Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

- Use alpha_atomic_{add,sub}_q() to update the pmap statistics. We now no 

longer need to lock the kernel pmap in pmap_kenter_pa() and pmap_kremove().
- Since locking the kernel pmap in interrupt context is no longer required,
  don't go to splimp() when locking the kernel pmap.
- Implement a new pmap_remove() function, not yet enabled by default.  It
  is structured like pmap_protect, and should be *much* faster.  This was
  actually written quite some time ago, but never committed because it
  didn't work properly.  Given the recent bugfix to pmap_protect(), "duh,
  of course it didn't work properly before...".  It seems to work fine now
  (have done several builds and run the UVM regression tests using the new
  code), but it is currently run-time optional so that some performance
  measurements can be easily run against the old and new code.
This commit is contained in:
thorpej 1999-11-28 19:53:11 +00:00
parent 1710fad449
commit 94552ad5d6
2 changed files with 224 additions and 93 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

288
sys/arch/alpha/alpha/pmap.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: pmap.c,v 1.117 1999/11/27 00:25:15 mycroft Exp $ */
/* $NetBSD: pmap.c,v 1.118 1999/11/28 19:53:11 thorpej Exp $ */
/*-
* Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
@ -154,7 +154,7 @@
#include <sys/cdefs.h> /* RCS ID & Copyright macro defns */
__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.117 1999/11/27 00:25:15 mycroft Exp $");
__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.118 1999/11/28 19:53:11 thorpej Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@ -707,6 +707,14 @@ do { \
#define PMAP_KERNEL_PTE(va) (&VPT[VPT_INDEX((va))])
#endif
/*
* PMAP_STAT_{INCR,DECR}:
*
* Increment or decrement a pmap statistic.
*/
#define PMAP_STAT_INCR(s, v) alpha_atomic_add_q((unsigned long *)(&(s)), (v))
#define PMAP_STAT_DECR(s, v) alpha_atomic_sub_q((unsigned long *)(&(s)), (v))
/*
* pmap_bootstrap:
*
@ -1164,7 +1172,7 @@ void
pmap_destroy(pmap)
pmap_t pmap;
{
int ps, refs;
int refs;
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
@ -1173,9 +1181,9 @@ pmap_destroy(pmap)
if (pmap == NULL)
return;
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
refs = --pmap->pm_count;
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
if (refs > 0)
return;
@ -1220,19 +1228,22 @@ void
pmap_reference(pmap)
pmap_t pmap;
{
int ps;
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
printf("pmap_reference(%p)\n", pmap);
#endif
if (pmap != NULL) {
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
pmap->pm_count++;
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
}
}
/* XXX testing purposes, for now. --thorpej */
void pmap_remove_old __P((pmap_t, vaddr_t, vaddr_t));
int pmap_remove_new = 0;
/*
* pmap_remove: [ INTERFACE ]
*
@ -1240,19 +1251,176 @@ pmap_reference(pmap)
*
* It is assumed that the start and end are properly
* rounded to the page size.
*
* XXX This routine could be optimized with regard to its
* handling of level 3 PTEs.
*/
void
pmap_remove(pmap, sva, eva)
pmap_t pmap;
vaddr_t sva, eva;
{
pt_entry_t *l1pte, *l2pte, *l3pte;
pt_entry_t *saved_l1pte, *saved_l2pte, *saved_l3pte;
vaddr_t l1eva, l2eva, vptva;
boolean_t needisync = FALSE;
long cpu_id = alpha_pal_whami();
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_REMOVE|PDB_PROTECT))
printf("pmap_remove(%p, %lx, %lx)\n", pmap, sva, eva);
#endif
if (pmap == NULL)
return;
/* XXX testing purposes, for now. --thorpej */
if (pmap_remove_new == 0)
pmap_remove_old(pmap, sva, eva);
/*
* If this is the kernel pmap, we can use a faster method
* for accessing the PTEs (since the PT pages are always
* resident).
*
* Note that this routine should NEVER be called from an
* interrupt context; pmap_kremove() is used for that.
