NetBSD/sys/vm/vm_glue.c

503 lines
14 KiB
C
Raw Normal View History

/* $NetBSD: vm_glue.c,v 1.75 1998/05/09 15:04:40 kleink Exp $ */
1993-03-21 12:45:37 +03:00
/*
1994-05-23 07:11:20 +04:00
* Copyright (c) 1991, 1993
* The Regents of the University of California. All rights reserved.
1993-03-21 12:45:37 +03:00
*
* 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.
*
1998-03-01 05:20:01 +03:00
* @(#)vm_glue.c 8.9 (Berkeley) 3/4/95
1993-03-21 12:45:37 +03:00
*
*
* 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.
*/
1993-12-17 10:56:32 +03:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/resourcevar.h>
#include <sys/buf.h>
#include <sys/user.h>
#ifdef SYSVSHM
#include <sys/shm.h>
#endif
1993-03-21 12:45:37 +03:00
1993-12-17 10:56:32 +03:00
#include <vm/vm.h>
#include <vm/vm_page.h>
#include <vm/vm_kern.h>
1993-03-21 12:45:37 +03:00
1993-12-20 15:39:55 +03:00
#include <machine/cpu.h>
1993-03-21 12:45:37 +03:00
int avefree = 0; /* XXX */
1994-05-19 12:08:46 +04:00
unsigned maxdmap = MAXDSIZ; /* XXX */
unsigned maxsmap = MAXSSIZ; /* XXX */
1993-03-21 12:45:37 +03:00
int readbuffers = 0; /* XXX allow kgdb to read kernel buffer pool */
1994-01-08 01:48:40 +03:00
int
1993-03-21 12:45:37 +03:00
kernacc(addr, len, rw)
caddr_t addr;
size_t len;
int rw;
1993-03-21 12:45:37 +03:00
{
boolean_t rv;
vm_offset_t saddr, eaddr;
1994-05-23 07:11:20 +04:00
vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
1993-03-21 12:45:37 +03:00
saddr = trunc_page(addr);
1993-07-15 19:42:17 +04:00
eaddr = round_page(addr+len);
1993-03-21 12:45:37 +03:00
rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
/*
* XXX there are still some things (e.g. the buffer cache) that
* are managed behind the VM system's back so even though an
* address is accessible in the mind of the VM system, there may
* not be physical pages where the VM thinks there is. This can
* lead to bogus allocation of pages in the kernel address space
* or worse, inconsistencies at the pmap level. We only worry
* about the buffer cache for now.
*/
if (!readbuffers && rv && (eaddr > (vm_offset_t)buffers &&
saddr < (vm_offset_t)buffers + MAXBSIZE * nbuf))
rv = FALSE;
return(rv == TRUE);
}
1994-01-08 01:48:40 +03:00
int
1993-03-21 12:45:37 +03:00
useracc(addr, len, rw)
caddr_t addr;
size_t len;
int rw;
1993-03-21 12:45:37 +03:00
{
boolean_t rv;
vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
1994-05-23 07:11:20 +04:00
#if defined(i386) || defined(pc532)
/*
* XXX - specially disallow access to user page tables - they are
1994-05-23 07:11:20 +04:00
* in the map. This is here until i386 & pc532 pmaps are fixed...
*/
1993-06-29 17:47:05 +04:00
if ((vm_offset_t) addr >= VM_MAXUSER_ADDRESS
1993-07-15 19:42:17 +04:00
|| (vm_offset_t) addr + len > VM_MAXUSER_ADDRESS
|| (vm_offset_t) addr + len <= (vm_offset_t) addr)
return (FALSE);
1994-05-23 07:11:20 +04:00
#endif
1993-03-21 12:45:37 +03:00
rv = vm_map_check_protection(&curproc->p_vmspace->vm_map,
1993-07-15 19:42:17 +04:00
trunc_page(addr), round_page(addr+len), prot);
1993-03-21 12:45:37 +03:00
return(rv == TRUE);
}
#ifdef KGDB
/*
* Change protections on kernel pages from addr to addr+len
* (presumably so debugger can plant a breakpoint).
