Annotation of Net2/vm/vm_glue.c, revision 1.1.1.1

1.1       root        1: /* 
                      2:  * Copyright (c) 1991 Regents of the University of California.
                      3:  * All rights reserved.
                      4:  *
                      5:  * This code is derived from software contributed to Berkeley by
                      6:  * The Mach Operating System project at Carnegie-Mellon University.
                      7:  *
                      8:  * Redistribution and use in source and binary forms, with or without
                      9:  * modification, are permitted provided that the following conditions
                     10:  * are met:
                     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 the
                     15:  *    documentation and/or other materials provided with the distribution.
                     16:  * 3. All advertising materials mentioning features or use of this software
                     17:  *    must display the following acknowledgement:
                     18:  *     This product includes software developed by the University of
                     19:  *     California, Berkeley and its contributors.
                     20:  * 4. Neither the name of the University nor the names of its contributors
                     21:  *    may be used to endorse or promote products derived from this software
                     22:  *    without specific prior written permission.
                     23:  *
                     24:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
                     25:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
                     26:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
                     27:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
                     28:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
                     29:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
                     30:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
                     31:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
                     32:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
                     33:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
                     34:  * SUCH DAMAGE.
                     35:  *
                     36:  *     @(#)vm_glue.c   7.8 (Berkeley) 5/15/91
                     37:  *
                     38:  *
                     39:  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
                     40:  * All rights reserved.
                     41:  * 
                     42:  * Permission to use, copy, modify and distribute this software and
                     43:  * its documentation is hereby granted, provided that both the copyright
                     44:  * notice and this permission notice appear in all copies of the
                     45:  * software, derivative works or modified versions, and any portions
                     46:  * thereof, and that both notices appear in supporting documentation.
                     47:  * 
                     48:  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 
                     49:  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 
                     50:  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
                     51:  * 
                     52:  * Carnegie Mellon requests users of this software to return to
                     53:  *
                     54:  *  Software Distribution Coordinator  or  [email protected]
                     55:  *  School of Computer Science
                     56:  *  Carnegie Mellon University
                     57:  *  Pittsburgh PA 15213-3890
                     58:  *
                     59:  * any improvements or extensions that they make and grant Carnegie the
                     60:  * rights to redistribute these changes.
                     61:  */
                     62: 
                     63: #include "param.h"
                     64: #include "systm.h"
                     65: #include "proc.h"
                     66: #include "resourcevar.h"
                     67: #include "buf.h"
                     68: #include "user.h"
                     69: 
                     70: #include "vm.h"
                     71: #include "vm_page.h"
                     72: #include "vm_kern.h"
                     73: 
                     74: int    avefree = 0;            /* XXX */
                     75: unsigned maxdmap = MAXDSIZ;    /* XXX */
                     76: int    readbuffers = 0;        /* XXX allow kgdb to read kernel buffer pool */
                     77: 
                     78: kernacc(addr, len, rw)
                     79:        caddr_t addr;
                     80:        int len, rw;
                     81: {
                     82:        boolean_t rv;
                     83:        vm_offset_t saddr, eaddr;
                     84:        vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
                     85: 
                     86:        saddr = trunc_page(addr);
                     87:        eaddr = round_page(addr+len-1);
                     88:        rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
                     89:        /*
                     90:         * XXX there are still some things (e.g. the buffer cache) that
                     91:         * are managed behind the VM system's back so even though an
                     92:         * address is accessible in the mind of the VM system, there may
                     93:         * not be physical pages where the VM thinks there is.  This can
                     94:         * lead to bogus allocation of pages in the kernel address space
                     95:         * or worse, inconsistencies at the pmap level.  We only worry
                     96:         * about the buffer cache for now.
                     97:         */
                     98:        if (!readbuffers && rv && (eaddr > (vm_offset_t)buffers &&
                     99:                   saddr < (vm_offset_t)buffers + MAXBSIZE * nbuf))
                    100:                rv = FALSE;
                    101:        return(rv == TRUE);
                    102: }
                    103: 
                    104: useracc(addr, len, rw)
                    105:        caddr_t addr;
                    106:        int len, rw;
                    107: {
                    108:        boolean_t rv;
                    109:        vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
                    110: 
                    111:        rv = vm_map_check_protection(&curproc->p_vmspace->vm_map,
                    112:            trunc_page(addr), round_page(addr+len-1), prot);
                    113:        return(rv == TRUE);
                    114: }
                    115: 
                    116: #ifdef KGDB
                    117: /*
                    118:  * Change protections on kernel pages from addr to addr+len
                    119:  * (presumably so debugger can plant a breakpoint).