*/
if (pmap == pmap_kernel()) {
PMAP_MAP_TO_HEAD_LOCK();
PMAP_LOCK(pmap);
while (sva < eva) {
l3pte = PMAP_KERNEL_PTE(sva);
if (pmap_pte_v(l3pte)) {
#ifdef DIAGNOSTIC
if (PAGE_IS_MANAGED(pmap_pte_pa(l3pte)) &&
pmap_pte_pv(l3pte) == 0)
panic("pmap_remove: managed page "
"without PG_PVLIST for 0x%lx",
sva);
#endif
needisync |= pmap_remove_mapping(pmap, sva,
l3pte, TRUE, cpu_id, NULL);
}
sva += PAGE_SIZE;
}
PMAP_UNLOCK(pmap);
PMAP_MAP_TO_HEAD_UNLOCK();
if (needisync) {
alpha_pal_imb();
#if defined(MULTIPROCESSOR) && 0
alpha_broadcast_ipi(ALPHA_IPI_IMB);
#endif
}
return;
}
#ifdef DIAGNOSTIC
if (sva > VM_MAXUSER_ADDRESS || eva > VM_MAXUSER_ADDRESS)
panic("pmap_remove: (0x%lx - 0x%lx) user pmap, kernel "
"address range", sva, eva);
#endif
PMAP_MAP_TO_HEAD_LOCK();
PMAP_LOCK(pmap);
/*
* If we're already referencing the kernel_lev1map, there
* is no work for us to do.
*/
if (pmap->pm_lev1map == kernel_lev1map)
goto out;
saved_l1pte = l1pte = pmap_l1pte(pmap, sva);
/*
* Add a reference to the L1 table to it won't get
* removed from under us.
*/
pmap_physpage_addref(saved_l1pte);
for (; sva < eva; sva = l1eva, l1pte++) {
l1eva = alpha_trunc_l1seg(sva) + ALPHA_L1SEG_SIZE;
if (pmap_pte_v(l1pte)) {
saved_l2pte = l2pte = pmap_l2pte(pmap, sva, l1pte);
/*
* Add a reference to the L2 table so it won't
* get removed from under us.
*/
pmap_physpage_addref(saved_l2pte);
for (; sva < l1eva && sva < eva; sva = l2eva, l2pte++) {
l2eva =
alpha_trunc_l2seg(sva) + ALPHA_L2SEG_SIZE;
if (pmap_pte_v(l2pte)) {
saved_l3pte = l3pte =
pmap_l3pte(pmap, sva, l2pte);
/*
* Add a reference to the L3 table so
* it won't get removed from under us.
*/
pmap_physpage_addref(saved_l3pte);
/*
* Remember this sva; if the L3 table
* gets removed, we need to invalidate
* the VPT TLB entry for it.
*/
vptva = sva;
for (; sva < l2eva && sva < eva;
sva += PAGE_SIZE, l3pte++) {
if (pmap_pte_v(l3pte)) {
needisync |=
pmap_remove_mapping(
pmap, sva,
l3pte, TRUE,
cpu_id, NULL);
}
}
/*
* Remove the reference to the L3
* table that we added above. This
* may free the L3 table.
*/
pmap_l3pt_delref(pmap, vptva,
saved_l3pte, cpu_id, NULL);
}
}
/*
* Remove the reference to the L2 table that we
* added above. This may free the L2 table.
*/
pmap_l2pt_delref(pmap, l1pte, saved_l2pte, cpu_id);
}
}
/*
* Remove the reference to the L1 table that we added above.
* This may free the L1 table.
*/
pmap_l1pt_delref(pmap, saved_l1pte, cpu_id);
if (needisync)
alpha_pal_imb();
out:
PMAP_UNLOCK(pmap);
PMAP_MAP_TO_HEAD_UNLOCK();
}
void
pmap_remove_old(pmap, sva, eva)
pmap_t pmap;
vaddr_t sva, eva;
{
pt_entry_t *l1pte, *l2pte, *l3pte;
boolean_t needisync = FALSE;
long cpu_id = alpha_pal_whami();
int ps;
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_REMOVE|PDB_PROTECT))
@ -1272,7 +1440,7 @@ pmap_remove(pmap, sva, eva)
*/
if (pmap == pmap_kernel()) {
PMAP_MAP_TO_HEAD_LOCK();
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
while (sva < eva) {
l3pte = PMAP_KERNEL_PTE(sva);
@ -1290,7 +1458,7 @@ pmap_remove(pmap, sva, eva)
sva += PAGE_SIZE;
}
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
PMAP_MAP_TO_HEAD_UNLOCK();
if (needisync) {
