Home | History | Annotate | Line # | Download | only in uvm
uvm_glue.c revision 1.71
      1  1.71  jdolecek /*	$NetBSD: uvm_glue.c,v 1.71 2003/11/02 16:53:43 jdolecek 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  1.71  jdolecek __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.71 2003/11/02 16:53:43 jdolecek 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  1.62   thorpej static void uvm_swapout __P((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   1.1       mrg /*
    103   1.1       mrg  * XXXCDC: do these really belong here?
    104   1.1       mrg  */
    105   1.1       mrg 
    106   1.1       mrg int readbuffers = 0;		/* allow KGDB to read kern buffer pool */
    107   1.1       mrg 				/* XXX: see uvm_kernacc */
    108   1.1       mrg 
    109  1.28   thorpej 
    110  1.28   thorpej /*
    111   1.1       mrg  * uvm_kernacc: can the kernel access a region of memory
    112   1.1       mrg  *
    113   1.1       mrg  * - called from malloc [DIAGNOSTIC], and /dev/kmem driver (mem.c)
    114   1.1       mrg  */
    115   1.1       mrg 
    116   1.6       mrg boolean_t
    117   1.6       mrg uvm_kernacc(addr, len, rw)
    118   1.6       mrg 	caddr_t addr;
    119  1.11    kleink 	size_t len;
    120  1.11    kleink 	int rw;
    121   1.6       mrg {
    122   1.6       mrg 	boolean_t rv;
    123  1.13       eeh 	vaddr_t saddr, eaddr;
    124   1.6       mrg 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
    125   1.6       mrg 
    126  1.31    kleink 	saddr = trunc_page((vaddr_t)addr);
    127  1.43       chs 	eaddr = round_page((vaddr_t)addr + len);
    128   1.6       mrg 	vm_map_lock_read(kernel_map);
    129   1.6       mrg 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
    130   1.6       mrg 	vm_map_unlock_read(kernel_map);
    131   1.6       mrg 
    132   1.6       mrg 	/*
    133   1.6       mrg 	 * XXX there are still some things (e.g. the buffer cache) that
    134   1.6       mrg 	 * are managed behind the VM system's back so even though an
    135   1.6       mrg 	 * address is accessible in the mind of the VM system, there may
    136   1.6       mrg 	 * not be physical pages where the VM thinks there is.  This can
    137   1.6       mrg 	 * lead to bogus allocation of pages in the kernel address space
    138   1.6       mrg 	 * or worse, inconsistencies at the pmap level.  We only worry
    139   1.6       mrg 	 * about the buffer cache for now.
    140   1.6       mrg 	 */
    141  1.13       eeh 	if (!readbuffers && rv && (eaddr > (vaddr_t)buffers &&
    142  1.13       eeh 			     saddr < (vaddr_t)buffers + MAXBSIZE * nbuf))
    143   1.6       mrg 		rv = FALSE;
    144   1.6       mrg 	return(rv);
    145   1.1       mrg }
    146   1.1       mrg 
    147   1.1       mrg /*
    148   1.1       mrg  * uvm_useracc: can the user access it?
    149   1.1       mrg  *
    150   1.1       mrg  * - called from physio() and sys___sysctl().
    151   1.1       mrg  */
    152   1.1       mrg 
    153   1.6       mrg boolean_t
    154   1.6       mrg uvm_useracc(addr, len, rw)
    155   1.6       mrg 	caddr_t addr;
    156  1.11    kleink 	size_t len;
    157  1.11    kleink 	int rw;
    158   1.1       mrg {
    159  1.50       chs 	struct vm_map *map;
    160   1.6       mrg 	boolean_t rv;
    161   1.6       mrg 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
    162   1.1       mrg 
    163  1.25   thorpej 	/* XXX curproc */
    164  1.25   thorpej 	map = &curproc->p_vmspace->vm_map;
    165  1.25   thorpej 
    166  1.25   thorpej 	vm_map_lock_read(map);
    167  1.31    kleink 	rv = uvm_map_checkprot(map, trunc_page((vaddr_t)addr),
    168  1.43       chs 	    round_page((vaddr_t)addr + len), prot);
    169  1.25   thorpej 	vm_map_unlock_read(map);
    170  1.25   thorpej 
    171   1.6       mrg 	return(rv);
    172   1.1       mrg }
    173   1.1       mrg 
    174   1.1       mrg #ifdef KGDB
    175   1.1       mrg /*
    176   1.1       mrg  * Change protections on kernel pages from addr to addr+len
    177   1.1       mrg  * (presumably so debugger can plant a breakpoint).
