Home | History | Annotate | Line # | Download | only in uvm
uvm_km.c revision 1.7
      1 /*	$NetBSD: uvm_km.c,v 1.7 1998/02/24 15:58:09 chuck Exp $	*/
      2 
      3 /*
      4  * XXXCDC: "ROUGH DRAFT" QUALITY UVM PRE-RELEASE FILE!
      5  *         >>>USE AT YOUR OWN RISK, WORK IS NOT FINISHED<<<
      6  */
      7 /*
      8  * Copyright (c) 1997 Charles D. Cranor and Washington University.
      9  * Copyright (c) 1991, 1993, The Regents of the University of California.
     10  *
     11  * All rights reserved.
     12  *
     13  * This code is derived from software contributed to Berkeley by
     14  * The Mach Operating System project at Carnegie-Mellon University.
     15  *
     16  * Redistribution and use in source and binary forms, with or without
     17  * modification, are permitted provided that the following conditions
     18  * are met:
     19  * 1. Redistributions of source code must retain the above copyright
     20  *    notice, this list of conditions and the following disclaimer.
     21  * 2. Redistributions in binary form must reproduce the above copyright
     22  *    notice, this list of conditions and the following disclaimer in the
     23  *    documentation and/or other materials provided with the distribution.
     24  * 3. All advertising materials mentioning features or use of this software
     25  *    must display the following acknowledgement:
     26  *	This product includes software developed by Charles D. Cranor,
     27  *      Washington University, the University of California, Berkeley and
     28  *      its contributors.
     29  * 4. Neither the name of the University nor the names of its contributors
     30  *    may be used to endorse or promote products derived from this software
     31  *    without specific prior written permission.
     32  *
     33  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     34  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     35  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     36  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     37  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     38  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     39  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     40  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     41  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     42  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     43  * SUCH DAMAGE.
     44  *
     45  *	@(#)vm_kern.c   8.3 (Berkeley) 1/12/94
     46  * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp
     47  *
     48  *
     49  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
     50  * All rights reserved.
     51  *
     52  * Permission to use, copy, modify and distribute this software and
     53  * its documentation is hereby granted, provided that both the copyright
     54  * notice and this permission notice appear in all copies of the
     55  * software, derivative works or modified versions, and any portions
     56  * thereof, and that both notices appear in supporting documentation.
     57  *
     58  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     59  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     60  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     61  *
     62  * Carnegie Mellon requests users of this software to return to
     63  *
     64  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     65  *  School of Computer Science
     66  *  Carnegie Mellon University
     67  *  Pittsburgh PA 15213-3890
     68  *
     69  * any improvements or extensions that they make and grant Carnegie the
     70  * rights to redistribute these changes.
     71  */
     72 
     73 #include "opt_uvmhist.h"
     74 #include "opt_pmap_new.h"
     75 
     76 /*
     77  * uvm_km.c: handle kernel memory allocation and management
     78  */
     79 
     80 /*
     81  * overview of kernel memory management:
     82  *
     83  * the kernel virtual address space is mapped by "kernel_map."   kernel_map
     84  * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
     85  * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
     86  *
     87  * the kernel_map has several "submaps."   submaps can only appear in
     88  * the kernel_map (user processes can't use them).   submaps "take over"
     89  * the management of a sub-range of the kernel's address space.  submaps
     90  * are typically allocated at boot time and are never released.   kernel
     91  * virtual address space that is mapped by a submap is locked by the
     92  * submap's lock -- not the kernel_map's lock.
     93  *
     94  * thus, the useful feature of submaps is that they allow us to break
     95  * up the locking and protection of the kernel address space into smaller
     96  * chunks.
     97  *
     98  * the vm system has several standard kernel submaps, including:
     99  *   kmem_map => contains only wired kernel memory for the kernel
    100  *		malloc.   *** access to kmem_map must be protected
    101  *		by splimp() because we are allowed to call malloc()
    102  *		at interrupt time ***
    103  *   mb_map => memory for large mbufs,  *** protected by splimp ***
    104  *   pager_map => used to map "buf" structures into kernel space
    105  *   exec_map => used during exec to handle exec args
    106  *   etc...