                    120:  * All addresses are assumed to reside in the Sysmap,
                    121:  */
                    122: chgkprot(addr, len, rw)
                    123:        register caddr_t addr;
                    124:        int len, rw;
                    125: {
                    126:        vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
                    127: 
                    128:        vm_map_protect(kernel_map, trunc_page(addr),
                    129:                       round_page(addr+len-1), prot, FALSE);
                    130: }
                    131: #endif
                    132: 
                    133: vslock(addr, len)
                    134:        caddr_t addr;
                    135:        u_int   len;
                    136: {
                    137:        vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
                    138:                        round_page(addr+len-1), FALSE);
                    139: }
                    140: 
                    141: vsunlock(addr, len, dirtied)
                    142:        caddr_t addr;
                    143:        u_int   len;
                    144:        int dirtied;
                    145: {
                    146: #ifdef lint
                    147:        dirtied++;
                    148: #endif lint
                    149:        vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
                    150:                        round_page(addr+len-1), TRUE);
                    151: }
                    152: 
                    153: /*
                    154:  * Implement fork's actions on an address space.
                    155:  * Here we arrange for the address space to be copied or referenced,
                    156:  * allocate a user struct (pcb and kernel stack), then call the
                    157:  * machine-dependent layer to fill those in and make the new process
                    158:  * ready to run.
                    159:  * NOTE: the kernel stack may be at a different location in the child
                    160:  * process, and thus addresses of automatic variables may be invalid
                    161:  * after cpu_fork returns in the child process.  We do nothing here
                    162:  * after cpu_fork returns.
                    163:  */
                    164: vm_fork(p1, p2, isvfork)
                    165:        register struct proc *p1, *p2;
                    166:        int isvfork;
                    167: {
                    168:        register struct user *up;
                    169:        vm_offset_t addr;
                    170: 
                    171: #ifdef i386
                    172:        /*
                    173:         * avoid copying any of the parent's pagetables or other per-process
                    174:         * objects that reside in the map by marking all of them non-inheritable
                    175:         */
                    176:        (void)vm_map_inherit(&p1->p_vmspace->vm_map,
                    177:                UPT_MIN_ADDRESS-UPAGES*NBPG, VM_MAX_ADDRESS, VM_INHERIT_NONE);
                    178: #endif
                    179:        p2->p_vmspace = vmspace_fork(p1->p_vmspace);
                    180: 
                    181: #ifdef SYSVSHM
                    182:        if (p1->p_vmspace->vm_shm)
                    183:                shmfork(p1, p2, isvfork);
                    184: #endif
                    185: 
                    186:        /*
                    187:         * Allocate a wired-down (for now) pcb and kernel stack for the process
                    188:         */
                    189:        addr = kmem_alloc_pageable(kernel_map, ctob(UPAGES));
                    190:        vm_map_pageable(kernel_map, addr, addr + ctob(UPAGES), FALSE);
                    191:        up = (struct user *)addr;
                    192:        p2->p_addr = up;
                    193: 
                    194:        /*
                    195:         * p_stats and p_sigacts currently point at fields
                    196:         * in the user struct but not at &u, instead at p_addr.
                    197:         * Copy p_sigacts and parts of p_stats; zero the rest
                    198:         * of p_stats (statistics).
