NetBSD/sys/uvm/uvm_km.c

825 lines
23 KiB
C
Raw Normal View History

2007-07-22 00:52:59 +04:00
/* $NetBSD: uvm_km.c,v 1.96 2007/07/21 20:52:59 ad Exp $ */
2001-05-25 08:06:11 +04:00
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
2001-05-25 08:06:11 +04:00
* Copyright (c) 1991, 1993, The Regents of the University of California.
*
* All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* The Mach Operating System project at Carnegie-Mellon University.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Charles D. Cranor,
2001-05-25 08:06:11 +04:00
* Washington University, the University of California, Berkeley and
* its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
1998-02-07 14:07:38 +03:00
* from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp
*
*
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
* All rights reserved.
2001-05-25 08:06:11 +04:00
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
2001-05-25 08:06:11 +04:00
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
2001-05-25 08:06:11 +04:00
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/*
* uvm_km.c: handle kernel memory allocation and management
*/
/*
* overview of kernel memory management:
*
* the kernel virtual address space is mapped by "kernel_map." kernel_map
* starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
* note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
*
2001-05-25 08:06:11 +04:00
* the kernel_map has several "submaps." submaps can only appear in
* the kernel_map (user processes can't use them). submaps "take over"
* the management of a sub-range of the kernel's address space. submaps
* are typically allocated at boot time and are never released. kernel
2001-05-25 08:06:11 +04:00
* virtual address space that is mapped by a submap is locked by the
* submap's lock -- not the kernel_map's lock.
*
* thus, the useful feature of submaps is that they allow us to break
* up the locking and protection of the kernel address space into smaller
* chunks.
*
* the vm system has several standard kernel submaps, including:
* kmem_map => contains only wired kernel memory for the kernel
* malloc. *** access to kmem_map must be protected
2001-01-14 05:10:01 +03:00
* by splvm() because we are allowed to call malloc()
* at interrupt time ***
2001-01-14 05:10:01 +03:00
* mb_map => memory for large mbufs, *** protected by splvm ***
* pager_map => used to map "buf" structures into kernel space
* exec_map => used during exec to handle exec args
* etc...
*
* the kernel allocates its private memory out of special uvm_objects whose
* reference count is set to UVM_OBJ_KERN (thus indicating that the objects
* are "special" and never die). all kernel objects should be thought of
2001-05-25 08:06:11 +04:00
* as large, fixed-sized, sparsely populated uvm_objects. each kernel
* object is equal to the size of kernel virtual address space (i.e. the
* value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
*
* note that just because a kernel object spans the entire kernel virutal
* address space doesn't mean that it has to be mapped into the entire space.
2001-05-25 08:06:11 +04:00
* large chunks of a kernel object's space go unused either because
* that area of kernel VM is unmapped, or there is some other type of
* object mapped into that range (e.g. a vnode). for submap's kernel
* objects, the only part of the object that can ever be populated is the
* offsets that are managed by the submap.
*
* note that the "offset" in a kernel object is always the kernel virtual
* address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
* example:
* suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
* uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
* kernel map]. if uvm_km_alloc returns virtual address 0xf8235000,
* then that means that the page at offset 0x235000 in kernel_object is
2001-05-25 08:06:11 +04:00
* mapped at 0xf8235000.
*
* kernel object have one other special property: when the kernel virtual
* memory mapping them is unmapped, the backing memory in the object is
* freed right away. this is done with the uvm_km_pgremove() function.
* this has to be done because there is no backing store for kernel pages
* and no need to save them after they are no longer referenced.
*/
#include <sys/cdefs.h>
2007-07-22 00:52:59 +04:00
__KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.96 2007/07/21 20:52:59 ad Exp $");
#include "opt_uvmhist.h"
#include <sys/param.h>
#include <sys/malloc.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/pool.h>
#include <uvm/uvm.h>
/*
* global data structures
*/
struct vm_map *kernel_map = NULL;
/*
* local data structues
*/
static struct vm_map_kernel kernel_map_store;
static struct vm_map_entry kernel_first_mapent_store;
#if !defined(PMAP_MAP_POOLPAGE)
/*
* kva cache
*
* XXX maybe it's better to do this at the uvm_map layer.