    107  *
    108  * the kernel allocates its private memory out of special uvm_objects whose
    109  * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
    110  * are "special" and never die).   all kernel objects should be thought of
    111  * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
    112  * object is equal to the size of kernel virtual address space (i.e. the
    113  * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
    114  *
    115  * most kernel private memory lives in kernel_object.   the only exception
    116  * to this is for memory that belongs to submaps that must be protected
    117  * by splimp().    each of these submaps has their own private kernel
    118  * object (e.g. kmem_object, mb_object).
    119  *
    120  * note that just because a kernel object spans the entire kernel virutal
    121  * address space doesn't mean that it has to be mapped into the entire space.
    122  * large chunks of a kernel object's space go unused either because
    123  * that area of kernel VM is unmapped, or there is some other type of
    124  * object mapped into that range (e.g. a vnode).    for submap's kernel
    125  * objects, the only part of the object that can ever be populated is the
    126  * offsets that are managed by the submap.
    127  *
    128  * note that the "offset" in a kernel object is always the kernel virtual
    129  * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
    130  * example:
    131  *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
    132  *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
    133  *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
    134  *   then that means that the page at offset 0x235000 in kernel_object is
    135  *   mapped at 0xf8235000.
    136  *
    137  * note that the offsets in kmem_object and mb_object also follow this
    138  * rule.   this means that the offsets for kmem_object must fall in the
    139  * range of [vm_map_min(kmem_object) - vm_map_min(kernel_map)] to
    140  * [vm_map_max(kmem_object) - vm_map_min(kernel_map)], so the offsets
    141  * in those objects will typically not start at zero.
    142  *
    143  * kernel object have one other special property: when the kernel virtual
    144  * memory mapping them is unmapped, the backing memory in the object is
    145  * freed right away.   this is done with the uvm_km_pgremove() function.
    146  * this has to be done because there is no backing store for kernel pages
    147  * and no need to save them after they are no longer referenced.
    148  */
    149 
    150 #include <sys/param.h>
    151 #include <sys/systm.h>
    152 #include <sys/proc.h>
    153 
    154 #include <vm/vm.h>
    155 #include <vm/vm_page.h>
    156 #include <vm/vm_kern.h>
    157 
    158 #include <uvm/uvm.h>
    159 
    160 /*
    161  * global data structures
    162  */
    163 
    164 vm_map_t kernel_map = NULL;
    165 
    166 /*
    167  * local functions
    168  */
    169 
    170 static int uvm_km_get __P((struct uvm_object *, vm_offset_t,
    171                            vm_page_t *, int *, int, vm_prot_t, int, int));
    172 /*
    173  * local data structues
    174  */
    175 
    176 static struct vm_map		kernel_map_store;
    177 static struct uvm_object	kmem_object_store;
    178 static struct uvm_object	mb_object_store;
    179 
    180 static struct uvm_pagerops km_pager = {
    181   NULL,	/* init */
    182   NULL, /* attach */
    183   NULL, /* reference */
    184   NULL, /* detach */
    185   NULL, /* fault */
    186   NULL, /* flush */
    187   uvm_km_get, /* get */
    188   /* ... rest are NULL */
    189 };
    190 
    191 /*
    192  * uvm_km_get: pager get function for kernel objects
    193  *
    194  * => currently we do not support pageout to the swap area, so this
    195  *    pager is very simple.    eventually we may want an anonymous
    196  *    object pager which will do paging.
    197  * => XXXCDC: this pager should be phased out in favor of the aobj pager
    198  */
    199 
    200 
    201 static int uvm_km_get(uobj, offset, pps, npagesp, centeridx, access_type,
    202         	      advice, flags)
    203 
    204 struct uvm_object *uobj;
    205 vm_offset_t offset;
    206 struct vm_page **pps;
    207 int *npagesp;
    208 int centeridx, advice, flags;
    209 vm_prot_t access_type;
    210 
    211 {
    212   vm_offset_t current_offset;
    213   vm_page_t ptmp;
    214   int lcv, gotpages, maxpages;
    215   boolean_t done;
    216   UVMHIST_FUNC("uvm_km_get"); UVMHIST_CALLED(maphist);
    217 
    218   UVMHIST_LOG(maphist, "flags=%d", flags,0,0,0);
    219 
    220   /*
    221    * get number of pages
    222    */
    223 
    224   maxpages = *npagesp;
    225 
    226   /*
    227    * step 1: handled the case where fault data structures are locked.
