xref: /plan9-contrib/sys/src/9/pc/mmu.c (revision 9aba0e67f3fdf114973a71f5ffa0b6cd3cc40cfa)

/*
 * Memory mappings.  Life was easier when 2G of memory was enough.
 *
 * The kernel memory starts at KZERO, with the text loaded at KZERO+1M
 * (9load sits under 1M during the load).  The memory from KZERO to the
 * top of memory is mapped 1-1 with physical memory, starting at physical
 * address 0.  All kernel memory and data structures (i.e., the entries stored
 * into conf.mem) must sit in this physical range: if KZERO is at 0xF0000000,
 * then the kernel can only have 256MB of memory for itself.
 *
 * The 256M below KZERO comprises three parts.  The lowest 4M is the
 * virtual page table, a virtual address representation of the current
 * page table tree.  The second 4M is used for temporary per-process
 * mappings managed by kmap and kunmap.  The remaining 248M is used
 * for global (shared by all procs and all processors) device memory
 * mappings and managed by vmap and vunmap.  The total amount (256M)
 * could probably be reduced somewhat if desired.  The largest device
 * mapping is that of the video card, and even though modern video cards
 * have embarrassing amounts of memory, the video drivers only use one
 * frame buffer worth (at most 16M).  Each is described in more detail below.
 *
 * The VPT is a 4M frame constructed by inserting the pdb into itself.
 * This short-circuits one level of the page tables, with the result that
 * the contents of second-level page tables can be accessed at VPT.
 * We use the VPT to edit the page tables (see mmu) after inserting them
 * into the page directory.  It is a convenient mechanism for mapping what
 * might be otherwise-inaccessible pages.  The idea was borrowed from
 * the Exokernel.
 *
 * The VPT doesn't solve all our problems, because we still need to
 * prepare page directories before we can install them.  For that, we
 * use tmpmap/tmpunmap, which map a single page at TMPADDR.
 */

#include	"u.h"
#include	"../port/lib.h"
#include	"mem.h"
#include	"dat.h"
#include	"fns.h"
#include	"io.h"

/*
 * Simple segment descriptors with no translation.
 */
#define	DATASEGM(p) 	{ 0xFFFF, SEGG|SEGB|(0xF<<16)|SEGP|SEGPL(p)|SEGDATA|SEGW }
#define	EXECSEGM(p) 	{ 0xFFFF, SEGG|SEGD|(0xF<<16)|SEGP|SEGPL(p)|SEGEXEC|SEGR }
#define	EXEC16SEGM(p) 	{ 0xFFFF, SEGG|(0xF<<16)|SEGP|SEGPL(p)|SEGEXEC|SEGR }
#define	TSSSEGM(b,p)	{ ((b)<<16)|sizeof(Tss),\
			  ((b)&0xFF000000)|(((b)>>16)&0xFF)|SEGTSS|SEGPL(p)|SEGP }

Segdesc gdt[NGDT] =
{
[NULLSEG]	{ 0, 0},		/* null descriptor */
[KDSEG]		DATASEGM(0),		/* kernel data/stack */
[KESEG]		EXECSEGM(0),		/* kernel code */
[UDSEG]		DATASEGM(3),		/* user data/stack */
[UESEG]		EXECSEGM(3),		/* user code */
[TSSSEG]	TSSSEGM(0,0),		/* tss segment */
[KESEG16]		EXEC16SEGM(0),	/* kernel code 16-bit */
};

static int didmmuinit;
static void taskswitch(ulong, ulong);
static void memglobal(void);

#define	vpt ((ulong*)VPT)
#define	VPTX(va)		(((ulong)(va))>>12)
#define	vpd (vpt+VPTX(VPT))

void
mmuinit0(void)
{
	memmove(m->gdt, gdt, sizeof gdt);
}

void
mmuinit(void)
{
	ulong x, *p;
	ushort ptr[3];

	didmmuinit = 1;

	if(0) print("vpt=%#.8ux vpd=%#p kmap=%#.8ux\n",
		VPT, vpd, KMAP);

	memglobal();
	m->pdb[PDX(VPT)] = PADDR(m->pdb)|PTEWRITE|PTEVALID;

	m->tss = malloc(sizeof(Tss));
	memset(m->tss, 0, sizeof(Tss));
	m->tss->iomap = 0xDFFF<<16;