    178   1.1       mrg  *
    179   1.1       mrg  * We force the protection change at the pmap level.  If we were
    180   1.1       mrg  * to use vm_map_protect a change to allow writing would be lazily-
    181   1.1       mrg  * applied meaning we would still take a protection fault, something
    182   1.1       mrg  * we really don't want to do.  It would also fragment the kernel
    183   1.1       mrg  * map unnecessarily.  We cannot use pmap_protect since it also won't
    184   1.1       mrg  * enforce a write-enable request.  Using pmap_enter is the only way
    185   1.1       mrg  * we can ensure the change takes place properly.
    186   1.1       mrg  */
    187   1.6       mrg void
    188   1.6       mrg uvm_chgkprot(addr, len, rw)
    189  1.32  augustss 	caddr_t addr;
    190  1.11    kleink 	size_t len;
    191  1.11    kleink 	int rw;
    192   1.6       mrg {
    193   1.6       mrg 	vm_prot_t prot;
    194  1.13       eeh 	paddr_t pa;
    195  1.13       eeh 	vaddr_t sva, eva;
    196   1.6       mrg 
    197   1.6       mrg 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
    198  1.31    kleink 	eva = round_page((vaddr_t)addr + len);
    199  1.31    kleink 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
    200   1.6       mrg 		/*
    201   1.6       mrg 		 * Extract physical address for the page.
    202   1.6       mrg 		 */
    203  1.27   thorpej 		if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
    204   1.6       mrg 			panic("chgkprot: invalid page");
    205  1.30   thorpej 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
    206   1.6       mrg 	}
    207  1.51     chris 	pmap_update(pmap_kernel());
    208   1.1       mrg }
    209   1.1       mrg #endif
    210   1.1       mrg 
    211   1.1       mrg /*
    212  1.52       chs  * uvm_vslock: wire user memory for I/O
    213   1.1       mrg  *
    214   1.1       mrg  * - called from physio and sys___sysctl
    215   1.1       mrg  * - XXXCDC: consider nuking this (or making it a macro?)
    216   1.1       mrg  */
    217   1.1       mrg 
    218  1.26   thorpej int
    219  1.22   thorpej uvm_vslock(p, addr, len, access_type)
    220   1.9   thorpej 	struct proc *p;
    221   1.6       mrg 	caddr_t	addr;
    222  1.11    kleink 	size_t	len;
    223  1.22   thorpej 	vm_prot_t access_type;
    224   1.1       mrg {
    225  1.50       chs 	struct vm_map *map;
    226  1.26   thorpej 	vaddr_t start, end;
    227  1.45       chs 	int error;
    228  1.26   thorpej 
    229  1.26   thorpej 	map = &p->p_vmspace->vm_map;
    230  1.31    kleink 	start = trunc_page((vaddr_t)addr);
    231  1.31    kleink 	end = round_page((vaddr_t)addr + len);
    232  1.57       chs 	error = uvm_fault_wire(map, start, end, VM_FAULT_WIRE, access_type);
    233  1.45       chs 	return error;
    234   1.1       mrg }
    235   1.1       mrg 
    236   1.1       mrg /*
    237  1.52       chs  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
    238   1.1       mrg  *
    239   1.1       mrg  * - called from physio and sys___sysctl
    240   1.1       mrg  * - XXXCDC: consider nuking this (or making it a macro?)
    241   1.1       mrg  */
    242   1.1       mrg 
    243   1.6       mrg void
    244   1.9   thorpej uvm_vsunlock(p, addr, len)
    245   1.9   thorpej 	struct proc *p;
    246   1.6       mrg 	caddr_t	addr;
    247  1.11    kleink 	size_t	len;
    248   1.1       mrg {
    249  1.43       chs 	uvm_fault_unwire(&p->p_vmspace->vm_map, trunc_page((vaddr_t)addr),
    250  1.43       chs 		round_page((vaddr_t)addr + len));
    251   1.1       mrg }
    252   1.1       mrg 
    253   1.1       mrg /*
    254  1.62   thorpej  * uvm_proc_fork: fork a virtual address space
    255   1.1       mrg  *
    256   1.1       mrg  * - the address space is copied as per parent map's inherit values
    257  1.62   thorpej  */
    258  1.62   thorpej void
    259  1.62   thorpej uvm_proc_fork(p1, p2, shared)
    260  1.62   thorpej 	struct proc *p1, *p2;
    261  1.62   thorpej 	boolean_t shared;
    262  1.62   thorpej {
    263  1.62   thorpej 
    264  1.62   thorpej 	if (shared == TRUE) {
    265  1.62   thorpej 		p2->p_vmspace = NULL;
    266  1.62   thorpej 		uvmspace_share(p1, p2);
    267  1.62   thorpej 	} else {
    268  1.62   thorpej 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
    269  1.62   thorpej 	}
    270  1.62   thorpej 
    271  1.62   thorpej 	cpu_proc_fork(p1, p2);
    272  1.62   thorpej }
    273  1.62   thorpej 
    274  1.62   thorpej 
    275  1.62   thorpej /*
    276  1.62   thorpej  * uvm_lwp_fork: fork a thread
    277  1.62   thorpej  *
    278   1.1       mrg  * - a new "user" structure is allocated for the child process
    279   1.1       mrg  *	[filled in by MD layer...]