    228    */
    229 
    230   if (flags & PGO_LOCKED) {
    231 
    232     /*
    233      * step 1a: get pages that are already resident.   only do this
    234      * if the data structures are locked (i.e. the first time through).
    235      */
    236 
    237     done = TRUE;	/* be optimistic */
    238     gotpages = 0;	/* # of pages we got so far */
    239 
    240     for (lcv = 0, current_offset = offset ;
    241 	 lcv < maxpages ; lcv++, current_offset += PAGE_SIZE) {
    242 
    243       /* do we care about this page?  if not, skip it */
    244       if (pps[lcv] == PGO_DONTCARE)
    245 	continue;
    246 
    247       /* lookup page */
    248       ptmp = uvm_pagelookup(uobj, current_offset);
    249 
    250       /* null?  attempt to allocate the page */
    251       if (ptmp == NULL) {
    252 	ptmp = uvm_pagealloc(uobj, current_offset, NULL);
    253 	if (ptmp) {
    254 	  ptmp->flags &= ~(PG_BUSY|PG_FAKE);	/* new page */
    255           UVM_PAGE_OWN(ptmp, NULL);
    256 	  ptmp->wire_count = 1;		/* XXX: prevents pageout attempts */
    257 	  uvm_pagezero(ptmp);
    258 	}
    259       }
    260 
    261       /* to be useful must get a non-busy, non-released page */
    262       if (ptmp == NULL || (ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) {
    263 	if (lcv == centeridx || (flags & PGO_ALLPAGES) != 0)
    264 	  done = FALSE;		/* need to do a wait or I/O! */
    265 	continue;
    266       }
    267 
    268       /* useful page: busy/lock it and plug it in our result array */
    269       ptmp->flags |= PG_BUSY;		/* caller must un-busy this page */
    270       UVM_PAGE_OWN(ptmp, "uvm_km_get1");
    271       pps[lcv] = ptmp;
    272       gotpages++;
    273 
    274     }	/* "for" lcv loop */
    275 
    276     /*
    277      * step 1b: now we've either done everything needed or we to unlock
    278      * and do some waiting or I/O.
    279      */
    280 
    281     UVMHIST_LOG(maphist, "<- done (done=%d)", done, 0,0,0);
    282 
    283     *npagesp = gotpages;
    284     if (done)
    285       return(VM_PAGER_OK);		/* bingo! */
    286     else
    287       return(VM_PAGER_UNLOCK);		/* EEK!   Need to unlock and I/O */
    288   }
    289 
    290   /*
    291    * step 2: get non-resident or busy pages.
    292    * object is locked.   data structures are unlocked.
    293    */
    294 
    295   for (lcv = 0, current_offset = offset ;
    296        lcv < maxpages ; lcv++, current_offset += PAGE_SIZE) {
    297 
    298     /* skip over pages we've already gotten or don't want */
    299     /* skip over pages we don't _have_ to get */
    300     if (pps[lcv] != NULL ||
    301 	(lcv != centeridx && (flags & PGO_ALLPAGES) == 0))
    302       continue;
    303 
    304     /*
    305      * we have yet to locate the current page (pps[lcv]).   we first
    306      * look for a page that is already at the current offset.   if we
    307      * find a page, we check to see if it is busy or released.  if that
    308      * is the case, then we sleep on the page until it is no longer busy
    309      * or released and repeat the lookup.    if the page we found is
    310      * neither busy nor released, then we busy it (so we own it) and
    311      * plug it into pps[lcv].   this 'break's the following while loop
    312      * and indicates we are ready to move on to the next page in the
    313      * "lcv" loop above.
    314      *
    315      * if we exit the while loop with pps[lcv] still set to NULL, then
    316      * it means that we allocated a new busy/fake/clean page ptmp in the
    317      * object and we need to do I/O to fill in the data.
    318      */
    319 
    320     while (pps[lcv] == NULL) {		/* top of "pps" while loop */
    321 
    322       /* look for a current page */
    323       ptmp = uvm_pagelookup(uobj, current_offset);
    324 
    325       /* nope?   allocate one now (if we can) */
    326       if (ptmp == NULL) {
    327 
    328 	ptmp = uvm_pagealloc(uobj, current_offset, NULL);	/* alloc */
    329 
    330 	/* out of RAM? */
    331 	if (ptmp == NULL) {
    332 	  simple_unlock(&uobj->vmobjlock);
    333 	  uvm_wait("kmgetwait1");
    334 	  simple_lock(&uobj->vmobjlock);
    335 	  continue;		/* goto top of pps while loop */
    336 	}
    337 
    338 	/*
    339 	 * got new page ready for I/O.  break pps while loop.  pps[lcv] is
    340 	 * still NULL.
