Home | History | Annotate | Line # | Download | only in uvm
uvm_glue.c revision 1.92.10.1
      1  1.92.10.1      elad /*	$NetBSD: uvm_glue.c,v 1.92.10.1 2006/04/19 03:58:21 elad Exp $	*/
      2        1.1       mrg 
      3       1.48       chs /*
      4        1.1       mrg  * Copyright (c) 1997 Charles D. Cranor and Washington University.
      5       1.48       chs  * Copyright (c) 1991, 1993, The Regents of the University of California.
      6        1.1       mrg  *
      7        1.1       mrg  * All rights reserved.
      8        1.1       mrg  *
      9        1.1       mrg  * This code is derived from software contributed to Berkeley by
     10        1.1       mrg  * The Mach Operating System project at Carnegie-Mellon University.
     11        1.1       mrg  *
     12        1.1       mrg  * Redistribution and use in source and binary forms, with or without
     13        1.1       mrg  * modification, are permitted provided that the following conditions
     14        1.1       mrg  * are met:
     15        1.1       mrg  * 1. Redistributions of source code must retain the above copyright
     16        1.1       mrg  *    notice, this list of conditions and the following disclaimer.
     17        1.1       mrg  * 2. Redistributions in binary form must reproduce the above copyright
     18        1.1       mrg  *    notice, this list of conditions and the following disclaimer in the
     19        1.1       mrg  *    documentation and/or other materials provided with the distribution.
     20        1.1       mrg  * 3. All advertising materials mentioning features or use of this software
     21        1.1       mrg  *    must display the following acknowledgement:
     22        1.1       mrg  *	This product includes software developed by Charles D. Cranor,
     23       1.48       chs  *      Washington University, the University of California, Berkeley and
     24        1.1       mrg  *      its contributors.
     25        1.1       mrg  * 4. Neither the name of the University nor the names of its contributors
     26        1.1       mrg  *    may be used to endorse or promote products derived from this software
     27        1.1       mrg  *    without specific prior written permission.
     28        1.1       mrg  *
     29        1.1       mrg  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     30        1.1       mrg  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     31        1.1       mrg  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     32        1.1       mrg  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     33        1.1       mrg  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     34        1.1       mrg  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     35        1.1       mrg  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     36        1.1       mrg  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     37        1.1       mrg  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     38        1.1       mrg  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     39        1.1       mrg  * SUCH DAMAGE.
     40        1.1       mrg  *
     41        1.1       mrg  *	@(#)vm_glue.c	8.6 (Berkeley) 1/5/94
     42        1.4       mrg  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
     43        1.1       mrg  *
     44        1.1       mrg  *
     45        1.1       mrg  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
     46        1.1       mrg  * All rights reserved.
     47       1.48       chs  *
     48        1.1       mrg  * Permission to use, copy, modify and distribute this software and
     49        1.1       mrg  * its documentation is hereby granted, provided that both the copyright
     50        1.1       mrg  * notice and this permission notice appear in all copies of the
     51        1.1       mrg  * software, derivative works or modified versions, and any portions
     52        1.1       mrg  * thereof, and that both notices appear in supporting documentation.
     53       1.48       chs  *
     54       1.48       chs  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     55       1.48       chs  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     56        1.1       mrg  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     57       1.48       chs  *
     58        1.1       mrg  * Carnegie Mellon requests users of this software to return to
     59        1.1       mrg  *
     60        1.1       mrg  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     61        1.1       mrg  *  School of Computer Science
     62        1.1       mrg  *  Carnegie Mellon University
     63        1.1       mrg  *  Pittsburgh PA 15213-3890
     64        1.1       mrg  *
     65        1.1       mrg  * any improvements or extensions that they make and grant Carnegie the
     66        1.1       mrg  * rights to redistribute these changes.
     67        1.1       mrg  */
     68       1.55     lukem 
     69       1.55     lukem #include <sys/cdefs.h>
     70  1.92.10.1      elad __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.92.10.1 2006/04/19 03:58:21 elad Exp $");
     71        1.1       mrg 
     72       1.49     lukem #include "opt_kgdb.h"
     73       1.59      yamt #include "opt_kstack.h"
     74        1.5       mrg #include "opt_uvmhist.h"
     75        1.5       mrg 
     76        1.1       mrg /*
     77        1.1       mrg  * uvm_glue.c: glue functions
     78        1.1       mrg  */
     79        1.1       mrg 
     80        1.1       mrg #include <sys/param.h>
     81        1.1       mrg #include <sys/systm.h>
     82        1.1       mrg #include <sys/proc.h>
     83        1.1       mrg #include <sys/resourcevar.h>
     84        1.1       mrg #include <sys/buf.h>
     85        1.1       mrg #include <sys/user.h>
     86        1.1       mrg 
     87        1.1       mrg #include <uvm/uvm.h>
     88        1.1       mrg 
     89        1.1       mrg #include <machine/cpu.h>
     90        1.1       mrg 
     91        1.1       mrg /*
     92        1.1       mrg  * local prototypes
     93        1.1       mrg  */
     94        1.1       mrg 
     95       1.78  junyoung static void uvm_swapout(struct lwp *);
     96        1.1       mrg 
     97       1.60       chs #define UVM_NUAREA_MAX 16
     98       1.60       chs void *uvm_uareas;
     99       1.60       chs int uvm_nuarea;
    100       1.62   thorpej struct simplelock uvm_uareas_slock = SIMPLELOCK_INITIALIZER;
    101       1.60       chs 
    102       1.75  jdolecek static void uvm_uarea_free(vaddr_t);
    103       1.75  jdolecek 
    104        1.1       mrg /*
    105        1.1       mrg  * XXXCDC: do these really belong here?
