Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.20
      1 /*	$NetBSD: pmap.c,v 1.20 2003/05/08 18:13:21 thorpej Exp $	*/
      2 
      3 /*
      4  * Copyright 2001 Wasabi Systems, Inc.
      5  * All rights reserved.
      6  *
      7  * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, Inc.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed for the NetBSD Project by
     20  *      Wasabi Systems, Inc.
     21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     22  *    or promote products derived from this software without specific prior
     23  *    written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     35  * POSSIBILITY OF SUCH DAMAGE.
     36  */
     37 
     38 /*
     39  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
     40  * Copyright (C) 1995, 1996 TooLs GmbH.
     41  * All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by TooLs GmbH.
     54  * 4. The name of TooLs GmbH may not be used to endorse or promote products
     55  *    derived from this software without specific prior written permission.
     56  *
     57  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
     58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     60  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     61  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
     62  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
     63  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     64  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
     65  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
     66  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     67  */
     68 
     69 #include <sys/param.h>
     70 #include <sys/malloc.h>
     71 #include <sys/proc.h>
     72 #include <sys/user.h>
     73 #include <sys/queue.h>
     74 #include <sys/systm.h>
     75 #include <sys/pool.h>
     76 #include <sys/device.h>
     77 
     78 #include <uvm/uvm.h>
     79 
     80 #include <machine/cpu.h>
     81 #include <machine/pcb.h>
     82 #include <machine/powerpc.h>
     83 
     84 #include <powerpc/spr.h>
     85 #include <machine/tlb.h>
     86 
     87 /*
     88  * kernmap is an array of PTEs large enough to map in
     89  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     90  */
     91 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
     92 caddr_t kernmap;
     93 
     94 #define MINCTX		2
     95 #define NUMCTX		256
     96 volatile struct pmap *ctxbusy[NUMCTX];
     97 
     98 #define TLBF_USED	0x1
     99 #define	TLBF_REF	0x2
    100 #define	TLBF_LOCKED	0x4
    101 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    102 typedef struct tlb_info_s {
    103 	char	ti_flags;
    104 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    105 	u_int	ti_va;
    106 } tlb_info_t;
    107 
    108 volatile tlb_info_t tlb_info[NTLB];
    109 /* We'll use a modified FIFO replacement policy cause it's cheap */
    110 volatile int tlbnext = TLB_NRESERVED;
    111 
    112 u_long dtlb_miss_count = 0;
    113 u_long itlb_miss_count = 0;
    114 u_long ktlb_miss_count = 0;
    115 u_long utlb_miss_count = 0;
    116 
    117 /* Event counters */
    118 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    119 	NULL, "cpu", "tlbmiss");
    120 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    121 	NULL, "cpu", "tlbhit");
    122 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    123 	NULL, "cpu", "tlbflush");
    124 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    125 	NULL, "cpu", "tlbenter");
    126 
    127 struct pmap kernel_pmap_;
    128 
    129 int physmem;
    130 static int npgs;
    131 static u_int nextavail;
    132 #ifndef MSGBUFADDR
    133 extern paddr_t msgbuf_paddr;
    134 #endif
    135 
    136 static struct mem_region *mem, *avail;
    137 
    138 /*
    139  * This is a cache of referenced/modified bits.
    140  * Bits herein are shifted by ATTRSHFT.
