NetBSD/sys/uvm/uvm_pager.c

465 lines
11 KiB
C
Raw Normal View History

/* $NetBSD: uvm_pager.c,v 1.65 2005/01/01 21:00:06 yamt Exp $ */
/*
*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
* All rights reserved.
*
* 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 and
* Washington University.
* 4. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
1998-02-07 14:07:38 +03:00
*
* from: Id: uvm_pager.c,v 1.1.2.23 1998/02/02 20:38:06 chuck Exp
*/
/*
* uvm_pager.c: generic functions used to assist the pagers.
*/
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: uvm_pager.c,v 1.65 2005/01/01 21:00:06 yamt Exp $");
#include "opt_uvmhist.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/pool.h>
#include <sys/vnode.h>
#define UVM_PAGER
#include <uvm/uvm.h>
struct pool *uvm_aiobuf_pool;
/*
* list of uvm pagers in the system
*/
struct uvm_pagerops * const uvmpagerops[] = {
&aobj_pager,
1998-03-09 03:58:55 +03:00
&uvm_deviceops,
&uvm_vnodeops,
&ubc_pager,
};
/*
* the pager map: provides KVA for I/O
*/
struct vm_map *pager_map; /* XXX */
struct simplelock pager_map_wanted_lock;
boolean_t pager_map_wanted; /* locked by pager map */
static vaddr_t emergva;
static boolean_t emerginuse;
/*
* uvm_pager_init: init pagers (at boot time)
*/
1998-03-09 03:58:55 +03:00
void
uvm_pager_init()
{
u_int lcv;
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
vaddr_t sva, eva;
1998-03-09 03:58:55 +03:00
/*
* init pager map
*/
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
sva = 0;
pager_map = uvm_km_suballoc(kernel_map, &sva, &eva, PAGER_MAP_SIZE, 0,
FALSE, NULL);
simple_lock_init(&pager_map_wanted_lock);
pager_map_wanted = FALSE;
emergva = uvm_km_valloc(kernel_map, round_page(MAXPHYS));
emerginuse = FALSE;
1998-03-09 03:58:55 +03:00
/*
* init ASYNC I/O queue
*/
2001-05-25 08:06:11 +04:00
1998-03-09 03:58:55 +03:00
TAILQ_INIT(&uvm.aio_done);
/*
* call pager init functions
*/
for (lcv = 0 ; lcv < sizeof(uvmpagerops)/sizeof(struct uvm_pagerops *);
lcv++) {
if (uvmpagerops[lcv]->pgo_init)
uvmpagerops[lcv]->pgo_init();
}
}
/*
* uvm_pagermapin: map pages into KVA (pager_map) for I/O that needs mappings
*
* we basically just map in a blank map entry to reserve the space in the
* map and then use pmap_enter() to put the mappings in by hand.
*/
vaddr_t
uvm_pagermapin(pps, npages, flags)
1998-03-09 03:58:55 +03:00
struct vm_page **pps;
int npages;
int flags;
{
vsize_t size;
vaddr_t kva;
vaddr_t cva;
1998-03-09 03:58:55 +03:00
struct vm_page *pp;
vm_prot_t prot;
1998-03-09 03:58:55 +03:00
UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist);
UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d)", pps, npages,0,0);
/*
* compute protection. outgoing I/O only needs read
* access to the page, whereas incoming needs read/write.
*/
prot = VM_PROT_READ;
if (flags & UVMPAGER_MAPIN_READ)
prot |= VM_PROT_WRITE;
ReStart:
size = npages << PAGE_SHIFT;
kva = 0; /* let system choose VA */
1998-03-09 03:58:55 +03:00
2001-05-25 08:06:11 +04:00
if (uvm_map(pager_map, &kva, size, NULL,
UVM_UNKNOWN_OFFSET, 0, UVM_FLAG_NOMERGE) != 0) {
if (curproc == uvm.pagedaemon_proc) {
simple_lock(&pager_map_wanted_lock);
if (emerginuse) {
UVM_UNLOCK_AND_WAIT(&emergva,
&pager_map_wanted_lock, FALSE,
"emergva", 0);
goto ReStart;
}
emerginuse = TRUE;
simple_unlock(&pager_map_wanted_lock);
kva = emergva;
/* The shift implicitly truncates to PAGE_SIZE */
KASSERT(npages <= (MAXPHYS >> PAGE_SHIFT));
goto enter;
}
if ((flags & UVMPAGER_MAPIN_WAITOK) == 0) {
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0);
return(0);
1998-03-09 03:58:55 +03:00
}
simple_lock(&pager_map_wanted_lock);
2001-05-25 08:06:11 +04:00
pager_map_wanted = TRUE;
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist, " SLEEPING on pager_map",0,0,0,0);
2001-05-25 08:06:11 +04:00
UVM_UNLOCK_AND_WAIT(pager_map, &pager_map_wanted_lock, FALSE,
"pager_map", 0);
1998-03-09 03:58:55 +03:00
goto ReStart;
}
enter:
1998-03-09 03:58:55 +03:00
/* got it */
for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) {
pp = *pps++;
KASSERT(pp);
KASSERT(pp->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
pmap_kenter_pa(cva, VM_PAGE_TO_PHYS(pp), prot);
1998-03-09 03:58:55 +03:00
}
pmap_update(vm_map_pmap(pager_map));
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0);
return(kva);
}
/*
* uvm_pagermapout: remove pager_map mapping
*
* we remove our mappings by hand and then remove the mapping (waking
* up anyone wanting space).