    106        1.1       mrg  */
    107        1.1       mrg 
    108       1.28   thorpej /*
    109        1.1       mrg  * uvm_kernacc: can the kernel access a region of memory
    110        1.1       mrg  *
    111       1.83      yamt  * - used only by /dev/kmem driver (mem.c)
    112        1.1       mrg  */
    113        1.1       mrg 
    114        1.6       mrg boolean_t
    115       1.89   thorpej uvm_kernacc(caddr_t addr, size_t len, int rw)
    116        1.6       mrg {
    117        1.6       mrg 	boolean_t rv;
    118       1.13       eeh 	vaddr_t saddr, eaddr;
    119        1.6       mrg 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
    120        1.6       mrg 
    121       1.31    kleink 	saddr = trunc_page((vaddr_t)addr);
    122       1.43       chs 	eaddr = round_page((vaddr_t)addr + len);
    123        1.6       mrg 	vm_map_lock_read(kernel_map);
    124        1.6       mrg 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
    125        1.6       mrg 	vm_map_unlock_read(kernel_map);
    126        1.6       mrg 
    127        1.6       mrg 	return(rv);
    128        1.1       mrg }
    129        1.1       mrg 
    130        1.1       mrg #ifdef KGDB
    131        1.1       mrg /*
    132        1.1       mrg  * Change protections on kernel pages from addr to addr+len
    133        1.1       mrg  * (presumably so debugger can plant a breakpoint).
    134        1.1       mrg  *
    135        1.1       mrg  * We force the protection change at the pmap level.  If we were
    136        1.1       mrg  * to use vm_map_protect a change to allow writing would be lazily-
    137        1.1       mrg  * applied meaning we would still take a protection fault, something
    138        1.1       mrg  * we really don't want to do.  It would also fragment the kernel
    139        1.1       mrg  * map unnecessarily.  We cannot use pmap_protect since it also won't
    140        1.1       mrg  * enforce a write-enable request.  Using pmap_enter is the only way
    141        1.1       mrg  * we can ensure the change takes place properly.
    142        1.1       mrg  */
    143        1.6       mrg void
    144       1.89   thorpej uvm_chgkprot(caddr_t addr, size_t len, int rw)
    145        1.6       mrg {
    146        1.6       mrg 	vm_prot_t prot;
    147       1.13       eeh 	paddr_t pa;
    148       1.13       eeh 	vaddr_t sva, eva;
    149        1.6       mrg 
    150        1.6       mrg 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
    151       1.31    kleink 	eva = round_page((vaddr_t)addr + len);
    152       1.31    kleink 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
    153        1.6       mrg 		/*
    154        1.6       mrg 		 * Extract physical address for the page.
    155        1.6       mrg 		 */
    156       1.27   thorpej 		if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
    157        1.6       mrg 			panic("chgkprot: invalid page");
    158       1.30   thorpej 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
    159        1.6       mrg 	}
    160       1.51     chris 	pmap_update(pmap_kernel());
    161        1.1       mrg }
    162        1.1       mrg #endif
    163        1.1       mrg 
    164        1.1       mrg /*
    165       1.52       chs  * uvm_vslock: wire user memory for I/O
    166        1.1       mrg  *
    167        1.1       mrg  * - called from physio and sys___sysctl
    168        1.1       mrg  * - XXXCDC: consider nuking this (or making it a macro?)
    169        1.1       mrg  */
    170        1.1       mrg 
    171       1.26   thorpej int
    172       1.89   thorpej uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type)
    173        1.1       mrg {
    174       1.50       chs 	struct vm_map *map;
    175       1.26   thorpej 	vaddr_t start, end;
    176       1.45       chs 	int error;
    177       1.26   thorpej 
    178       1.26   thorpej 	map = &p->p_vmspace->vm_map;
    179       1.31    kleink 	start = trunc_page((vaddr_t)addr);
    180       1.31    kleink 	end = round_page((vaddr_t)addr + len);
    181  1.92.10.1      elad 	error = uvm_fault_wire(map, start, end, access_type, 0);
    182       1.45       chs 	return error;
    183        1.1       mrg }
    184        1.1       mrg 
    185        1.1       mrg /*
    186       1.52       chs  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
    187        1.1       mrg  *
    188        1.1       mrg  * - called from physio and sys___sysctl
    189        1.1       mrg  * - XXXCDC: consider nuking this (or making it a macro?)
