Home | History | Annotate | Line # | Download | only in uvm
uvm_km.c revision 1.52
      1  1.52       chs /*	$NetBSD: uvm_km.c,v 1.52 2001/09/15 20:36:46 chs Exp $	*/
      2   1.1       mrg 
      3  1.47       chs /*
      4   1.1       mrg  * Copyright (c) 1997 Charles D. Cranor and Washington University.
      5  1.47       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  1.47       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_kern.c   8.3 (Berkeley) 1/12/94
     42   1.4       mrg  * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 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  1.47       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  1.47       chs  *
     54  1.47       chs  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     55  1.47       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  1.47       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   1.6       mrg 
     69   1.6       mrg #include "opt_uvmhist.h"
     70   1.1       mrg 
     71   1.1       mrg /*
     72   1.1       mrg  * uvm_km.c: handle kernel memory allocation and management
     73   1.1       mrg  */
     74   1.1       mrg 
     75   1.7     chuck /*
     76   1.7     chuck  * overview of kernel memory management:
     77   1.7     chuck  *
     78   1.7     chuck  * the kernel virtual address space is mapped by "kernel_map."   kernel_map
     79   1.7     chuck  * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
     80   1.7     chuck  * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
     81   1.7     chuck  *
     82  1.47       chs  * the kernel_map has several "submaps."   submaps can only appear in
     83   1.7     chuck  * the kernel_map (user processes can't use them).   submaps "take over"
     84   1.7     chuck  * the management of a sub-range of the kernel's address space.  submaps
     85   1.7     chuck  * are typically allocated at boot time and are never released.   kernel
     86  1.47       chs  * virtual address space that is mapped by a submap is locked by the
     87   1.7     chuck  * submap's lock -- not the kernel_map's lock.
     88   1.7     chuck  *
     89   1.7     chuck  * thus, the useful feature of submaps is that they allow us to break
     90   1.7     chuck  * up the locking and protection of the kernel address space into smaller
     91   1.7     chuck  * chunks.
     92   1.7     chuck  *
     93   1.7     chuck  * the vm system has several standard kernel submaps, including:
     94   1.7     chuck  *   kmem_map => contains only wired kernel memory for the kernel
     95   1.7     chuck  *		malloc.   *** access to kmem_map must be protected
     96  1.42   thorpej  *		by splvm() because we are allowed to call malloc()
     97   1.7     chuck  *		at interrupt time ***
     98  1.42   thorpej  *   mb_map => memory for large mbufs,  *** protected by splvm ***
     99   1.7     chuck  *   pager_map => used to map "buf" structures into kernel space
    100   1.7     chuck  *   exec_map => used during exec to handle exec args
    101   1.7     chuck  *   etc...
    102   1.7     chuck  *
    103   1.7     chuck  * the kernel allocates its private memory out of special uvm_objects whose
    104   1.7     chuck  * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
    105   1.7     chuck  * are "special" and never die).   all kernel objects should be thought of
    106  1.47       chs  * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
    107   1.7     chuck  * object is equal to the size of kernel virtual address space (i.e. the
    108   1.7     chuck  * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
    109   1.7     chuck  *
    110   1.7     chuck  * most kernel private memory lives in kernel_object.   the only exception
    111   1.7     chuck  * to this is for memory that belongs to submaps that must be protected
    112  1.52       chs  * by splvm().  pages in these submaps are not assigned to an object.
    113   1.7     chuck  *
    114   1.7     chuck  * note that just because a kernel object spans the entire kernel virutal
    115   1.7     chuck  * address space doesn't mean that it has to be mapped into the entire space.
    116  1.47       chs  * large chunks of a kernel object's space go unused either because
    117  1.47       chs  * that area of kernel VM is unmapped, or there is some other type of
    118   1.7     chuck  * object mapped into that range (e.g. a vnode).    for submap's kernel
    119   1.7     chuck  * objects, the only part of the object that can ever be populated is the
    120   1.7     chuck  * offsets that are managed by the submap.
    121   1.7     chuck  *
    122   1.7     chuck  * note that the "offset" in a kernel object is always the kernel virtual
    123   1.7     chuck  * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
    124   1.7     chuck  * example:
    125   1.7     chuck  *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
    126   1.7     chuck  *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
    127   1.7     chuck  *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
    128   1.7     chuck  *   then that means that the page at offset 0x235000 in kernel_object is
    129  1.47       chs  *   mapped at 0xf8235000.
    130   1.7     chuck  *
    131   1.7     chuck  * kernel object have one other special property: when the kernel virtual
    132   1.7     chuck  * memory mapping them is unmapped, the backing memory in the object is
    133   1.7     chuck  * freed right away.   this is done with the uvm_km_pgremove() function.
    134   1.7     chuck  * this has to be done because there is no backing store for kernel pages
    135   1.7     chuck  * and no need to save them after they are no longer referenced.