1994-05-23 07:11:20 +04:00
*
* We force the protection change at the pmap level. If we were
* to use vm_map_protect a change to allow writing would be lazily-
* applied meaning we would still take a protection fault, something
* we really don't want to do. It would also fragment the kernel
* map unnecessarily. We cannot use pmap_protect since it also won't
* enforce a write-enable request. Using pmap_enter is the only way
* we can ensure the change takes place properly.
1993-03-21 12:45:37 +03:00
*/
1994-05-23 07:11:20 +04:00
void
1993-03-21 12:45:37 +03:00
chgkprot(addr, len, rw)
register caddr_t addr;
size_t len;
int rw;
1993-03-21 12:45:37 +03:00
{
1994-05-23 07:11:20 +04:00
vm_prot_t prot;
vm_offset_t pa, sva, eva;
prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
eva = round_page(addr + len);
for (sva = trunc_page(addr); sva < eva; sva += PAGE_SIZE) {
/*
* Extract physical address for the page.
* We use a cheezy hack to differentiate physical
* page 0 from an invalid mapping, not that it
* really matters...
*/
1995-04-10 20:48:27 +04:00
pa = pmap_extract(pmap_kernel(), sva|1);
1994-05-23 07:11:20 +04:00
if (pa == 0)
panic("chgkprot: invalid page");
1995-04-10 20:48:27 +04:00
pmap_enter(pmap_kernel(), sva, pa&~1, prot, TRUE);
1994-05-23 07:11:20 +04:00
}
1993-03-21 12:45:37 +03:00
}
#endif
1994-01-08 01:46:05 +03:00
void
vslock(p, addr, len)
struct proc *p;
1993-03-21 12:45:37 +03:00
caddr_t addr;
size_t len;
1993-03-21 12:45:37 +03:00
{
vm_map_pageable(&p->p_vmspace->vm_map, trunc_page(addr),
1993-07-15 19:42:17 +04:00
round_page(addr+len), FALSE);
1993-03-21 12:45:37 +03:00
}
1994-01-08 01:46:05 +03:00
void
vsunlock(p, addr, len)
struct proc *p;
1993-03-21 12:45:37 +03:00
caddr_t addr;
size_t len;
1993-03-21 12:45:37 +03:00
{
vm_map_pageable(&p->p_vmspace->vm_map, trunc_page(addr),
1993-07-15 19:42:17 +04:00
round_page(addr+len), TRUE);
1993-03-21 12:45:37 +03:00
}
/*
* 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.
*/
void
vm_fork(p1, p2, shared)
1993-03-21 12:45:37 +03:00
register struct proc *p1, *p2;
boolean_t shared;
1993-03-21 12:45:37 +03:00
{
struct user *up = p2->p_addr;
1993-03-21 12:45:37 +03:00
/*
* Share the address space if we've been directed to.
1993-03-21 12:45:37 +03:00
*/
if (shared == TRUE)
vmspace_share(p1, p2);
else
p2->p_vmspace = vmspace_fork(p1->p_vmspace);
1993-03-21 12:45:37 +03:00
/*
* Wire down the U-area for the process, which contains the PCB
* and the kernel stack.
1993-03-21 12:45:37 +03:00
*/
vm_map_pageable(kernel_map, (vm_offset_t)up,
(vm_offset_t)up + USPACE, FALSE);
1993-03-21 12:45:37 +03:00
/*
* 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. The child will exit
* directly to user mode on its first time slice, and will
* not return here.
*/
cpu_fork(p1, p2);
1993-03-21 12:45:37 +03:00
}
/*
* Set default limits for VM system.
* Called for proc 0, and then inherited by all others.
*/
1993-06-27 10:27:29 +04:00
void
1993-03-21 12:45:37 +03:00
vm_init_limits(p)
register struct proc *p;
{
/*
* Set up the initial limits on process VM.
* Set the maximum resident set size to be all
* of (reasonably) available memory. This causes
* any single, large process to start random page
* replacement once it fills memory.