    190        1.1       mrg  */
    191        1.1       mrg 
    192        1.6       mrg void
    193       1.89   thorpej uvm_vsunlock(struct proc *p, caddr_t addr, size_t len)
    194        1.1       mrg {
    195       1.43       chs 	uvm_fault_unwire(&p->p_vmspace->vm_map, trunc_page((vaddr_t)addr),
    196       1.43       chs 		round_page((vaddr_t)addr + len));
    197        1.1       mrg }
    198        1.1       mrg 
    199        1.1       mrg /*
    200       1.62   thorpej  * uvm_proc_fork: fork a virtual address space
    201        1.1       mrg  *
    202        1.1       mrg  * - the address space is copied as per parent map's inherit values
    203       1.62   thorpej  */
    204       1.62   thorpej void
    205       1.89   thorpej uvm_proc_fork(struct proc *p1, struct proc *p2, boolean_t shared)
    206       1.62   thorpej {
    207       1.62   thorpej 
    208       1.62   thorpej 	if (shared == TRUE) {
    209       1.62   thorpej 		p2->p_vmspace = NULL;
    210       1.62   thorpej 		uvmspace_share(p1, p2);
    211       1.62   thorpej 	} else {
    212       1.62   thorpej 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
    213       1.62   thorpej 	}
    214       1.62   thorpej 
    215       1.62   thorpej 	cpu_proc_fork(p1, p2);
    216       1.62   thorpej }
    217       1.62   thorpej 
    218       1.62   thorpej 
    219       1.62   thorpej /*
    220       1.62   thorpej  * uvm_lwp_fork: fork a thread
    221       1.62   thorpej  *
    222        1.1       mrg  * - a new "user" structure is allocated for the child process
    223        1.1       mrg  *	[filled in by MD layer...]
    224       1.20   thorpej  * - if specified, the child gets a new user stack described by
    225       1.20   thorpej  *	stack and stacksize
    226        1.1       mrg  * - NOTE: the kernel stack may be at a different location in the child
    227        1.1       mrg  *	process, and thus addresses of automatic variables may be invalid
    228       1.62   thorpej  *	after cpu_lwp_fork returns in the child process.  We do nothing here
    229       1.62   thorpej  *	after cpu_lwp_fork returns.
    230        1.1       mrg  * - XXXCDC: we need a way for this to return a failure value rather
    231        1.1       mrg  *   than just hang
    232        1.1       mrg  */
    233        1.6       mrg void
    234       1.89   thorpej uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
    235       1.89   thorpej     void (*func)(void *), void *arg)
    236        1.6       mrg {
    237       1.62   thorpej 	struct user *up = l2->l_addr;
    238       1.45       chs 	int error;
    239        1.6       mrg 
    240        1.6       mrg 	/*
    241        1.7   thorpej 	 * Wire down the U-area for the process, which contains the PCB
    242       1.62   thorpej 	 * and the kernel stack.  Wired state is stored in l->l_flag's
    243       1.62   thorpej 	 * L_INMEM bit rather than in the vm_map_entry's wired count
    244       1.61       chs 	 * to prevent kernel_map fragmentation.  If we reused a cached U-area,
    245       1.62   thorpej 	 * L_INMEM will already be set and we don't need to do anything.
    246       1.21   thorpej 	 *
    247       1.61       chs 	 * Note the kernel stack gets read/write accesses right off the bat.
    248        1.6       mrg 	 */
    249       1.61       chs 
    250       1.62   thorpej 	if ((l2->l_flag & L_INMEM) == 0) {
    251       1.61       chs 		error = uvm_fault_wire(kernel_map, (vaddr_t)up,
    252  1.92.10.1      elad 		    (vaddr_t)up + USPACE, VM_PROT_READ | VM_PROT_WRITE, 0);
    253       1.61       chs 		if (error)
    254       1.62   thorpej 			panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
    255       1.67       scw #ifdef PMAP_UAREA
    256       1.67       scw 		/* Tell the pmap this is a u-area mapping */
    257       1.67       scw 		PMAP_UAREA((vaddr_t)up);
    258       1.67       scw #endif
    259       1.62   thorpej 		l2->l_flag |= L_INMEM;
    260       1.61       chs 	}
    261       1.59      yamt 
    262       1.59      yamt #ifdef KSTACK_CHECK_MAGIC
    263       1.59      yamt 	/*
    264       1.59      yamt 	 * fill stack with magic number
    265       1.59      yamt 	 */
    266       1.63      yamt 	kstack_setup_magic(l2);
    267       1.59      yamt #endif
    268        1.6       mrg 
    269        1.6       mrg 	/*
    270       1.62   thorpej 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
    271       1.62   thorpej  	 * to run.  If this is a normal user fork, the child will exit
    272       1.34   thorpej 	 * directly to user mode via child_return() on its first time
    273       1.34   thorpej 	 * slice and will not return here.  If this is a kernel thread,
    274       1.34   thorpej 	 * the specified entry point will be executed.