	/*
	 * We used to keep the GDT in the Mach structure, but it
	 * turns out that that slows down access to the rest of the
	 * page.  Since the Mach structure is accessed quite often,
	 * it pays off anywhere from a factor of 1.25 to 2 on real
	 * hardware to separate them (the AMDs are more sensitive
	 * than Intels in this regard).  Under VMware it pays off
	 * a factor of about 10 to 100.
	 */
	memmove(m->gdt, gdt, sizeof gdt);
	x = (ulong)m->tss;
	m->gdt[TSSSEG].d0 = (x<<16)|sizeof(Tss);
	m->gdt[TSSSEG].d1 = (x&0xFF000000)|((x>>16)&0xFF)|SEGTSS|SEGPL(0)|SEGP;

	ptr[0] = sizeof(gdt)-1;
	x = (ulong)m->gdt;
	ptr[1] = x & 0xFFFF;
	ptr[2] = (x>>16) & 0xFFFF;
	lgdt(ptr);

	ptr[0] = sizeof(Segdesc)*256-1;
	x = IDTADDR;
	ptr[1] = x & 0xFFFF;
	ptr[2] = (x>>16) & 0xFFFF;
	lidt(ptr);

	/* make kernel text unwritable */
	for(x = KTZERO; x < (ulong)etext; x += BY2PG){
		p = mmuwalk(m->pdb, x, 2, 0);
		if(p == nil)
			panic("mmuinit");
		*p &= ~PTEWRITE;
	}

	taskswitch(PADDR(m->pdb),  (ulong)m + BY2PG);
	ltr(TSSSEL);
}

/*
 * On processors that support it, we set the PTEGLOBAL bit in
 * page table and page directory entries that map kernel memory.
 * Doing this tells the processor not to bother flushing them
 * from the TLB when doing the TLB flush associated with a
 * context switch (write to CR3).  Since kernel memory mappings
 * are never removed, this is safe.  (If we ever remove kernel memory
 * mappings, we can do a full flush by turning off the PGE bit in CR4,
 * writing to CR3, and then turning the PGE bit back on.)
 *
 * See also mmukmap below.
 *
 * Processor support for the PTEGLOBAL bit is enabled in devarch.c.
 */
static void
memglobal(void)
{
	int i, j;
	ulong *pde, *pte;

	/* only need to do this once, on bootstrap processor */
	if(m->machno != 0)
		return;

	if(!m->havepge)
		return;

	pde = m->pdb;
	for(i=PDX(KZERO); i<1024; i++){
		if(pde[i] & PTEVALID){
			pde[i] |= PTEGLOBAL;
			if(!(pde[i] & PTESIZE)){
				pte = KADDR(pde[i]&~(BY2PG-1));
				for(j=0; j<1024; j++)
					if(pte[j] & PTEVALID)
						pte[j] |= PTEGLOBAL;
			}
		}
	}
}

/*
 * Flush all the user-space and device-mapping mmu info
 * for this process, because something has been deleted.
 * It will be paged back in on demand.
 */
void
flushmmu(void)
{
	int s;

	s = splhi();
	up->newtlb = 1;
	mmuswitch(up);
	splx(s);
}

/*
 * Flush a single page mapping from the tlb.
 */
void
flushpg(ulong va)
{
	if(X86FAMILY(m->cpuidax) >= 4)
		invlpg(va);
	else
		putcr3(getcr3());
}

/*
 * Allocate a new page for a page directory.
 * We keep a small cache of pre-initialized
 * page directories in each mach.
 */
static Page*
mmupdballoc(void)
{
	int s;
	Page *page;
	ulong *pdb;

	s = splhi();
	m->pdballoc++;
	if(m->pdbpool == 0){
		spllo();
		page = newpage(0, 0, 0);
		page->va = (ulong)vpd;
		splhi();
		pdb = tmpmap(page);
		memmove(pdb, m->pdb, BY2PG);
		pdb[PDX(VPT)] = page->pa|PTEWRITE|PTEVALID;	/* set up VPT */
		tmpunmap(pdb);
	}else{
		page = m->pdbpool;
		m->pdbpool = page->next;
		m->pdbcnt--;
	}
	splx(s);
	return page;
}

static void
mmupdbfree(Proc *proc, Page *p)
{
	if(islo())
		panic("mmupdbfree: islo");
	m->pdbfree++;
	if(m->pdbcnt >= 10){
		p->next = proc->mmufree;
		proc->mmufree = p;
	}else{
		p->next = m->pdbpool;
		m->pdbpool = p;
		m->pdbcnt++;
	}
}