*/
#define KM_VACACHE_SIZE (32 * PAGE_SIZE) /* XXX tune */
static void *km_vacache_alloc(struct pool *, int);
static void km_vacache_free(struct pool *, void *);
static void km_vacache_init(struct vm_map *, const char *, size_t);
/* XXX */
#define KM_VACACHE_POOL_TO_MAP(pp) \
((struct vm_map *)((char *)(pp) - \
offsetof(struct vm_map_kernel, vmk_vacache)))
static void *
km_vacache_alloc(struct pool *pp, int flags)
{
vaddr_t va;
size_t size;
struct vm_map *map;
size = pp->pr_alloc->pa_pagesz;
map = KM_VACACHE_POOL_TO_MAP(pp);
va = vm_map_min(map); /* hint */
if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
return NULL;
return (void *)va;
}
static void
km_vacache_free(struct pool *pp, void *v)
{
vaddr_t va = (vaddr_t)v;
size_t size = pp->pr_alloc->pa_pagesz;
struct vm_map *map;
map = KM_VACACHE_POOL_TO_MAP(pp);
uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
}
/*
* km_vacache_init: initialize kva cache.
*/
static void
km_vacache_init(struct vm_map *map, const char *name, size_t size)
{
struct vm_map_kernel *vmk;
struct pool *pp;
struct pool_allocator *pa;
int ipl;
KASSERT(VM_MAP_IS_KERNEL(map));
KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
vmk = vm_map_to_kernel(map);
pp = &vmk->vmk_vacache;
pa = &vmk->vmk_vacache_allocator;
memset(pa, 0, sizeof(*pa));
pa->pa_alloc = km_vacache_alloc;
pa->pa_free = km_vacache_free;
pa->pa_pagesz = (unsigned int)size;
pa->pa_backingmap = map;
pa->pa_backingmapptr = NULL;
if ((map->flags & VM_MAP_INTRSAFE) != 0)
ipl = IPL_VM;
else
ipl = IPL_NONE;
pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa,
ipl);
}
void
uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
{
map->flags |= VM_MAP_VACACHE;
if (size == 0)
size = KM_VACACHE_SIZE;
km_vacache_init(map, name, size);
}
#else /* !defined(PMAP_MAP_POOLPAGE) */
void
uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
{
/* nothing */
}
#endif /* !defined(PMAP_MAP_POOLPAGE) */
void
uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags)
{
struct vm_map_kernel *vmk = vm_map_to_kernel(map);
const bool intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
int s = 0xdeadbeaf; /* XXX: gcc */
if (intrsafe) {
s = splvm();
}
callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
if (intrsafe) {
splx(s);
}
}
/*
* uvm_km_init: init kernel maps and objects to reflect reality (i.e.
* KVM already allocated for text, data, bss, and static data structures).
*
* => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
* we assume that [vmin -> start] has already been allocated and that
* "end" is the end.
*/
1998-03-09 03:58:55 +03:00
void
2005-06-27 06:19:48 +04:00
uvm_km_init(vaddr_t start, vaddr_t end)
{
vaddr_t base = VM_MIN_KERNEL_ADDRESS;
/*
* next, init kernel memory objects.
1998-03-09 03:58:55 +03:00
*/
1998-03-09 03:58:55 +03:00
/* kernel_object: for pageable anonymous kernel memory */
uao_init();
uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
1998-03-09 03:58:55 +03:00
/*
* init the map and reserve any space that might already
* have been allocated kernel space before installing.
1998-03-09 03:58:55 +03:00
*/
uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
kernel_map_store.vmk_map.pmap = pmap_kernel();
if (start != base) {
int error;
struct uvm_map_args args;
error = uvm_map_prepare(&kernel_map_store.vmk_map,
base, start - base,
NULL, UVM_UNKNOWN_OFFSET, 0,
UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
if (!error) {
kernel_first_mapent_store.flags =
UVM_MAP_KERNEL | UVM_MAP_FIRST;
error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
&kernel_first_mapent_store);
}
if (error)
panic(
"uvm_km_init: could not reserve space for kernel");
}
2001-05-25 08:06:11 +04:00
1998-03-09 03:58:55 +03:00
/*
* install!
*/
kernel_map = &kernel_map_store.vmk_map;
uvm_km_vacache_init(kernel_map, "kvakernel", 0);
}
/*
* uvm_km_suballoc: allocate a submap in the kernel map. once a submap
* is allocated all references to that area of VM must go through it. this
* allows the locking of VAs in kernel_map to be broken up into regions.