*/
1998-03-09 03:58:55 +03:00
void
uvm_pagermapout(kva, npages)
vaddr_t kva;
1998-03-09 03:58:55 +03:00
int npages;
{
vsize_t size = npages << PAGE_SHIFT;
struct vm_map_entry *entries;
1998-03-09 03:58:55 +03:00
UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist);
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist, " (kva=0x%x, npages=%d)", kva, npages,0,0);
/*
* duplicate uvm_unmap, but add in pager_map_wanted handling.
*/
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
pmap_kremove(kva, npages << PAGE_SHIFT);
if (kva == emergva) {
simple_lock(&pager_map_wanted_lock);
emerginuse = FALSE;
wakeup(&emergva);
simple_unlock(&pager_map_wanted_lock);
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
return;
}
1998-03-09 03:58:55 +03:00
vm_map_lock(pager_map);
uvm_unmap_remove(pager_map, kva, kva + size, &entries, NULL);
1998-03-09 03:58:55 +03:00
simple_lock(&pager_map_wanted_lock);
if (pager_map_wanted) {
pager_map_wanted = FALSE;
wakeup(pager_map);
}
simple_unlock(&pager_map_wanted_lock);
vm_map_unlock(pager_map);
if (entries)
uvm_unmap_detach(entries, 0);
pmap_update(pmap_kernel());
1998-03-09 03:58:55 +03:00
UVMHIST_LOG(maphist,"<- done",0,0,0,0);
}
/*
* interrupt-context iodone handler for nested i/o bufs.
*
* => must be at splbio().
*/
void
uvm_aio_biodone1(bp)
struct buf *bp;
{
struct buf *mbp = bp->b_private;
KASSERT(mbp != bp);
if (bp->b_flags & B_ERROR) {
mbp->b_flags |= B_ERROR;
mbp->b_error = bp->b_error;
}
mbp->b_resid -= bp->b_bcount;
pool_put(&bufpool, bp);
if (mbp->b_resid == 0) {
biodone(mbp);
}
}
/*
* interrupt-context iodone handler for single-buf i/os
* or the top-level buf of a nested-buf i/o.
*
* => must be at splbio().
*/
void
uvm_aio_biodone(bp)
struct buf *bp;
{
/* reset b_iodone for when this is a single-buf i/o. */
bp->b_iodone = uvm_aio_aiodone;
simple_lock(&uvm.aiodoned_lock); /* locks uvm.aio_done */
TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
wakeup(&uvm.aiodoned);
simple_unlock(&uvm.aiodoned_lock);
}
/*
* uvm_aio_aiodone: do iodone processing for async i/os.
* this should be called in thread context, not interrupt context.