/*
 * A user-space memory segment has been deleted, or the
 * process is exiting.  Clear all the pde entries for user-space
 * memory mappings and device mappings.  Any entries that
 * are needed will be paged back in as necessary.
 */
static void
mmuptefree(Proc* proc)
{
	int s;
	ulong *pdb;
	Page **last, *page;

	if(proc->mmupdb == nil || proc->mmuused == nil)
		return;
	s = splhi();
	pdb = tmpmap(proc->mmupdb);
	last = &proc->mmuused;
	for(page = *last; page; page = page->next){
		pdb[page->daddr] = 0;
		last = &page->next;
	}
	tmpunmap(pdb);
	splx(s);
	*last = proc->mmufree;
	proc->mmufree = proc->mmuused;
	proc->mmuused = 0;
}

static void
taskswitch(ulong pdb, ulong stack)
{
	Tss *tss;

	tss = m->tss;
	tss->ss0 = KDSEL;
	tss->esp0 = stack;
	tss->ss1 = KDSEL;
	tss->esp1 = stack;
	tss->ss2 = KDSEL;
	tss->esp2 = stack;
	putcr3(pdb);
}

void
mmuswitch(Proc* proc)
{
	ulong *pdb;

	if(proc->newtlb){
		mmuptefree(proc);
		proc->newtlb = 0;
	}

	if(proc->mmupdb){
		pdb = tmpmap(proc->mmupdb);
		pdb[PDX(MACHADDR)] = m->pdb[PDX(MACHADDR)];
		tmpunmap(pdb);
		taskswitch(proc->mmupdb->pa, (ulong)(proc->kstack+KSTACK));
	}else
		taskswitch(PADDR(m->pdb), (ulong)(proc->kstack+KSTACK));
}

/*
 * Release any pages allocated for a page directory base or page-tables
 * for this process:
 *   switch to the prototype pdb for this processor (m->pdb);
 *   call mmuptefree() to place all pages used for page-tables (proc->mmuused)
 *   onto the process' free list (proc->mmufree). This has the side-effect of
 *   cleaning any user entries in the pdb (proc->mmupdb);
 *   if there's a pdb put it in the cache of pre-initialised pdb's
 *   for this processor (m->pdbpool) or on the process' free list;
 *   finally, place any pages freed back into the free pool (palloc).
 * This routine is only called from schedinit() with palloc locked.
 */
void
mmurelease(Proc* proc)
{
	Page *page, *next;
	ulong *pdb;

	if(islo())
		panic("mmurelease: islo");
	taskswitch(PADDR(m->pdb), (ulong)m + BY2PG);
	if(proc->kmaptable){
		if(proc->mmupdb == nil)
			panic("mmurelease: no mmupdb");
		if(--proc->kmaptable->ref)
			panic("mmurelease: kmap ref %d\n", proc->kmaptable->ref);
		if(proc->nkmap)
			panic("mmurelease: nkmap %d\n", proc->nkmap);
		/*
		 * remove kmaptable from pdb before putting pdb up for reuse.
		 */
		pdb = tmpmap(proc->mmupdb);
		if(PPN(pdb[PDX(KMAP)]) != proc->kmaptable->pa)
			panic("mmurelease: bad kmap pde %#.8lux kmap %#.8lux",
				pdb[PDX(KMAP)], proc->kmaptable->pa);
		pdb[PDX(KMAP)] = 0;
		tmpunmap(pdb);
		/*
		 * move kmaptable to free list.
		 */
		pagechainhead(proc->kmaptable);
		proc->kmaptable = 0;
	}
	if(proc->mmupdb){
		mmuptefree(proc);
		mmupdbfree(proc, proc->mmupdb);
		proc->mmupdb = 0;
	}
	for(page = proc->mmufree; page; page = next){
		next = page->next;
		if(--page->ref)
			panic("mmurelease: page->ref %d\n", page->ref);
		pagechainhead(page);
	}
	if(proc->mmufree && palloc.r.p)
		wakeup(&palloc.r);
	proc->mmufree = 0;
}