    341 	 */
    342 	break;
    343       }
    344 
    345       /* page is there, see if we need to wait on it */
    346       if ((ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) {
    347 	ptmp->flags |= PG_WANTED;
    348 	UVM_UNLOCK_AND_WAIT(ptmp,&uobj->vmobjlock,0,"uvn_get",0);
    349 	simple_lock(&uobj->vmobjlock);
    350 	continue;		/* goto top of pps while loop */
    351       }
    352 
    353       /*
    354        * if we get here then the page has become resident and unbusy
    355        * between steps 1 and 2.  we busy it now (so we own it) and set
    356        * pps[lcv] (so that we exit the while loop).
    357        */
    358       ptmp->flags |= PG_BUSY;	/* we own it, caller must un-busy */
    359       UVM_PAGE_OWN(ptmp, "uvm_km_get2");
    360       pps[lcv] = ptmp;
    361     }
    362 
    363     /*
    364      * if we own the a valid page at the correct offset, pps[lcv] will
    365      * point to it.   nothing more to do except go to the next page.
    366      */
    367 
    368     if (pps[lcv])
    369       continue;			/* next lcv */
    370 
    371     /*
    372      * we have a "fake/busy/clean" page that we just allocated.
    373      * do the needed "i/o" (in this case that means zero it).
    374      */
    375 
    376     uvm_pagezero(ptmp);
    377     ptmp->flags &= ~(PG_FAKE);
    378     ptmp->wire_count = 1;		/* XXX: prevents pageout attempts */
    379     pps[lcv] = ptmp;
    380 
    381   }	/* lcv loop */
    382 
    383   /*
    384    * finally, unlock object and return.
    385    */
    386 
    387   simple_unlock(&uobj->vmobjlock);
    388   UVMHIST_LOG(maphist, "<- done (OK)",0,0,0,0);
    389   return(VM_PAGER_OK);
    390 }
    391 
    392 /*
    393  * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
    394  * KVM already allocated for text, data, bss, and static data structures).
    395  *
    396  * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
    397  *    we assume that [min -> start] has already been allocated and that
    398  *    "end" is the end.
    399  */
    400 
    401 void uvm_km_init(start, end)
    402 
    403 vm_offset_t start, end;
    404 
    405 {
    406   vm_offset_t base = VM_MIN_KERNEL_ADDRESS;
    407 
    408   /*
    409    * first, init kernel memory objects.
    410    */
    411 
    412   /* kernel_object: for pageable anonymous kernel memory */
    413   uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
    414 				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
    415 
    416   /* kmem_object: for malloc'd memory (always wired, protected by splimp) */
    417   simple_lock_init(&kmem_object_store.vmobjlock);
    418   kmem_object_store.pgops = &km_pager;
    419   TAILQ_INIT(&kmem_object_store.memq);
    420   kmem_object_store.uo_npages = 0;
    421   kmem_object_store.uo_refs = UVM_OBJ_KERN;
    422 					/* we are special.  we never die */
    423   uvmexp.kmem_object = &kmem_object_store;
    424 
    425   /* mb_object: for mbuf memory (always wired, protected by splimp) */
    426   simple_lock_init(&mb_object_store.vmobjlock);
    427   mb_object_store.pgops = &km_pager;
    428   TAILQ_INIT(&mb_object_store.memq);
    429   mb_object_store.uo_npages = 0;
    430   mb_object_store.uo_refs = UVM_OBJ_KERN;
    431 					/* we are special.  we never die */
    432   uvmexp.mb_object = &mb_object_store;
    433 
    434   /*
    435    * init the map and reserve allready allocated kernel space
    436    * before installing.
    437    */
    438 
    439   uvm_map_setup(&kernel_map_store, base, end, FALSE);
    440   kernel_map_store.pmap = pmap_kernel();
    441   if (uvm_map(&kernel_map_store, &base, start - base, NULL, UVM_UNKNOWN_OFFSET,
    442 	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    443 			  UVM_ADV_RANDOM,UVM_FLAG_FIXED)) != KERN_SUCCESS)
    444     panic("uvm_km_init: could not reserve space for kernel");
    445 
    446   /*
    447    * install!