    136   1.7     chuck  */
    137   1.7     chuck 
    138   1.1       mrg #include <sys/param.h>
    139   1.1       mrg #include <sys/systm.h>
    140   1.1       mrg #include <sys/proc.h>
    141   1.1       mrg 
    142   1.1       mrg #include <uvm/uvm.h>
    143   1.1       mrg 
    144   1.1       mrg /*
    145   1.1       mrg  * global data structures
    146   1.1       mrg  */
    147   1.1       mrg 
    148  1.49       chs struct vm_map *kernel_map = NULL;
    149   1.1       mrg 
    150   1.1       mrg /*
    151   1.1       mrg  * local data structues
    152   1.1       mrg  */
    153   1.1       mrg 
    154   1.1       mrg static struct vm_map		kernel_map_store;
    155   1.1       mrg 
    156   1.1       mrg /*
    157   1.1       mrg  * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
    158   1.1       mrg  * KVM already allocated for text, data, bss, and static data structures).
    159   1.1       mrg  *
    160   1.1       mrg  * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
    161   1.1       mrg  *    we assume that [min -> start] has already been allocated and that
    162   1.1       mrg  *    "end" is the end.
    163   1.1       mrg  */
    164   1.1       mrg 
    165   1.8       mrg void
    166   1.8       mrg uvm_km_init(start, end)
    167  1.14       eeh 	vaddr_t start, end;
    168   1.1       mrg {
    169  1.14       eeh 	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
    170  1.27   thorpej 
    171  1.27   thorpej 	/*
    172  1.27   thorpej 	 * next, init kernel memory objects.
    173   1.8       mrg 	 */
    174   1.1       mrg 
    175   1.8       mrg 	/* kernel_object: for pageable anonymous kernel memory */
    176  1.34       chs 	uao_init();
    177   1.8       mrg 	uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
    178   1.3       chs 				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
    179   1.1       mrg 
    180  1.24   thorpej 	/*
    181  1.47       chs 	 * init the map and reserve allready allocated kernel space
    182   1.8       mrg 	 * before installing.
    183   1.8       mrg 	 */
    184   1.1       mrg 
    185  1.25   thorpej 	uvm_map_setup(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
    186   1.8       mrg 	kernel_map_store.pmap = pmap_kernel();
    187   1.8       mrg 	if (uvm_map(&kernel_map_store, &base, start - base, NULL,
    188  1.39   thorpej 	    UVM_UNKNOWN_OFFSET, 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
    189  1.43       chs 	    UVM_INH_NONE, UVM_ADV_RANDOM,UVM_FLAG_FIXED)) != 0)
    190   1.8       mrg 		panic("uvm_km_init: could not reserve space for kernel");
    191  1.47       chs 
    192   1.8       mrg 	/*
    193   1.8       mrg 	 * install!
    194   1.8       mrg 	 */
    195   1.8       mrg 
    196   1.8       mrg 	kernel_map = &kernel_map_store;
    197   1.1       mrg }
    198   1.1       mrg 
    199   1.1       mrg /*
    200   1.1       mrg  * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
    201   1.1       mrg  * is allocated all references to that area of VM must go through it.  this
    202   1.1       mrg  * allows the locking of VAs in kernel_map to be broken up into regions.
    203   1.1       mrg  *
    204   1.5   thorpej  * => if `fixed' is true, *min specifies where the region described
    205   1.5   thorpej  *      by the submap must start
    206   1.1       mrg  * => if submap is non NULL we use that as the submap, otherwise we
    207   1.1       mrg  *	alloc a new map
    208   1.1       mrg  */
    209   1.8       mrg struct vm_map *
    210  1.25   thorpej uvm_km_suballoc(map, min, max, size, flags, fixed, submap)
    211   1.8       mrg 	struct vm_map *map;
    212  1.52       chs 	vaddr_t *min, *max;		/* IN/OUT, OUT */
    213  1.14       eeh 	vsize_t size;
    214  1.25   thorpej 	int flags;
    215   1.8       mrg 	boolean_t fixed;
    216   1.8       mrg 	struct vm_map *submap;
    217   1.8       mrg {
    218   1.8       mrg 	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
    219   1.1       mrg 
    220   1.8       mrg 	size = round_page(size);	/* round up to pagesize */
    221   1.1       mrg 
    222   1.8       mrg 	/*
    223   1.8       mrg 	 * first allocate a blank spot in the parent map
    224   1.8       mrg 	 */
    225   1.8       mrg 
    226  1.39   thorpej 	if (uvm_map(map, min, size, NULL, UVM_UNKNOWN_OFFSET, 0,
    227   1.8       mrg 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    228  1.43       chs 	    UVM_ADV_RANDOM, mapflags)) != 0) {
    229   1.8       mrg 	       panic("uvm_km_suballoc: unable to allocate space in parent map");
    230   1.8       mrg 	}
    231   1.8       mrg 
    232   1.8       mrg 	/*
    233   1.8       mrg 	 * set VM bounds (min is filled in by uvm_map)
    234   1.8       mrg 	 */
    235   1.1       mrg 
    236   1.8       mrg 	*max = *min + size;
    237   1.5   thorpej 
    238   1.8       mrg 	/*
    239   1.8       mrg 	 * add references to pmap and create or init the submap
    240   1.8       mrg 	 */
    241   1.1       mrg 
    242   1.8       mrg 	pmap_reference(vm_map_pmap(map));
    243   1.8       mrg 	if (submap == NULL) {
    244  1.25   thorpej 		submap = uvm_map_create(vm_map_pmap(map), *min, *max, flags);
    245   1.8       mrg 		if (submap == NULL)
    246   1.8       mrg 			panic("uvm_km_suballoc: unable to create submap");
    247   1.8       mrg 	} else {
    248  1.25   thorpej 		uvm_map_setup(submap, *min, *max, flags);
    249   1.8       mrg 		submap->pmap = vm_map_pmap(map);
    250   1.8       mrg 	}
    251   1.1       mrg 
    252   1.8       mrg 	/*
    253   1.8       mrg 	 * now let uvm_map_submap plug in it...