/*
 * Allocate and install pdb for the current process.
 */
static void
upallocpdb(void)
{
	int s;
	ulong *pdb;
	Page *page;

	if(up->mmupdb != nil)
		return;
	page = mmupdballoc();
	s = splhi();
	if(up->mmupdb != nil){
		/*
		 * Perhaps we got an interrupt while
		 * mmupdballoc was sleeping and that
		 * interrupt allocated an mmupdb?
		 * Seems unlikely.
		 */
		mmupdbfree(up, page);
		splx(s);
		return;
	}
	pdb = tmpmap(page);
	pdb[PDX(MACHADDR)] = m->pdb[PDX(MACHADDR)];
	tmpunmap(pdb);
	up->mmupdb = page;
	putcr3(up->mmupdb->pa);
	splx(s);
}

/*
 * Update the mmu in response to a user fault.  pa may have PTEWRITE set.
 */
void
putmmu(ulong va, ulong pa, Page*)
{
	int old, s;
	Page *page;

	if(up->mmupdb == nil)
		upallocpdb();

	/*
	 * We should be able to get through this with interrupts
	 * turned on (if we get interrupted we'll just pick up
	 * where we left off) but we get many faults accessing
	 * vpt[] near the end of this function, and they always happen
	 * after the process has been switched out and then
	 * switched back, usually many times in a row (perhaps
	 * it cannot switch back successfully for some reason).
	 *
	 * In any event, I'm tired of searching for this bug.
	 * Turn off interrupts during putmmu even though
	 * we shouldn't need to.		- rsc
	 */

	s = splhi();
	if(!(vpd[PDX(va)]&PTEVALID)){
		if(up->mmufree == 0){
			spllo();
			page = newpage(0, 0, 0);
			splhi();
		}
		else{
			page = up->mmufree;
			up->mmufree = page->next;
		}
		vpd[PDX(va)] = PPN(page->pa)|PTEUSER|PTEWRITE|PTEVALID;
		/* page is now mapped into the VPT - clear it */
		memset((void*)(VPT+PDX(va)*BY2PG), 0, BY2PG);
		page->daddr = PDX(va);
		page->next = up->mmuused;
		up->mmuused = page;
	}
	old = vpt[VPTX(va)];
	vpt[VPTX(va)] = pa|PTEUSER|PTEVALID;
	if(old&PTEVALID)
		flushpg(va);
	if(getcr3() != up->mmupdb->pa)
		print("bad cr3 %#.8lux %#.8lux\n", getcr3(), up->mmupdb->pa);
	splx(s);
}

/*
 * Double-check the user MMU.
 * Error checking only.
 */
void
checkmmu(ulong va, ulong pa)
{
	if(up->mmupdb == 0)
		return;
	if(!(vpd[PDX(va)]&PTEVALID) || !(vpt[VPTX(va)]&PTEVALID))
		return;
	if(PPN(vpt[VPTX(va)]) != pa)
		print("%ld %s: va=%#08lux pa=%#08lux pte=%#08lux\n",
			up->pid, up->text,
			va, pa, vpt[VPTX(va)]);
}

/*
 * Walk the page-table pointed to by pdb and return a pointer
 * to the entry for virtual address va at the requested level.
 * If the entry is invalid and create isn't requested then bail
 * out early. Otherwise, for the 2nd level walk, allocate a new
 * page-table page and register it in the 1st level.  This is used
 * only to edit kernel mappings, which use pages from kernel memory,
 * so it's okay to use KADDR to look at the tables.
 */
ulong*
mmuwalk(ulong* pdb, ulong va, int level, int create)
{
	ulong *table;
	void *map;

	table = &pdb[PDX(va)];
	if(!(*table & PTEVALID) && create == 0)
		return 0;

	switch(level){

	default:
		return 0;

	case 1:
		return table;

	case 2:
		if(*table & PTESIZE)
			panic("mmuwalk2: va %luX entry %luX\n", va, *table);
		if(!(*table & PTEVALID)){
			/*
			 * Have to call low-level allocator from
			 * memory.c if we haven't set up the xalloc
			 * tables yet.
			 */
			if(didmmuinit)
				map = xspanalloc(BY2PG, BY2PG, 0);
			else
				map = rampage();
			if(map == nil)
				panic("mmuwalk xspanalloc failed");
			*table = PADDR(map)|PTEWRITE|PTEVALID;
		}
		table = KADDR(PPN(*table));
		return &table[PTX(va)];
	}
}