    448    */
    449 
    450   kernel_map = &kernel_map_store;
    451 }
    452 
    453 /*
    454  * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
    455  * is allocated all references to that area of VM must go through it.  this
    456  * allows the locking of VAs in kernel_map to be broken up into regions.
    457  *
    458  * => if `fixed' is true, *min specifies where the region described
    459  *      by the submap must start
    460  * => if submap is non NULL we use that as the submap, otherwise we
    461  *	alloc a new map
    462  */
    463 
    464 struct vm_map *uvm_km_suballoc(map, min, max, size, pageable, fixed, submap)
    465 
    466 struct vm_map *map;
    467 vm_offset_t *min, *max;		/* OUT, OUT */
    468 vm_size_t size;
    469 boolean_t pageable;
    470 boolean_t fixed;
    471 struct vm_map *submap;
    472 
    473 {
    474   int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
    475 
    476   size = round_page(size);	/* round up to pagesize */
    477 
    478   /*
    479    * first allocate a blank spot in the parent map
    480    */
    481 
    482   if (uvm_map(map, min, size, NULL, UVM_UNKNOWN_OFFSET,
    483 	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    484 			  UVM_ADV_RANDOM, mapflags)) != KERN_SUCCESS) {
    485     panic("uvm_km_suballoc: unable to allocate space in parent map");
    486   }
    487 
    488   /*
    489    * set VM bounds (min is filled in by uvm_map)
    490    */
    491 
    492   *max = *min + size;
    493 
    494   /*
    495    * add references to pmap and create or init the submap
    496    */
    497 
    498   pmap_reference(vm_map_pmap(map));
    499   if (submap == NULL) {
    500     submap = uvm_map_create(vm_map_pmap(map), *min, *max, pageable);
    501     if (submap == NULL)
    502       panic("uvm_km_suballoc: unable to create submap");
    503   } else {
    504       uvm_map_setup(submap, *min, *max, pageable);
    505       submap->pmap = vm_map_pmap(map);
    506   }
    507 
    508   /*
    509    * now let uvm_map_submap plug in it...
    510    */
    511 
    512   if (uvm_map_submap(map, *min, *max, submap) != KERN_SUCCESS)
    513     panic("uvm_km_suballoc: submap allocation failed");
    514 
    515   return(submap);
    516 }
    517 
    518 /*
    519  * uvm_km_pgremove: remove pages from a kernel uvm_object.
    520  *
    521  * => when you unmap a part of anonymous kernel memory you want to toss
    522  *    the pages right away.    (this gets called from uvm_unmap_...).
    523  */
    524 
    525 #define UKM_HASH_PENALTY 4      /* a guess */
    526 
    527 void uvm_km_pgremove(uobj, start, end)
    528 
    529 struct uvm_object *uobj;
    530 vm_offset_t start, end;
    531 
    532 {
    533   boolean_t by_list, is_aobj;
    534   struct vm_page *pp, *ppnext;
    535   vm_offset_t curoff;
    536   UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
    537 
    538   simple_lock(&uobj->vmobjlock);		/* lock object */
    539 
    540   /* is uobj an aobj? */
    541   is_aobj = uobj->pgops == &aobj_pager;
    542 
    543   /* choose cheapest traversal */
    544   by_list = (uobj->uo_npages <=
    545 	     ((end - start) / PAGE_SIZE) * UKM_HASH_PENALTY);
    546 
    547   if (by_list)
    548     goto loop_by_list;
    549 
    550   /* by hash */
    551 
    552   for (curoff = start ; curoff < end ; curoff += PAGE_SIZE) {
    553     pp = uvm_pagelookup(uobj, curoff);
    554     if (pp == NULL)
    555       continue;
    556 
    557     UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,pp->flags & PG_BUSY,0,0);
    558     /* now do the actual work */
    559     if (pp->flags & PG_BUSY)
    560       pp->flags |= PG_RELEASED;	/* owner must check for this when done */
    561     else {
    562       pmap_page_protect(PMAP_PGARG(pp), VM_PROT_NONE);
    563 
    564       /*
    565        * if this kernel object is an aobj, free the swap slot.
