xref: /freebsd-10.0-release/sys/vm/vm_glue.c

/*
 * 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 1.3 1994/08/02 07:55:19 davidg Exp $
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/resourcevar.h>
#include <sys/buf.h>
#include <sys/user.h>

#include <sys/kernel.h>
#include <sys/dkstat.h>

#include <vm/vm.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_kern.h>

#include <machine/stdarg.h>

extern char kstack[];
int	avefree = 0;		/* XXX */
int	readbuffers = 0;	/* XXX allow kgdb to read kernel buffer pool */
/* vm_map_t upages_map; */

void swapout(struct proc *p);
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;
	vmmeter();

	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) {
				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) {
			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 <machine/stdarg.h>		/* 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);
}