    275        1.6       mrg 	 */
    276       1.62   thorpej 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
    277       1.14   thorpej }
    278       1.14   thorpej 
    279       1.14   thorpej /*
    280       1.60       chs  * uvm_uarea_alloc: allocate a u-area
    281       1.60       chs  */
    282       1.60       chs 
    283       1.61       chs boolean_t
    284       1.61       chs uvm_uarea_alloc(vaddr_t *uaddrp)
    285       1.60       chs {
    286       1.60       chs 	vaddr_t uaddr;
    287       1.60       chs 
    288       1.60       chs #ifndef USPACE_ALIGN
    289       1.60       chs #define USPACE_ALIGN    0
    290       1.60       chs #endif
    291       1.60       chs 
    292       1.62   thorpej 	simple_lock(&uvm_uareas_slock);
    293       1.75  jdolecek 	if (uvm_nuarea > 0) {
    294       1.75  jdolecek 		uaddr = (vaddr_t)uvm_uareas;
    295       1.60       chs 		uvm_uareas = *(void **)uvm_uareas;
    296       1.60       chs 		uvm_nuarea--;
    297       1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    298       1.61       chs 		*uaddrp = uaddr;
    299       1.61       chs 		return TRUE;
    300       1.60       chs 	} else {
    301       1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    302       1.84      yamt 		*uaddrp = uvm_km_alloc(kernel_map, USPACE, USPACE_ALIGN,
    303       1.84      yamt 		    UVM_KMF_PAGEABLE);
    304       1.61       chs 		return FALSE;
    305       1.60       chs 	}
    306       1.60       chs }
    307       1.60       chs 
    308       1.60       chs /*
    309       1.75  jdolecek  * uvm_uarea_free: free a u-area; never blocks
    310       1.75  jdolecek  */
    311       1.75  jdolecek 
    312       1.92     perry static inline void
    313       1.75  jdolecek uvm_uarea_free(vaddr_t uaddr)
    314       1.75  jdolecek {
    315       1.75  jdolecek 	simple_lock(&uvm_uareas_slock);
    316       1.75  jdolecek 	*(void **)uaddr = uvm_uareas;
    317       1.75  jdolecek 	uvm_uareas = (void *)uaddr;
    318       1.75  jdolecek 	uvm_nuarea++;
    319       1.75  jdolecek 	simple_unlock(&uvm_uareas_slock);
    320       1.75  jdolecek }
    321       1.75  jdolecek 
    322       1.75  jdolecek /*
    323       1.75  jdolecek  * uvm_uarea_drain: return memory of u-areas over limit
    324       1.75  jdolecek  * back to system
    325       1.60       chs  */
    326       1.60       chs 
    327       1.60       chs void
    328       1.75  jdolecek uvm_uarea_drain(boolean_t empty)
    329       1.60       chs {
    330       1.75  jdolecek 	int leave = empty ? 0 : UVM_NUAREA_MAX;
    331       1.75  jdolecek 	vaddr_t uaddr;
    332       1.75  jdolecek 
    333       1.75  jdolecek 	if (uvm_nuarea <= leave)
    334       1.75  jdolecek 		return;
    335       1.60       chs 
    336       1.62   thorpej 	simple_lock(&uvm_uareas_slock);
    337       1.75  jdolecek 	while(uvm_nuarea > leave) {
    338       1.75  jdolecek 		uaddr = (vaddr_t)uvm_uareas;
    339       1.75  jdolecek 		uvm_uareas = *(void **)uvm_uareas;
    340       1.75  jdolecek 		uvm_nuarea--;
    341       1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    342       1.84      yamt 		uvm_km_free(kernel_map, uaddr, USPACE, UVM_KMF_PAGEABLE);
    343       1.75  jdolecek 		simple_lock(&uvm_uareas_slock);
    344       1.60       chs 	}
    345       1.75  jdolecek 	simple_unlock(&uvm_uareas_slock);
    346       1.60       chs }
    347       1.60       chs 
    348       1.60       chs /*
    349       1.80        pk  * uvm_exit: exit a virtual address space
    350       1.80        pk  *
    351       1.80        pk  * - the process passed to us is a dead (pre-zombie) process; we
    352       1.80        pk  *   are running on a different context now (the reaper).
    353       1.80        pk  * - borrow proc0's address space because freeing the vmspace
    354       1.80        pk  *   of the dead process may block.
    355       1.80        pk  */
    356       1.80        pk 
    357       1.80        pk void
    358       1.89   thorpej uvm_proc_exit(struct proc *p)
    359       1.80        pk {
    360       1.80        pk 	struct lwp *l = curlwp; /* XXX */
    361       1.80        pk 	struct vmspace *ovm;
    362       1.80        pk 
    363       1.80        pk 	KASSERT(p == l->l_proc);
    364       1.80        pk 	ovm = p->p_vmspace;
    365       1.80        pk 
    366       1.80        pk 	/*
    367       1.80        pk 	 * borrow proc0's address space.
    368       1.80        pk 	 */
    369       1.80        pk 	pmap_deactivate(l);
    370       1.80        pk 	p->p_vmspace = proc0.p_vmspace;
    371       1.80        pk 	pmap_activate(l);
    372       1.80        pk 
    373       1.80        pk 	uvmspace_free(ovm);
    374       1.80        pk }
    375       1.80        pk 
    376       1.80        pk void
    377       1.80        pk uvm_lwp_exit(struct lwp *l)
    378       1.80        pk {
    379       1.80        pk 	vaddr_t va = (vaddr_t)l->l_addr;
    380       1.80        pk 
    381       1.80        pk 	l->l_flag &= ~L_INMEM;
    382       1.80        pk 	uvm_uarea_free(va);
    383       1.80        pk 	l->l_addr = NULL;
    384       1.80        pk }
    385       1.80        pk 
    386       1.80        pk /*
    387        1.1       mrg  * uvm_init_limit: init per-process VM limits
    388        1.1       mrg  *
    389        1.1       mrg  * - called for process 0 and then inherited by all others.
    390        1.1       mrg  */
    391       1.60       chs 
    392        1.6       mrg void
    393       1.89   thorpej uvm_init_limits(struct proc *p)
    394        1.6       mrg {
    395        1.6       mrg 
    396        1.6       mrg 	/*
    397        1.6       mrg 	 * Set up the initial limits on process VM.  Set the maximum
    398        1.6       mrg 	 * resident set size to be all of (reasonably) available memory.