                    199:         */
                    200:        p2->p_stats = &up->u_stats;
                    201:        p2->p_sigacts = &up->u_sigacts;
                    202:        up->u_sigacts = *p1->p_sigacts;
                    203:        bzero(&up->u_stats.pstat_startzero,
                    204:            (unsigned) ((caddr_t)&up->u_stats.pstat_endzero -
                    205:            (caddr_t)&up->u_stats.pstat_startzero));
                    206:        bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
                    207:            ((caddr_t)&up->u_stats.pstat_endcopy -
                    208:             (caddr_t)&up->u_stats.pstat_startcopy));
                    209: 
                    210: #ifdef i386
                    211:        { u_int addr = UPT_MIN_ADDRESS - UPAGES*NBPG; struct vm_map *vp;
                    212: 
                    213:        vp = &p2->p_vmspace->vm_map;
                    214:        (void)vm_map_pageable(vp, addr, 0xfe000000 - addr, TRUE);
                    215:        (void)vm_deallocate(vp, addr, 0xfe000000 - addr);
                    216:        (void)vm_allocate(vp, &addr, UPT_MAX_ADDRESS - addr, FALSE);
                    217:        (void)vm_map_inherit(vp, addr, UPT_MAX_ADDRESS, VM_INHERIT_NONE);
                    218:        }
                    219: #endif
                    220:        /*
                    221:         * cpu_fork will copy and update the kernel stack and pcb,
                    222:         * and make the child ready to run.  It marks the child
                    223:         * so that it can return differently than the parent.
                    224:         * It returns twice, once in the parent process and
                    225:         * once in the child.
                    226:         */
                    227:        return (cpu_fork(p1, p2));
                    228: }
                    229: 
                    230: /*
                    231:  * Set default limits for VM system.
                    232:  * Called for proc 0, and then inherited by all others.
                    233:  */
                    234: vm_init_limits(p)
                    235:        register struct proc *p;
                    236: {
                    237: 
                    238:        /*
                    239:         * Set up the initial limits on process VM.
                    240:         * Set the maximum resident set size to be all
                    241:         * of (reasonably) available memory.  This causes
                    242:         * any single, large process to start random page
                    243:         * replacement once it fills memory.
                    244:         */
                    245:         p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
                    246:         p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
                    247:         p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
                    248:         p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
                    249:        p->p_rlimit[RLIMIT_RSS].rlim_cur = p->p_rlimit[RLIMIT_RSS].rlim_max =
                    250:                ptoa(vm_page_free_count);
                    251: }
                    252: 
                    253: #include "../vm/vm_pageout.h"
                    254: 
                    255: #ifdef DEBUG
                    256: int    enableswap = 1;
                    257: int    swapdebug = 0;
                    258: #define        SDB_FOLLOW      1
                    259: #define SDB_SWAPIN     2
                    260: #define SDB_SWAPOUT    4
                    261: #endif
                    262: 
                    263: /*
                    264:  * Brutally simple:
                    265:  *     1. Attempt to swapin every swaped-out, runnable process in
                    266:  *        order of priority.
                    267:  *     2. If not enough memory, wake the pageout daemon and let it
                    268:  *        clear some space.
                    269:  */
                    270: sched()
                    271: {
                    272:        register struct proc *p;
                    273:        register int pri;
                    274:        struct proc *pp;
                    275:        int ppri;
                    276:        vm_offset_t addr;
                    277:        vm_size_t size;
                    278: 
                    279: loop:
                    280: #ifdef DEBUG
                    281:        if (!enableswap) {
                    282:                pp = NULL;
                    283:                goto noswap;
                    284:        }
                    285: #endif
                    286:        pp = NULL;
                    287:        ppri = INT_MIN;
                    288:        for (p = allproc; p != NULL; p = p->p_nxt)
                    289:                if (p->p_stat == SRUN && (p->p_flag & SLOAD) == 0) {
                    290:                        pri = p->p_time + p->p_slptime - p->p_nice * 8;
                    291:                        if (pri > ppri) {
                    292:                                pp = p;
                    293:                                ppri = pri;
                    294:                        }
                    295:                }
                    296: #ifdef DEBUG
                    297:        if (swapdebug & SDB_FOLLOW)
                    298:                printf("sched: running, procp %x pri %d\n", pp, ppri);
                    299: noswap:
                    300: #endif
                    301:        /*
                    302:         * Nothing to do, back to sleep
                    303:         */
                    304:        if ((p = pp) == NULL) {
                    305:                sleep((caddr_t)&proc0, PVM);
                    306:                goto loop;
                    307:        }
                    308: 
                    309:        /*
                    310:         * We would like to bring someone in.