*
* => if `fixed' is true, *vmin specifies where the region described
* by the submap must start
* => if submap is non NULL we use that as the submap, otherwise we
* alloc a new map
*/
1998-03-09 03:58:55 +03:00
struct vm_map *
2005-06-27 06:19:48 +04:00
uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed,
2005-06-27 06:19:48 +04:00
struct vm_map_kernel *submap)
1998-03-09 03:58:55 +03:00
{
int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
KASSERT(vm_map_pmap(map) == pmap_kernel());
1998-03-09 03:58:55 +03:00
size = round_page(size); /* round up to pagesize */
size += uvm_mapent_overhead(size, flags);
1998-03-09 03:58:55 +03:00
/*
* first allocate a blank spot in the parent map
*/
if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
1998-03-09 03:58:55 +03:00
UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM, mapflags)) != 0) {
1998-03-09 03:58:55 +03:00
panic("uvm_km_suballoc: unable to allocate space in parent map");
}
1998-03-09 03:58:55 +03:00
/*
* set VM bounds (vmin is filled in by uvm_map)
1998-03-09 03:58:55 +03:00
*/
*vmax = *vmin + size;
1998-03-09 03:58:55 +03:00
/*
* add references to pmap and create or init the submap
*/
pmap_reference(vm_map_pmap(map));
if (submap == NULL) {
submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
1998-03-09 03:58:55 +03:00
if (submap == NULL)
panic("uvm_km_suballoc: unable to create submap");
}
uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
submap->vmk_map.pmap = vm_map_pmap(map);
1998-03-09 03:58:55 +03:00
/*
* now let uvm_map_submap plug in it...
*/
if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
1998-03-09 03:58:55 +03:00
panic("uvm_km_suballoc: submap allocation failed");
return(&submap->vmk_map);
}
/*
* uvm_km_pgremove: remove pages from a kernel uvm_object.
*
* => when you unmap a part of anonymous kernel memory you want to toss
* the pages right away. (this gets called from uvm_unmap_...).
*/
1998-03-09 03:58:55 +03:00
void
2005-06-27 06:19:48 +04:00
uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
{
struct uvm_object * const uobj = uvm_kernel_object;
const voff_t start = startva - vm_map_min(kernel_map);
const voff_t end = endva - vm_map_min(kernel_map);
struct vm_page *pg;
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
voff_t curoff, nextoff;
int swpgonlydelta = 0;
1998-03-09 03:58:55 +03:00
UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
KASSERT(startva < endva);
KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
simple_lock(&uobj->vmobjlock);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
for (curoff = start; curoff < end; curoff = nextoff) {
nextoff = curoff + PAGE_SIZE;
pg = uvm_pagelookup(uobj, curoff);
if (pg != NULL && pg->flags & PG_BUSY) {
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
pg->flags |= PG_WANTED;
UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
"km_pgrm", 0);
simple_lock(&uobj->vmobjlock);
nextoff = curoff;
continue;
}
1998-03-09 03:58:55 +03:00
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
/*
* free the swap slot, then the page.
*/
if (pg == NULL &&
uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
swpgonlydelta++;
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
}
uao_dropswap(uobj, curoff >> PAGE_SHIFT);
if (pg != NULL) {
uvm_lock_pageq();
uvm_pagefree(pg);
uvm_unlock_pageq();
1998-03-09 03:58:55 +03:00
}
}
simple_unlock(&uobj->vmobjlock);
if (swpgonlydelta > 0) {
mutex_enter(&uvm_swap_data_lock);
KASSERT(uvmexp.swpgonly >= swpgonlydelta);
uvmexp.swpgonly -= swpgonlydelta;
mutex_exit(&uvm_swap_data_lock);
}
}
/*
* uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
* regions.
*
* => when you unmap a part of anonymous kernel memory you want to toss
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
* the pages right away. (this is called from uvm_unmap_...).
* => none of the pages will ever be busy, and none of them will ever
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
* be on the active or inactive queues (because they have no object).