    280  1.20   thorpej  * - if specified, the child gets a new user stack described by
    281  1.20   thorpej  *	stack and stacksize
    282   1.1       mrg  * - NOTE: the kernel stack may be at a different location in the child
    283   1.1       mrg  *	process, and thus addresses of automatic variables may be invalid
    284  1.62   thorpej  *	after cpu_lwp_fork returns in the child process.  We do nothing here
    285  1.62   thorpej  *	after cpu_lwp_fork returns.
    286   1.1       mrg  * - XXXCDC: we need a way for this to return a failure value rather
    287   1.1       mrg  *   than just hang
    288   1.1       mrg  */
    289   1.6       mrg void
    290  1.62   thorpej uvm_lwp_fork(l1, l2, stack, stacksize, func, arg)
    291  1.62   thorpej 	struct lwp *l1, *l2;
    292  1.20   thorpej 	void *stack;
    293  1.20   thorpej 	size_t stacksize;
    294  1.34   thorpej 	void (*func) __P((void *));
    295  1.34   thorpej 	void *arg;
    296   1.6       mrg {
    297  1.62   thorpej 	struct user *up = l2->l_addr;
    298  1.45       chs 	int error;
    299   1.6       mrg 
    300   1.6       mrg 	/*
    301   1.7   thorpej 	 * Wire down the U-area for the process, which contains the PCB
    302  1.62   thorpej 	 * and the kernel stack.  Wired state is stored in l->l_flag's
    303  1.62   thorpej 	 * L_INMEM bit rather than in the vm_map_entry's wired count
    304  1.61       chs 	 * to prevent kernel_map fragmentation.  If we reused a cached U-area,
    305  1.62   thorpej 	 * L_INMEM will already be set and we don't need to do anything.
    306  1.21   thorpej 	 *
    307  1.61       chs 	 * Note the kernel stack gets read/write accesses right off the bat.
    308   1.6       mrg 	 */
    309  1.61       chs 
    310  1.62   thorpej 	if ((l2->l_flag & L_INMEM) == 0) {
    311  1.61       chs 		error = uvm_fault_wire(kernel_map, (vaddr_t)up,
    312  1.61       chs 		    (vaddr_t)up + USPACE, VM_FAULT_WIRE,
    313  1.61       chs 		    VM_PROT_READ | VM_PROT_WRITE);
    314  1.61       chs 		if (error)
    315  1.62   thorpej 			panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
    316  1.67       scw #ifdef PMAP_UAREA
    317  1.67       scw 		/* Tell the pmap this is a u-area mapping */
    318  1.67       scw 		PMAP_UAREA((vaddr_t)up);
    319  1.67       scw #endif
    320  1.62   thorpej 		l2->l_flag |= L_INMEM;
    321  1.61       chs 	}
    322  1.59      yamt 
    323  1.59      yamt #ifdef KSTACK_CHECK_MAGIC
    324  1.59      yamt 	/*
    325  1.59      yamt 	 * fill stack with magic number
    326  1.59      yamt 	 */
    327  1.63      yamt 	kstack_setup_magic(l2);
    328  1.59      yamt #endif
    329   1.6       mrg 
    330   1.6       mrg 	/*
    331  1.62   thorpej 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
    332  1.62   thorpej  	 * to run.  If this is a normal user fork, the child will exit
    333  1.34   thorpej 	 * directly to user mode via child_return() on its first time
    334  1.34   thorpej 	 * slice and will not return here.  If this is a kernel thread,
    335  1.34   thorpej 	 * the specified entry point will be executed.
    336   1.6       mrg 	 */
    337  1.62   thorpej 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
    338  1.14   thorpej }
    339  1.14   thorpej 
    340  1.14   thorpej /*
    341  1.14   thorpej  * uvm_exit: exit a virtual address space
    342  1.14   thorpej  *
    343  1.14   thorpej  * - the process passed to us is a dead (pre-zombie) process; we
    344  1.14   thorpej  *   are running on a different context now (the reaper).
    345  1.14   thorpej  * - we must run in a separate thread because freeing the vmspace
    346  1.14   thorpej  *   of the dead process may block.