    399        1.6       mrg 	 * This causes any single, large process to start random page
    400        1.6       mrg 	 * replacement once it fills memory.
    401        1.6       mrg 	 */
    402        1.6       mrg 
    403        1.6       mrg 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
    404       1.79        pk 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
    405        1.6       mrg 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
    406       1.79        pk 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
    407        1.6       mrg 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
    408        1.1       mrg }
    409        1.1       mrg 
    410        1.1       mrg #ifdef DEBUG
    411        1.1       mrg int	enableswap = 1;
    412        1.1       mrg int	swapdebug = 0;
    413        1.1       mrg #define	SDB_FOLLOW	1
    414        1.1       mrg #define SDB_SWAPIN	2
    415        1.1       mrg #define SDB_SWAPOUT	4
    416        1.1       mrg #endif
    417        1.1       mrg 
    418        1.1       mrg /*
    419        1.1       mrg  * uvm_swapin: swap in a process's u-area.
    420        1.1       mrg  */
    421        1.1       mrg 
    422        1.6       mrg void
    423       1.89   thorpej uvm_swapin(struct lwp *l)
    424        1.6       mrg {
    425       1.13       eeh 	vaddr_t addr;
    426       1.52       chs 	int s, error;
    427        1.6       mrg 
    428       1.62   thorpej 	addr = (vaddr_t)l->l_addr;
    429       1.62   thorpej 	/* make L_INMEM true */
    430  1.92.10.1      elad 	error = uvm_fault_wire(kernel_map, addr, addr + USPACE,
    431  1.92.10.1      elad 	    VM_PROT_READ | VM_PROT_WRITE, 0);
    432       1.52       chs 	if (error) {
    433       1.52       chs 		panic("uvm_swapin: rewiring stack failed: %d", error);
    434       1.52       chs 	}
    435        1.6       mrg 
    436        1.6       mrg 	/*
    437        1.6       mrg 	 * Some architectures need to be notified when the user area has
    438        1.6       mrg 	 * moved to new physical page(s) (e.g.  see mips/mips/vm_machdep.c).
    439        1.6       mrg 	 */
    440       1.62   thorpej 	cpu_swapin(l);
    441       1.41     enami 	SCHED_LOCK(s);
    442       1.62   thorpej 	if (l->l_stat == LSRUN)
    443       1.62   thorpej 		setrunqueue(l);
    444       1.62   thorpej 	l->l_flag |= L_INMEM;
    445       1.41     enami 	SCHED_UNLOCK(s);
    446       1.62   thorpej 	l->l_swtime = 0;
    447        1.6       mrg 	++uvmexp.swapins;
    448        1.1       mrg }
    449        1.1       mrg 
    450        1.1       mrg /*
    451        1.1       mrg  * uvm_scheduler: process zero main loop
    452        1.1       mrg  *
    453        1.1       mrg  * - attempt to swapin every swaped-out, runnable process in order of
    454        1.1       mrg  *	priority.
    455        1.1       mrg  * - if not enough memory, wake the pagedaemon and let it clear space.
    456        1.1       mrg  */
    457        1.1       mrg 
    458        1.6       mrg void
    459       1.89   thorpej uvm_scheduler(void)
    460        1.1       mrg {
    461       1.62   thorpej 	struct lwp *l, *ll;
    462       1.32  augustss 	int pri;
    463        1.6       mrg 	int ppri;
    464        1.1       mrg 
    465        1.1       mrg loop:
    466        1.1       mrg #ifdef DEBUG
    467        1.6       mrg 	while (!enableswap)
    468       1.43       chs 		tsleep(&proc0, PVM, "noswap", 0);
    469        1.1       mrg #endif
    470       1.62   thorpej 	ll = NULL;		/* process to choose */
    471        1.6       mrg 	ppri = INT_MIN;	/* its priority */
    472       1.29   thorpej 	proclist_lock_read();
    473        1.6       mrg 
    474       1.62   thorpej 	LIST_FOREACH(l, &alllwp, l_list) {
    475        1.6       mrg 		/* is it a runnable swapped out process? */
    476       1.62   thorpej 		if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
    477       1.62   thorpej 			pri = l->l_swtime + l->l_slptime -
    478       1.62   thorpej 			    (l->l_proc->p_nice - NZERO) * 8;
    479        1.6       mrg 			if (pri > ppri) {   /* higher priority?  remember it. */
    480       1.62   thorpej 				ll = l;
    481        1.6       mrg 				ppri = pri;
    482        1.6       mrg 			}
    483        1.6       mrg 		}
    484        1.6       mrg 	}
    485       1.39  sommerfe 	/*
    486       1.39  sommerfe 	 * XXXSMP: possible unlock/sleep race between here and the
    487       1.39  sommerfe 	 * "scheduler" tsleep below..
    488       1.39  sommerfe 	 */
    489       1.28   thorpej 	proclist_unlock_read();
    490        1.1       mrg 
    491        1.1       mrg #ifdef DEBUG
    492        1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    493       1.62   thorpej 		printf("scheduler: running, procp %p pri %d\n", ll, ppri);
    494        1.1       mrg #endif
    495        1.6       mrg 	/*
    496        1.6       mrg 	 * Nothing to do, back to sleep
    497        1.6       mrg 	 */
    498       1.62   thorpej 	if ((l = ll) == NULL) {
    499       1.43       chs 		tsleep(&proc0, PVM, "scheduler", 0);
    500        1.6       mrg 		goto loop;
    501        1.6       mrg 	}
    502        1.6       mrg 
    503        1.6       mrg 	/*
    504        1.6       mrg 	 * we have found swapped out process which we would like to bring
    505        1.6       mrg 	 * back in.
