Rewrite how mod/ref emulation is done. In particular, make

modified attribute checking actually work.  Previously, a
page could be reported as modified even if it wasn't!  This
is very bad, and e.g. caused the VM system to attempt to
"clean" pages of read-only files.

Only lip-services is paid to referenced attribute emulation.
This could be improved.
This commit is contained in:
thorpej 2001-02-02 02:28:18 +00:00
parent dda0d50982
commit ef2315d256
2 changed files with 205 additions and 127 deletions

View File

@ -1,4 +1,4 @@
/* $NetBSD: pmap.h,v 1.6 2000/09/28 14:03:38 is Exp $ */
/* $NetBSD: pmap.h,v 1.7 2001/02/02 02:28:18 thorpej Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -279,6 +279,10 @@ struct pv_head {
struct pv_entry *pvh_list; /* head of list (locked by pvh_lock) */
};
/* These are kept in the vm_physseg array. */
#define PGA_REFERENCED 0x01 /* page is referenced */
#define PGA_MODIFIED 0x02 /* page is modified */
struct pv_entry { /* locked by its list's pvh_lock */
struct pv_entry *pv_next; /* next entry */
struct pmap *pv_pmap; /* the pmap */
@ -364,15 +368,10 @@ extern int pmap_pg_g; /* do we support PG_G? */
#define pmap_update() tlbflush()
#endif
#define pmap_clear_modify(pg) pmap_change_attrs(pg, 0, PG_M)
#define pmap_clear_reference(pg) pmap_change_attrs(pg, 0, PG_U)
#define pmap_is_referenced(pg) pmap_test_attrs(pg, PGA_REFERENCED)
#define pmap_is_modified(pg) pmap_test_attrs(pg, PGA_MODIFIED)
#define pmap_copy(DP,SP,D,L,S) pmap_transfer(DP,SP,D,L,S, FALSE)
#define pmap_is_modified(pg) pmap_test_attrs(pg, PG_M)
#ifdef notyet
#define pmap_is_referenced(pg) pmap_test_attrs(pg, PG_U)
#else
#define pmap_is_referenced(pg) 1
#endif
#define pmap_move(DP,SP,D,L,S) pmap_transfer(DP,SP,D,L,S, TRUE)
#define pmap_phys_address(ppn) sh3_ptob(ppn)
#define pmap_valid_entry(E) ((E) & PG_V) /* is PDE or PTE valid? */
@ -386,7 +385,6 @@ void pmap_activate __P((struct proc *));
void pmap_bootstrap __P((vaddr_t));
boolean_t pmap_change_attrs __P((struct vm_page *, int, int));
void pmap_deactivate __P((struct proc *));
static void pmap_page_protect __P((struct vm_page *, vm_prot_t));
void pmap_page_remove __P((struct vm_page *));
static void pmap_protect __P((struct pmap *, vaddr_t,
vaddr_t, vm_prot_t));
@ -459,29 +457,6 @@ pmap_update_2pg(va, vb)
#endif
}
/*
* pmap_page_protect: change the protection of all recorded mappings
* of a managed page
*
* => this function is a frontend for pmap_page_remove/pmap_change_attrs
* => we only have to worry about making the page more protected.
* unprotecting a page is done on-demand at fault time.
*/
__inline static void
pmap_page_protect(pg, prot)
struct vm_page *pg;
vm_prot_t prot;
{
if ((prot & VM_PROT_WRITE) == 0) {
if (prot & (VM_PROT_READ|VM_PROT_EXECUTE)) {
(void) pmap_change_attrs(pg, PG_RO, PG_RW);
} else {
pmap_page_remove(pg);
}
}
}
/*
* pmap_protect: change the protection of pages in a pmap
*

View File

@ -1,4 +1,4 @@
/* $NetBSD: pmap.c,v 1.13 2001/01/14 03:25:46 thorpej Exp $ */
/* $NetBSD: pmap.c,v 1.14 2001/02/02 02:28:18 thorpej Exp $ */
/*
* Copyright (c) 1997 Charles D. Cranor and Washington University.