    347  1.14   thorpej  */
    348  1.60       chs 
    349  1.14   thorpej void
    350  1.62   thorpej uvm_proc_exit(p)
    351  1.14   thorpej 	struct proc *p;
    352  1.14   thorpej {
    353  1.62   thorpej 	uvmspace_free(p->p_vmspace);
    354  1.62   thorpej }
    355  1.14   thorpej 
    356  1.62   thorpej void
    357  1.71  jdolecek uvm_lwp_exit(struct lwp *l)
    358  1.62   thorpej {
    359  1.62   thorpej 	vaddr_t va = (vaddr_t)l->l_addr;
    360  1.62   thorpej 
    361  1.62   thorpej 	l->l_flag &= ~L_INMEM;
    362  1.60       chs 	uvm_uarea_free(va);
    363  1.62   thorpej 	l->l_addr = NULL;
    364   1.1       mrg }
    365   1.1       mrg 
    366   1.1       mrg /*
    367  1.60       chs  * uvm_uarea_alloc: allocate a u-area
    368  1.60       chs  */
    369  1.60       chs 
    370  1.61       chs boolean_t
    371  1.61       chs uvm_uarea_alloc(vaddr_t *uaddrp)
    372  1.60       chs {
    373  1.60       chs 	vaddr_t uaddr;
    374  1.60       chs 
    375  1.60       chs #ifndef USPACE_ALIGN
    376  1.60       chs #define USPACE_ALIGN    0
    377  1.60       chs #endif
    378  1.60       chs 
    379  1.62   thorpej 	simple_lock(&uvm_uareas_slock);
    380  1.60       chs 	uaddr = (vaddr_t)uvm_uareas;
    381  1.60       chs 	if (uaddr) {
    382  1.60       chs 		uvm_uareas = *(void **)uvm_uareas;
    383  1.60       chs 		uvm_nuarea--;
    384  1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    385  1.61       chs 		*uaddrp = uaddr;
    386  1.61       chs 		return TRUE;
    387  1.60       chs 	} else {
    388  1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    389  1.61       chs 		*uaddrp = uvm_km_valloc_align(kernel_map, USPACE, USPACE_ALIGN);
    390  1.61       chs 		return FALSE;
    391  1.60       chs 	}
    392  1.60       chs }
    393  1.60       chs 
    394  1.60       chs /*
    395  1.60       chs  * uvm_uarea_free: free a u-area
    396  1.60       chs  */
    397  1.60       chs 
    398  1.60       chs void
    399  1.60       chs uvm_uarea_free(vaddr_t uaddr)
    400  1.60       chs {
    401  1.60       chs 
    402  1.62   thorpej 	simple_lock(&uvm_uareas_slock);
    403  1.60       chs 	if (uvm_nuarea < UVM_NUAREA_MAX) {
    404  1.60       chs 		*(void **)uaddr = uvm_uareas;
    405  1.60       chs 		uvm_uareas = (void *)uaddr;
    406  1.60       chs 		uvm_nuarea++;
    407  1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    408  1.60       chs 	} else {
    409  1.62   thorpej 		simple_unlock(&uvm_uareas_slock);
    410  1.60       chs 		uvm_km_free(kernel_map, uaddr, USPACE);
    411  1.60       chs 	}
    412  1.60       chs }
    413  1.60       chs 
    414  1.60       chs /*
    415   1.1       mrg  * uvm_init_limit: init per-process VM limits
    416   1.1       mrg  *
    417   1.1       mrg  * - called for process 0 and then inherited by all others.
    418   1.1       mrg  */
    419  1.60       chs 
    420   1.6       mrg void
    421   1.6       mrg uvm_init_limits(p)
    422   1.6       mrg 	struct proc *p;
    423   1.6       mrg {
    424   1.6       mrg 
    425   1.6       mrg 	/*
    426   1.6       mrg 	 * Set up the initial limits on process VM.  Set the maximum
    427   1.6       mrg 	 * resident set size to be all of (reasonably) available memory.
    428   1.6       mrg 	 * This causes any single, large process to start random page
    429   1.6       mrg 	 * replacement once it fills memory.
    430   1.6       mrg 	 */
    431   1.6       mrg 
    432   1.6       mrg 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
    433   1.6       mrg 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
    434   1.6       mrg 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
    435   1.6       mrg 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
    436   1.6       mrg 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
    437   1.1       mrg }
    438   1.1       mrg 
    439   1.1       mrg #ifdef DEBUG
    440   1.1       mrg int	enableswap = 1;
    441   1.1       mrg int	swapdebug = 0;
    442   1.1       mrg #define	SDB_FOLLOW	1
    443   1.1       mrg #define SDB_SWAPIN	2
    444   1.1       mrg #define SDB_SWAPOUT	4
    445   1.1       mrg #endif
    446   1.1       mrg 
    447   1.1       mrg /*
    448   1.1       mrg  * uvm_swapin: swap in a process's u-area.