    506        1.6       mrg 	 *
    507        1.6       mrg 	 * XXX: this part is really bogus cuz we could deadlock on memory
    508        1.6       mrg 	 * despite our feeble check
    509        1.6       mrg 	 */
    510        1.6       mrg 	if (uvmexp.free > atop(USPACE)) {
    511        1.1       mrg #ifdef DEBUG
    512        1.6       mrg 		if (swapdebug & SDB_SWAPIN)
    513        1.6       mrg 			printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
    514       1.62   thorpej 	     l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
    515        1.1       mrg #endif
    516       1.62   thorpej 		uvm_swapin(l);
    517        1.6       mrg 		goto loop;
    518        1.6       mrg 	}
    519        1.6       mrg 	/*
    520        1.6       mrg 	 * not enough memory, jab the pageout daemon and wait til the coast
    521        1.6       mrg 	 * is clear
    522        1.6       mrg 	 */
    523        1.1       mrg #ifdef DEBUG
    524        1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    525        1.6       mrg 		printf("scheduler: no room for pid %d(%s), free %d\n",
    526       1.62   thorpej 	   l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
    527        1.1       mrg #endif
    528        1.6       mrg 	uvm_wait("schedpwait");
    529        1.1       mrg #ifdef DEBUG
    530        1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    531        1.6       mrg 		printf("scheduler: room again, free %d\n", uvmexp.free);
    532        1.1       mrg #endif
    533        1.6       mrg 	goto loop;
    534        1.1       mrg }
    535        1.1       mrg 
    536        1.1       mrg /*
    537       1.62   thorpej  * swappable: is LWP "l" swappable?
    538        1.1       mrg  */
    539        1.1       mrg 
    540       1.62   thorpej #define	swappable(l)							\
    541       1.62   thorpej 	(((l)->l_flag & (L_INMEM)) &&					\
    542       1.62   thorpej 	 ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) &&	\
    543       1.62   thorpej 	 (l)->l_holdcnt == 0)
    544        1.1       mrg 
    545        1.1       mrg /*
    546        1.1       mrg  * swapout_threads: find threads that can be swapped and unwire their
    547        1.1       mrg  *	u-areas.
    548        1.1       mrg  *
    549        1.1       mrg  * - called by the pagedaemon
    550        1.1       mrg  * - try and swap at least one processs
    551        1.1       mrg  * - processes that are sleeping or stopped for maxslp or more seconds
    552        1.1       mrg  *   are swapped... otherwise the longest-sleeping or stopped process
    553        1.1       mrg  *   is swapped, otherwise the longest resident process...
    554        1.1       mrg  */
    555       1.60       chs 
    556        1.6       mrg void
    557       1.89   thorpej uvm_swapout_threads(void)
    558        1.1       mrg {
    559       1.62   thorpej 	struct lwp *l;
    560       1.62   thorpej 	struct lwp *outl, *outl2;
    561        1.6       mrg 	int outpri, outpri2;
    562        1.6       mrg 	int didswap = 0;
    563       1.48       chs 	extern int maxslp;
    564        1.6       mrg 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
    565        1.1       mrg 
    566        1.1       mrg #ifdef DEBUG
    567        1.6       mrg 	if (!enableswap)
    568        1.6       mrg 		return;
    569        1.1       mrg #endif
    570        1.1       mrg 
    571        1.6       mrg 	/*
    572       1.62   thorpej 	 * outl/outpri  : stop/sleep thread with largest sleeptime < maxslp
    573       1.62   thorpej 	 * outl2/outpri2: the longest resident thread (its swap time)
    574        1.6       mrg 	 */
    575       1.62   thorpej 	outl = outl2 = NULL;
    576        1.6       mrg 	outpri = outpri2 = 0;
    577       1.29   thorpej 	proclist_lock_read();
    578       1.62   thorpej 	LIST_FOREACH(l, &alllwp, l_list) {
    579       1.81      yamt 		KASSERT(l->l_proc != NULL);
    580       1.62   thorpej 		if (!swappable(l))
    581        1.6       mrg 			continue;
    582       1.62   thorpej 		switch (l->l_stat) {
    583       1.68        cl 		case LSONPROC:
    584       1.69        cl 			continue;
    585       1.69        cl 
    586       1.62   thorpej 		case LSRUN:
    587       1.62   thorpej 			if (l->l_swtime > outpri2) {
    588       1.62   thorpej 				outl2 = l;
    589       1.62   thorpej 				outpri2 = l->l_swtime;
    590        1.6       mrg 			}
    591        1.6       mrg 			continue;
    592       1.48       chs 
    593       1.62   thorpej 		case LSSLEEP:
    594       1.62   thorpej 		case LSSTOP:
    595       1.62   thorpej 			if (l->l_slptime >= maxslp) {
    596       1.62   thorpej 				uvm_swapout(l);
    597        1.6       mrg 				didswap++;
    598       1.62   thorpej 			} else if (l->l_slptime > outpri) {
    599       1.62   thorpej 				outl = l;
    600       1.62   thorpej 				outpri = l->l_slptime;
    601        1.6       mrg 			}
    602        1.6       mrg 			continue;
    603        1.6       mrg 		}
    604        1.6       mrg 	}
    605       1.28   thorpej 	proclist_unlock_read();
    606        1.6       mrg 
    607        1.6       mrg 	/*
    608        1.6       mrg 	 * If we didn't get rid of any real duds, toss out the next most
    609        1.6       mrg 	 * likely sleeping/stopped or running candidate.  We only do this
    610        1.6       mrg 	 * if we are real low on memory since we don't gain much by doing
    611        1.6       mrg 	 * it (USPACE bytes).