    254   1.8       mrg 	 */
    255   1.1       mrg 
    256  1.43       chs 	if (uvm_map_submap(map, *min, *max, submap) != 0)
    257   1.8       mrg 		panic("uvm_km_suballoc: submap allocation failed");
    258   1.1       mrg 
    259   1.8       mrg 	return(submap);
    260   1.1       mrg }
    261   1.1       mrg 
    262   1.1       mrg /*
    263   1.1       mrg  * uvm_km_pgremove: remove pages from a kernel uvm_object.
    264   1.1       mrg  *
    265   1.1       mrg  * => when you unmap a part of anonymous kernel memory you want to toss
    266   1.1       mrg  *    the pages right away.    (this gets called from uvm_unmap_...).
    267   1.1       mrg  */
    268   1.1       mrg 
    269   1.8       mrg void
    270   1.8       mrg uvm_km_pgremove(uobj, start, end)
    271   1.8       mrg 	struct uvm_object *uobj;
    272  1.14       eeh 	vaddr_t start, end;
    273   1.1       mrg {
    274  1.24   thorpej 	boolean_t by_list;
    275  1.52       chs 	struct vm_page *pg, *nextpg;
    276  1.52       chs 	voff_t curoff, nextoff;
    277   1.8       mrg 	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
    278   1.1       mrg 
    279  1.40       chs 	KASSERT(uobj->pgops == &aobj_pager);
    280  1.40       chs 	simple_lock(&uobj->vmobjlock);
    281   1.3       chs 
    282   1.8       mrg 	/* choose cheapest traversal */
    283   1.8       mrg 	by_list = (uobj->uo_npages <=
    284  1.52       chs 	     ((end - start) >> PAGE_SHIFT) * UVM_PAGE_HASH_PENALTY);
    285   1.8       mrg 	if (by_list)
    286   1.8       mrg 		goto loop_by_list;
    287   1.1       mrg 
    288  1.52       chs 	for (curoff = start; curoff < end; curoff = nextoff) {
    289  1.52       chs 		nextoff = curoff + PAGE_SIZE;
    290  1.52       chs 		pg = uvm_pagelookup(uobj, curoff);
    291  1.52       chs 		if (pg == NULL) {
    292  1.52       chs 			continue;
    293  1.52       chs 		}
    294  1.52       chs 		if (pg->flags & PG_BUSY) {
    295  1.52       chs 			pg->flags |= PG_WANTED;
    296  1.52       chs 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
    297  1.52       chs 				    "km_pgrm", 0);
    298  1.52       chs 			simple_lock(&uobj->vmobjlock);
    299  1.52       chs 			nextoff = curoff;
    300   1.8       mrg 			continue;
    301  1.52       chs 		}
    302   1.8       mrg 
    303  1.52       chs 		/*
    304  1.52       chs 		 * free the swap slot, then the page.
    305  1.52       chs 		 */
    306   1.8       mrg 
    307  1.52       chs 		uao_dropswap(uobj, curoff >> PAGE_SHIFT);
    308  1.52       chs 		uvm_lock_pageq();
    309  1.52       chs 		uvm_pagefree(pg);
    310  1.52       chs 		uvm_unlock_pageq();
    311   1.8       mrg 	}
    312   1.8       mrg 	simple_unlock(&uobj->vmobjlock);
    313   1.8       mrg 	return;
    314   1.1       mrg 
    315   1.1       mrg loop_by_list:
    316  1.52       chs 	for (pg = TAILQ_FIRST(&uobj->memq); pg != NULL; pg = nextpg) {
    317  1.52       chs 		nextpg = TAILQ_NEXT(pg, listq);
    318  1.52       chs 		if (pg->offset < start || pg->offset >= end) {
    319  1.52       chs 			continue;
    320  1.52       chs 		}
    321  1.52       chs 		if (pg->flags & PG_BUSY) {
    322  1.52       chs 			pg->flags |= PG_WANTED;
    323  1.52       chs 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
    324  1.52       chs 				    "km_pgrm", 0);
    325  1.52       chs 			simple_lock(&uobj->vmobjlock);
    326  1.52       chs 			nextpg = TAILQ_FIRST(&uobj->memq);
    327   1.8       mrg 			continue;
    328   1.8       mrg 		}
    329   1.8       mrg 
    330  1.52       chs 		/*
    331  1.52       chs 		 * free the swap slot, then the page.