    449   1.1       mrg  */
    450   1.1       mrg 
    451   1.6       mrg void
    452  1.62   thorpej uvm_swapin(l)
    453  1.62   thorpej 	struct lwp *l;
    454   1.6       mrg {
    455  1.13       eeh 	vaddr_t addr;
    456  1.52       chs 	int s, error;
    457   1.6       mrg 
    458  1.62   thorpej 	addr = (vaddr_t)l->l_addr;
    459  1.62   thorpej 	/* make L_INMEM true */
    460  1.57       chs 	error = uvm_fault_wire(kernel_map, addr, addr + USPACE, VM_FAULT_WIRE,
    461  1.21   thorpej 	    VM_PROT_READ | VM_PROT_WRITE);
    462  1.52       chs 	if (error) {
    463  1.52       chs 		panic("uvm_swapin: rewiring stack failed: %d", error);
    464  1.52       chs 	}
    465   1.6       mrg 
    466   1.6       mrg 	/*
    467   1.6       mrg 	 * Some architectures need to be notified when the user area has
    468   1.6       mrg 	 * moved to new physical page(s) (e.g.  see mips/mips/vm_machdep.c).
    469   1.6       mrg 	 */
    470  1.62   thorpej 	cpu_swapin(l);
    471  1.41     enami 	SCHED_LOCK(s);
    472  1.62   thorpej 	if (l->l_stat == LSRUN)
    473  1.62   thorpej 		setrunqueue(l);
    474  1.62   thorpej 	l->l_flag |= L_INMEM;
    475  1.41     enami 	SCHED_UNLOCK(s);
    476  1.62   thorpej 	l->l_swtime = 0;
    477   1.6       mrg 	++uvmexp.swapins;
    478   1.1       mrg }
    479   1.1       mrg 
    480   1.1       mrg /*
    481   1.1       mrg  * uvm_scheduler: process zero main loop
    482   1.1       mrg  *
    483   1.1       mrg  * - attempt to swapin every swaped-out, runnable process in order of
    484   1.1       mrg  *	priority.
    485   1.1       mrg  * - if not enough memory, wake the pagedaemon and let it clear space.
    486   1.1       mrg  */
    487   1.1       mrg 
    488   1.6       mrg void
    489   1.6       mrg uvm_scheduler()
    490   1.1       mrg {
    491  1.62   thorpej 	struct lwp *l, *ll;
    492  1.32  augustss 	int pri;
    493   1.6       mrg 	int ppri;
    494   1.1       mrg 
    495   1.1       mrg loop:
    496   1.1       mrg #ifdef DEBUG
    497   1.6       mrg 	while (!enableswap)
    498  1.43       chs 		tsleep(&proc0, PVM, "noswap", 0);
    499   1.1       mrg #endif
    500  1.62   thorpej 	ll = NULL;		/* process to choose */
    501   1.6       mrg 	ppri = INT_MIN;	/* its priority */
    502  1.29   thorpej 	proclist_lock_read();
    503   1.6       mrg 
    504  1.62   thorpej 	LIST_FOREACH(l, &alllwp, l_list) {
    505   1.6       mrg 		/* is it a runnable swapped out process? */
    506  1.62   thorpej 		if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
    507  1.62   thorpej 			pri = l->l_swtime + l->l_slptime -
    508  1.62   thorpej 			    (l->l_proc->p_nice - NZERO) * 8;
    509   1.6       mrg 			if (pri > ppri) {   /* higher priority?  remember it. */
    510  1.62   thorpej 				ll = l;
    511   1.6       mrg 				ppri = pri;
    512   1.6       mrg 			}
    513   1.6       mrg 		}
    514   1.6       mrg 	}
    515  1.39  sommerfe 	/*
    516  1.39  sommerfe 	 * XXXSMP: possible unlock/sleep race between here and the
    517  1.39  sommerfe 	 * "scheduler" tsleep below..
    518  1.39  sommerfe 	 */
    519  1.28   thorpej 	proclist_unlock_read();
    520   1.1       mrg 
    521   1.1       mrg #ifdef DEBUG
    522   1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    523  1.62   thorpej 		printf("scheduler: running, procp %p pri %d\n", ll, ppri);
    524   1.1       mrg #endif
    525   1.6       mrg 	/*
    526   1.6       mrg 	 * Nothing to do, back to sleep
    527   1.6       mrg 	 */
    528  1.62   thorpej 	if ((l = ll) == NULL) {
    529  1.43       chs 		tsleep(&proc0, PVM, "scheduler", 0);
    530   1.6       mrg 		goto loop;
    531   1.6       mrg 	}
    532   1.6       mrg 
    533   1.6       mrg 	/*
    534   1.6       mrg 	 * we have found swapped out process which we would like to bring
    535   1.6       mrg 	 * back in.