@ -1309,7 +1477,7 @@ pmap_remove(pmap, sva, eva)
#endif
PMAP_MAP_TO_HEAD_LOCK();
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
while (sva < eva) {
/*
@ -1343,7 +1511,7 @@ pmap_remove(pmap, sva, eva)
if (needisync)
alpha_pal_imb();
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
PMAP_MAP_TO_HEAD_UNLOCK();
}
@ -1363,7 +1531,6 @@ pmap_page_protect(pg, prot)
boolean_t needisync = FALSE;
long cpu_id = alpha_pal_whami();
paddr_t pa = VM_PAGE_TO_PHYS(pg);
int ps;
#ifdef DEBUG
if ((pmapdebug & (PDB_FOLLOW|PDB_PROTECT)) ||
@ -1403,10 +1570,8 @@ pmap_page_protect(pg, prot)
PMAP_HEAD_TO_MAP_LOCK();
simple_lock(&pvh->pvh_slock);
for (pv = LIST_FIRST(&pvh->pvh_list); pv != NULL; pv = nextpv) {
struct pmap *pmap = pv->pv_pmap;
nextpv = LIST_NEXT(pv, pv_list);
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pv->pv_pmap);
#ifdef DEBUG
if (pmap_pte_v(pmap_l2pte(pv->pv_pmap, pv->pv_va, NULL)) == 0 ||
pmap_pte_pa(pv->pv_pte) != pa)
@ -1425,7 +1590,7 @@ pmap_page_protect(pg, prot)
}
}
#endif
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pv->pv_pmap);
}
if (needisync)
@ -1452,7 +1617,6 @@ pmap_protect(pmap, sva, eva, prot)
boolean_t hadasm;
vaddr_t l1eva, l2eva;
long cpu_id = alpha_pal_whami();
int ps;
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_PROTECT))
@ -1471,7 +1635,7 @@ pmap_protect(pmap, sva, eva, prot)
if (prot & VM_PROT_WRITE)
return;
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
bits = pte_prot(pmap, prot);
isactive = PMAP_ISACTIVE(pmap);
@ -1520,7 +1684,7 @@ pmap_protect(pmap, sva, eva, prot)
#endif
}
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
}
/*
@ -1554,7 +1718,7 @@ pmap_enter(pmap, va, pa, prot, flags)
boolean_t isactive;
boolean_t wired;
long cpu_id = alpha_pal_whami();
int ps, error;
int error;
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_ENTER))
@ -1570,7 +1734,7 @@ pmap_enter(pmap, va, pa, prot, flags)
wired = (flags & PMAP_WIRED) != 0;
PMAP_MAP_TO_HEAD_LOCK();
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
if (pmap == pmap_kernel()) {
#ifdef DIAGNOSTIC
@ -1712,9 +1876,9 @@ pmap_enter(pmap, va, pa, prot, flags)
* Adjust the wiring count.
*/
if (wired)
pmap->pm_stats.wired_count++;
PMAP_STAT_INCR(pmap->pm_stats.wired_count, 1);
else
pmap->pm_stats.wired_count--;
PMAP_STAT_DECR(pmap->pm_stats.wired_count, 1);
}
/*
@ -1760,9 +1924,9 @@ pmap_enter(pmap, va, pa, prot, flags)
/*
* Increment counters.
*/
pmap->pm_stats.resident_count++;
PMAP_STAT_INCR(pmap->pm_stats.resident_count, 1);
if (wired)
pmap->pm_stats.wired_count++;
PMAP_STAT_INCR(pmap->pm_stats.wired_count, 1);
validate:
/*
@ -1834,7 +1998,7 @@ pmap_enter(pmap, va, pa, prot, flags)
#endif
}
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
PMAP_MAP_TO_HEAD_UNLOCK();
return (KERN_SUCCESS);
@ -1858,7 +2022,6 @@ pmap_kenter_pa(va, pa, prot)
long cpu_id = alpha_pal_whami();
boolean_t needisync = FALSE;
pmap_t pmap = pmap_kernel();
int ps;
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_ENTER))
@ -1876,17 +2039,10 @@ pmap_kenter_pa(va, pa, prot)
pte = PMAP_KERNEL_PTE(va);
/*
* Update stats; must lock the kernel pmap to do this.