/*
 * Device mappings are shared by all procs and processors and
 * live in the virtual range VMAP to VMAP+VMAPSIZE.  The master
 * copy of the mappings is stored in mach0->pdb, and they are
 * paged in from there as necessary by vmapsync during faults.
 */

static Lock vmaplock;

static int findhole(ulong *a, int n, int count);
static ulong vmapalloc(ulong size);
static void pdbunmap(ulong*, ulong, int);

/*
 * Add a device mapping to the vmap range.
 */
void*
vmap(ulong pa, int size)
{
	int osize;
	ulong o, va;

	/*
	 * might be asking for less than a page.
	 */
	osize = size;
	o = pa & (BY2PG-1);
	pa -= o;
	size += o;

	size = ROUND(size, BY2PG);
	if(pa == 0){
		print("vmap pa=0 pc=%#p\n", getcallerpc(&pa));
		return nil;
	}
	ilock(&vmaplock);
	if((va = vmapalloc(size)) == 0
	|| pdbmap(MACHP(0)->pdb, pa|PTEUNCACHED|PTEWRITE, va, size) < 0){
		iunlock(&vmaplock);
		return 0;
	}
	iunlock(&vmaplock);
	/* avoid trap on local processor
	for(i=0; i<size; i+=4*MB)
		vmapsync(va+i);
	*/
	USED(osize);
//	print("  vmap %#.8lux %d => %#.8lux\n", pa+o, osize, va+o);
	return (void*)(va + o);
}

static int
findhole(ulong *a, int n, int count)
{
	int have, i;

	have = 0;
	for(i=0; i<n; i++){
		if(a[i] == 0)
			have++;
		else
			have = 0;
		if(have >= count)
			return i+1 - have;
	}
	return -1;
}

/*
 * Look for free space in the vmap.
 */
static ulong
vmapalloc(ulong size)
{
	int i, n, o;
	ulong *vpdb;
	int vpdbsize;

	vpdb = &MACHP(0)->pdb[PDX(VMAP)];
	vpdbsize = VMAPSIZE/(4*MB);

	if(size >= 4*MB){
		n = (size+4*MB-1) / (4*MB);
		if((o = findhole(vpdb, vpdbsize, n)) != -1)
			return VMAP + o*4*MB;
		return 0;
	}
	n = (size+BY2PG-1) / BY2PG;
	for(i=0; i<vpdbsize; i++)
		if((vpdb[i]&PTEVALID) && !(vpdb[i]&PTESIZE))
			if((o = findhole(KADDR(PPN(vpdb[i])), WD2PG, n)) != -1)
				return VMAP + i*4*MB + o*BY2PG;
	if((o = findhole(vpdb, vpdbsize, 1)) != -1)
		return VMAP + o*4*MB;

	/*
	 * could span page directory entries, but not worth the trouble.
	 * not going to be very much contention.
	 */
	return 0;
}

/*
 * Remove a device mapping from the vmap range.
 * Since pdbunmap does not remove page tables, just entries,
 * the call need not be interlocked with vmap.
 */
void
vunmap(void *v, int size)
{
	int i;
	ulong va, o;
	Mach *nm;
	Proc *p;

	/*
	 * might not be aligned
	 */
	va = (ulong)v;
	o = va&(BY2PG-1);
	va -= o;
	size += o;
	size = ROUND(size, BY2PG);

	if(size < 0 || va < VMAP || va+size > VMAP+VMAPSIZE)
		panic("vunmap va=%#.8lux size=%#x pc=%#.8lux\n",
			va, size, getcallerpc(&va));

	pdbunmap(MACHP(0)->pdb, va, size);