    536   1.6       mrg 	 *
    537   1.6       mrg 	 * XXX: this part is really bogus cuz we could deadlock on memory
    538   1.6       mrg 	 * despite our feeble check
    539   1.6       mrg 	 */
    540   1.6       mrg 	if (uvmexp.free > atop(USPACE)) {
    541   1.1       mrg #ifdef DEBUG
    542   1.6       mrg 		if (swapdebug & SDB_SWAPIN)
    543   1.6       mrg 			printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
    544  1.62   thorpej 	     l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
    545   1.1       mrg #endif
    546  1.62   thorpej 		uvm_swapin(l);
    547   1.6       mrg 		goto loop;
    548   1.6       mrg 	}
    549   1.6       mrg 	/*
    550   1.6       mrg 	 * not enough memory, jab the pageout daemon and wait til the coast
    551   1.6       mrg 	 * is clear
    552   1.6       mrg 	 */
    553   1.1       mrg #ifdef DEBUG
    554   1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    555   1.6       mrg 		printf("scheduler: no room for pid %d(%s), free %d\n",
    556  1.62   thorpej 	   l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
    557   1.1       mrg #endif
    558   1.6       mrg 	uvm_wait("schedpwait");
    559   1.1       mrg #ifdef DEBUG
    560   1.6       mrg 	if (swapdebug & SDB_FOLLOW)
    561   1.6       mrg 		printf("scheduler: room again, free %d\n", uvmexp.free);
    562   1.1       mrg #endif
    563   1.6       mrg 	goto loop;
    564   1.1       mrg }
    565   1.1       mrg 
    566   1.1       mrg /*
    567  1.62   thorpej  * swappable: is LWP "l" swappable?
    568   1.1       mrg  */
    569   1.1       mrg 
    570  1.62   thorpej #define	swappable(l)							\
    571  1.62   thorpej 	(((l)->l_flag & (L_INMEM)) &&					\
    572  1.62   thorpej 	 ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) &&	\
    573  1.62   thorpej 	 (l)->l_holdcnt == 0)
    574   1.1       mrg 
    575   1.1       mrg /*
    576   1.1       mrg  * swapout_threads: find threads that can be swapped and unwire their
    577   1.1       mrg  *	u-areas.
    578   1.1       mrg  *
    579   1.1       mrg  * - called by the pagedaemon
    580   1.1       mrg  * - try and swap at least one processs
    581   1.1       mrg  * - processes that are sleeping or stopped for maxslp or more seconds
    582   1.1       mrg  *   are swapped... otherwise the longest-sleeping or stopped process
    583   1.1       mrg  *   is swapped, otherwise the longest resident process...
    584   1.1       mrg  */
    585  1.60       chs 
    586   1.6       mrg void
    587   1.6       mrg uvm_swapout_threads()
    588   1.1       mrg {
    589  1.62   thorpej 	struct lwp *l;
    590  1.62   thorpej 	struct lwp *outl, *outl2;
    591   1.6       mrg 	int outpri, outpri2;
    592   1.6       mrg 	int didswap = 0;
    593  1.48       chs 	extern int maxslp;
    594   1.6       mrg 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
    595   1.1       mrg 
    596   1.1       mrg #ifdef DEBUG
    597   1.6       mrg 	if (!enableswap)
    598   1.6       mrg 		return;
    599   1.1       mrg #endif
    600   1.1       mrg 
    601   1.6       mrg 	/*
    602  1.62   thorpej 	 * outl/outpri  : stop/sleep thread with largest sleeptime < maxslp
    603  1.62   thorpej 	 * outl2/outpri2: the longest resident thread (its swap time)
    604   1.6       mrg 	 */
    605  1.62   thorpej 	outl = outl2 = NULL;
    606   1.6       mrg 	outpri = outpri2 = 0;
    607  1.29   thorpej 	proclist_lock_read();
    608  1.62   thorpej 	LIST_FOREACH(l, &alllwp, l_list) {
    609  1.62   thorpej 		if (!swappable(l))
    610   1.6       mrg 			continue;
    611  1.62   thorpej 		switch (l->l_stat) {
    612  1.68        cl 		case LSONPROC:
    613  1.69        cl 			KDASSERT(l->l_cpu != curcpu());
    614  1.69        cl 			continue;
    615  1.69        cl 
    616  1.62   thorpej 		case LSRUN:
    617  1.62   thorpej 			if (l->l_swtime > outpri2) {
    618  1.62   thorpej 				outl2 = l;
    619  1.62   thorpej 				outpri2 = l->l_swtime;
    620   1.6       mrg 			}
    621   1.6       mrg 			continue;
    622  1.48       chs 
    623  1.62   thorpej 		case LSSLEEP:
    624  1.62   thorpej 		case LSSTOP:
    625  1.62   thorpej 			if (l->l_slptime >= maxslp) {
    626  1.62   thorpej 				uvm_swapout(l);
    627   1.6       mrg 				didswap++;
    628  1.62   thorpej 			} else if (l->l_slptime > outpri) {
    629  1.62   thorpej 				outl = l;
    630  1.62   thorpej 				outpri = l->l_slptime;
    631   1.6       mrg 			}
    632   1.6       mrg 			continue;
    633   1.6       mrg 		}
    634   1.6       mrg 	}
    635  1.28   thorpej 	proclist_unlock_read();
    636   1.6       mrg 
    637   1.6       mrg 	/*
    638   1.6       mrg 	 * If we didn't get rid of any real duds, toss out the next most
    639   1.6       mrg 	 * likely sleeping/stopped or running candidate.  We only do this
    640   1.6       mrg 	 * if we are real low on memory since we don't gain much by doing
    641   1.6       mrg 	 * it (USPACE bytes).