    141  */
    142 static char *pmap_attrib;
    143 
    144 #define PV_WIRED	0x1
    145 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    146 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    147 
    148 struct pv_entry {
    149 	struct pv_entry *pv_next;	/* Linked list of mappings */
    150 	vaddr_t pv_va;			/* virtual address of mapping */
    151 	struct pmap *pv_pm;
    152 };
    153 
    154 struct pv_entry *pv_table;
    155 static struct pool pv_pool;
    156 
    157 static int pmap_initialized;
    158 
    159 static int ctx_flush(int);
    160 
    161 inline struct pv_entry *pa_to_pv(paddr_t);
    162 static inline char *pa_to_attr(paddr_t);
    163 
    164 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    165 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    166 
    167 static void pmap_pinit(pmap_t);
    168 static void pmap_release(pmap_t);
    169 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    170 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    171 
    172 
    173 inline struct pv_entry *
    174 pa_to_pv(paddr_t pa)
    175 {
    176 	int bank, pg;
    177 
    178 	bank = vm_physseg_find(atop(pa), &pg);
    179 	if (bank == -1)
    180 		return NULL;
    181 	return &vm_physmem[bank].pmseg.pvent[pg];
    182 }
    183 
    184 static inline char *
    185 pa_to_attr(paddr_t pa)
    186 {
    187 	int bank, pg;
    188 
    189 	bank = vm_physseg_find(atop(pa), &pg);
    190 	if (bank == -1)
    191 		return NULL;
    192 	return &vm_physmem[bank].pmseg.attrs[pg];
    193 }
    194 
    195 /*
    196  * Insert PTE into page table.
    197  */
    198 int
    199 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    200 {
    201 	int seg = STIDX(va);
    202 	int ptn = PTIDX(va);
    203 	paddr_t pa;
    204 
    205 	if (!pm->pm_ptbl[seg]) {
    206 		/* Don't allocate a page to clear a non-existent mapping. */
    207 		if (!pte) return (1);
    208 		/* Allocate a page XXXX this will sleep! */
    209 		pa = 0;
    210 		pm->pm_ptbl[seg] =
    211 		    (uint *)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
    212 	}
    213 	pm->pm_ptbl[seg][ptn] = pte;
    214 
    215 	/* Flush entry. */
    216 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    217 	return (1);
    218 }
    219 
    220 /*
    221  * Get a pointer to a PTE in a page table.
    222  */
    223 volatile u_int *
    224 pte_find(struct pmap *pm, vaddr_t va)
    225 {
    226 	int seg = STIDX(va);
    227 	int ptn = PTIDX(va);
    228 
    229 	if (pm->pm_ptbl[seg])
    230 		return (&pm->pm_ptbl[seg][ptn]);
    231 
    232 	return (NULL);
    233 }
    234 
    235 /*
    236  * This is called during initppc, before the system is really initialized.
    237  */
    238 void
    239 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    240 {
    241 	struct mem_region *mp, *mp1;
    242 	int cnt, i;
    243 	u_int s, e, sz;
    244 
    245 	/*
    246 	 * Allocate the kernel page table at the end of
    247 	 * kernel space so it's in the locked TTE.
    248 	 */
    249 	kernmap = (caddr_t)kernelend;
    250 
    251 	/*
    252 	 * Initialize kernel page table.
    253 	 */
    254 	for (i = 0; i < STSZ; i++) {
    255 		pmap_kernel()->pm_ptbl[i] = 0;
    256 	}
    257 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    258 
    259 	/*
    260 	 * Announce page-size to the VM-system
    261 	 */
    262 	uvmexp.pagesize = NBPG;
    263 	uvm_setpagesize();
    264 
    265 	/*
    266 	 * Get memory.
    267 	 */
    268 	mem_regions(&mem, &avail);
    269 	for (mp = mem; mp->size; mp++) {
    270 		physmem += btoc(mp->size);
    271 		printf("+%lx,",mp->size);
    272 	}
    273 	printf("\n");
    274 	ppc4xx_tlb_init();
    275 	/*
    276 	 * Count the number of available entries.
    277 	 */
    278 	for (cnt = 0, mp = avail; mp->size; mp++)
    279 		cnt++;
    280 
    281 	/*
    282 	 * Page align all regions.
    283 	 * Non-page aligned memory isn't very interesting to us.
    284 	 * Also, sort the entries for ascending addresses.
    285 	 */
    286 	kernelstart &= ~PGOFSET;
    287 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    288 	for (mp = avail; mp->size; mp++) {
    289 		s = mp->start;
    290 		e = mp->start + mp->size;
    291 		printf("%08x-%08x -> ",s,e);
    292 		/*
    293 		 * Check whether this region holds all of the kernel.
    294 		 */
    295 		if (s < kernelstart && e > kernelend) {
    296 			avail[cnt].start = kernelend;
    297 			avail[cnt++].size = e - kernelend;
    298 			e = kernelstart;
    299 		}
    300 		/*
    301 		 * Look whether this regions starts within the kernel.