    332  1.52       chs 		 */
    333   1.8       mrg 
    334  1.52       chs 		uao_dropswap(uobj, pg->offset >> PAGE_SHIFT);
    335  1.52       chs 		uvm_lock_pageq();
    336  1.52       chs 		uvm_pagefree(pg);
    337  1.52       chs 		uvm_unlock_pageq();
    338  1.24   thorpej 	}
    339  1.24   thorpej 	simple_unlock(&uobj->vmobjlock);
    340  1.24   thorpej }
    341  1.24   thorpej 
    342  1.24   thorpej 
    343  1.24   thorpej /*
    344  1.24   thorpej  * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for "intrsafe"
    345  1.52       chs  *    maps
    346  1.24   thorpej  *
    347  1.24   thorpej  * => when you unmap a part of anonymous kernel memory you want to toss
    348  1.52       chs  *    the pages right away.    (this is called from uvm_unmap_...).
    349  1.24   thorpej  * => none of the pages will ever be busy, and none of them will ever
    350  1.52       chs  *    be on the active or inactive queues (because they have no object).
    351  1.24   thorpej  */
    352  1.24   thorpej 
    353  1.24   thorpej void
    354  1.52       chs uvm_km_pgremove_intrsafe(start, end)
    355  1.24   thorpej 	vaddr_t start, end;
    356  1.24   thorpej {
    357  1.52       chs 	struct vm_page *pg;
    358  1.52       chs 	paddr_t pa;
    359  1.24   thorpej 	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
    360  1.24   thorpej 
    361  1.52       chs 	for (; start < end; start += PAGE_SIZE) {
    362  1.52       chs 		if (!pmap_extract(pmap_kernel(), start, &pa)) {
    363  1.24   thorpej 			continue;
    364  1.40       chs 		}
    365  1.52       chs 		pg = PHYS_TO_VM_PAGE(pa);
    366  1.52       chs 		KASSERT(pg);
    367  1.52       chs 		KASSERT(pg->uobject == NULL && pg->uanon == NULL);
    368  1.52       chs 		uvm_pagefree(pg);
    369  1.24   thorpej 	}
    370   1.1       mrg }
    371   1.1       mrg 
    372   1.1       mrg 
    373   1.1       mrg /*
    374   1.1       mrg  * uvm_km_kmemalloc: lower level kernel memory allocator for malloc()
    375   1.1       mrg  *
    376   1.1       mrg  * => we map wired memory into the specified map using the obj passed in
    377   1.1       mrg  * => NOTE: we can return NULL even if we can wait if there is not enough
    378   1.1       mrg  *	free VM space in the map... caller should be prepared to handle
    379   1.1       mrg  *	this case.
    380   1.1       mrg  * => we return KVA of memory allocated
    381   1.1       mrg  * => flags: NOWAIT, VALLOC - just allocate VA, TRYLOCK - fail if we can't
    382   1.1       mrg  *	lock the map
    383   1.1       mrg  */
    384   1.1       mrg 
    385  1.14       eeh vaddr_t
    386   1.8       mrg uvm_km_kmemalloc(map, obj, size, flags)
    387  1.49       chs 	struct vm_map *map;
    388   1.8       mrg 	struct uvm_object *obj;
    389  1.14       eeh 	vsize_t size;
    390   1.8       mrg 	int flags;
    391   1.1       mrg {
    392  1.14       eeh 	vaddr_t kva, loopva;
    393  1.14       eeh 	vaddr_t offset;
    394  1.44   thorpej 	vsize_t loopsize;
    395   1.8       mrg 	struct vm_page *pg;
    396   1.8       mrg 	UVMHIST_FUNC("uvm_km_kmemalloc"); UVMHIST_CALLED(maphist);
    397   1.1       mrg 
    398   1.8       mrg 	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
    399  1.40       chs 		    map, obj, size, flags);
    400  1.40       chs 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    401   1.1       mrg 
    402   1.8       mrg 	/*
    403   1.8       mrg 	 * setup for call
    404   1.8       mrg 	 */
    405   1.8       mrg 
    406   1.8       mrg 	size = round_page(size);
    407   1.8       mrg 	kva = vm_map_min(map);	/* hint */
    408   1.1       mrg 
    409   1.8       mrg 	/*
    410   1.8       mrg 	 * allocate some virtual space
    411   1.8       mrg 	 */
    412   1.8       mrg 
    413  1.35   thorpej 	if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
    414  1.39   thorpej 	      0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    415  1.47       chs 			  UVM_ADV_RANDOM, (flags & UVM_KMF_TRYLOCK)))
    416  1.43       chs 			!= 0)) {
    417   1.8       mrg 		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
    418   1.8       mrg 		return(0);
    419   1.8       mrg 	}
    420   1.8       mrg 
    421   1.8       mrg 	/*
    422   1.8       mrg 	 * if all we wanted was VA, return now
    423   1.8       mrg 	 */
    424   1.8       mrg 
    425   1.8       mrg 	if (flags & UVM_KMF_VALLOC) {
    426   1.8       mrg 		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
    427   1.8       mrg 		return(kva);
    428   1.8       mrg 	}
    429  1.40       chs 
    430   1.8       mrg 	/*
    431   1.8       mrg 	 * recover object offset from virtual address
    432   1.8       mrg 	 */
    433   1.8       mrg 
    434   1.8       mrg 	offset = kva - vm_map_min(kernel_map);
    435   1.8       mrg 	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
    436   1.8       mrg 
    437   1.8       mrg 	/*
    438   1.8       mrg 	 * now allocate and map in the memory... note that we are the only ones
    439   1.8       mrg 	 * whom should ever get a handle on this area of VM.