*/
void
uvm_aio_aiodone(bp)
struct buf *bp;
{
int npages = bp->b_bufsize >> PAGE_SHIFT;
struct vm_page *pg, *pgs[npages];
struct uvm_object *uobj;
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 simplelock *slock;
int s, i, error, swslot;
boolean_t write, swap;
UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "bp %p", bp, 0,0,0);
error = (bp->b_flags & B_ERROR) ? (bp->b_error ? bp->b_error : EIO) : 0;
write = (bp->b_flags & B_READ) == 0;
/* XXXUBC B_NOCACHE is for swap pager, should be done differently */
if (write && !(bp->b_flags & B_NOCACHE) && bioops.io_pageiodone) {
(*bioops.io_pageiodone)(bp);
}
uobj = NULL;
for (i = 0; i < npages; i++) {
pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0);
}
uvm_pagermapout((vaddr_t)bp->b_data, npages);
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
swslot = 0;
slock = NULL;
pg = pgs[0];
swap = (pg->uanon != NULL && pg->uobject == NULL) ||
(pg->pqflags & PQ_AOBJ) != 0;
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
if (!swap) {
uobj = pg->uobject;
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
slock = &uobj->vmobjlock;
simple_lock(slock);
uvm_lock_pageq();
} else if (error) {
if (pg->uobject != NULL) {
swslot = uao_find_swslot(pg->uobject,
pg->offset >> PAGE_SHIFT);
} else {
swslot = pg->uanon->an_swslot;
}
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
KASSERT(swslot);
}
for (i = 0; i < npages; i++) {
pg = pgs[i];
KASSERT(swap || pg->uobject == uobj);
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
UVMHIST_LOG(ubchist, "pg %p", pg, 0,0,0);
/*
* for swap i/os, lock each page's object (or anon)
* individually since each page may need a different lock.
*/
if (swap) {
if (pg->uobject != NULL) {
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
slock = &pg->uobject->vmobjlock;
} else {
slock = &pg->uanon->an_lock;
}
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
simple_lock(slock);
uvm_lock_pageq();
}
/*
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
* process errors. for reads, just mark the page to be freed.
* for writes, if the error was ENOMEM, we assume this was
* a transient failure so we mark the page dirty so that
* we'll try to write it again later. for all other write
* errors, we assume the error is permanent, thus the data
* in the page is lost. bummer.
*/
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
if (error) {
int slot;
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
if (!write) {
pg->flags |= PG_RELEASED;
continue;
} else if (error == ENOMEM) {
if (pg->flags & PG_PAGEOUT) {
pg->flags &= ~PG_PAGEOUT;
uvmexp.paging--;
}
pg->flags &= ~PG_CLEAN;
uvm_pageactivate(pg);
slot = 0;
} else
slot = SWSLOT_BAD;
if (swap) {
if (pg->uobject != NULL) {
int oldslot;
oldslot = uao_set_swslot(pg->uobject,
pg->offset >> PAGE_SHIFT, slot);
KASSERT(oldslot == swslot + i);
} else {
KASSERT(pg->uanon->an_swslot ==
swslot + i);
pg->uanon->an_swslot = slot;
}
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
}
}
/*
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
* if the page is PG_FAKE, this must have been a read to
* initialize the page. clear PG_FAKE and activate the page.
* we must also clear the pmap "modified" flag since it may
* still be set from the page's previous identity.
*/
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
if (pg->flags & PG_FAKE) {
KASSERT(!write);
pg->flags &= ~PG_FAKE;
uvm_pageactivate(pg);
pmap_clear_modify(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
/*
* do accounting for pagedaemon i/o and arrange to free
* the pages instead of just unbusying them.
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
*/
if (pg->flags & PG_PAGEOUT) {
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_PAGEOUT;
uvmexp.paging--;
2004-10-03 11:59:02 +04:00
uvmexp.pdfreed++;
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_RELEASED;
}
/*
* for swap pages, unlock everything for this page now.
*/
if (swap) {
if (pg->uobject == NULL && pg->uanon->an_ref == 0 &&
(pg->flags & PG_RELEASED) != 0) {
uvm_unlock_pageq();
uvm_anon_release(pg->uanon);
} else {
uvm_page_unbusy(&pg, 1);
uvm_unlock_pageq();
simple_unlock(slock);
}
}
}
if (!swap) {
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
uvm_page_unbusy(pgs, npages);
uvm_unlock_pageq();
simple_unlock(slock);
} else {
KASSERT(write);
/* these pages are now only in swap. */
simple_lock(&uvm.swap_data_lock);
KASSERT(uvmexp.swpgonly + npages <= uvmexp.swpginuse);
if (error != ENOMEM)
uvmexp.swpgonly += npages;
simple_unlock(&uvm.swap_data_lock);
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
if (error) {
if (error != ENOMEM)
uvm_swap_markbad(swslot, npages);
else
uvm_swap_free(swslot, npages);
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
}
uvmexp.pdpending--;
}
s = splbio();
if (write && (bp->b_flags & B_AGE) != 0) {
vwakeup(bp);
}
pool_put(&bufpool, bp);
splx(s);
}