    612        1.6       mrg 	 */
    613        1.6       mrg 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
    614       1.62   thorpej 		if ((l = outl) == NULL)
    615       1.62   thorpej 			l = outl2;
    616        1.1       mrg #ifdef DEBUG
    617        1.6       mrg 		if (swapdebug & SDB_SWAPOUT)
    618       1.62   thorpej 			printf("swapout_threads: no duds, try procp %p\n", l);
    619        1.1       mrg #endif
    620       1.62   thorpej 		if (l)
    621       1.62   thorpej 			uvm_swapout(l);
    622        1.6       mrg 	}
    623        1.1       mrg }
    624        1.1       mrg 
    625        1.1       mrg /*
    626       1.62   thorpej  * uvm_swapout: swap out lwp "l"
    627        1.1       mrg  *
    628       1.48       chs  * - currently "swapout" means "unwire U-area" and "pmap_collect()"
    629        1.1       mrg  *   the pmap.
    630        1.1       mrg  * - XXXCDC: should deactivate all process' private anonymous memory
    631        1.1       mrg  */
    632        1.1       mrg 
    633        1.6       mrg static void
    634       1.89   thorpej uvm_swapout(struct lwp *l)
    635        1.1       mrg {
    636       1.13       eeh 	vaddr_t addr;
    637        1.6       mrg 	int s;
    638       1.62   thorpej 	struct proc *p = l->l_proc;
    639        1.1       mrg 
    640        1.1       mrg #ifdef DEBUG
    641        1.6       mrg 	if (swapdebug & SDB_SWAPOUT)
    642       1.62   thorpej 		printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
    643       1.62   thorpej 	   p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
    644       1.62   thorpej 	   l->l_slptime, uvmexp.free);
    645        1.1       mrg #endif
    646        1.1       mrg 
    647        1.6       mrg 	/*
    648        1.6       mrg 	 * Mark it as (potentially) swapped out.
    649        1.6       mrg 	 */
    650       1.41     enami 	SCHED_LOCK(s);
    651       1.69        cl 	if (l->l_stat == LSONPROC) {
    652       1.69        cl 		KDASSERT(l->l_cpu != curcpu());
    653       1.68        cl 		SCHED_UNLOCK(s);
    654       1.68        cl 		return;
    655       1.68        cl 	}
    656       1.62   thorpej 	l->l_flag &= ~L_INMEM;
    657       1.62   thorpej 	if (l->l_stat == LSRUN)
    658       1.62   thorpej 		remrunqueue(l);
    659       1.41     enami 	SCHED_UNLOCK(s);
    660       1.62   thorpej 	l->l_swtime = 0;
    661       1.53       chs 	p->p_stats->p_ru.ru_nswap++;
    662        1.6       mrg 	++uvmexp.swapouts;
    663       1.68        cl 
    664       1.68        cl 	/*
    665       1.68        cl 	 * Do any machine-specific actions necessary before swapout.
    666       1.68        cl 	 * This can include saving floating point state, etc.
    667       1.68        cl 	 */
    668       1.68        cl 	cpu_swapout(l);
    669       1.43       chs 
    670       1.43       chs 	/*
    671       1.43       chs 	 * Unwire the to-be-swapped process's user struct and kernel stack.
    672       1.43       chs 	 */
    673       1.62   thorpej 	addr = (vaddr_t)l->l_addr;
    674       1.62   thorpej 	uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
    675       1.43       chs 	pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
    676        1.1       mrg }
    677        1.1       mrg 
    678       1.56   thorpej /*
    679       1.56   thorpej  * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
    680       1.56   thorpej  * a core file.