    642   1.6       mrg 	 */
    643   1.6       mrg 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
    644  1.62   thorpej 		if ((l = outl) == NULL)
    645  1.62   thorpej 			l = outl2;
    646   1.1       mrg #ifdef DEBUG
    647   1.6       mrg 		if (swapdebug & SDB_SWAPOUT)
    648  1.62   thorpej 			printf("swapout_threads: no duds, try procp %p\n", l);
    649   1.1       mrg #endif
    650  1.62   thorpej 		if (l)
    651  1.62   thorpej 			uvm_swapout(l);
    652   1.6       mrg 	}
    653   1.1       mrg }
    654   1.1       mrg 
    655   1.1       mrg /*
    656  1.62   thorpej  * uvm_swapout: swap out lwp "l"
    657   1.1       mrg  *
    658  1.48       chs  * - currently "swapout" means "unwire U-area" and "pmap_collect()"
    659   1.1       mrg  *   the pmap.
    660   1.1       mrg  * - XXXCDC: should deactivate all process' private anonymous memory
    661   1.1       mrg  */
    662   1.1       mrg 
    663   1.6       mrg static void
    664  1.62   thorpej uvm_swapout(l)
    665  1.62   thorpej 	struct lwp *l;
    666   1.1       mrg {
    667  1.13       eeh 	vaddr_t addr;
    668   1.6       mrg 	int s;
    669  1.62   thorpej 	struct proc *p = l->l_proc;
    670   1.1       mrg 
    671   1.1       mrg #ifdef DEBUG
    672   1.6       mrg 	if (swapdebug & SDB_SWAPOUT)
    673  1.62   thorpej 		printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
    674  1.62   thorpej 	   p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
    675  1.62   thorpej 	   l->l_slptime, uvmexp.free);
    676   1.1       mrg #endif
    677   1.1       mrg 
    678   1.6       mrg 	/*
    679   1.6       mrg 	 * Mark it as (potentially) swapped out.
    680   1.6       mrg 	 */
    681  1.41     enami 	SCHED_LOCK(s);
    682  1.69        cl 	if (l->l_stat == LSONPROC) {
    683  1.69        cl 		KDASSERT(l->l_cpu != curcpu());
    684  1.68        cl 		SCHED_UNLOCK(s);
    685  1.68        cl 		return;
    686  1.68        cl 	}
    687  1.62   thorpej 	l->l_flag &= ~L_INMEM;
    688  1.62   thorpej 	if (l->l_stat == LSRUN)
    689  1.62   thorpej 		remrunqueue(l);
    690  1.41     enami 	SCHED_UNLOCK(s);
    691  1.62   thorpej 	l->l_swtime = 0;
    692  1.53       chs 	p->p_stats->p_ru.ru_nswap++;
    693   1.6       mrg 	++uvmexp.swapouts;
    694  1.68        cl 
    695  1.68        cl 	/*
    696  1.68        cl 	 * Do any machine-specific actions necessary before swapout.
    697  1.68        cl 	 * This can include saving floating point state, etc.
    698  1.68        cl 	 */
    699  1.68        cl 	cpu_swapout(l);
    700  1.43       chs 
    701  1.43       chs 	/*
    702  1.43       chs 	 * Unwire the to-be-swapped process's user struct and kernel stack.