@ -395,6 +395,8 @@ static void pmap_unmap_ptes __P((struct pmap *));
void pmap_pinit __P((pmap_t));
void pmap_release __P((pmap_t));
void pmap_changebit(struct pv_head *, pt_entry_t,
pt_entry_t);
/*
* p m a p i n l i n e h e l p e r f u n c t i o n s
@ -2318,7 +2320,7 @@ pmap_page_remove(pg)
/*
* pmap_test_attrs: test a page's attributes
*
* => we set pv_head => pmap locking
* => no need for any locking here
*/
boolean_t
@ -2326,11 +2328,7 @@ pmap_test_attrs(pg, testbits)
struct vm_page *pg;
int testbits;
{
int bank, off;
char *myattrs;
struct pv_head *pvh;
struct pv_entry *pve;
pt_entry_t *ptes, pte;
int bank, off, attr;
/* XXX: vm_page should either contain pv_head or have a pointer to it */
bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
@ -2339,68 +2337,30 @@ pmap_test_attrs(pg, testbits)
return(FALSE);
}
/*
* before locking: see if attributes are already set and if so,
* return!
*/
attr = vm_physmem[bank].pmseg.attrs[off];
myattrs = &vm_physmem[bank].pmseg.attrs[off];
if (*myattrs & testbits)
return(TRUE);
/* test to see if there is a list before bothering to lock */
pvh = &vm_physmem[bank].pmseg.pvhead[off];
if (pvh->pvh_list == NULL) {
return(FALSE);
}
/* nope, gonna have to do it the hard way */
PMAP_HEAD_TO_MAP_LOCK();
/* XXX: needed if we hold head->map lock? */
simple_lock(&pvh->pvh_lock);
for (pve = pvh->pvh_list; pve != NULL && (*myattrs & testbits) == 0;
pve = pve->pv_next) {
ptes = pmap_map_ptes(pve->pv_pmap);
pte = ptes[sh3_btop(pve->pv_va)];
pmap_unmap_ptes(pve->pv_pmap);
*myattrs |= pte;
}
/*
* note that we will exit the for loop with a non-null pve if
* we have found the bits we are testing for.
*/
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
return((*myattrs & testbits) != 0);
return ((attr & testbits) != 0 ? TRUE : FALSE);
}
/*
* pmap_change_attrs: change a page's attributes
* pmap_clear_modify: clear the "modified" attribute on a page.
*
* => we set pv_head => pmap locking
* => we return TRUE if we cleared one of the bits we were asked to
* => we return TRUE if the page was marked "modified".
*/
boolean_t
pmap_change_attrs(pg, setbits, clearbits)
struct vm_page *pg;
int setbits, clearbits;
pmap_clear_modify(vm_page_t pg)
{
u_int32_t result;
int bank, off;
struct pv_head *pvh;
struct pv_entry *pve;
pt_entry_t *ptes, npte;
char *myattrs;
boolean_t rv;
/* XXX: vm_page should either contain pv_head or have a pointer to it */
bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
if (bank == -1) {
printf("pmap_change_attrs: unmanaged page?\n");
return(FALSE);
printf("pmap_clear_modify: unmanged page?\n");
return (FALSE);
}
PMAP_HEAD_TO_MAP_LOCK();
@ -2408,24 +2368,87 @@ pmap_change_attrs(pg, setbits, clearbits)
/* XXX: needed if we hold head->map lock? */
simple_lock(&pvh->pvh_lock);
myattrs = &vm_physmem[bank].pmseg.attrs[off];
result = *myattrs & clearbits;
*myattrs = (*myattrs | setbits) & ~clearbits;
if (vm_physmem[bank].pmseg.attrs[off] & PGA_MODIFIED) {
rv = TRUE;
pmap_changebit(pvh, 0, ~PG_M);
vm_physmem[bank].pmseg.attrs[off] &= ~PGA_MODIFIED;
} else
rv = FALSE;
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
return (rv);
}
/*
* pmap_clear_reference: clear the "referenced" attribute on a page.