    566        */
    567       if (is_aobj) {
    568 	int slot = uao_set_swslot(uobj, curoff / PAGE_SIZE, 0);
    569 
    570 	if (slot)
    571 	  uvm_swap_free(slot, 1);
    572       }
    573 
    574       uvm_lock_pageq();
    575       uvm_pagefree(pp);
    576       uvm_unlock_pageq();
    577     }
    578     /* done */
    579 
    580   }
    581   simple_unlock(&uobj->vmobjlock);
    582   return;
    583 
    584 loop_by_list:
    585 
    586   for (pp = uobj->memq.tqh_first ; pp != NULL ; pp = ppnext) {
    587 
    588     ppnext = pp->listq.tqe_next;
    589     if (pp->offset < start || pp->offset >= end) {
    590       continue;
    591     }
    592 
    593     UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,pp->flags & PG_BUSY,0,0);
    594     /* now do the actual work */
    595     if (pp->flags & PG_BUSY)
    596       pp->flags |= PG_RELEASED;	/* owner must check for this when done */
    597     else {
    598       pmap_page_protect(PMAP_PGARG(pp), VM_PROT_NONE);
    599 
    600       /*
    601        * if this kernel object is an aobj, free the swap slot.
    602        */
    603       if (is_aobj) {
    604 	int slot = uao_set_swslot(uobj, pp->offset / PAGE_SIZE, 0);
    605 
    606 	if (slot)
    607 	  uvm_swap_free(slot, 1);
    608       }
    609 
    610       uvm_lock_pageq();
    611       uvm_pagefree(pp);
    612       uvm_unlock_pageq();
    613     }
    614     /* done */
    615 
    616   }
    617   simple_unlock(&uobj->vmobjlock);
    618   return;
    619 }
    620 
    621 
    622 /*
    623  * uvm_km_kmemalloc: lower level kernel memory allocator for malloc()
    624  *
    625  * => we map wired memory into the specified map using the obj passed in
    626  * => NOTE: we can return NULL even if we can wait if there is not enough
    627  *	free VM space in the map... caller should be prepared to handle
    628  *	this case.
    629  * => we return KVA of memory allocated
    630  * => flags: NOWAIT, VALLOC - just allocate VA, TRYLOCK - fail if we can't
    631  *	lock the map
    632  */
    633 
    634 vm_offset_t uvm_km_kmemalloc(map, obj, size, flags)
    635 
    636 vm_map_t map;
    637 struct uvm_object *obj;
    638 vm_size_t size;
    639 int flags;
    640 
    641 {
    642   vm_offset_t kva, loopva;
    643   vm_offset_t offset;
    644   struct vm_page *pg;
    645   UVMHIST_FUNC("uvm_km_kmemalloc"); UVMHIST_CALLED(maphist);
    646 
    647 
    648   UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
    649 	map, obj, size, flags);
    650 #ifdef DIAGNOSTIC
    651   /* sanity check */
    652   if (vm_map_pmap(map) != pmap_kernel())
    653     panic("uvm_km_kmemalloc: invalid map");
    654 #endif
    655 
    656   /*
    657    * setup for call
    658    */
    659 
    660   size = round_page(size);
    661   kva = vm_map_min(map);	/* hint */
    662 
    663   /*
    664    * allocate some virtual space
    665    */
    666 
    667   if (uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
    668 	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    669 			  UVM_ADV_RANDOM, (flags & UVM_KMF_TRYLOCK)))
    670       != KERN_SUCCESS) {
    671     UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
    672     return(0);
    673   }
    674 
    675   /*
    676    * if all we wanted was VA, return now
    677    */
    678 
    679   if (flags & UVM_KMF_VALLOC) {
    680     UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
    681     return(kva);
    682   }
    683   /*
    684    * recover object offset from virtual address
    685    */
    686 
    687   offset = kva - vm_map_min(kernel_map);
    688   UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
    689 
    690   /*
    691    * now allocate and map in the memory... note that we are the only ones
    692    * whom should ever get a handle on this area of VM.
    693    */
    694 
    695   loopva = kva;
    696   while (size) {
    697     simple_lock(&obj->vmobjlock);
    698     pg = uvm_pagealloc(obj, offset, NULL);
    699     if (pg) {
    700       pg->flags &= ~PG_BUSY;	/* new page */
    701       UVM_PAGE_OWN(pg, NULL);
    702 
    703       pg->wire_count = 1;
    704       uvmexp.wired++;
    705     }
    706     simple_unlock(&obj->vmobjlock);
    707 
    708     /*
    709      * out of memory?