    302 		 */
    303 		if (s >= kernelstart && s < kernelend) {
    304 			if (e <= kernelend)
    305 				goto empty;
    306 			s = kernelend;
    307 		}
    308 		/*
    309 		 * Now look whether this region ends within the kernel.
    310 		 */
    311 		if (e > kernelstart && e <= kernelend) {
    312 			if (s >= kernelstart)
    313 				goto empty;
    314 			e = kernelstart;
    315 		}
    316 		/*
    317 		 * Now page align the start and size of the region.
    318 		 */
    319 		s = round_page(s);
    320 		e = trunc_page(e);
    321 		if (e < s)
    322 			e = s;
    323 		sz = e - s;
    324 		printf("%08x-%08x = %x\n",s,e,sz);
    325 		/*
    326 		 * Check whether some memory is left here.
    327 		 */
    328 		if (sz == 0) {
    329 		empty:
    330 			memmove(mp, mp + 1,
    331 				(cnt - (mp - avail)) * sizeof *mp);
    332 			cnt--;
    333 			mp--;
    334 			continue;
    335 		}
    336 		/*
    337 		 * Do an insertion sort.
    338 		 */
    339 		npgs += btoc(sz);
    340 		for (mp1 = avail; mp1 < mp; mp1++)
    341 			if (s < mp1->start)
    342 				break;
    343 		if (mp1 < mp) {
    344 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    345 			mp1->start = s;
    346 			mp1->size = sz;
    347 		} else {
    348 			mp->start = s;
    349 			mp->size = sz;
    350 		}
    351 	}
    352 
    353 	/*
    354 	 * We cannot do pmap_steal_memory here,
    355 	 * since we don't run with translation enabled yet.
    356 	 */
    357 #ifndef MSGBUFADDR
    358 	/*
    359 	 * allow for msgbuf
    360 	 */
    361 	sz = round_page(MSGBUFSIZE);
    362 	mp = NULL;
    363 	for (mp1 = avail; mp1->size; mp1++)
    364 		if (mp1->size >= sz)
    365 			mp = mp1;
    366 	if (mp == NULL)
    367 		panic("not enough memory?");
    368 
    369 	npgs -= btoc(sz);
    370 	msgbuf_paddr = mp->start + mp->size - sz;
    371 	mp->size -= sz;
    372 	if (mp->size <= 0)
    373 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    374 #endif
    375 
    376 	printf("Loading pages\n");
    377 	for (mp = avail; mp->size; mp++)
    378 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    379 			atop(mp->start), atop(mp->start + mp->size),
    380 			VM_FREELIST_DEFAULT);
    381 
    382 	/*
    383 	 * Initialize kernel pmap and hardware.
    384 	 */
    385 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    386 	pmap_kernel()->pm_ctx = KERNEL_PID;
    387 	nextavail = avail->start;
    388 
    389 	/*
    390 	 * Define the boundaries of the managed kernel virtual
    391 	 * address space.
    392 	 */
    393 	virtual_avail = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    394 	virtual_end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    395 
    396 	evcnt_attach_static(&tlbhit_ev);
    397 	evcnt_attach_static(&tlbmiss_ev);
    398 	evcnt_attach_static(&tlbflush_ev);
    399 	evcnt_attach_static(&tlbenter_ev);
    400 	printf("Done\n");
    401 }
    402 
    403 /*
    404  * Restrict given range to physical memory
    405  *
    406  * (Used by /dev/mem)
    407  */
    408 void
    409 pmap_real_memory(paddr_t *start, psize_t *size)
    410 {
    411 	struct mem_region *mp;
    412 
    413 	for (mp = mem; mp->size; mp++) {
    414 		if (*start + *size > mp->start &&
    415 		    *start < mp->start + mp->size) {
    416 			if (*start < mp->start) {
    417 				*size -= mp->start - *start;
    418 				*start = mp->start;
    419 			}
    420 			if (*start + *size > mp->start + mp->size)
    421 				*size = mp->start + mp->size - *start;
    422 			return;
    423 		}
    424 	}
    425 	*size = 0;
    426 }
    427 
    428 /*
    429  * Initialize anything else for pmap handling.