*
* => we set pv_head => pmap locking
* => we return TRUE if the page was marked "modified".
* => XXX Note, referenced emulation isn't complete.
*/
boolean_t
pmap_clear_reference(vm_page_t pg)
{
int bank, off;
struct pv_head *pvh;
boolean_t rv;
/* XXX: vm_page should either contain pv_head or have a pointer to it */
bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
if (bank == -1) {
printf("pmap_clear_modify: unmanged page?\n");
return (FALSE);
}
PMAP_HEAD_TO_MAP_LOCK();
pvh = &vm_physmem[bank].pmseg.pvhead[off];
/* XXX: needed if we hold head->map lock? */
simple_lock(&pvh->pvh_lock);
if (vm_physmem[bank].pmseg.attrs[off] & PGA_REFERENCED) {
rv = TRUE;
#if 0 /* XXX notyet */
pmap_changebit(pvh, 0, ~PG_V);
#endif
vm_physmem[bank].pmseg.attrs[off] &= ~PGA_REFERENCED;
} else
rv = FALSE;
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
return (rv);
}
/*
* pmap_changebit: set or clear the specified PTEL bits for all
* mappings on the specified page.
*
* => We expect the pv_head -> map lock to be held
*/
void
pmap_changebit(struct pv_head *pvh, pt_entry_t set, pt_entry_t mask)
{
struct pv_entry *pve;
pt_entry_t *ptes, npte;
for (pve = pvh->pvh_list; pve != NULL; pve = pve->pv_next) {
#ifdef DIAGNOSTIC
if (pve->pv_va >= uvm.pager_sva && pve->pv_va < uvm.pager_eva) {
printf("pmap_change_attrs: found pager VA on pv_list\n");
/*
* XXX Don't write protect pager mappings.
*/
if (pve->pv_pmap == pmap_kernel() &&
/* XXX */ (mask == ~(PG_RW))) {
if (pve->pv_va >= uvm.pager_sva &&
pve->pv_va < uvm.pager_eva)
continue;
}
if (!pmap_valid_entry(pve->pv_pmap->pm_pdir[pdei(pve->pv_va)]))
panic("pmap_change_attrs: mapping without PTP "
"detected");
#endif
ptes = pmap_map_ptes(pve->pv_pmap); /* locks pmap */
npte = ptes[sh3_btop(pve->pv_va)];
result |= (npte & clearbits);
npte = (npte | setbits) & ~clearbits;
npte = (ptes[sh3_btop(pve->pv_va)] | set) & mask;
if (ptes[sh3_btop(pve->pv_va)] != npte) {
ptes[sh3_btop(pve->pv_va)] = npte; /* zap! */
@ -2435,11 +2458,6 @@ pmap_change_attrs(pg, setbits, clearbits)
}
pmap_unmap_ptes(pve->pv_pmap); /* unlocks pmap */
}
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
return(result != 0);
}
/*
@ -2449,10 +2467,37 @@ pmap_change_attrs(pg, setbits, clearbits)
/*
* pmap_page_protect: change the protection of all recorded mappings
* of a managed page
*
* => NOTE: this is an inline function in pmap.h
*/
void
pmap_page_protect(vm_page_t pg, vm_prot_t prot)
{
int bank, off;
struct pv_head *pvh;
if ((prot & VM_PROT_WRITE) == 0) {
if (prot & VM_PROT_ALL) {
bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
if (bank == -1) {
printf("pmap_page_protect: unmanged page?\n");
return;
}
PMAP_HEAD_TO_MAP_LOCK();
pvh = &vm_physmem[bank].pmseg.pvhead[off];
/* XXX: needed if we hold head->map lock? */
simple_lock(&pvh->pvh_lock);
pmap_changebit(pvh, 0, ~PG_RW);
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
} else {
pmap_page_remove(pg);
}
}
}
/* see pmap.h */
/*
@ -3270,20 +3315,49 @@ enter_now:
*/
npte = pa | protection_codes[prot] | PG_V | PG_N | PG_4K;
if (pvh)
if (pvh) {
/*
* Page is managed. Deal with modified/referenced
* attributes.