    710      */
    711 
    712     if (pg == NULL) {
    713       if (flags & UVM_KMF_NOWAIT) {
    714 	uvm_unmap(map, kva, kva + size, 0); /* free everything! */
    715 	return(0);
    716       } else {
    717 	uvm_wait("km_getwait2");		/* sleep here */
    718 	continue;
    719       }
    720     }
    721 
    722     /*
    723      * map it in: note that we call pmap_enter with the map and object
    724      * unlocked in case we are kmem_map/kmem_object (because if pmap_enter
    725      * wants to allocate out of kmem_object it will need to lock it itself!)
    726      */
    727 #if defined(PMAP_NEW)
    728     pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), VM_PROT_ALL);
    729 #else
    730     pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), UVM_PROT_ALL, TRUE);
    731 #endif
    732     loopva += PAGE_SIZE;
    733     offset += PAGE_SIZE;
    734     size -= PAGE_SIZE;
    735   }
    736 
    737   UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    738   return(kva);
    739 }
    740 
    741 /*
    742  * uvm_km_free: free an area of kernel memory
    743  */
    744 
    745 void uvm_km_free(map, addr, size)
    746 
    747 vm_map_t map;
    748 vm_offset_t addr;
    749 vm_size_t size;
    750 
    751 {
    752   uvm_unmap(map, trunc_page(addr), round_page(addr+size), 1);
    753 }
    754 
    755 /*
    756  * uvm_km_free_wakeup: free an area of kernel memory and wake up
    757  * anyone waiting for vm space.
    758  *
    759  * => XXX: "wanted" bit + unlock&wait on other end?
    760  */
    761 
    762 void uvm_km_free_wakeup(map, addr, size)
    763 
    764 vm_map_t map;
    765 vm_offset_t addr;
    766 vm_size_t size;
    767 
    768 {
    769   vm_map_entry_t dead_entries;
    770 
    771   vm_map_lock(map);
    772   (void)uvm_unmap_remove(map, trunc_page(addr), round_page(addr+size), 1,
    773 			 &dead_entries);
    774   thread_wakeup(map);
    775   vm_map_unlock(map);
    776 
    777   if (dead_entries != NULL)
    778     uvm_unmap_detach(dead_entries, 0);
    779 }
    780 
    781 /*
    782  * uvm_km_alloc1: allocate wired down memory in the kernel map.
    783  *
    784  * => we can sleep if needed
    785  */
    786 
    787 vm_offset_t uvm_km_alloc1(map, size, zeroit)
    788 
    789 vm_map_t map;
    790 vm_size_t size;
    791 boolean_t zeroit;
    792 
    793 {
    794   vm_offset_t kva, loopva, offset;
    795   struct vm_page *pg;
    796   UVMHIST_FUNC("uvm_km_alloc1"); UVMHIST_CALLED(maphist);
    797 
    798   UVMHIST_LOG(maphist,"(map=0x%x, size=0x%x)", map, size,0,0);
    799 
    800 #ifdef DIAGNOSTIC
    801   if (vm_map_pmap(map) != pmap_kernel())
    802     panic("uvm_km_alloc1");
    803 #endif
    804 
    805   size = round_page(size);
    806   kva = vm_map_min(map);		/* hint */
    807 
    808   /*
    809    * allocate some virtual space
    810    */
    811 
    812   if (uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET,
    813 	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    814 			  UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) {
    815     UVMHIST_LOG(maphist,"<- done (no VM)",0,0,0,0);
    816     return(0);
    817   }
    818 
    819   /*
    820    * recover object offset from virtual address
    821    */
    822 
    823   offset = kva - vm_map_min(kernel_map);
    824   UVMHIST_LOG(maphist,"  kva=0x%x, offset=0x%x", kva, offset,0,0);
    825 
    826   /*
    827    * now allocate the memory.  we must be careful about released pages.