*/
PMAP_LOCK(pmap, ps);
if (pmap_pte_v(pte) == 0)
pmap->pm_stats.resident_count++;
PMAP_STAT_INCR(pmap->pm_stats.resident_count, 1);
if (pmap_pte_w(pte) == 0)
pmap->pm_stats.wired_count++;
PMAP_UNLOCK(pmap, ps);
PMAP_STAT_DECR(pmap->pm_stats.wired_count, 1);
if ((prot & VM_PROT_EXECUTE) != 0 || pmap_pte_exec(pte))
needisync = TRUE;
@ -1962,7 +2118,6 @@ pmap_kremove(va, size)
boolean_t needisync = FALSE;
long cpu_id = alpha_pal_whami();
pmap_t pmap = pmap_kernel();
int ps;
#ifdef DEBUG
if (pmapdebug & (PDB_FOLLOW|PDB_ENTER))
@ -1975,8 +2130,6 @@ pmap_kremove(va, size)
panic("pmap_kremove: user address");
#endif
PMAP_LOCK(pmap, ps);
for (; size != 0; size -= PAGE_SIZE, va += PAGE_SIZE) {
pte = PMAP_KERNEL_PTE(va);
if (pmap_pte_v(pte)) {
@ -1995,13 +2148,11 @@ pmap_kremove(va, size)
pmap_tlb_shootdown(pmap, va, PG_ASM);
#endif
/* Update stats. */
pmap->pm_stats.resident_count--;
pmap->pm_stats.wired_count--;
PMAP_STAT_DECR(pmap->pm_stats.resident_count, 1);
PMAP_STAT_DECR(pmap->pm_stats.wired_count, 1);
}
}
PMAP_UNLOCK(pmap, ps);
if (needisync) {
alpha_pal_imb();
#if defined(MULTIPROCESSOR) && 0
@ -2023,7 +2174,6 @@ pmap_unwire(pmap, va)
vaddr_t va;
{
pt_entry_t *pte;
int ps;
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
@ -2032,7 +2182,7 @@ pmap_unwire(pmap, va)
if (pmap == NULL)
return;
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
pte = pmap_l3pte(pmap, va, NULL);
#ifdef DIAGNOSTIC
@ -2047,7 +2197,7 @@ pmap_unwire(pmap, va)
*/
if (pmap_pte_w_chg(pte, 0)) {
pmap_pte_set_w(pte, FALSE);
pmap->pm_stats.wired_count--;
PMAP_STAT_DECR(pmap->pm_stats.wired_count, 1);
}
#ifdef DIAGNOSTIC
else {
@ -2056,7 +2206,7 @@ pmap_unwire(pmap, va)
}
#endif
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
}
/*
@ -2072,7 +2222,6 @@ pmap_extract(pmap, va, pap)
paddr_t *pap;
{
pt_entry_t *l1pte, *l2pte, *l3pte;
int ps;
paddr_t pa;
boolean_t rv = FALSE;
@ -2080,7 +2229,7 @@ pmap_extract(pmap, va, pap)
if (pmapdebug & PDB_FOLLOW)
printf("pmap_extract(%p, %lx) -> ", pmap, va);
#endif
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
l1pte = pmap_l1pte(pmap, va);
if (pmap_pte_v(l1pte) == 0)
@ -2100,7 +2249,7 @@ pmap_extract(pmap, va, pap)
rv = TRUE;
out:
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW) {
if (rv)
@ -2202,7 +2351,6 @@ pmap_activate(p)
{
struct pmap *pmap = p->p_vmspace->vm_map.pmap;
long cpu_id = alpha_pal_whami();
int ps;
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
@ -2222,7 +2370,7 @@ pmap_activate(p)
TAILQ_INSERT_TAIL(&pmap_all_pmaps, pmap, pm_list);
simple_unlock(&pmap_all_pmaps_slock);
PMAP_LOCK(pmap, ps);
PMAP_LOCK(pmap);
/*
* Allocate an ASN.
@ -2231,7 +2379,7 @@ pmap_activate(p)
PMAP_ACTIVATE(pmap, p, cpu_id);
PMAP_UNLOCK(pmap, ps);
PMAP_UNLOCK(pmap);
}
/*
@ -2575,8 +2723,8 @@ pmap_remove_mapping(pmap, va, pte, dolock, cpu_id, prmt)
* Update statistics
*/
if (pmap_pte_w(pte))
pmap->pm_stats.wired_count--;
pmap->pm_stats.resident_count--;
PMAP_STAT_DECR(pmap->pm_stats.wired_count, 1);
PMAP_STAT_DECR(pmap->pm_stats.resident_count, 1);
/*
* Invalidate the PTE after saving the reference modify info.