	/*
	 * Flush mapping from all the tlbs and copied pdbs.
	 * This can be (and is) slow, since it is called only rarely.
	 * It is possible for vunmap to be called with up == nil,
	 * e.g. from the reset/init driver routines during system
	 * boot. In that case it suffices to flush the MACH(0) TLB
	 * and return.
	 */
	if(!active.thunderbirdsarego){
		putcr3(PADDR(MACHP(0)->pdb));
		return;
	}
	for(i=0; i<conf.nproc; i++){
		p = proctab(i);
		if(p->state == Dead)
			continue;
		if(p != up)
			p->newtlb = 1;
	}
	for(i=0; i<conf.nmach; i++){
		nm = MACHP(i);
		if(nm != m)
			nm->flushmmu = 1;
	}
	flushmmu();
	for(i=0; i<conf.nmach; i++){
		nm = MACHP(i);
		if(nm != m)
			while((active.machs&(1<<nm->machno)) && nm->flushmmu)
				;
	}
}

/*
 * Add kernel mappings for pa -> va for a section of size bytes.
 */
int
pdbmap(ulong *pdb, ulong pa, ulong va, int size)
{
	int pse;
	ulong pgsz, *pte, *table;
	ulong flag, off;

	flag = pa&0xFFF;
	pa &= ~0xFFF;

	if((MACHP(0)->cpuiddx & 0x08) && (getcr4() & 0x10))
		pse = 1;
	else
		pse = 0;

	for(off=0; off<size; off+=pgsz){
		table = &pdb[PDX(va+off)];
		if((*table&PTEVALID) && (*table&PTESIZE))
			panic("vmap: va=%#.8lux pa=%#.8lux pde=%#.8lux",
				va+off, pa+off, *table);

		/*
		 * Check if it can be mapped using a 4MB page:
		 * va, pa aligned and size >= 4MB and processor can do it.
		 */
		if(pse && (pa+off)%(4*MB) == 0 && (va+off)%(4*MB) == 0 && (size-off) >= 4*MB){
			*table = (pa+off)|flag|PTESIZE|PTEVALID;
			pgsz = 4*MB;
		}else{
			pte = mmuwalk(pdb, va+off, 2, 1);
			if(*pte&PTEVALID)
				panic("vmap: va=%#.8lux pa=%#.8lux pte=%#.8lux",
					va+off, pa+off, *pte);
			*pte = (pa+off)|flag|PTEVALID;
			pgsz = BY2PG;
		}
	}
	return 0;
}

/*
 * Remove mappings.  Must already exist, for sanity.
 * Only used for kernel mappings, so okay to use KADDR.
 */
static void
pdbunmap(ulong *pdb, ulong va, int size)
{
	ulong vae;
	ulong *table;

	vae = va+size;
	while(va < vae){
		table = &pdb[PDX(va)];
		if(!(*table & PTEVALID)){
			panic("vunmap: not mapped");
			/*
			va = (va+4*MB-1) & ~(4*MB-1);
			continue;
			*/
		}
		if(*table & PTESIZE){
			*table = 0;
			va = (va+4*MB-1) & ~(4*MB-1);
			continue;
		}
		table = KADDR(PPN(*table));
		if(!(table[PTX(va)] & PTEVALID))
			panic("vunmap: not mapped");
		table[PTX(va)] = 0;
		va += BY2PG;
	}
}

/*
 * Handle a fault by bringing vmap up to date.
 * Only copy pdb entries and they never go away,
 * so no locking needed.
 */
int
vmapsync(ulong va)
{
	ulong entry, *table;

	if(va < VMAP || va >= VMAP+VMAPSIZE)
		return 0;

	entry = MACHP(0)->pdb[PDX(va)];
	if(!(entry&PTEVALID))
		return 0;
	if(!(entry&PTESIZE)){
		/* make sure entry will help the fault */
		table = KADDR(PPN(entry));
		if(!(table[PTX(va)]&PTEVALID))
			return 0;
	}
	vpd[PDX(va)] = entry;
	/*
	 * TLB doesn't cache negative results, so no flush needed.
	 */
	return 1;
}


/*
 * KMap is used to map individual pages into virtual memory.
 * It is rare to have more than a few KMaps at a time (in the
 * absence of interrupts, only two at a time are ever used,
 * but interrupts can stack).  The mappings are local to a process,
 * so we can use the same range of virtual address space for
 * all processes without any coordination.
 */
#define kpt (vpt+VPTX(KMAP))
#define NKPT (KMAPSIZE/BY2PG)