    681       1.56   thorpej  */
    682       1.56   thorpej 
    683       1.56   thorpej int
    684       1.89   thorpej uvm_coredump_walkmap(struct proc *p, void *iocookie,
    685       1.89   thorpej     int (*func)(struct proc *, void *, struct uvm_coredump_state *),
    686       1.89   thorpej     void *cookie)
    687       1.56   thorpej {
    688       1.56   thorpej 	struct uvm_coredump_state state;
    689       1.56   thorpej 	struct vmspace *vm = p->p_vmspace;
    690       1.56   thorpej 	struct vm_map *map = &vm->vm_map;
    691       1.56   thorpej 	struct vm_map_entry *entry;
    692       1.56   thorpej 	int error;
    693       1.56   thorpej 
    694       1.64    atatat 	entry = NULL;
    695       1.64    atatat 	vm_map_lock_read(map);
    696       1.87      matt 	state.end = 0;
    697       1.64    atatat 	for (;;) {
    698       1.64    atatat 		if (entry == NULL)
    699       1.64    atatat 			entry = map->header.next;
    700       1.64    atatat 		else if (!uvm_map_lookup_entry(map, state.end, &entry))
    701       1.64    atatat 			entry = entry->next;
    702       1.64    atatat 		if (entry == &map->header)
    703       1.64    atatat 			break;
    704       1.64    atatat 
    705       1.56   thorpej 		state.cookie = cookie;
    706       1.86      matt 		if (state.end > entry->start) {
    707       1.86      matt 			state.start = state.end;
    708       1.86      matt 		} else {
    709       1.86      matt 			state.start = entry->start;
    710       1.86      matt 		}
    711       1.86      matt 		state.realend = entry->end;
    712       1.56   thorpej 		state.end = entry->end;
    713       1.56   thorpej 		state.prot = entry->protection;
    714       1.56   thorpej 		state.flags = 0;
    715       1.56   thorpej 
    716       1.82       chs 		/*
    717       1.82       chs 		 * Dump the region unless one of the following is true:
    718       1.82       chs 		 *
    719       1.82       chs 		 * (1) the region has neither object nor amap behind it
    720       1.82       chs 		 *     (ie. it has never been accessed).
    721       1.82       chs 		 *
    722       1.82       chs 		 * (2) the region has no amap and is read-only
    723       1.82       chs 		 *     (eg. an executable text section).
    724       1.82       chs 		 *
    725       1.82       chs 		 * (3) the region's object is a device.
    726       1.85   nathanw 		 *
    727       1.85   nathanw 		 * (4) the region is unreadable by the process.
    728       1.82       chs 		 */
    729       1.56   thorpej 
    730       1.82       chs 		KASSERT(!UVM_ET_ISSUBMAP(entry));
    731       1.82       chs 		KASSERT(state.start < VM_MAXUSER_ADDRESS);
    732       1.82       chs 		KASSERT(state.end <= VM_MAXUSER_ADDRESS);
    733       1.82       chs 		if (entry->object.uvm_obj == NULL &&
    734       1.82       chs 		    entry->aref.ar_amap == NULL) {
    735       1.86      matt 			state.realend = state.start;
    736       1.86      matt 		} else if ((entry->protection & VM_PROT_WRITE) == 0 &&
    737       1.82       chs 		    entry->aref.ar_amap == NULL) {
    738       1.86      matt 			state.realend = state.start;
    739       1.86      matt 		} else if (entry->object.uvm_obj != NULL &&
    740       1.82       chs 		    UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
    741       1.86      matt 			state.realend = state.start;
    742       1.86      matt 		} else if ((entry->protection & VM_PROT_READ) == 0) {
    743       1.86      matt 			state.realend = state.start;
    744       1.86      matt 		} else {
    745       1.86      matt 			if (state.start >= (vaddr_t)vm->vm_maxsaddr)
    746       1.86      matt 				state.flags |= UVM_COREDUMP_STACK;
    747       1.86      matt 
    748       1.86      matt 			/*
    749       1.86      matt 			 * If this an anonymous entry, only dump instantiated
    750       1.86      matt 			 * pages.
    751       1.86      matt 			 */
    752       1.86      matt 			if (entry->object.uvm_obj == NULL) {
    753       1.86      matt 				vaddr_t end;
    754       1.86      matt 
    755       1.86      matt 				amap_lock(entry->aref.ar_amap);
    756       1.86      matt 				for (end = state.start;
    757       1.86      matt 				     end < state.end; end += PAGE_SIZE) {
    758       1.86      matt 					struct vm_anon *anon;
    759       1.86      matt 					anon = amap_lookup(&entry->aref,
    760       1.86      matt 					    end - entry->start);
    761       1.86      matt 					/*
    762       1.86      matt 					 * If we have already encountered an
    763       1.86      matt 					 * uninstantiated page, stop at the
    764       1.86      matt 					 * first instantied page.
    765       1.86      matt 					 */
    766       1.86      matt 					if (anon != NULL &&
    767       1.86      matt 					    state.realend != state.end) {
    768       1.86      matt 						state.end = end;
    769       1.86      matt 						break;
    770       1.86      matt 					}
    771       1.86      matt 
    772       1.86      matt 					/*
    773       1.86      matt 					 * If this page is the first
    774       1.86      matt 					 * uninstantiated page, mark this as
    775       1.86      matt 					 * the real ending point.  Continue to
    776       1.86      matt 					 * counting uninstantiated pages.
    777       1.86      matt 					 */
    778       1.86      matt 					if (anon == NULL &&
    779       1.86      matt 					    state.realend == state.end) {
    780       1.86      matt 						state.realend = end;
    781       1.86      matt 					}
    782       1.86      matt 				}
    783       1.86      matt 				amap_unlock(entry->aref.ar_amap);
    784       1.86      matt 			}
    785       1.82       chs 		}
    786       1.86      matt 
    787       1.56   thorpej 
    788       1.64    atatat 		vm_map_unlock_read(map);
    789       1.88      matt 		error = (*func)(p, iocookie, &state);
    790       1.56   thorpej 		if (error)
    791       1.56   thorpej 			return (error);
    792       1.64    atatat 		vm_map_lock_read(map);
    793       1.56   thorpej 	}
    794       1.64    atatat 	vm_map_unlock_read(map);
    795       1.56   thorpej 
    796       1.56   thorpej 	return (0);
    797       1.56   thorpej }
    798