@ -2646,7 +2794,6 @@ pmap_changebit(pa, set, mask, cpu_id)
vaddr_t va;
boolean_t hadasm, isactive;
boolean_t needisync = FALSE;
int ps;
#ifdef DEBUG
if (pmapdebug & PDB_BITS)
@ -2673,7 +2820,7 @@ pmap_changebit(pa, set, mask, cpu_id)
continue;
}
PMAP_LOCK(pv->pv_pmap, ps);
PMAP_LOCK(pv->pv_pmap);
pte = pv->pv_pte;
npte = (*pte | set) & mask;
@ -2689,7 +2836,7 @@ pmap_changebit(pa, set, mask, cpu_id)
hadasm ? PG_ASM : 0);
#endif
}
PMAP_UNLOCK(pv->pv_pmap, ps);
PMAP_UNLOCK(pv->pv_pmap);
}
if (needisync) {
@ -2717,7 +2864,6 @@ pmap_emulate_reference(p, v, user, write)
struct pv_head *pvh;
boolean_t didlock = FALSE;
long cpu_id = alpha_pal_whami();
int ps;
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
@ -2742,7 +2888,7 @@ pmap_emulate_reference(p, v, user, write)
if (p->p_vmspace == NULL)
panic("pmap_emulate_reference: bad p_vmspace");
#endif
PMAP_LOCK(p->p_vmspace->vm_map.pmap, ps);
PMAP_LOCK(p->p_vmspace->vm_map.pmap);
didlock = TRUE;
pte = pmap_l3pte(p->p_vmspace->vm_map.pmap, v, NULL);
/*
@ -2785,7 +2931,7 @@ pmap_emulate_reference(p, v, user, write)
* it now.
*/
if (didlock)
PMAP_UNLOCK(p->p_vmspace->vm_map.pmap, ps);
PMAP_UNLOCK(p->p_vmspace->vm_map.pmap);
#ifdef DEBUG
if (pmapdebug & PDB_FOLLOW)
@ -3489,7 +3635,7 @@ pmap_ptpage_steal(pmap, usage, pap)
{
struct pv_head *pvh;
pmap_t spmap;
int ps, l1idx, l2idx, l3idx;
int l1idx, l2idx, l3idx;
pt_entry_t *lev2map, *lev3map;
vaddr_t va;
paddr_t pa;
@ -3521,9 +3667,9 @@ pmap_ptpage_steal(pmap, usage, pap)
if (spmap == pmap_kernel() || spmap == pmap)
continue;
PMAP_LOCK(spmap, ps);
PMAP_LOCK(spmap);
if (spmap->pm_lev1map == kernel_lev1map) {
PMAP_UNLOCK(spmap, ps);
PMAP_UNLOCK(spmap);
continue;
}
@ -3566,7 +3712,7 @@ pmap_ptpage_steal(pmap, usage, pap)
* Didn't find something we could steal in this
* pmap, try the next one.
*/
PMAP_UNLOCK(spmap, ps);
PMAP_UNLOCK(spmap);
continue;
found_one:
@ -3592,7 +3738,7 @@ pmap_ptpage_steal(pmap, usage, pap)
}
}
PMAP_UNLOCK(spmap, ps);
PMAP_UNLOCK(spmap);
if (needisync) {
alpha_pal_imb();

29
sys/arch/alpha/include/pmap.h
View File

@ -1,4 +1,4 @@
/* $NetBSD: pmap.h,v 1.31 1999/05/24 20:11:58 thorpej Exp $ */
/* $NetBSD: pmap.h,v 1.32 1999/11/28 19:53:11 thorpej Exp $ */
/*-
* Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
@ -292,28 +292,13 @@ pmap_l3pte(pmap, v, l2pte)
/*
* Macros for locking pmap structures.
*
* Note that the kernel pmap can be accessed from interrupt context,
* so when we have the kernel pmap lock asserted, we must block any
* interrupts that can cause memory allocation, otherwise we can deadlock
* if an interrupt occurs and causes us to recurse into the pmap and
* attempt to try to assert the lock while it is already held.
*
* No other pmap can be accessed from interrupt context, so we do not
* need to block interrupts in any other case.
* Note that we if we access the kernel pmap in interrupt context, it
* is only to update statistics. Since stats are updated using atomic
* operations, locking the kernel pmap is not necessary. Therefore,
* it is not necessary to block interrupts when locking pmap strucutres.
*/
#define PMAP_LOCK(pmap, s) \
do { \
if ((pmap) == pmap_kernel()) \
(s) = splimp(); \
simple_lock(&(pmap)->pm_slock); \
} while (0)
#define PMAP_UNLOCK(pmap, s) \
do { \
simple_unlock(&(pmap)->pm_slock); \
if ((pmap) == pmap_kernel()) \
splx((s)); \
} while (0)
#define PMAP_LOCK(pmap) simple_lock(&(pmap)->pm_slock)
#define PMAP_UNLOCK(pmap) simple_unlock(&(pmap)->pm_slock)
#endif /* _KERNEL */