    828    */
    829 
    830   loopva = kva;
    831   while (size) {
    832     simple_lock(&uvm.kernel_object->vmobjlock);
    833     pg = uvm_pagelookup(uvm.kernel_object, offset);
    834 
    835     /* if we found a page in an unallocated region, it must be released */
    836     if (pg) {
    837       if ((pg->flags & PG_RELEASED) == 0)
    838 	panic("uvm_km_alloc1: non-released page");
    839       pg->flags |= PG_WANTED;
    840       UVM_UNLOCK_AND_WAIT(pg, &uvm.kernel_object->vmobjlock,0,"km_alloc",0);
    841       continue;   /* retry */
    842     }
    843 
    844     /* allocate ram */
    845     pg = uvm_pagealloc(uvm.kernel_object, offset, NULL);
    846     if (pg) {
    847       pg->flags &= ~PG_BUSY;	/* new page */
    848       UVM_PAGE_OWN(pg, NULL);
    849     }
    850     simple_unlock(&uvm.kernel_object->vmobjlock);
    851     if (pg == NULL) {
    852       uvm_wait("km_alloc1w");	/* wait for memory */
    853       continue;
    854     }
    855 
    856     /* map it in */
    857 #if defined(PMAP_NEW)
    858     pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), UVM_PROT_ALL);
    859 #else
    860     pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg), UVM_PROT_ALL, TRUE);
    861 #endif
    862     loopva += PAGE_SIZE;
    863     offset += PAGE_SIZE;
    864     size -= PAGE_SIZE;
    865   }
    866 
    867   /*
    868    * zero on request (note that "size" is now zero due to the above loop
    869    * so we need to subtract kva from loopva to reconstruct the size).
    870    */
    871 
    872   if (zeroit)
    873     bzero((caddr_t)kva, loopva - kva);
    874 
    875   UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    876   return(kva);
    877 }
    878 
    879 /*
    880  * uvm_km_valloc: allocate zero-fill memory in the kernel's address space
    881  *
    882  * => memory is not allocated until fault time
    883  */
    884 
    885 vm_offset_t uvm_km_valloc(map, size)
    886 
    887 vm_map_t map;
    888 vm_size_t size;
    889 
    890 {
    891   vm_offset_t kva;
    892   UVMHIST_FUNC("uvm_km_valloc"); UVMHIST_CALLED(maphist);
    893 
    894   UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
    895 
    896 #ifdef DIAGNOSTIC
    897   if (vm_map_pmap(map) != pmap_kernel())
    898     panic("uvm_km_valloc");
    899 #endif
    900 
    901   size = round_page(size);
    902   kva = vm_map_min(map);		/* hint */
    903 
    904   /*
    905    * allocate some virtual space.   will be demand filled by kernel_object.
    906    */
    907 
    908   if (uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET,
    909 	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    910 			  UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) {
    911     UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
    912     return(0);
    913   }
    914 
    915   UVMHIST_LOG(maphist, "<- done (kva=0x%x)", kva,0,0,0);
    916   return(kva);
    917 }
    918 
    919 /*
    920  * uvm_km_valloc_wait: allocate zero-fill memory in the kernel's address space
    921  *
    922  * => memory is not allocated until fault time
    923  * => if no room in map, wait for space to free, unless requested size
    924  *    is larger than map (in which case we return 0)
    925  */
    926 
    927 vm_offset_t uvm_km_valloc_wait(map, size)
    928 
    929 vm_map_t map;
    930 vm_size_t size;
    931 
    932 {
    933   vm_offset_t kva;
    934   UVMHIST_FUNC("uvm_km_valloc_wait"); UVMHIST_CALLED(maphist);
    935 
    936   UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
    937 
    938 #ifdef DIAGNOSTIC
    939   if (vm_map_pmap(map) != pmap_kernel())
    940     panic("uvm_km_valloc_wait");
    941 #endif
    942 
    943   size = round_page(size);
    944   if (size > vm_map_max(map) - vm_map_min(map))
    945     return(0);
    946 
    947   while (1) {
    948     kva = vm_map_min(map);		/* hint */
    949 
    950     /*
    951      * allocate some virtual space.   will be demand filled by kernel_object.
    952      */
    953 
    954     if (uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET,
    955 		UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    956 			    UVM_ADV_RANDOM, 0)) == KERN_SUCCESS){
    957       UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
    958       return(kva);
    959     }
    960 
    961     /*
    962      * failed.  sleep for a while (on map)
    963      */
    964 
    965     UVMHIST_LOG(maphist,"<<<sleeping>>>",0,0,0,0);
    966     tsleep((caddr_t)map, PVM, "vallocwait", 0);
    967   }
    968   /*NOTREACHED*/
    969 }
    970