*
* XXX Referenced emulation could be improved.
*/
int attrs;
simple_lock(&pvh->pvh_lock);
if (flags & VM_PROT_WRITE)
vm_physmem[bank].pmseg.attrs[off] |=
(PGA_REFERENCED|PGA_MODIFIED);
else if (flags & VM_PROT_ALL)
vm_physmem[bank].pmseg.attrs[off] |=
PGA_REFERENCED;
attrs = vm_physmem[bank].pmseg.attrs[off];
simple_unlock(&pvh->pvh_lock);
/*
* If the page is dirty, don't fault on write.
*/
if (attrs & PGA_MODIFIED)
npte |= PG_M;
/*
* Mapping was entered on the PV list.
*/
npte |= PG_PVLIST;
} else {
/*
* Non-managed page -- mark it "dirty" anyway.
*/
npte |= PG_M;
}
if (wired)
npte |= PG_W;
if (va < VM_MAXUSER_ADDRESS)
npte |= PG_u;
else if (va < VM_MAX_ADDRESS)
npte |= (PG_u | PG_RW); /* XXXCDC: no longer needed? */
if (pmap == pmap_kernel()) {
if (pmap == pmap_kernel())
npte |= pmap_pg_g;
npte |= PG_M; /* XXX */
}
if (flags & VM_PROT_WRITE)
npte |= PG_M;
ptes[sh3_btop(va)] = npte; /* zap! */
@ -3446,23 +3520,52 @@ pmap_emulate_reference(p, va, user, write)
vaddr_t va;
int user, write;
{
struct pmap *pmap = p->p_vmspace->vm_map.pmap;
pt_entry_t *ptes;
struct pmap *pmap;
pt_entry_t *ptes, pte;
paddr_t pa;
struct pv_head *pvh;
int bank, off;
if (pmap_valid_entry(pmap->pm_pdir[pdei(va)])) {
ptes = pmap_map_ptes(pmap); /* locks pmap */
/*
* Convert process and virtual address to physical address.
*/
if (user)
pmap = p->p_vmspace->vm_map.pmap;
else
pmap = pmap_kernel();
if (!pmap_valid_entry(ptes[sh3_btop(va)]))
panic("pmap_set_modify: invalid (unmapped) va");
ptes[sh3_btop(va)] |= PG_M;
pmap_update_pg(va);
pmap_unmap_ptes(pmap); /* unlocks map */
}
#ifdef DIAGNOSTIC
else {
panic("pmap_emulate_reference: invalid PDE");
}
#endif
ptes = pmap_map_ptes(pmap); /* locks pmap */
pte = ptes[sh3_btop(va)];
if (pmap_valid_entry(pte) == FALSE)
panic("pmap_emulate_reference: invalid pte");
pa = pte & PG_FRAME;
pmap_unmap_ptes(pmap); /* unlocks pmap */
/*
* Twiddle the appropriate bits to reflect the reference
* and/or modification.
*
* The rules:
* (1) always mark the page as referenced, and
* (2) if it was a write fault, mark page as modified.
*/
bank = vm_physseg_find(sh3_btop(pa), &off);
if (bank == -1)
panic("pmap_emulate_reference: unmanaged page?");
pvh = &vm_physmem[bank].pmseg.pvhead[off];
pte = PG_V; /* referenced */
if (write)
pte |= PG_M;
PMAP_HEAD_TO_MAP_LOCK();
simple_lock(&pvh->pvh_lock);
pmap_changebit(pvh, pte, ~0);
simple_unlock(&pvh->pvh_lock);
PMAP_HEAD_TO_MAP_UNLOCK();
}
paddr_t