Index: head/sys/vm/vm_glue.c =================================================================== --- head/sys/vm/vm_glue.c (revision 2691) +++ head/sys/vm/vm_glue.c (revision 2692) @@ -1,683 +1,691 @@ /* * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_glue.c,v 1.5 1994/08/09 10:42:41 davidg Exp $ + * $Id: vm_glue.c,v 1.6 1994/08/18 22:36:01 wollman Exp $ */ #include #include #include #include #include #include #include #include #include #include #include #include #include extern char kstack[]; int avefree = 0; /* XXX */ int readbuffers = 0; /* XXX allow kgdb to read kernel buffer pool */ /* vm_map_t upages_map; */ int kernacc(addr, len, rw) caddr_t addr; int len, rw; { boolean_t rv; vm_offset_t saddr, eaddr; vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; saddr = trunc_page(addr); eaddr = round_page(addr+len); rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); return(rv == TRUE); } int useracc(addr, len, rw) caddr_t addr; int len, rw; { boolean_t rv; vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; /* * XXX - check separately to disallow access to user area and user * page tables - they are in the map. * * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was * once only used (as an end address) in trap.c. Use it as an end * address here too. This bogusness has spread. I just fixed * where it was used as a max in vm_mmap.c. */ if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS || (vm_offset_t) addr + len < (vm_offset_t) addr) { return (FALSE); } rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, trunc_page(addr), round_page(addr+len), prot); return(rv == TRUE); } #ifdef KGDB /* * Change protections on kernel pages from addr to addr+len * (presumably so debugger can plant a breakpoint). * All addresses are assumed to reside in the Sysmap, */ chgkprot(addr, len, rw) register caddr_t addr; int len, rw; { vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; vm_map_protect(kernel_map, trunc_page(addr), round_page(addr+len), prot, FALSE); } #endif void vslock(addr, len) caddr_t addr; u_int len; { vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), round_page(addr+len), FALSE); } void vsunlock(addr, len, dirtied) caddr_t addr; u_int len; int dirtied; { #ifdef lint dirtied++; #endif lint vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), round_page(addr+len), TRUE); } /* * Implement fork's actions on an address space. * Here we arrange for the address space to be copied or referenced, * allocate a user struct (pcb and kernel stack), then call the * machine-dependent layer to fill those in and make the new process * ready to run. * NOTE: the kernel stack may be at a different location in the child * process, and thus addresses of automatic variables may be invalid * after cpu_fork returns in the child process. We do nothing here * after cpu_fork returns. */ int vm_fork(p1, p2, isvfork) register struct proc *p1, *p2; int isvfork; { register struct user *up; vm_offset_t addr, ptaddr; int i; struct vm_map *vp; while( cnt.v_free_count < cnt.v_free_min) VM_WAIT; /* * avoid copying any of the parent's pagetables or other per-process * objects that reside in the map by marking all of them non-inheritable */ (void)vm_map_inherit(&p1->p_vmspace->vm_map, UPT_MIN_ADDRESS - UPAGES * NBPG, VM_MAX_ADDRESS, VM_INHERIT_NONE); p2->p_vmspace = vmspace_fork(p1->p_vmspace); #ifdef SYSVSHM if (p1->p_vmspace->vm_shm) shmfork(p1, p2, isvfork); #endif /* * Allocate a wired-down (for now) pcb and kernel stack for the process */ addr = (vm_offset_t) kstack; vp = &p2->p_vmspace->vm_map; /* ream out old pagetables and kernel stack */ (void)vm_deallocate(vp, addr, UPT_MAX_ADDRESS - addr); /* get new pagetables and kernel stack */ (void)vm_allocate(vp, &addr, UPT_MAX_ADDRESS - addr, FALSE); /* force in the page table encompassing the UPAGES */ ptaddr = trunc_page((u_int)vtopte(addr)); vm_map_pageable(vp, ptaddr, ptaddr + NBPG, FALSE); /* and force in (demand-zero) the UPAGES */ vm_map_pageable(vp, addr, addr + UPAGES * NBPG, FALSE); /* get a kernel virtual address for the UPAGES for this proc */ up = (struct user *)kmem_alloc_pageable(kernel_map, UPAGES * NBPG); /* and force-map the upages into the kernel pmap */ for (i = 0; i < UPAGES; i++) pmap_enter(vm_map_pmap(kernel_map), ((vm_offset_t) up) + NBPG * i, pmap_extract(vp->pmap, addr + NBPG * i), VM_PROT_READ|VM_PROT_WRITE, 1); /* and allow the UPAGES page table entry to be paged (at the vm system level) */ vm_map_pageable(vp, ptaddr, ptaddr + NBPG, TRUE); p2->p_addr = up; /* * p_stats and p_sigacts currently point at fields * in the user struct but not at &u, instead at p_addr. * Copy p_sigacts and parts of p_stats; zero the rest * of p_stats (statistics). */ p2->p_stats = &up->u_stats; p2->p_sigacts = &up->u_sigacts; up->u_sigacts = *p1->p_sigacts; bzero(&up->u_stats.pstat_startzero, (unsigned) ((caddr_t)&up->u_stats.pstat_endzero - (caddr_t)&up->u_stats.pstat_startzero)); bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, ((caddr_t)&up->u_stats.pstat_endcopy - (caddr_t)&up->u_stats.pstat_startcopy)); /* * cpu_fork will copy and update the kernel stack and pcb, * and make the child ready to run. It marks the child * so that it can return differently than the parent. * It returns twice, once in the parent process and * once in the child. */ return (cpu_fork(p1, p2)); } /* * Set default limits for VM system. * Called for proc 0, and then inherited by all others. */ void vm_init_limits(p) register struct proc *p; { int rss_limit; /* * Set up the initial limits on process VM. * Set the maximum resident set size to be half * of (reasonably) available memory. Since this * is a soft limit, it comes into effect only * when the system is out of memory - half of * main memory helps to favor smaller processes, * and reduces thrashing of the object cache. */ p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; /* limit the limit to no less than 128K */ rss_limit = max(cnt.v_free_count / 2, 32); p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; } #ifdef DEBUG int enableswap = 1; int swapdebug = 0; #define SDB_FOLLOW 1 #define SDB_SWAPIN 2 #define SDB_SWAPOUT 4 #endif void faultin(p) struct proc *p; { vm_offset_t i; vm_offset_t vaddr, ptaddr; vm_offset_t v, v1; struct user *up; int s; int opflag; if ((p->p_flag & P_INMEM) == 0) { int rv0, rv1; vm_map_t map; ++p->p_lock; map = &p->p_vmspace->vm_map; /* force the page table encompassing the kernel stack (upages) */ ptaddr = trunc_page((u_int)vtopte(kstack)); vm_map_pageable(map, ptaddr, ptaddr + NBPG, FALSE); /* wire in the UPAGES */ vm_map_pageable(map, (vm_offset_t) kstack, (vm_offset_t) kstack + UPAGES * NBPG, FALSE); /* and map them nicely into the kernel pmap */ for (i = 0; i < UPAGES; i++) { vm_offset_t off = i * NBPG; vm_offset_t pa = (vm_offset_t) pmap_extract(&p->p_vmspace->vm_pmap, (vm_offset_t) kstack + off); pmap_enter(vm_map_pmap(kernel_map), ((vm_offset_t)p->p_addr) + off, pa, VM_PROT_READ|VM_PROT_WRITE, 1); } /* and let the page table pages go (at least above pmap level) */ vm_map_pageable(map, ptaddr, ptaddr + NBPG, TRUE); s = splhigh(); if (p->p_stat == SRUN) setrunqueue(p); p->p_flag |= P_INMEM; /* undo the effect of setting SLOCK above */ --p->p_lock; splx(s); } } int swapinreq; int percentactive; /* * This swapin algorithm attempts to swap-in processes only if there * is enough space for them. Of course, if a process waits for a long * time, it will be swapped in anyway. */ void scheduler() { register struct proc *p; register int pri; struct proc *pp; int ppri; vm_offset_t addr; int lastidle, lastrun; int curidle, currun; int forceload; int percent; int ntries; lastidle = 0; lastrun = 0; loop: ntries = 0; curidle = cp_time[CP_IDLE]; currun = cp_time[CP_USER] + cp_time[CP_SYS] + cp_time[CP_NICE]; percent = (100*(currun-lastrun)) / ( 1 + (currun-lastrun) + (curidle-lastidle)); lastrun = currun; lastidle = curidle; if( percent > 100) percent = 100; percentactive = percent; if( percentactive < 25) forceload = 1; else forceload = 0; loop1: pp = NULL; ppri = INT_MIN; for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { if (p->p_stat == SRUN && (p->p_flag & P_INMEM) == 0) { int mempri; pri = p->p_swtime + p->p_slptime - p->p_nice * 8; mempri = pri > 0 ? pri : 0; /* * if this process is higher priority and there is * enough space, then select this process instead * of the previous selection. */ if (pri > ppri && (((cnt.v_free_count + (mempri * (4*PAGE_SIZE) / PAGE_SIZE) >= (p->p_vmspace->vm_swrss)) || (ntries > 0 && forceload)))) { pp = p; ppri = pri; } } } if ((pp == NULL) && (ntries == 0) && forceload) { ++ntries; goto loop1; } /* * Nothing to do, back to sleep */ if ((p = pp) == NULL) { tsleep((caddr_t)&proc0, PVM, "sched", 0); goto loop; } /* * We would like to bring someone in. (only if there is space). */ /* printf("swapin: %d, free: %d, res: %d, min: %d\n", p->p_pid, cnt.v_free_count, cnt.v_free_reserved, cnt.v_free_min); */ (void) splhigh(); if ((forceload && (cnt.v_free_count > (cnt.v_free_reserved + UPAGES + 1))) || (cnt.v_free_count >= cnt.v_free_min)) { spl0(); faultin(p); p->p_swtime = 0; goto loop; } /* * log the memory shortage */ swapinreq += p->p_vmspace->vm_swrss; /* * Not enough memory, jab the pageout daemon and wait til the * coast is clear. */ if( cnt.v_free_count < cnt.v_free_min) { VM_WAIT; } else { tsleep((caddr_t)&proc0, PVM, "sched", 0); } (void) spl0(); goto loop; } #define swappable(p) \ (((p)->p_lock == 0) && \ ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO)) == P_INMEM) extern int vm_pageout_free_min; /* * Swapout is driven by the pageout daemon. Very simple, we find eligible * procs and unwire their u-areas. We try to always "swap" at least one * process in case we need the room for a swapin. * If any procs have been sleeping/stopped for at least maxslp seconds, * they are swapped. Else, we swap the longest-sleeping or stopped process, * if any, otherwise the longest-resident process. */ void swapout_threads() { register struct proc *p; struct proc *outp, *outp2; int outpri, outpri2; int tpri; int didswap = 0; int swapneeded = swapinreq; extern int maxslp; int runnablenow; int s; swapmore: runnablenow = 0; outp = outp2 = NULL; outpri = outpri2 = INT_MIN; for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { if (!swappable(p)) continue; switch (p->p_stat) { case SRUN: ++runnablenow; /* * count the process as being in a runnable state */ if ((tpri = p->p_swtime + p->p_nice * 8) > outpri2) { outp2 = p; outpri2 = tpri; } continue; case SSLEEP: case SSTOP: /* * do not swapout a process that is waiting for VM datastructures * there is a possible deadlock. */ if (!lock_try_write( &p->p_vmspace->vm_map.lock)) { continue; } vm_map_unlock( &p->p_vmspace->vm_map); - if (p->p_slptime > maxslp) { + /* + * If the process has been asleep for awhile and had most + * of its pages taken away already, swap it out. + */ + if ((p->p_slptime > maxslp) && (p->p_vmspace->vm_pmap.pm_stats.resident_count <= 6)) { swapout(p); didswap++; } else if ((tpri = p->p_slptime + p->p_nice * 8) > outpri) { outp = p; outpri = tpri ; } continue; } } /* * We swapout only if there are more than two runnable processes or if * another process needs some space to swapin. */ if ((swapinreq || ((percentactive > 90) && (runnablenow > 2))) && (((cnt.v_free_count + cnt.v_inactive_count) <= (cnt.v_free_target + cnt.v_inactive_target)) || (cnt.v_free_count < cnt.v_free_min))) { if ((p = outp) == 0) { p = outp2; } - if (p) { + /* + * Only swapout processes that have already had most + * of their pages taken away. + */ + if (p && (p->p_vmspace->vm_pmap.pm_stats.resident_count <= 6)) { swapout(p); didswap = 1; } } /* * if we previously had found a process to swapout, and we need to swapout * more then try again. */ #if 0 if( p && swapinreq) goto swapmore; #endif /* * If we swapped something out, and another process needed memory, * then wakeup the sched process. */ if (didswap) { if (swapneeded) wakeup((caddr_t)&proc0); swapinreq = 0; } } void swapout(p) register struct proc *p; { vm_offset_t addr; struct pmap *pmap = &p->p_vmspace->vm_pmap; vm_map_t map = &p->p_vmspace->vm_map; vm_offset_t ptaddr; int i; ++p->p_stats->p_ru.ru_nswap; /* * remember the process resident count */ p->p_vmspace->vm_swrss = p->p_vmspace->vm_pmap.pm_stats.resident_count; /* * and decrement the amount of needed space */ swapinreq -= min(swapinreq, p->p_vmspace->vm_pmap.pm_stats.resident_count); (void) splhigh(); p->p_flag &= ~P_INMEM; if (p->p_stat == SRUN) remrq(p); (void) spl0(); ++p->p_lock; /* let the upages be paged */ pmap_remove(vm_map_pmap(kernel_map), (vm_offset_t) p->p_addr, ((vm_offset_t) p->p_addr) + UPAGES * NBPG); vm_map_pageable(map, (vm_offset_t) kstack, (vm_offset_t) kstack + UPAGES * NBPG, TRUE); --p->p_lock; p->p_swtime = 0; } /* * The rest of these routines fake thread handling */ #ifndef assert_wait void assert_wait(event, ruptible) int event; boolean_t ruptible; { #ifdef lint ruptible++; #endif curproc->p_thread = event; } #endif void thread_block(char *msg) { if (curproc->p_thread) tsleep((caddr_t)curproc->p_thread, PVM, msg, 0); } void thread_sleep_(event, lock, wmesg) int event; simple_lock_t lock; char *wmesg; { curproc->p_thread = event; simple_unlock(lock); if (curproc->p_thread) { tsleep((caddr_t)event, PVM, wmesg, 0); } } #ifndef thread_wakeup void thread_wakeup(event) int event; { wakeup((caddr_t)event); } #endif /* * DEBUG stuff */ int indent = 0; #include /* see subr_prf.c */ /*ARGSUSED2*/ void #if __STDC__ iprintf(const char *fmt, ...) #else iprintf(fmt /* , va_alist */) char *fmt; /* va_dcl */ #endif { register int i; va_list ap; for (i = indent; i >= 8; i -= 8) printf("\t"); while (--i >= 0) printf(" "); va_start(ap, fmt); printf("%r", fmt, ap); va_end(ap); } Index: head/sys/vm/vm_pageout.c =================================================================== --- head/sys/vm/vm_pageout.c (revision 2691) +++ head/sys/vm/vm_pageout.c (revision 2692) @@ -1,864 +1,863 @@ /* * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1994 John S. Dyson * All rights reserved. * Copyright (c) 1994 David Greenman * All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_pageout.c 7.4 (Berkeley) 5/7/91 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_pageout.c,v 1.10 1994/09/06 11:28:46 davidg Exp $ + * $Id: vm_pageout.c,v 1.11 1994/09/12 11:31:36 davidg Exp $ */ /* * The proverbial page-out daemon. */ #include #include #include #include #include #include #include #include extern vm_map_t kmem_map; int vm_pages_needed; /* Event on which pageout daemon sleeps */ int vm_pagescanner; /* Event on which pagescanner sleeps */ int vm_pageout_free_min = 0; /* Stop pageout to wait for pagers at this free level */ int vm_pageout_pages_needed = 0; /* flag saying that the pageout daemon needs pages */ int vm_page_pagesfreed; int vm_desired_cache_size; extern int npendingio; extern int hz; int vm_pageout_proc_limit; extern int nswiodone; extern int swap_pager_full; extern int swap_pager_ready(); #define MAXREF 32767 #define MAXSCAN 512 /* maximum number of pages to scan in active queue */ /* set the "clock" hands to be (MAXSCAN * 4096) Bytes */ #define ACT_DECLINE 1 #define ACT_ADVANCE 3 #define ACT_MAX 100 #define LOWATER ((2048*1024)/NBPG) #define VM_PAGEOUT_PAGE_COUNT 8 int vm_pageout_page_count = VM_PAGEOUT_PAGE_COUNT; static vm_offset_t vm_space_needed; int vm_pageout_req_do_stats; int vm_page_max_wired = 0; /* XXX max # of wired pages system-wide */ /* * vm_pageout_clean: * cleans a vm_page */ int vm_pageout_clean(m, sync) register vm_page_t m; int sync; { /* * Clean the page and remove it from the * laundry. * * We set the busy bit to cause * potential page faults on this page to * block. * * And we set pageout-in-progress to keep * the object from disappearing during * pageout. This guarantees that the * page won't move from the inactive * queue. (However, any other page on * the inactive