*/
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;
1994-05-23 07:11:20 +04:00
p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(cnt.v_free_count);
1993-03-21 12:45:37 +03:00
}
1993-12-17 10:56:32 +03:00
#include <vm/vm_pageout.h>
1993-03-21 12:45:37 +03:00
#ifdef DEBUG
int enableswap = 1;
int swapdebug = 0;
#define SDB_FOLLOW 1
#define SDB_SWAPIN 2
#define SDB_SWAPOUT 4
#endif
/*
* Swap in a process's u-area.
*/
void
swapin(p)
struct proc *p;
{
vm_offset_t addr;
int s;
addr = (vm_offset_t)p->p_addr;
vm_map_pageable(kernel_map, addr, addr + USPACE, FALSE);
/*
* Some architectures need to be notified when the
* user area has moved to new physical page(s) (e.g.
* see pmax/pmax/vm_machdep.c).
*/
cpu_swapin(p);
s = splstatclock();
if (p->p_stat == SRUN)
setrunqueue(p);
p->p_flag |= P_INMEM;
splx(s);
p->p_swtime = 0;
1996-05-30 01:20:11 +04:00
++cnt.v_swpin;
}
1993-03-21 12:45:37 +03:00
/*
* Brutally simple:
* 1. Attempt to swapin every swaped-out, runnable process in
* order of priority.
* 2. If not enough memory, wake the pageout daemon and let it
* clear some space.
*/
1993-06-27 10:27:29 +04:00
void
scheduler()
1993-03-21 12:45:37 +03:00
{
register struct proc *p;
register int pri;
struct proc *pp;
int ppri;
loop:
#ifdef DEBUG
1994-05-23 07:11:20 +04:00
while (!enableswap)
tsleep((caddr_t)&proc0, PVM, "noswap", 0);
1993-03-21 12:45:37 +03:00
#endif
pp = NULL;
ppri = INT_MIN;
for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
1994-05-04 07:41:12 +04:00
if (p->p_stat == SRUN && (p->p_flag & P_INMEM) == 0) {
1998-03-01 05:20:01 +03:00
/* XXX should also penalize based on vm_swrss */
pri = p->p_swtime + p->p_slptime
- (p->p_nice - NZERO) * 8;
1993-03-21 12:45:37 +03:00
if (pri > ppri) {
pp = p;
ppri = pri;
}
}
1994-05-23 07:11:20 +04:00
}
1993-03-21 12:45:37 +03:00
#ifdef DEBUG
if (swapdebug & SDB_FOLLOW)
1996-10-13 01:50:05 +04:00
printf("scheduler: running, procp %p pri %d\n", pp, ppri);
1993-03-21 12:45:37 +03:00
#endif
/*
* Nothing to do, back to sleep
*/
if ((p = pp) == NULL) {
tsleep((caddr_t)&proc0, PVM, "scheduler", 0);
1993-03-21 12:45:37 +03:00
goto loop;
}
/*
* We would like to bring someone in.
* This part is really bogus cuz we could deadlock on memory
* despite our feeble check.
1998-03-01 05:20:01 +03:00
* XXX should require at least vm_swrss / 2
1993-03-21 12:45:37 +03:00
*/
if (cnt.v_free_count > atop(USPACE)) {
1993-03-21 12:45:37 +03:00
#ifdef DEBUG
if (swapdebug & SDB_SWAPIN)
1996-10-13 01:50:05 +04:00
printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
1996-10-10 21:16:17 +04:00
p->p_pid, p->p_comm, p->p_addr,
ppri, cnt.v_free_count);
1993-03-21 12:45:37 +03:00
#endif
swapin(p);
1993-03-21 12:45:37 +03:00
goto loop;
}
/*
* Not enough memory, jab the pageout daemon and wait til the
* coast is clear.
*/
#ifdef DEBUG
if (swapdebug & SDB_FOLLOW)
1996-10-13 01:50:05 +04:00
printf("scheduler: no room for pid %d(%s), free %d\n",
1996-10-10 21:16:17 +04:00
p->p_pid, p->p_comm, cnt.v_free_count);
1993-03-21 12:45:37 +03:00
#endif
(void) splhigh();
vm_wait("fLowmem");
1993-03-21 12:45:37 +03:00
(void) spl0();
#ifdef DEBUG
if (swapdebug & SDB_FOLLOW)
1996-10-13 01:50:05 +04:00
printf("scheduler: room again, free %d\n", cnt.v_free_count);
1993-03-21 12:45:37 +03:00
#endif
goto loop;
}
1994-05-23 07:11:20 +04:00
#define swappable(p) \
(((p)->p_flag & (P_SYSTEM | P_INMEM | P_WEXIT)) == P_INMEM && \
(p)->p_holdcnt == 0)
1993-03-21 12:45:37 +03:00
/*
* 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.