    440   1.8       mrg 	 */
    441   1.8       mrg 
    442   1.8       mrg 	loopva = kva;
    443  1.44   thorpej 	loopsize = size;
    444  1.44   thorpej 	while (loopsize) {
    445  1.52       chs 		if (obj) {
    446  1.52       chs 			simple_lock(&obj->vmobjlock);
    447  1.52       chs 		}
    448  1.52       chs 		pg = uvm_pagealloc(obj, offset, NULL, UVM_PGA_USERESERVE);
    449  1.45   thorpej 		if (__predict_true(pg != NULL)) {
    450   1.8       mrg 			pg->flags &= ~PG_BUSY;	/* new page */
    451   1.8       mrg 			UVM_PAGE_OWN(pg, NULL);
    452   1.8       mrg 		}
    453  1.52       chs 		if (obj) {
    454  1.52       chs 			simple_unlock(&obj->vmobjlock);
    455  1.52       chs 		}
    456  1.47       chs 
    457   1.8       mrg 		/*
    458   1.8       mrg 		 * out of memory?
    459   1.8       mrg 		 */
    460   1.8       mrg 
    461  1.35   thorpej 		if (__predict_false(pg == NULL)) {
    462   1.8       mrg 			if (flags & UVM_KMF_NOWAIT) {
    463   1.8       mrg 				/* free everything! */
    464  1.17     chuck 				uvm_unmap(map, kva, kva + size);
    465   1.8       mrg 				return(0);
    466   1.8       mrg 			} else {
    467   1.8       mrg 				uvm_wait("km_getwait2");	/* sleep here */
    468   1.8       mrg 				continue;
    469   1.8       mrg 			}
    470   1.8       mrg 		}
    471  1.47       chs 
    472   1.8       mrg 		/*
    473  1.52       chs 		 * map it in
    474   1.8       mrg 		 */
    475  1.40       chs 
    476  1.52       chs 		if (obj == NULL) {
    477  1.24   thorpej 			pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
    478  1.24   thorpej 			    VM_PROT_ALL);
    479  1.24   thorpej 		} else {
    480  1.24   thorpej 			pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg),
    481  1.33   thorpej 			    UVM_PROT_ALL,
    482  1.33   thorpej 			    PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE);
    483  1.24   thorpej 		}
    484   1.8       mrg 		loopva += PAGE_SIZE;
    485   1.8       mrg 		offset += PAGE_SIZE;
    486  1.44   thorpej 		loopsize -= PAGE_SIZE;
    487   1.8       mrg 	}
    488  1.51     chris 
    489  1.51     chris        	pmap_update(pmap_kernel());
    490  1.51     chris 
    491   1.8       mrg 	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    492   1.8       mrg 	return(kva);
    493   1.1       mrg }
    494   1.1       mrg 
    495   1.1       mrg /*
    496   1.1       mrg  * uvm_km_free: free an area of kernel memory
    497   1.1       mrg  */
    498   1.1       mrg 
    499   1.8       mrg void
    500   1.8       mrg uvm_km_free(map, addr, size)
    501  1.49       chs 	struct vm_map *map;
    502  1.14       eeh 	vaddr_t addr;
    503  1.14       eeh 	vsize_t size;
    504   1.8       mrg {
    505  1.17     chuck 	uvm_unmap(map, trunc_page(addr), round_page(addr+size));
    506   1.1       mrg }
    507   1.1       mrg 
    508   1.1       mrg /*
    509   1.1       mrg  * uvm_km_free_wakeup: free an area of kernel memory and wake up
    510   1.1       mrg  * anyone waiting for vm space.
    511   1.1       mrg  *
    512   1.1       mrg  * => XXX: "wanted" bit + unlock&wait on other end?
    513   1.1       mrg  */
    514   1.1       mrg 
    515   1.8       mrg void
    516   1.8       mrg uvm_km_free_wakeup(map, addr, size)
    517  1.49       chs 	struct vm_map *map;
    518  1.14       eeh 	vaddr_t addr;
    519  1.14       eeh 	vsize_t size;
    520   1.1       mrg {
    521  1.49       chs 	struct vm_map_entry *dead_entries;
    522   1.1       mrg 
    523   1.8       mrg 	vm_map_lock(map);
    524  1.47       chs 	uvm_unmap_remove(map, trunc_page(addr), round_page(addr + size),
    525  1.43       chs 	    &dead_entries);
    526  1.31   thorpej 	wakeup(map);
    527   1.8       mrg 	vm_map_unlock(map);
    528   1.8       mrg 	if (dead_entries != NULL)
    529   1.8       mrg 		uvm_unmap_detach(dead_entries, 0);
    530   1.1       mrg }
    531   1.1       mrg 
    532   1.1       mrg /*
    533   1.1       mrg  * uvm_km_alloc1: allocate wired down memory in the kernel map.