queue may move!) */ register vm_object_t object; register vm_pager_t pager; int pageout_status[VM_PAGEOUT_PAGE_COUNT]; vm_page_t ms[VM_PAGEOUT_PAGE_COUNT]; int pageout_count; int anyok=0; int i; vm_offset_t offset = m->offset; object = m->object; if (!object) { printf("pager: object missing\n"); return 0; } /* * Try to collapse the object before * making a pager for it. We must * unlock the page queues first. * We try to defer the creation of a pager * until all shadows are not paging. This * allows vm_object_collapse to work better and * helps control swap space size. * (J. Dyson 11 Nov 93) */ if (!object->pager && cnt.v_free_count < vm_pageout_free_min) return 0; if (!object->pager && object->shadow && object->shadow->paging_in_progress) return 0; if( !sync) { if (object->shadow) { vm_object_collapse(object); if (!vm_page_lookup(object, offset)) return 0; } if ((m->busy != 0) || (m->flags & PG_BUSY) || (m->hold_count != 0)) { return 0; } } pageout_count = 1; ms[0] = m; if (pager = object->pager) { for (i = 1; i < vm_pageout_page_count; i++) { if (ms[i] = vm_page_lookup(object, offset+i*NBPG)) { if (( ((ms[i]->flags & (PG_CLEAN|PG_INACTIVE|PG_BUSY)) == PG_INACTIVE) || ( (ms[i]->flags & PG_CLEAN|PG_BUSY) == 0 && sync == VM_PAGEOUT_FORCE)) && (ms[i]->wire_count == 0) && (ms[i]->busy == 0) && (ms[i]->hold_count == 0)) pageout_count++; else break; } else break; } for(i=0;iflags |= PG_BUSY; pmap_page_protect(VM_PAGE_TO_PHYS(ms[i]), VM_PROT_READ); } object->paging_in_progress += pageout_count; cnt.v_pageouts++; cnt.v_pgpgout += pageout_count; } else { m->flags |= PG_BUSY; pmap_page_protect(VM_PAGE_TO_PHYS(m), VM_PROT_READ); cnt.v_pageouts++; cnt.v_pgpgout++; object->paging_in_progress++; pager = vm_pager_allocate(PG_DFLT, (caddr_t)0, object->size, VM_PROT_ALL, 0); if (pager != NULL) { vm_object_setpager(object, pager, 0, FALSE); } } /* * If there is no pager for the page, * use the default pager. If there's * no place to put the page at the * moment, leave it in the laundry and * hope that there will be paging space * later. */ if ((pager && pager->pg_type == PG_SWAP) || cnt.v_free_count >= vm_pageout_free_min) { if( pageout_count == 1) { pageout_status[0] = pager ? vm_pager_put(pager, m, ((sync || (object == kernel_object)) ? TRUE: FALSE)) : VM_PAGER_FAIL; } else { if( !pager) { for(i=0;iflags &= ~PG_LAUNDRY; ++anyok; break; case VM_PAGER_PEND: ms[i]->flags &= ~PG_LAUNDRY; ++anyok; break; case VM_PAGER_BAD: /* * Page outside of range of object. * Right now we essentially lose the * changes by pretending it worked. */ ms[i]->flags &= ~PG_LAUNDRY; ms[i]->flags |= PG_CLEAN; pmap_clear_modify(VM_PAGE_TO_PHYS(ms[i])); break; case VM_PAGER_ERROR: case VM_PAGER_FAIL: /* * If page couldn't be paged out, then * reactivate the page so it doesn't * clog the inactive list. (We will * try paging out it again later). */ if (ms[i]->flags & PG_INACTIVE) vm_page_activate(ms[i]); break; case VM_PAGER_AGAIN: break; } /* * If the operation is still going, leave * the page busy to block all other accesses. * Also, leave the paging in progress * indicator set so that we don't attempt an * object collapse. */ if (pageout_status[i] != VM_PAGER_PEND) { PAGE_WAKEUP(ms[i]); if (--object->paging_in_progress == 0) wakeup((caddr_t) object); if ((ms[i]->flags & PG_REFERENCED) || pmap_is_referenced(VM_PAGE_TO_PHYS(ms[i]))) { pmap_clear_reference(VM_PAGE_TO_PHYS(ms[i])); ms[i]->flags &= ~PG_REFERENCED; if( ms[i]->flags & PG_INACTIVE) vm_page_activate(ms[i]); } } } return anyok; } /* * vm_pageout_object_deactivate_pages * * deactivate enough pages to satisfy the inactive target * requirements or if vm_page_proc_limit is set, then * deactivate all of the pages in the object and its * shadows. * * The object and map must be locked. */ int vm_pageout_object_deactivate_pages(map, object, count) vm_map_t map; vm_object_t object; int count; { register vm_page_t p, next; int rcount; int s; int dcount; dcount = 0; if (count == 0) count = 1; if (object->shadow) { int scount = count; if( object->shadow->ref_count > 1) scount /= object->shadow->ref_count; if( scount) dcount += vm_pageout_object_deactivate_pages(map, object->shadow, scount); } if (object->paging_in_progress) return dcount; /* * scan the objects entire memory queue */ rcount = object->resident_page_count; p = object->memq.tqh_first; while (p && (rcount-- > 0)) { next = p->listq.tqe_next; vm_page_lock_queues(); /* * if a page is active, not wired and is in the processes pmap, * then deactivate the page. */ if ((p->flags & (PG_ACTIVE|PG_BUSY)) == PG_ACTIVE && p->wire_count == 0 && p->hold_count == 0 && p->busy == 0 && pmap_page_exists(vm_map_pmap(map), VM_PAGE_TO_PHYS(p))) { if (!pmap_is_referenced(VM_PAGE_TO_PHYS(p)) && (p->flags & PG_REFERENCED) == 0) { p->act_count -= min(p->act_count, ACT_DECLINE); /* * if the page act_count is zero -- then we deactivate */ if (!p->act_count) { vm_page_deactivate(p); pmap_page_protect(VM_PAGE_TO_PHYS(p), VM_PROT_NONE); /* * else if on the next go-around we will deactivate the page * we need to place the page on the end of the queue to age * the other pages in memory. */ } else { TAILQ_REMOVE(&vm_page_queue_active, p, pageq); TAILQ_INSERT_TAIL(&vm_page_queue_active, p, pageq); TAILQ_REMOVE(&object->memq, p, listq); TAILQ_INSERT_TAIL(&object->memq, p, listq); } /* * see if we are done yet */ if (p->flags & PG_INACTIVE) { --count; ++dcount; if (count <= 0 && cnt.v_inactive_count > cnt.v_inactive_target) { vm_page_unlock_queues(); return dcount; } } } else { /* * Move the page to the bottom of the queue. */ pmap_clear_reference(VM_PAGE_TO_PHYS(p)); p->flags &= ~PG_REFERENCED; if (p->act_count < ACT_MAX) p->act_count += ACT_ADVANCE; TAILQ_REMOVE(&vm_page_queue_active, p, pageq); TAILQ_INSERT_TAIL(&vm_page_queue_active, p, pageq); TAILQ_REMOVE(&object->memq, p, listq); TAILQ_INSERT_TAIL(&object->memq, p, listq); } } vm_page_unlock_queues(); p = next; } return dcount; } /* * deactivate some number of pages in a map, try to do it fairly, but * that is really hard to do. */ void vm_pageout_map_deactivate_pages(map, entry, count, freeer) vm_map_t map; vm_map_entry_t entry; int *count; int (*freeer)(vm_map_t, vm_object_t, int); { vm_map_t tmpm; vm_map_entry_t tmpe; vm_object_t obj; if (*count <= 0) return; vm_map_reference(map); if (!lock_try_read(&map->lock)) { vm_map_deallocate(map); return; } if (entry == 0) { tmpe = map->header.next; while (tmpe != &map->header && *count > 0) { vm_pageout_map_deactivate_pages(map, tmpe, count, freeer); tmpe = tmpe->next; }; } else if (entry->is_sub_map || entry->is_a_map) { tmpm = entry->object.share_map; tmpe = tmpm->header.next; while (tmpe != &tmpm->header && *count > 0) { vm_pageout_map_deactivate_pages(tmpm, tmpe, count, freeer); tmpe = tmpe->next; }; } else if (obj = entry->object.vm_object) { *count -= (*freeer)(map, obj, *count); } lock_read_done(&map->lock); vm_map_deallocate(map); return; } /* * vm_pageout_scan does the dirty work for the pageout daemon. */ int vm_pageout_scan() { vm_page_t m; int page_shortage, maxscan, maxlaunder; int pages_freed, free, nproc; int desired_free; vm_page_t next; struct proc *p; vm_object_t object; int s; int force_wakeup = 0; int cache_size, orig_cache_size; /* * We manage the cached memory by attempting to keep it * at about the desired level. * We deactivate the pages for the oldest cached objects * first. This keeps pages that are "cached" from hogging * physical memory. */ orig_cache_size = 0; object = vm_object_cached_list.tqh_first; /* calculate the total cached size */ while( object) { orig_cache_size += object->resident_page_count; object = object->cached_list.tqe_next; } redeact: cache_size = orig_cache_size; object = vm_object_cached_list.tqh_first; vm_object_cache_lock(); while ( object && (cnt.v_inactive_count < cnt.v_inactive_target) && (cache_size >= vm_desired_cache_size)) { vm_object_cache_unlock(); if (object != vm_object_lookup(object->pager)) panic("vm_object_deactivate: I'm sooo confused."); /* * if there are no resident pages -- get rid of the object */ if( object->resident_page_count == 0) { pager_cache(object, FALSE); goto redeact; } else { /* * if there are resident pages -- deactivate them */ vm_object_deactivate_pages(object); cache_size -= object->resident_page_count; object = object->cached_list.tqe_next; } vm_object_cache_lock(); } vm_object_cache_unlock(); morefree: /* - * now check malloc area or swap processes out if we are in low - * memory conditions + * now swap processes out if we are in low memory conditions */ if (cnt.v_free_count <= cnt.v_free_min) { /* * swap out inactive processes */ swapout_threads(); } + /* * scan the processes for exceeding their rlimits or if process * is swapped out -- deactivate pages */ rescanproc1: for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { vm_offset_t size; int overage; quad_t limit; /* * if this is a system process or if we have already * looked at this process, skip it. */ if (p->p_flag & (P_SYSTEM|P_WEXIT)) { continue; } /* * if the process is in a non-running type state, * don't touch it. */ if (p->p_stat != SRUN && p->p_stat != SSLEEP) { continue; } /* * get a limit */ limit = qmin(p->p_rlimit[RLIMIT_RSS].rlim_cur, p->p_rlimit[RLIMIT_RSS].rlim_max); /* * let processes that are swapped out really be swapped out * set the limit to nothing (will force a swap-out.) */ if ((p->p_flag & P_INMEM) == 0) limit = 0; size = p->p_vmspace->vm_pmap.pm_stats.resident_count * NBPG; - if (limit > 0 && size >= limit) { + if (limit >= 0 && size >= limit) { overage = (size - limit) / NBPG; vm_pageout_map_deactivate_pages(&p->p_vmspace->vm_map, (vm_map_entry_t) 0, &overage, vm_pageout_object_deactivate_pages); } } if (((cnt.v_free_count + cnt.v_inactive_count) >= (cnt.v_inactive_target + cnt.v_free_target)) && (cnt.v_free_count >= cnt.v_free_target)) return force_wakeup; pages_freed = 0; desired_free = cnt.v_free_target; /* * Start scanning the inactive queue for pages we can free. * We keep scanning until we have enough free pages or * we have scanned through the entire queue. If we * encounter dirty pages, we start cleaning them. */ maxlaunder = (cnt.v_free_target - cnt.v_free_count); maxscan = cnt.v_inactive_count; rescan1: m = vm_page_queue_inactive.tqh_first; while (m && (maxscan-- > 0) && (cnt.v_free_count < desired_free) ) { vm_page_t next; next = m->pageq.tqe_next; if( (m->flags & PG_INACTIVE) == 0) { printf("vm_pageout_scan: page not inactive?"); continue; } /* * activate held pages */ if (m->hold_count != 0) { vm_page_activate(m); m = next; continue; } /* * dont mess with busy pages */ if (m->busy || (m->flags & PG_BUSY)) { m = next; continue; } /* * if page is clean and but the page has been referenced, * then reactivate the page, but if we are very low on memory * or the page has not been referenced, then we free it to the * vm system. */ if (m->flags & PG_CLEAN) { if ((cnt.v_free_count > vm_pageout_free_min) /* XXX */ && ((pmap_is_referenced(VM_PAGE_TO_PHYS(m)) || (m->flags & PG_REFERENCED) != 0))) { m->flags &= ~PG_REFERENCED; vm_page_activate(m); } else if (!m->act_count) { pmap_page_protect(VM_PAGE_TO_PHYS(m), VM_PROT_NONE); vm_page_free(m); ++cnt.v_dfree; ++pages_freed; } else { m->act_count -= min(m->act_count, ACT_DECLINE); TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq); TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq); } } else if ((m->flags & PG_LAUNDRY) && maxlaunder > 0) { int written; if (pmap_is_referenced(VM_PAGE_TO_PHYS(m)) || ((m->flags & PG_REFERENCED) != 0)) { pmap_clear_reference(VM_PAGE_TO_PHYS(m)); vm_page_activate(m); m->flags &= ~PG_REFERENCED; m = next; continue; } /* * If a page is dirty, then it is either * being washed (but not yet cleaned) * or it is still in the laundry. If it is * still in the laundry, then we start the * cleaning operation. */ if (written = vm_pageout_clean(m,0)) { maxlaunder -= written; } if (!next) break; /* * if the next page has been re-activated, start