    430  * Called during vm_init().
    431  */
    432 void
    433 pmap_init(void)
    434 {
    435 	struct pv_entry *pv;
    436 	vsize_t sz;
    437 	vaddr_t addr;
    438 	int i, s;
    439 	int bank;
    440 	char *attr;
    441 
    442 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    443 	sz = round_page(sz);
    444 	addr = uvm_km_zalloc(kernel_map, sz);
    445 	s = splvm();
    446 	pv = pv_table = (struct pv_entry *)addr;
    447 	for (i = npgs; --i >= 0;)
    448 		pv++->pv_pm = NULL;
    449 	pmap_attrib = (char *)pv;
    450 	memset(pv, 0, npgs);
    451 
    452 	pv = pv_table;
    453 	attr = pmap_attrib;
    454 	for (bank = 0; bank < vm_nphysseg; bank++) {
    455 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    456 		vm_physmem[bank].pmseg.pvent = pv;
    457 		vm_physmem[bank].pmseg.attrs = attr;
    458 		pv += sz;
    459 		attr += sz;
    460 	}
    461 
    462 	pmap_initialized = 1;
    463 	splx(s);
    464 
    465 	/* Setup a pool for additional pvlist structures */
    466 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    467 }
    468 
    469 #ifdef PMAP_GROWKERNEL
    470 /*
    471  * Preallocate kernel page tables to a specified VA.
    472  * This simply loops through the first TTE for each
    473  * page table from the beginning of the kernel pmap,
    474  * reads the entry, and if the result is
    475  * zero (either invalid entry or no page table) it stores
    476  * a zero there, populating page tables in the process.
    477  * This is not the most efficient technique but i don't
    478  * expect it to be called that often.
    479  */
    480 extern struct vm_page *vm_page_alloc1 __P((void));
    481 extern void vm_page_free1 __P((struct vm_page *));
    482 
    483 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    484 
    485 vaddr_t
    486 pmap_growkernel(maxkvaddr)
    487 	vaddr_t maxkvaddr;
    488 {
    489 	int s;
    490 	int seg;
    491 	paddr_t pg;
    492 	struct pmap *pm = pmap_kernel();
    493 
    494 	s = splvm();
    495 
    496 	/* Align with the start of a page table */
    497 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    498 	     kbreak += PTMAP) {
    499 		seg = STIDX(kbreak);
    500 
    501 		if (pte_find(pm, kbreak)) continue;
    502 
    503 		if (uvm.page_init_done) {
    504 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    505 		} else {
    506 			if (!uvm_page_physget(&pg))
    507 				panic("pmap_growkernel: no memory");
    508 		}
    509 		if (!pg) panic("pmap_growkernel: no pages");
    510 		pmap_zero_page((paddr_t)pg);
    511 
    512 		/* XXX This is based on all phymem being addressable */
    513 		pm->pm_ptbl[seg] = (u_int *)pg;
    514 	}
    515 	splx(s);
    516 	return (kbreak);
    517 }
    518 
    519 /*
    520  *	vm_page_alloc1:
    521  *
    522  *	Allocate and return a memory cell with no associated object.
    523  */
    524 struct vm_page *
    525 vm_page_alloc1()
    526 {
    527 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    528 	if (pg) {
    529 		pg->wire_count = 1;	/* no mappings yet */
    530 		pg->flags &= ~PG_BUSY;	/* never busy */
    531 	}
    532 	return pg;
    533 }
    534 
    535 /*
    536  *	vm_page_free1:
    537  *
    538  *	Returns the given page to the free list,
    539  *	disassociating it with any VM object.
    540  *
    541  *	Object and page must be locked prior to entry.
    542  */
    543 void
    544 vm_page_free1(mem)
    545 	struct vm_page *mem;
    546 {
    547 #ifdef DIAGNOSTIC
    548 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    549 		printf("Freeing invalid page %p\n", mem);
    550 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    551 #ifdef DDB
    552 		Debugger();
    553 #endif
    554 		return;
    555 	}
    556 #endif
    557 	mem->flags |= PG_BUSY;
    558 	mem->wire_count = 0;
    559 	uvm_pagefree(mem);
    560 }
    561 #endif
    562 
    563 /*
    564  * Create and return a physical map.