    534   1.1       mrg  *
    535   1.1       mrg  * => we can sleep if needed
    536   1.1       mrg  */
    537   1.1       mrg 
    538  1.14       eeh vaddr_t
    539   1.8       mrg uvm_km_alloc1(map, size, zeroit)
    540  1.49       chs 	struct vm_map *map;
    541  1.14       eeh 	vsize_t size;
    542   1.8       mrg 	boolean_t zeroit;
    543   1.1       mrg {
    544  1.14       eeh 	vaddr_t kva, loopva, offset;
    545   1.8       mrg 	struct vm_page *pg;
    546   1.8       mrg 	UVMHIST_FUNC("uvm_km_alloc1"); UVMHIST_CALLED(maphist);
    547   1.1       mrg 
    548   1.8       mrg 	UVMHIST_LOG(maphist,"(map=0x%x, size=0x%x)", map, size,0,0);
    549  1.40       chs 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    550   1.1       mrg 
    551   1.8       mrg 	size = round_page(size);
    552   1.8       mrg 	kva = vm_map_min(map);		/* hint */
    553   1.1       mrg 
    554   1.8       mrg 	/*
    555   1.8       mrg 	 * allocate some virtual space
    556   1.8       mrg 	 */
    557   1.1       mrg 
    558  1.35   thorpej 	if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object,
    559  1.39   thorpej 	      UVM_UNKNOWN_OFFSET, 0, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
    560  1.35   thorpej 					      UVM_INH_NONE, UVM_ADV_RANDOM,
    561  1.43       chs 					      0)) != 0)) {
    562   1.8       mrg 		UVMHIST_LOG(maphist,"<- done (no VM)",0,0,0,0);
    563   1.8       mrg 		return(0);
    564   1.8       mrg 	}
    565   1.8       mrg 
    566   1.8       mrg 	/*
    567   1.8       mrg 	 * recover object offset from virtual address
    568   1.8       mrg 	 */
    569   1.8       mrg 
    570   1.8       mrg 	offset = kva - vm_map_min(kernel_map);
    571   1.8       mrg 	UVMHIST_LOG(maphist,"  kva=0x%x, offset=0x%x", kva, offset,0,0);
    572   1.8       mrg 
    573   1.8       mrg 	/*
    574  1.52       chs 	 * now allocate the memory.
    575   1.8       mrg 	 */
    576   1.8       mrg 
    577   1.8       mrg 	loopva = kva;
    578   1.8       mrg 	while (size) {
    579   1.8       mrg 		simple_lock(&uvm.kernel_object->vmobjlock);
    580  1.52       chs 		KASSERT(uvm_pagelookup(uvm.kernel_object, offset) == NULL);
    581  1.23       chs 		pg = uvm_pagealloc(uvm.kernel_object, offset, NULL, 0);
    582   1.8       mrg 		if (pg) {
    583  1.52       chs 			pg->flags &= ~PG_BUSY;
    584   1.8       mrg 			UVM_PAGE_OWN(pg, NULL);
    585   1.8       mrg 		}
    586   1.8       mrg 		simple_unlock(&uvm.kernel_object->vmobjlock);
    587  1.52       chs 		if (pg == NULL) {
    588  1.52       chs 			uvm_wait("km_alloc1w");
    589   1.8       mrg 			continue;
    590   1.8       mrg 		}
    591   1.8       mrg 		pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg),
    592  1.33   thorpej 		    UVM_PROT_ALL, PMAP_WIRED | VM_PROT_READ | VM_PROT_WRITE);
    593   1.8       mrg 		loopva += PAGE_SIZE;
    594   1.8       mrg 		offset += PAGE_SIZE;
    595   1.8       mrg 		size -= PAGE_SIZE;
    596   1.8       mrg 	}
    597  1.51     chris 	pmap_update(map->pmap);
    598  1.46   thorpej 
    599   1.8       mrg 	/*
    600   1.8       mrg 	 * zero on request (note that "size" is now zero due to the above loop
    601   1.8       mrg 	 * so we need to subtract kva from loopva to reconstruct the size).
