
 200         switch (uap->cmd) {
 201         case USCHED_SET_SCHEDULER:
 202                 if ((error = priv_check(curthread, PRIV_SCHED_SET)) != 0)
 203                         break;
 204                 error = copyinstr(uap->data, buffer, sizeof(buffer), NULL);
 205                 if (error)
 206                         break;
 207                 TAILQ_FOREACH(item, &usched_list, entry) {
 208                         if ((strcmp(item->name, buffer) == 0))
 209                                 break;
 210                 }
 211
 212                 /*
 213                  * If the scheduler for a process is being changed, disassociate
 214                  * the old scheduler before switching to the new one.
 215                  *
 216                  * XXX we might have to add an additional ABI call to do a 'full
 217                  * disassociation' and another ABI call to do a 'full
 218                  * reassociation'
 219                  */
 220                 /* XXX lwp have to deal with multiple lwps here */
 221                 if (p->p_nthreads != 1) {
 222                         error = EINVAL;
 223                         break;
 224                 }
 225                 if (item && item != p->p_usched) {
 226                         /* XXX lwp */
 227                         p->p_usched->release_curproc(ONLY_LWP_IN_PROC(p));
 228                         p->p_usched->heuristic_exiting(ONLY_LWP_IN_PROC(p), p);
 229                         p->p_usched = item;
 230                 } else if (item == NULL) {
 231                         error = EINVAL;
 232                 }
 233                 break;
 234         case USCHED_SET_CPU:
 235                 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
 236                         break;
 237                 if (uap->bytes != sizeof(int)) {
 238                         error = EINVAL;
 239                         break;
 240                 }
 241                 error = copyin(uap->data, &cpuid, sizeof(int));
 242                 if (error)
 243                         break;
 244                 if (cpuid < 0 || cpuid >= ncpus) {
 245                         error = EFBIG;
 246                         break;
 247                 }
 248                 if (CPUMASK_TESTBIT(smp_active_mask, cpuid) == 0) {
 249                         error = EINVAL;
 250                         break;
 251                 }
 252                 CPUMASK_ASSBIT(lp->lwp_cpumask, cpuid);
 253                 if (cpuid != mycpu->gd_cpuid) {
 254                         lwkt_migratecpu(cpuid);
 255                         p->p_usched->changedcpu(lp);
 256                 }
 257                 break;
 258         case USCHED_GET_CPU:
 259                 /* USCHED_GET_CPU doesn't require special privileges. */
 260                 if (uap->bytes != sizeof(int)) {
 261                         error = EINVAL;
 262                         break;
 263                 }
 264                 error = copyout(&(mycpu->gd_cpuid), uap->data, sizeof(int));
 265                 break;
 266         case USCHED_GET_CPUMASK:
 267                 /* USCHED_GET_CPUMASK doesn't require special privileges. */
 268                 if (uap->bytes != sizeof(cpumask_t)) {
 269                         error = EINVAL;
 270                         break;
 271                 }
 272                 error = copyout(&lp->lwp_cpumask, uap->data, sizeof(cpumask_t));
 273                 break;
 274         case USCHED_ADD_CPU:
 275                 if ((error = priv_check(curthread, PRIV_SCHED_CPUSET)) != 0)
 276                         break;
 277                 if (uap->bytes != sizeof(int)) {
 278                         error = EINVAL;
 279                         break;
 280                 }
 281                 error = copyin(uap->data, &cpuid, sizeof(int));
 282                 if (error)
 283                         break;
 284                 if (cpuid < 0 || cpuid >= ncpus) {
 285                         error = EFBIG;
 286                         break;
 287                 }
 288                 if (CPUMASK_TESTBIT(smp_active_mask, cpuid) == 0) {
 289                         error = EINVAL;
 290                         break;
 291                 }
 292                 CPUMASK_ORBIT(lp->lwp_cpumask, cpuid);
 293                 break;
 294         case USCHED_DEL_CPU:
 295                 /* USCHED_DEL_CPU doesn't require special privileges. */
 296                 if (uap->bytes != sizeof(int)) {
 297                         error = EINVAL;
 298                         break;
 299                 }
 300                 error = copyin(uap->data, &cpuid, sizeof(int));
 301                 if (error)
 302                         break;
 303                 if (cpuid < 0 || cpuid >= ncpus) {
 304                         error = EFBIG;
 305                         break;
 306                 }
 307                 lp = curthread->td_lwp;
 308                 mask = lp->lwp_cpumask;
 309                 CPUMASK_ANDMASK(mask, smp_active_mask);
 310                 CPUMASK_NANDBIT(mask, cpuid);
 311                 if (CPUMASK_TESTZERO(mask)) {
 312                         error = EPERM;
 313                 } else {
 314                         CPUMASK_NANDBIT(lp->lwp_cpumask, cpuid);
 315                         if (CPUMASK_TESTMASK(lp->lwp_cpumask,
 316                                             mycpu->gd_cpumask) == 0) {
 317                                 mask = lp->lwp_cpumask;
 318                                 CPUMASK_ANDMASK(mask, smp_active_mask);
 319                                 cpuid = BSFCPUMASK(mask);
 320                                 lwkt_migratecpu(cpuid);
 321                                 p->p_usched->changedcpu(lp);
 322                         }
 323                 }
 324                 break;
 325         default:
 326                 error = EINVAL;
 327                 break;
 328         }
 329         rel_mplock();
 330         return (error);
 331 }
 332