                    311:         * This part is really bogus cuz we could deadlock on memory
                    312:         * despite our feeble check.
                    313:         */
                    314:        size = round_page(ctob(UPAGES));
                    315:        addr = (vm_offset_t) p->p_addr;
                    316:        if (vm_page_free_count > atop(size)) {
                    317: #ifdef DEBUG
                    318:                if (swapdebug & SDB_SWAPIN)
                    319:                        printf("swapin: pid %d(%s)@%x, pri %d free %d\n",
                    320:                               p->p_pid, p->p_comm, p->p_addr,
                    321:                               ppri, vm_page_free_count);
                    322: #endif
                    323:                vm_map_pageable(kernel_map, addr, addr+size, FALSE);
                    324:                (void) splclock();
                    325:                if (p->p_stat == SRUN)
                    326:                        setrq(p);
                    327:                p->p_flag |= SLOAD;
                    328:                (void) spl0();
                    329:                p->p_time = 0;
                    330:                goto loop;
                    331:        }
                    332:        /*
                    333:         * Not enough memory, jab the pageout daemon and wait til the
                    334:         * coast is clear.
                    335:         */
                    336: #ifdef DEBUG
                    337:        if (swapdebug & SDB_FOLLOW)
                    338:                printf("sched: no room for pid %d(%s), free %d\n",
                    339:                       p->p_pid, p->p_comm, vm_page_free_count);
                    340: #endif
                    341:        (void) splhigh();
                    342:        VM_WAIT;
                    343:        (void) spl0();
                    344: #ifdef DEBUG
                    345:        if (swapdebug & SDB_FOLLOW)
                    346:                printf("sched: room again, free %d\n", vm_page_free_count);
                    347: #endif
                    348:        goto loop;
                    349: }
                    350: 
                    351: #define        swappable(p) \
                    352:        (((p)->p_flag & (SSYS|SLOAD|SKEEP|SWEXIT|SPHYSIO)) == SLOAD)
                    353: 
                    354: /*
                    355:  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
                    356:  * procs and unwire their u-areas.  We try to always "swap" at least one
                    357:  * process in case we need the room for a swapin.
                    358:  * If any procs have been sleeping/stopped for at least maxslp seconds,
                    359:  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
                    360:  * if any, otherwise the longest-resident process.
                    361:  */
                    362: swapout_threads()
                    363: {
                    364:        register struct proc *p;
                    365:        struct proc *outp, *outp2;
                    366:        int outpri, outpri2;
                    367:        int didswap = 0;
                    368:        extern int maxslp;
                    369: 
                    370: #ifdef DEBUG
                    371:        if (!enableswap)
                    372:                return;
                    373: #endif
                    374:        outp = outp2 = NULL;
                    375:        outpri = outpri2 = 0;
                    376:        for (p = allproc; p != NULL; p = p->p_nxt) {
                    377:                if (!swappable(p))
                    378:                        continue;
                    379:                switch (p->p_stat) {
                    380:                case SRUN:
                    381:                        if (p->p_time > outpri2) {
                    382:                                outp2 = p;
                    383:                                outpri2 = p->p_time;
                    384:                        }
                    385:                        continue;
                    386:                        
                    387:                case SSLEEP:
                    388:                case SSTOP:
                    389:                        if (p->p_slptime > maxslp) {
                    390:                                swapout(p);
                    391:                                didswap++;
                    392:                        } else if (p->p_slptime > outpri) {
                    393:                                outp = p;
                    394:                                outpri = p->p_slptime;
                    395:                        }
                    396:                        continue;
                    397:                }
                    398:        }
                    399:        /*
                    400:         * If we didn't get rid of any real duds, toss out the next most
                    401:         * likely sleeping/stopped or running candidate.  We only do this
                    402:         * if we are real low on memory since we don't gain much by doing
                    403:         * it (UPAGES pages).