    602   1.8       mrg 	 */
    603   1.1       mrg 
    604   1.8       mrg 	if (zeroit)
    605  1.13     perry 		memset((caddr_t)kva, 0, loopva - kva);
    606   1.8       mrg 	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    607   1.8       mrg 	return(kva);
    608   1.1       mrg }
    609   1.1       mrg 
    610   1.1       mrg /*
    611   1.1       mrg  * uvm_km_valloc: allocate zero-fill memory in the kernel's address space
    612   1.1       mrg  *
    613   1.1       mrg  * => memory is not allocated until fault time
    614   1.1       mrg  */
    615   1.1       mrg 
    616  1.14       eeh vaddr_t
    617   1.8       mrg uvm_km_valloc(map, size)
    618  1.49       chs 	struct vm_map *map;
    619  1.14       eeh 	vsize_t size;
    620   1.1       mrg {
    621  1.41  nisimura 	return(uvm_km_valloc_align(map, size, 0));
    622  1.41  nisimura }
    623  1.41  nisimura 
    624  1.41  nisimura vaddr_t
    625  1.41  nisimura uvm_km_valloc_align(map, size, align)
    626  1.49       chs 	struct vm_map *map;
    627  1.41  nisimura 	vsize_t size;
    628  1.41  nisimura 	vsize_t align;
    629  1.41  nisimura {
    630  1.14       eeh 	vaddr_t kva;
    631   1.8       mrg 	UVMHIST_FUNC("uvm_km_valloc"); UVMHIST_CALLED(maphist);
    632   1.1       mrg 
    633   1.8       mrg 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
    634  1.40       chs 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    635   1.1       mrg 
    636   1.8       mrg 	size = round_page(size);
    637   1.8       mrg 	kva = vm_map_min(map);		/* hint */
    638   1.1       mrg 
    639   1.8       mrg 	/*
    640   1.8       mrg 	 * allocate some virtual space.  will be demand filled by kernel_object.
    641   1.8       mrg 	 */
    642   1.1       mrg 
    643  1.35   thorpej 	if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object,
    644  1.41  nisimura 	    UVM_UNKNOWN_OFFSET, align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
    645  1.35   thorpej 					    UVM_INH_NONE, UVM_ADV_RANDOM,
    646  1.43       chs 					    0)) != 0)) {
    647   1.8       mrg 		UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
    648   1.8       mrg 		return(0);
    649   1.8       mrg 	}
    650   1.1       mrg 
    651   1.8       mrg 	UVMHIST_LOG(maphist, "<- done (kva=0x%x)", kva,0,0,0);
    652   1.8       mrg 	return(kva);
    653   1.1       mrg }
    654   1.1       mrg 
    655   1.1       mrg /*
    656   1.1       mrg  * uvm_km_valloc_wait: allocate zero-fill memory in the kernel's address space
    657   1.1       mrg  *
    658   1.1       mrg  * => memory is not allocated until fault time
    659   1.1       mrg  * => if no room in map, wait for space to free, unless requested size
    660   1.1       mrg  *    is larger than map (in which case we return 0)
    661   1.1       mrg  */
    662   1.1       mrg 
    663  1.14       eeh vaddr_t
    664  1.38     jeffs uvm_km_valloc_prefer_wait(map, size, prefer)
    665  1.49       chs 	struct vm_map *map;
    666  1.14       eeh 	vsize_t size;
    667  1.38     jeffs 	voff_t prefer;
    668   1.1       mrg {
    669  1.14       eeh 	vaddr_t kva;
    670  1.38     jeffs 	UVMHIST_FUNC("uvm_km_valloc_prefer_wait"); UVMHIST_CALLED(maphist);
    671   1.1       mrg 
    672   1.8       mrg 	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
    673  1.40       chs 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    674   1.1       mrg 
    675   1.8       mrg 	size = round_page(size);
    676   1.8       mrg 	if (size > vm_map_max(map) - vm_map_min(map))
    677   1.8       mrg 		return(0);
    678   1.8       mrg 
    679  1.52       chs 	for (;;) {
    680   1.8       mrg 		kva = vm_map_min(map);		/* hint */
    681   1.8       mrg 
    682   1.8       mrg 		/*
    683   1.8       mrg 		 * allocate some virtual space.   will be demand filled
    684   1.8       mrg 		 * by kernel_object.
    685   1.8       mrg 		 */
    686   1.8       mrg 
    687  1.35   thorpej 		if (__predict_true(uvm_map(map, &kva, size, uvm.kernel_object,
    688  1.39   thorpej 		    prefer, 0, UVM_MAPFLAG(UVM_PROT_ALL,
    689   1.8       mrg 		    UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_RANDOM, 0))
    690  1.43       chs 		    == 0)) {
    691   1.8       mrg 			UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    692   1.8       mrg 			return(kva);
    693   1.8       mrg 		}
    694   1.8       mrg 
    695   1.8       mrg 		/*
    696   1.8       mrg 		 * failed.  sleep for a while (on map)
    697   1.8       mrg 		 */
    698   1.8       mrg 
    699   1.8       mrg 		UVMHIST_LOG(maphist,"<<<sleeping>>>",0,0,0,0);
    700   1.8       mrg 		tsleep((caddr_t)map, PVM, "vallocwait", 0);
    701   1.8       mrg 	}
    702   1.8       mrg 	/*NOTREACHED*/
    703  1.38     jeffs }
    704  1.38     jeffs 
    705  1.38     jeffs vaddr_t
    706  1.38     jeffs uvm_km_valloc_wait(map, size)
    707  1.49       chs 	struct vm_map *map;
    708  1.38     jeffs 	vsize_t size;
    709  1.38     jeffs {
    710  1.38     jeffs 	return uvm_km_valloc_prefer_wait(map, size, UVM_UNKNOWN_OFFSET);
    711  1.10   thorpej }
    712  1.10   thorpej 
    713  1.10   thorpej /* Sanity; must specify both or none. */
    714  1.10   thorpej #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
    715  1.10   thorpej     (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
    716  1.10   thorpej #error Must specify MAP and UNMAP together.