*/
void
2005-06-27 06:19:48 +04:00
uvm_km_pgremove_intrsafe(vaddr_t start, vaddr_t end)
{
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
struct vm_page *pg;
paddr_t pa;
UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
KASSERT(VM_MIN_KERNEL_ADDRESS <= start);
KASSERT(start < end);
KASSERT(end <= VM_MAX_KERNEL_ADDRESS);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
for (; start < end; start += PAGE_SIZE) {
if (!pmap_extract(pmap_kernel(), start, &pa)) {
continue;
}
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
pg = PHYS_TO_VM_PAGE(pa);
KASSERT(pg);
KASSERT(pg->uobject == NULL && pg->uanon == NULL);
uvm_pagefree(pg);
}
}
#if defined(DEBUG)
void
uvm_km_check_empty(vaddr_t start, vaddr_t end, bool intrsafe)
{
vaddr_t va;
paddr_t pa;
KDASSERT(VM_MIN_KERNEL_ADDRESS <= start);
KDASSERT(start < end);
2006-03-17 12:37:55 +03:00
KDASSERT(end <= VM_MAX_KERNEL_ADDRESS);
for (va = start; va < end; va += PAGE_SIZE) {
if (pmap_extract(pmap_kernel(), va, &pa)) {
panic("uvm_km_check_empty: va %p has pa 0x%llx",
(void *)va, (long long)pa);
}
if (!intrsafe) {
const struct vm_page *pg;
2007-07-22 00:52:59 +04:00
simple_lock(&uvm_kernel_object->vmobjlock);
pg = uvm_pagelookup(uvm_kernel_object,
va - vm_map_min(kernel_map));
2007-07-22 00:52:59 +04:00
simple_unlock(&uvm_kernel_object->vmobjlock);
if (pg) {
panic("uvm_km_check_empty: "
"has page hashed at %p", (const void *)va);
}
}
}
}
#endif /* defined(DEBUG) */
/*
* uvm_km_alloc: allocate an area of kernel memory.
*
* => NOTE: we can return 0 even if we can wait if there is not enough
* free VM space in the map... caller should be prepared to handle
* this case.
* => we return KVA of memory allocated
*/
vaddr_t
2005-06-27 06:19:48 +04:00
uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
{
vaddr_t kva, loopva;
vaddr_t offset;
vsize_t loopsize;
1998-03-09 03:58:55 +03:00
struct vm_page *pg;
struct uvm_object *obj;
int pgaflags;
vm_prot_t prot;
UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
KASSERT(vm_map_pmap(map) == pmap_kernel());
KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
1998-03-09 03:58:55 +03:00
/*
* setup for call
*/
1998-03-09 03:58:55 +03:00
kva = vm_map_min(map); /* hint */
size = round_page(size);
obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL;
UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
map, obj, size, flags);
1998-03-09 03:58:55 +03:00
/*
* allocate some virtual space
*/
if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
UVM_ADV_RANDOM,
(flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
| UVM_FLAG_QUANTUM)) != 0)) {
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
return(0);
}
/*
* if all we wanted was VA, return now
*/
if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
return(kva);
}
1998-03-09 03:58:55 +03:00
/*
* recover object offset from virtual address
*/
offset = kva - vm_map_min(kernel_map);
UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0);
/*
* now allocate and map in the memory... note that we are the only ones
* whom should ever get a handle on this area of VM.
*/
loopva = kva;
loopsize = size;
pgaflags = UVM_PGA_USERESERVE;
if (flags & UVM_KMF_ZERO)
pgaflags |= UVM_PGA_ZERO;
prot = VM_PROT_READ | VM_PROT_WRITE;
if (flags & UVM_KMF_EXEC)
prot |= VM_PROT_EXECUTE;
while (loopsize) {
KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
pg = uvm_pagealloc(NULL, offset, NULL, pgaflags);
2001-05-25 08:06:11 +04:00
1998-03-09 03:58:55 +03:00
/*
* out of memory?
*/
if (__predict_false(pg == NULL)) {
if ((flags & UVM_KMF_NOWAIT) ||
((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
1998-03-09 03:58:55 +03:00
/* free everything! */
uvm_km_free(map, kva, size,
flags & UVM_KMF_TYPEMASK);
return (0);
1998-03-09 03:58:55 +03:00
} else {
uvm_wait("km_getwait2"); /* sleep here */
continue;
}
}
2001-05-25 08:06:11 +04:00
pg->flags &= ~PG_BUSY; /* new page */
UVM_PAGE_OWN(pg, NULL);
1998-03-09 03:58:55 +03:00
/*
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
* map it in
1998-03-09 03:58:55 +03:00
*/
pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), prot);
1998-03-09 03:58:55 +03:00
loopva += PAGE_SIZE;
offset += PAGE_SIZE;
loopsize -= PAGE_SIZE;
1998-03-09 03:58:55 +03:00
}
2004-03-24 10:47:32 +03:00
pmap_update(pmap_kernel());
2004-03-24 10:47:32 +03:00
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
return(kva);
}
/*
* uvm_km_free: free an area of kernel memory
*/
1998-03-09 03:58:55 +03:00
void
2005-06-27 06:19:48 +04:00
uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
{
KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
KASSERT((addr & PAGE_MASK) == 0);
KASSERT(vm_map_pmap(map) == pmap_kernel());
1998-03-09 03:58:55 +03:00
size = round_page(size);
if (flags & UVM_KMF_PAGEABLE) {
uvm_km_pgremove(addr, addr + size);
pmap_remove(pmap_kernel(), addr, addr + size);
} else if (flags & UVM_KMF_WIRED) {
uvm_km_pgremove_intrsafe(addr, addr + size);
pmap_kremove(addr, size);
}
uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
}
/* Sanity; must specify both or none. */
#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
(!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
#error Must specify MAP and UNMAP together.