*/
1994-01-08 01:48:40 +03:00
void
1993-03-21 12:45:37 +03:00
swapout_threads()
{
register struct proc *p;
struct proc *outp, *outp2;
int outpri, outpri2;
int didswap = 0;
extern int maxslp;
#ifdef DEBUG
if (!enableswap)
return;
#endif
outp = outp2 = NULL;
outpri = outpri2 = 0;
for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
1993-03-21 12:45:37 +03:00
if (!swappable(p))
continue;
switch (p->p_stat) {
case SRUN:
if (p->p_swtime > outpri2) {
1993-03-21 12:45:37 +03:00
outp2 = p;
outpri2 = p->p_swtime;
1993-03-21 12:45:37 +03:00
}
continue;
case SSLEEP:
case SSTOP:
1994-05-23 07:11:20 +04:00
if (p->p_slptime >= maxslp) {
1993-03-21 12:45:37 +03:00
swapout(p);
didswap++;
} else if (p->p_slptime > outpri) {
outp = p;
outpri = p->p_slptime;
}
continue;
}
}
/*
* If we didn't get rid of any real duds, toss out the next most
* likely sleeping/stopped or running candidate. We only do this
* if we are real low on memory since we don't gain much by doing
* it (USPACE bytes).
1993-03-21 12:45:37 +03:00
*/
if (didswap == 0 &&
cnt.v_free_count <= atop(round_page(USPACE))) {
1993-03-21 12:45:37 +03:00
if ((p = outp) == 0)
p = outp2;
#ifdef DEBUG
if (swapdebug & SDB_SWAPOUT)
1996-10-13 01:50:05 +04:00
printf("swapout_threads: no duds, try procp %p\n", p);
1993-03-21 12:45:37 +03:00
#endif
if (p)
swapout(p);
}
}
1994-05-23 07:11:20 +04:00
void
1993-03-21 12:45:37 +03:00
swapout(p)
register struct proc *p;
{
vm_offset_t addr;
1995-12-21 07:44:27 +03:00
int s;
1993-03-21 12:45:37 +03:00
#ifdef DEBUG
if (swapdebug & SDB_SWAPOUT)
1996-10-13 01:50:05 +04:00
printf("swapout: pid %d(%s)@%p, stat %x pri %d free %d\n",
1996-10-10 21:16:17 +04:00
p->p_pid, p->p_comm, p->p_addr, p->p_stat,
p->p_slptime, cnt.v_free_count);
1993-03-21 12:45:37 +03:00
#endif
1993-03-21 12:45:37 +03:00
/*
* Do any machine-specific actions necessary before swapout.
* This can include saving floating point state, etc.
1993-03-21 12:45:37 +03:00
*/
cpu_swapout(p);
1993-03-21 12:45:37 +03:00
1994-05-23 07:11:20 +04:00
/*
* Unwire the to-be-swapped process's user struct and kernel stack.
1994-05-23 07:11:20 +04:00
*/
1995-12-21 07:44:27 +03:00
addr = (vm_offset_t)p->p_addr;
vm_map_pageable(kernel_map, addr, addr + USPACE, TRUE);
1993-03-21 12:45:37 +03:00
pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
/*
* Mark it as (potentially) swapped out.
*/
1995-12-21 07:44:27 +03:00
s = splstatclock();
1994-05-04 07:41:12 +04:00
p->p_flag &= ~P_INMEM;
1993-03-21 12:45:37 +03:00
if (p->p_stat == SRUN)
remrunqueue(p);
1995-12-21 07:44:27 +03:00
splx(s);
p->p_swtime = 0;
1996-05-30 01:20:11 +04:00
++cnt.v_swpout;
1993-03-21 12:45:37 +03:00
}