KMap*
kmap(Page *page)
{
	int i, o, s;

	if(up == nil)
		panic("kmap: up=0 pc=%#.8lux", getcallerpc(&page));
	if(up->mmupdb == nil)
		upallocpdb();
	if(up->nkmap < 0)
		panic("kmap %lud %s: nkmap=%d", up->pid, up->text, up->nkmap);

	/*
	 * Splhi shouldn't be necessary here, but paranoia reigns.
	 * See comment in putmmu above.
	 */
	s = splhi();
	up->nkmap++;
	if(!(vpd[PDX(KMAP)]&PTEVALID)){
		/* allocate page directory */
		if(KMAPSIZE > BY2XPG)
			panic("bad kmapsize");
		if(up->kmaptable != nil)
			panic("kmaptable");
		spllo();
		up->kmaptable = newpage(0, 0, 0);
		splhi();
		vpd[PDX(KMAP)] = up->kmaptable->pa|PTEWRITE|PTEVALID;
		flushpg((ulong)kpt);
		memset(kpt, 0, BY2PG);
		kpt[0] = page->pa|PTEWRITE|PTEVALID;
		up->lastkmap = 0;
		splx(s);
		return (KMap*)KMAP;
	}
	if(up->kmaptable == nil)
		panic("no kmaptable");
	o = up->lastkmap+1;
	for(i=0; i<NKPT; i++){
		if(kpt[(i+o)%NKPT] == 0){
			o = (i+o)%NKPT;
			kpt[o] = page->pa|PTEWRITE|PTEVALID;
			up->lastkmap = o;
			splx(s);
			return (KMap*)(KMAP+o*BY2PG);
		}
	}
	panic("out of kmap");
	return nil;
}

void
kunmap(KMap *k)
{
	ulong va;

	va = (ulong)k;
	if(up->mmupdb == nil || !(vpd[PDX(KMAP)]&PTEVALID))
		panic("kunmap: no kmaps");
	if(va < KMAP || va >= KMAP+KMAPSIZE)
		panic("kunmap: bad address %#.8lux pc=%#p", va, getcallerpc(&k));
	if(!(vpt[VPTX(va)]&PTEVALID))
		panic("kunmap: not mapped %#.8lux pc=%#p", va, getcallerpc(&k));
	up->nkmap--;
	if(up->nkmap < 0)
		panic("kunmap %lud %s: nkmap=%d", up->pid, up->text, up->nkmap);
	vpt[VPTX(va)] = 0;
	flushpg(va);
}

/*
 * Temporary one-page mapping used to edit page directories.
 *
 * The fasttmp #define controls whether the code optimizes
 * the case where the page is already mapped in the physical
 * memory window.
 */
#define fasttmp 1

void*
tmpmap(Page *p)
{
	ulong i;
	ulong *entry;

	if(islo())
		panic("tmpaddr: islo");

	if(fasttmp && p->pa < -KZERO)
		return KADDR(p->pa);

	/*
	 * PDX(TMPADDR) == PDX(MACHADDR), so this
	 * entry is private to the processor and shared
	 * between up->mmupdb (if any) and m->pdb.
	 */
	entry = &vpt[VPTX(TMPADDR)];
	if(!(*entry&PTEVALID)){
		for(i=KZERO; i<=CPU0MACH; i+=BY2PG)
			print("%#p: *%#p=%#p (vpt=%#p index=%#p)\n", i, &vpt[VPTX(i)], vpt[VPTX(i)], vpt, VPTX(i));
		panic("tmpmap: no entry");
	}
	if(PPN(*entry) != PPN(TMPADDR-KZERO))
		panic("tmpmap: already mapped entry=%#.8lux", *entry);
	*entry = p->pa|PTEWRITE|PTEVALID;
	flushpg(TMPADDR);
	return (void*)TMPADDR;
}

void
tmpunmap(void *v)
{
	ulong *entry;

	if(islo())
		panic("tmpaddr: islo");
	if(fasttmp && (ulong)v >= KZERO && v != (void*)TMPADDR)
		return;
	if(v != (void*)TMPADDR)
		panic("tmpunmap: bad address");
	entry = &vpt[VPTX(TMPADDR)];
	if(!(*entry&PTEVALID) || PPN(*entry) == PPN(PADDR(TMPADDR)))
		panic("tmpmap: not mapped entry=%#.8lux", *entry);
	*entry = PPN(TMPADDR-KZERO)|PTEWRITE|PTEVALID;
	flushpg(TMPADDR);
}