    717  1.10   thorpej #endif
    718  1.10   thorpej 
    719  1.10   thorpej /*
    720  1.10   thorpej  * uvm_km_alloc_poolpage: allocate a page for the pool allocator
    721  1.10   thorpej  *
    722  1.10   thorpej  * => if the pmap specifies an alternate mapping method, we use it.
    723  1.10   thorpej  */
    724  1.10   thorpej 
    725  1.11   thorpej /* ARGSUSED */
    726  1.14       eeh vaddr_t
    727  1.15   thorpej uvm_km_alloc_poolpage1(map, obj, waitok)
    728  1.49       chs 	struct vm_map *map;
    729  1.12   thorpej 	struct uvm_object *obj;
    730  1.15   thorpej 	boolean_t waitok;
    731  1.10   thorpej {
    732  1.10   thorpej #if defined(PMAP_MAP_POOLPAGE)
    733  1.10   thorpej 	struct vm_page *pg;
    734  1.14       eeh 	vaddr_t va;
    735  1.10   thorpej 
    736  1.15   thorpej  again:
    737  1.29       chs 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    738  1.35   thorpej 	if (__predict_false(pg == NULL)) {
    739  1.15   thorpej 		if (waitok) {
    740  1.15   thorpej 			uvm_wait("plpg");
    741  1.15   thorpej 			goto again;
    742  1.15   thorpej 		} else
    743  1.15   thorpej 			return (0);
    744  1.15   thorpej 	}
    745  1.10   thorpej 	va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
    746  1.35   thorpej 	if (__predict_false(va == 0))
    747  1.10   thorpej 		uvm_pagefree(pg);
    748  1.10   thorpej 	return (va);
    749  1.10   thorpej #else
    750  1.14       eeh 	vaddr_t va;
    751  1.10   thorpej 	int s;
    752  1.10   thorpej 
    753  1.16   thorpej 	/*
    754  1.42   thorpej 	 * NOTE: We may be called with a map that doens't require splvm
    755  1.16   thorpej 	 * protection (e.g. kernel_map).  However, it does not hurt to
    756  1.42   thorpej 	 * go to splvm in this case (since unprocted maps will never be
    757  1.16   thorpej 	 * accessed in interrupt context).
    758  1.16   thorpej 	 *
    759  1.16   thorpej 	 * XXX We may want to consider changing the interface to this
    760  1.16   thorpej 	 * XXX function.
    761  1.16   thorpej 	 */
    762  1.16   thorpej 
    763  1.42   thorpej 	s = splvm();
    764  1.15   thorpej 	va = uvm_km_kmemalloc(map, obj, PAGE_SIZE, waitok ? 0 : UVM_KMF_NOWAIT);
    765  1.10   thorpej 	splx(s);
    766  1.10   thorpej 	return (va);
    767  1.10   thorpej #endif /* PMAP_MAP_POOLPAGE */
    768  1.10   thorpej }
    769  1.10   thorpej 
    770  1.10   thorpej /*
    771  1.10   thorpej  * uvm_km_free_poolpage: free a previously allocated pool page
    772  1.10   thorpej  *
    773  1.10   thorpej  * => if the pmap specifies an alternate unmapping method, we use it.
    774  1.10   thorpej  */
    775  1.10   thorpej 
    776  1.11   thorpej /* ARGSUSED */
    777  1.10   thorpej void
    778  1.11   thorpej uvm_km_free_poolpage1(map, addr)
    779  1.49       chs 	struct vm_map *map;
    780  1.14       eeh 	vaddr_t addr;
    781  1.10   thorpej {
    782  1.10   thorpej #if defined(PMAP_UNMAP_POOLPAGE)
    783  1.14       eeh 	paddr_t pa;
    784  1.10   thorpej 
    785  1.10   thorpej 	pa = PMAP_UNMAP_POOLPAGE(addr);
    786  1.10   thorpej 	uvm_pagefree(PHYS_TO_VM_PAGE(pa));
    787  1.10   thorpej #else
    788  1.10   thorpej 	int s;
    789  1.16   thorpej 
    790  1.16   thorpej 	/*
    791  1.42   thorpej 	 * NOTE: We may be called with a map that doens't require splvm
    792  1.16   thorpej 	 * protection (e.g. kernel_map).  However, it does not hurt to
    793  1.42   thorpej 	 * go to splvm in this case (since unprocted maps will never be
    794  1.16   thorpej 	 * accessed in interrupt context).
    795  1.16   thorpej 	 *
    796  1.16   thorpej 	 * XXX We may want to consider changing the interface to this
    797  1.16   thorpej 	 * XXX function.
    798  1.16   thorpej 	 */
    799  1.10   thorpej 
    800  1.42   thorpej 	s = splvm();
    801  1.11   thorpej 	uvm_km_free(map, addr, PAGE_SIZE);
    802  1.10   thorpej 	splx(s);
    803  1.10   thorpej #endif /* PMAP_UNMAP_POOLPAGE */
    804   1.1       mrg }
    805