                    404:         */
                    405:        if (didswap == 0 &&
                    406:            vm_page_free_count <= atop(round_page(ctob(UPAGES)))) {
                    407:                if ((p = outp) == 0)
                    408:                        p = outp2;
                    409: #ifdef DEBUG
                    410:                if (swapdebug & SDB_SWAPOUT)
                    411:                        printf("swapout_threads: no duds, try procp %x\n", p);
                    412: #endif
                    413:                if (p)
                    414:                        swapout(p);
                    415:        }
                    416: }
                    417: 
                    418: swapout(p)
                    419:        register struct proc *p;
                    420: {
                    421:        vm_offset_t addr;
                    422:        vm_size_t size;
                    423: 
                    424: #ifdef DEBUG
                    425:        if (swapdebug & SDB_SWAPOUT)
                    426:                printf("swapout: pid %d(%s)@%x, stat %x pri %d free %d\n",
                    427:                       p->p_pid, p->p_comm, p->p_addr, p->p_stat,
                    428:                       p->p_slptime, vm_page_free_count);
                    429: #endif
                    430:        size = round_page(ctob(UPAGES));
                    431:        addr = (vm_offset_t) p->p_addr;
                    432: #ifdef hp300
                    433:        /*
                    434:         * Ugh!  u-area is double mapped to a fixed address behind the
                    435:         * back of the VM system and accesses are usually through that
                    436:         * address rather than the per-process address.  Hence reference
                    437:         * and modify information are recorded at the fixed address and
                    438:         * lost at context switch time.  We assume the u-struct and
                    439:         * kernel stack are always accessed/modified and force it to be so.
                    440:         */
                    441:        {
                    442:                register int i;
                    443:                volatile long tmp;
                    444: 
                    445:                for (i = 0; i < UPAGES; i++) {
                    446:                        tmp = *(long *)addr; *(long *)addr = tmp;
                    447:                        addr += NBPG;
                    448:                }
                    449:                addr = (vm_offset_t) p->p_addr;
                    450:        }
                    451: #endif
                    452:        vm_map_pageable(kernel_map, addr, addr+size, TRUE);
                    453:        pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
                    454:        (void) splhigh();
                    455:        p->p_flag &= ~SLOAD;
                    456:        if (p->p_stat == SRUN)
                    457:                remrq(p);
                    458:        (void) spl0();
                    459:        p->p_time = 0;
                    460: }
                    461: 
                    462: /*
                    463:  * The rest of these routines fake thread handling
                    464:  */
                    465: 
                    466: void
                    467: assert_wait(event, ruptible)
                    468:        int event;
                    469:        boolean_t ruptible;
                    470: {
                    471: #ifdef lint
                    472:        ruptible++;
                    473: #endif
                    474:        curproc->p_thread = event;
                    475: }
                    476: 
                    477: void
                    478: thread_block()
                    479: {
                    480:        int s = splhigh();
                    481: 
                    482:        if (curproc->p_thread)
                    483:                sleep((caddr_t)curproc->p_thread, PVM);
                    484:        splx(s);
                    485: }
                    486: 
                    487: thread_sleep(event, lock, ruptible)
                    488:        int event;
                    489:        simple_lock_t lock;
                    490:        boolean_t ruptible;
                    491: {
                    492: #ifdef lint
                    493:        ruptible++;
                    494: #endif
                    495:        int s = splhigh();
                    496: 
                    497:        curproc->p_thread = event;
                    498:        simple_unlock(lock);
                    499:        if (curproc->p_thread)
                    500:                sleep((caddr_t)event, PVM);
                    501:        splx(s);
                    502: }
                    503: 
                    504: thread_wakeup(event)
                    505:        int event;
                    506: {
                    507:        int s = splhigh();
                    508: 
                    509:        wakeup((caddr_t)event);
                    510:        splx(s);
                    511: }
                    512: 
                    513: /*
                    514:  * DEBUG stuff
                    515:  */
                    516: 
                    517: int indent = 0;
                    518: 
                    519: /*ARGSUSED2*/
                    520: iprintf(a, b, c, d, e, f, g, h)
                    521:        char *a;
                    522: {
                    523:        register int i;
                    524: 
                    525:        i = indent;
                    526:        while (i >= 8) {
                    527:                printf("\t");
                    528:                i -= 8;
                    529:        }
                    530:        for (; i > 0; --i)
                    531:                printf(" ");
                    532:        printf(a, b, c, d, e, f, g, h);
                    533: }

unix.superglobalmegacorp.com

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.