    703  1.43       chs 	 */
    704  1.62   thorpej 	addr = (vaddr_t)l->l_addr;
    705  1.62   thorpej 	uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
    706  1.43       chs 	pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
    707   1.1       mrg }
    708   1.1       mrg 
    709  1.56   thorpej /*
    710  1.56   thorpej  * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
    711  1.56   thorpej  * a core file.
    712  1.56   thorpej  */
    713  1.56   thorpej 
    714  1.56   thorpej int
    715  1.66      fvdl uvm_coredump_walkmap(p, vp, cred, func, cookie)
    716  1.66      fvdl 	struct proc *p;
    717  1.56   thorpej 	struct vnode *vp;
    718  1.56   thorpej 	struct ucred *cred;
    719  1.66      fvdl 	int (*func)(struct proc *, struct vnode *, struct ucred *,
    720  1.56   thorpej 	    struct uvm_coredump_state *);
    721  1.56   thorpej 	void *cookie;
    722  1.56   thorpej {
    723  1.56   thorpej 	struct uvm_coredump_state state;
    724  1.56   thorpej 	struct vmspace *vm = p->p_vmspace;
    725  1.56   thorpej 	struct vm_map *map = &vm->vm_map;
    726  1.56   thorpej 	struct vm_map_entry *entry;
    727  1.56   thorpej 	vaddr_t maxstack;
    728  1.56   thorpej 	int error;
    729  1.56   thorpej 
    730  1.56   thorpej 	maxstack = trunc_page(USRSTACK - ctob(vm->vm_ssize));
    731  1.56   thorpej 
    732  1.64    atatat 	entry = NULL;
    733  1.64    atatat 	vm_map_lock_read(map);
    734  1.64    atatat 	for (;;) {
    735  1.64    atatat 		if (entry == NULL)
    736  1.64    atatat 			entry = map->header.next;
    737  1.70      yamt 		else if (state.end < vm_map_min(map) ||
    738  1.70      yamt 		    vm_map_max(map) <= state.end)
    739  1.70      yamt 			break;
    740  1.64    atatat 		else if (!uvm_map_lookup_entry(map, state.end, &entry))
    741  1.64    atatat 			entry = entry->next;
    742  1.64    atatat 		if (entry == &map->header)
    743  1.64    atatat 			break;
    744  1.64    atatat 
    745  1.56   thorpej 		/* Should never happen for a user process. */
    746  1.56   thorpej 		if (UVM_ET_ISSUBMAP(entry))
    747  1.56   thorpej 			panic("uvm_coredump_walkmap: user process with "
    748  1.56   thorpej 			    "submap?");
    749  1.56   thorpej 
    750  1.56   thorpej 		state.cookie = cookie;
    751  1.56   thorpej 		state.start = entry->start;
    752  1.56   thorpej 		state.end = entry->end;
    753  1.56   thorpej 		state.prot = entry->protection;
    754  1.56   thorpej 		state.flags = 0;
    755  1.56   thorpej 
    756  1.56   thorpej 		if (state.start >= VM_MAXUSER_ADDRESS)
    757  1.56   thorpej 			continue;
    758  1.56   thorpej 
    759  1.56   thorpej 		if (state.end > VM_MAXUSER_ADDRESS)
    760  1.56   thorpej 			state.end = VM_MAXUSER_ADDRESS;
    761  1.56   thorpej 
    762  1.56   thorpej 		if (state.start >= (vaddr_t)vm->vm_maxsaddr) {
    763  1.56   thorpej 			if (state.end <= maxstack)
    764  1.56   thorpej 				continue;
    765  1.56   thorpej 			if (state.start < maxstack)
    766  1.56   thorpej 				state.start = maxstack;
    767  1.56   thorpej 			state.flags |= UVM_COREDUMP_STACK;
    768  1.56   thorpej 		}
    769  1.56   thorpej 
    770  1.56   thorpej 		if ((entry->protection & VM_PROT_WRITE) == 0)
    771  1.58      matt 			state.flags |= UVM_COREDUMP_NODUMP;
    772  1.58      matt 
    773  1.58      matt 		if (entry->object.uvm_obj != NULL &&
    774  1.58      matt 		    entry->object.uvm_obj->pgops == &uvm_deviceops)
    775  1.56   thorpej 			state.flags |= UVM_COREDUMP_NODUMP;
    776  1.56   thorpej 
    777  1.64    atatat 		vm_map_unlock_read(map);
    778  1.66      fvdl 		error = (*func)(p, vp, cred, &state);
    779  1.56   thorpej 		if (error)
    780  1.56   thorpej 			return (error);
    781  1.64    atatat 		vm_map_lock_read(map);
    782  1.56   thorpej 	}
    783  1.64    atatat 	vm_map_unlock_read(map);
    784  1.56   thorpej 
    785  1.56   thorpej 	return (0);
    786  1.56   thorpej }
    787