scanning again */ if ((next->flags & PG_INACTIVE) == 0) goto rescan1; } else if ((m->flags & PG_REFERENCED) || pmap_is_referenced(VM_PAGE_TO_PHYS(m))) { pmap_clear_reference(VM_PAGE_TO_PHYS(m)); m->flags &= ~PG_REFERENCED; vm_page_activate(m); } m = next; } - /* * Compute the page shortage. If we are still very low on memory * be sure that we will move a minimal amount of pages from active * to inactive. */ page_shortage = cnt.v_inactive_target - (cnt.v_free_count + cnt.v_inactive_count); if (page_shortage <= 0) { if (pages_freed == 0) { if( cnt.v_free_count < cnt.v_free_min) { page_shortage = cnt.v_free_min - cnt.v_free_count; } else if(((cnt.v_free_count + cnt.v_inactive_count) < (cnt.v_free_min + cnt.v_inactive_target))) { page_shortage = 1; } else { page_shortage = 0; } } } maxscan = cnt.v_active_count; m = vm_page_queue_active.tqh_first; while (m && maxscan-- && (page_shortage > 0)) { next = m->pageq.tqe_next; /* * Don't deactivate pages that are busy. */ if ((m->busy != 0) || (m->flags & PG_BUSY) || (m->hold_count != 0)) { m = next; continue; } if ((m->flags & PG_REFERENCED) || pmap_is_referenced(VM_PAGE_TO_PHYS(m))) { pmap_clear_reference(VM_PAGE_TO_PHYS(m)); m->flags &= ~PG_REFERENCED; if (m->act_count < ACT_MAX) m->act_count += ACT_ADVANCE; TAILQ_REMOVE(&vm_page_queue_active, m, pageq); TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); TAILQ_REMOVE(&m->object->memq, m, listq); TAILQ_INSERT_TAIL(&m->object->memq, m, listq); } else { m->act_count -= min(m->act_count, ACT_DECLINE); /* * if the page act_count is zero -- then we deactivate */ if (!m->act_count) { vm_page_deactivate(m); --page_shortage; /* * else if on the next go-around we will deactivate the page * we need to place the page on the end of the queue to age * the other pages in memory. */ } else { TAILQ_REMOVE(&vm_page_queue_active, m, pageq); TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq); TAILQ_REMOVE(&m->object->memq, m, listq); TAILQ_INSERT_TAIL(&m->object->memq, m, listq); } } m = next; } /* * if we have not freed any pages and we are desparate for memory * then we keep trying until we get some (any) memory. */ if( !force_wakeup && (swap_pager_full || !force_wakeup || (pages_freed == 0 && (cnt.v_free_count < cnt.v_free_min)))){ vm_pager_sync(); force_wakeup = 1; goto morefree; } vm_page_pagesfreed += pages_freed; return force_wakeup; } /* * vm_pageout is the high level pageout daemon. */ void vm_pageout() { extern swiopend; static int nowakeup; (void) spl0(); /* * Initialize some paging parameters. */ vmretry: cnt.v_free_min = 12; /* * free_reserved needs to include enough for the largest * swap pager structures plus enough for any pv_entry * structs when paging. */ cnt.v_free_reserved = 4 + cnt.v_page_count / 1024; if (cnt.v_free_min < 8) cnt.v_free_min = 8; if (cnt.v_free_min > 32) cnt.v_free_min = 32; vm_pageout_free_min = cnt.v_free_reserved; cnt.v_free_target = 2*cnt.v_free_min + cnt.v_free_reserved; cnt.v_inactive_target = cnt.v_free_count / 12; cnt.v_free_min += cnt.v_free_reserved; vm_desired_cache_size = cnt.v_page_count / 3; /* XXX does not really belong here */ if (vm_page_max_wired == 0) vm_page_max_wired = cnt.v_free_count / 3; (void) swap_pager_alloc(0, 0, 0, 0); /* * The pageout daemon is never done, so loop * forever. */ while (TRUE) { int force_wakeup; /* cnt.v_free_min = 12 + averunnable.ldavg[0] / 1024; cnt.v_free_target = 2*cnt.v_free_min + cnt.v_free_reserved; cnt.v_inactive_target = cnt.v_free_target*2; */ tsleep((caddr_t) &vm_pages_needed, PVM, "psleep", 0); vm_pager_sync(); /* * The force wakeup hack added to eliminate delays and potiential * deadlock. It was possible for the page daemon to indefintely * postpone waking up a process that it might be waiting for memory * on. The putmulti stuff seems to have aggravated the situation. */ force_wakeup = vm_pageout_scan(); vm_pager_sync(); if( force_wakeup) wakeup( (caddr_t) &cnt.v_free_count); cnt.v_scan++; wakeup((caddr_t) kmem_map); } }