/*
 * These could go back to being macros once the kernel is debugged,
 * but the extra checking is nice to have.
 */
void*
kaddr(ulong pa)
{
	if(pa > (ulong)-KZERO)
		panic("kaddr: pa=%#.8lux", pa);
	return (void*)(pa+KZERO);
}

ulong
paddr(void *v)
{
	ulong va;

	va = (ulong)v;
	if(va < KZERO)
		panic("paddr: va=%#.8lux pc=%#p", va, getcallerpc(&v));
	return va-KZERO;
}

/*
 * More debugging.
 */
void
countpagerefs(ulong *ref, int print)
{
	int i, n;
	Mach *mm;
	Page *pg;
	Proc *p;

	n = 0;
	for(i=0; i<conf.nproc; i++){
		p = proctab(i);
		if(p->mmupdb){
			if(print){
				if(ref[pagenumber(p->mmupdb)])
					iprint("page %#.8lux is proc %d (pid %lud) pdb\n",
						p->mmupdb->pa, i, p->pid);
				continue;
			}
			if(ref[pagenumber(p->mmupdb)]++ == 0)
				n++;
			else
				iprint("page %#.8lux is proc %d (pid %lud) pdb but has other refs!\n",
					p->mmupdb->pa, i, p->pid);
		}
		if(p->kmaptable){
			if(print){
				if(ref[pagenumber(p->kmaptable)])
					iprint("page %#.8lux is proc %d (pid %lud) kmaptable\n",
						p->kmaptable->pa, i, p->pid);
				continue;
			}
			if(ref[pagenumber(p->kmaptable)]++ == 0)
				n++;
			else
				iprint("page %#.8lux is proc %d (pid %lud) kmaptable but has other refs!\n",
					p->kmaptable->pa, i, p->pid);
		}
		for(pg=p->mmuused; pg; pg=pg->next){
			if(print){
				if(ref[pagenumber(pg)])
					iprint("page %#.8lux is on proc %d (pid %lud) mmuused\n",
						pg->pa, i, p->pid);
				continue;
			}
			if(ref[pagenumber(pg)]++ == 0)
				n++;
			else
				iprint("page %#.8lux is on proc %d (pid %lud) mmuused but has other refs!\n",
					pg->pa, i, p->pid);
		}
		for(pg=p->mmufree; pg; pg=pg->next){
			if(print){
				if(ref[pagenumber(pg)])
					iprint("page %#.8lux is on proc %d (pid %lud) mmufree\n",
						pg->pa, i, p->pid);
				continue;
			}
			if(ref[pagenumber(pg)]++ == 0)
				n++;
			else
				iprint("page %#.8lux is on proc %d (pid %lud) mmufree but has other refs!\n",
					pg->pa, i, p->pid);
		}
	}
	if(!print)
		iprint("%d pages in proc mmu\n", n);
	n = 0;
	for(i=0; i<conf.nmach; i++){
		mm = MACHP(i);
		for(pg=mm->pdbpool; pg; pg=pg->next){
			if(print){
				if(ref[pagenumber(pg)])
					iprint("page %#.8lux is in cpu%d pdbpool\n",
						pg->pa, i);
				continue;
			}
			if(ref[pagenumber(pg)]++ == 0)
				n++;
			else
				iprint("page %#.8lux is in cpu%d pdbpool but has other refs!\n",
					pg->pa, i);
		}
	}
	if(!print){
		iprint("%d pages in mach pdbpools\n", n);
		for(i=0; i<conf.nmach; i++)
			iprint("cpu%d: %d pdballoc, %d pdbfree\n",
				i, MACHP(i)->pdballoc, MACHP(i)->pdbfree);
	}
}

void
checkfault(ulong, ulong)
{
}

/*
 * Return the number of bytes that can be accessed via KADDR(pa).
 * If pa is not a valid argument to KADDR, return 0.
 */
ulong
cankaddr(ulong pa)
{
	if(pa >= -KZERO)
		return 0;
	return -KZERO - pa;
}