#endif
/*
* uvm_km_alloc_poolpage: allocate a page for the pool allocator
*
* => if the pmap specifies an alternate mapping method, we use it.
*/
/* ARGSUSED */
vaddr_t
uvm_km_alloc_poolpage_cache(struct vm_map *map, bool waitok)
{
#if defined(PMAP_MAP_POOLPAGE)
return uvm_km_alloc_poolpage(map, waitok);
#else
struct vm_page *pg;
struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
vaddr_t va;
int s = 0xdeadbeaf; /* XXX: gcc */
const bool intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
if ((map->flags & VM_MAP_VACACHE) == 0)
return uvm_km_alloc_poolpage(map, waitok);
if (intrsafe)
s = splvm();
va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
if (intrsafe)
splx(s);
if (va == 0)
return 0;
KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
again:
pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
if (__predict_false(pg == NULL)) {
if (waitok) {
uvm_wait("plpg");
goto again;
} else {
if (intrsafe)
s = splvm();
pool_put(pp, (void *)va);
if (intrsafe)
splx(s);
return 0;
}
}
2005-04-01 16:37:27 +04:00
pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
pmap_update(pmap_kernel());
return va;
#endif /* PMAP_MAP_POOLPAGE */
}
vaddr_t
uvm_km_alloc_poolpage(struct vm_map *map, bool waitok)
{
#if defined(PMAP_MAP_POOLPAGE)
struct vm_page *pg;
vaddr_t va;
again:
pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
if (__predict_false(pg == NULL)) {
if (waitok) {
uvm_wait("plpg");
goto again;
} else
return (0);
}
va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
if (__predict_false(va == 0))
uvm_pagefree(pg);
return (va);
#else
vaddr_t va;
int s = 0xdeadbeaf; /* XXX: gcc */
const bool intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
if (intrsafe)
s = splvm();
va = uvm_km_alloc(map, PAGE_SIZE, 0,
(waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
if (intrsafe)
splx(s);
return (va);
#endif /* PMAP_MAP_POOLPAGE */
}
/*
* uvm_km_free_poolpage: free a previously allocated pool page
*
* => if the pmap specifies an alternate unmapping method, we use it.
*/
/* ARGSUSED */
void
2005-06-27 06:19:48 +04:00
uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
{
#if defined(PMAP_UNMAP_POOLPAGE)
uvm_km_free_poolpage(map, addr);
#else
struct pool *pp;
int s = 0xdeadbeaf; /* XXX: gcc */
const bool intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
if ((map->flags & VM_MAP_VACACHE) == 0) {
uvm_km_free_poolpage(map, addr);
return;
}
KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
uvm_km_pgremove_intrsafe(addr, addr + PAGE_SIZE);
pmap_kremove(addr, PAGE_SIZE);
#if defined(DEBUG)
pmap_update(pmap_kernel());
#endif
KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
pp = &vm_map_to_kernel(map)->vmk_vacache;
if (intrsafe)
s = splvm();
pool_put(pp, (void *)addr);
if (intrsafe)
splx(s);
#endif
}
/* ARGSUSED */
void
2005-06-27 06:19:48 +04:00
uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
{
#if defined(PMAP_UNMAP_POOLPAGE)
paddr_t pa;
pa = PMAP_UNMAP_POOLPAGE(addr);
uvm_pagefree(PHYS_TO_VM_PAGE(pa));
#else
int s = 0xdeadbeaf; /* XXX: gcc */
const bool intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
if (intrsafe)
s = splvm();
uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
if (intrsafe)
splx(s);
#endif /* PMAP_UNMAP_POOLPAGE */
}