    565  */
    566 struct pmap *
    567 pmap_create(void)
    568 {
    569 	struct pmap *pm;
    570 
    571 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    572 	memset((caddr_t)pm, 0, sizeof *pm);
    573 	pmap_pinit(pm);
    574 	return pm;
    575 }
    576 
    577 /*
    578  * Initialize a preallocated and zeroed pmap structure.
    579  */
    580 void
    581 pmap_pinit(struct pmap *pm)
    582 {
    583 	int i;
    584 
    585 	/*
    586 	 * Allocate some segment registers for this pmap.
    587 	 */
    588 	pm->pm_refs = 1;
    589 	for (i = 0; i < STSZ; i++)
    590 		pm->pm_ptbl[i] = NULL;
    591 }
    592 
    593 /*
    594  * Add a reference to the given pmap.
    595  */
    596 void
    597 pmap_reference(struct pmap *pm)
    598 {
    599 
    600 	pm->pm_refs++;
    601 }
    602 
    603 /*
    604  * Retire the given pmap from service.
    605  * Should only be called if the map contains no valid mappings.
    606  */
    607 void
    608 pmap_destroy(struct pmap *pm)
    609 {
    610 
    611 	if (--pm->pm_refs == 0) {
    612 		pmap_release(pm);
    613 		free((caddr_t)pm, M_VMPMAP);
    614 	}
    615 }
    616 
    617 /*
    618  * Release any resources held by the given physical map.
    619  * Called when a pmap initialized by pmap_pinit is being released.
    620  */
    621 static void
    622 pmap_release(struct pmap *pm)
    623 {
    624 	int i;
    625 
    626 	for (i = 0; i < STSZ; i++)
    627 		if (pm->pm_ptbl[i]) {
    628 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
    629 			    PAGE_SIZE);
    630 			pm->pm_ptbl[i] = NULL;
    631 		}
    632 	if (pm->pm_ctx) ctx_free(pm);
    633 }
    634 
    635 /*
    636  * Copy the range specified by src_addr/len
    637  * from the source map to the range dst_addr/len
    638  * in the destination map.
    639  *
    640  * This routine is only advisory and need not do anything.
    641  */
    642 void
    643 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    644 	  vsize_t len, vaddr_t src_addr)
    645 {
    646 }
    647 
    648 /*
    649  * Require that all active physical maps contain no
    650  * incorrect entries NOW.
    651  */
    652 void
    653 pmap_update(struct pmap *pmap)
    654 {
    655 }
    656 
    657 /*
    658  * Garbage collects the physical map system for
    659  * pages which are no longer used.
    660  * Success need not be guaranteed -- that is, there
    661  * may well be pages which are not referenced, but
    662  * others may be collected.
    663  * Called by the pageout daemon when pages are scarce.
    664  */
    665 void
    666 pmap_collect(struct pmap *pm)
    667 {
    668 }
    669 
    670 /*
    671  * Fill the given physical page with zeroes.
    672  */
    673 void
    674 pmap_zero_page(paddr_t pa)
    675 {
    676 
    677 #ifdef PPC_4XX_NOCACHE
    678 	memset((caddr_t)pa, 0, PAGE_SIZE);
    679 #else
    680 	int i;
    681 
    682 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
    683 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    684 		pa += CACHELINESIZE;
    685 	}
    686 #endif
    687 }
    688 
    689 /*
    690  * Copy the given physical source page to its destination.
    691  */
    692 void
    693 pmap_copy_page(paddr_t src, paddr_t dst)
    694 {
    695 
    696 	memcpy((caddr_t)dst, (caddr_t)src, PAGE_SIZE);
    697 	dcache_flush_page(dst);
    698 }
    699 
    700 /*
    701  * This returns whether this is the first mapping of a page.
    702  */
    703 static inline int
    704 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    705 {
    706 	struct pv_entry *pv, *npv = NULL;
    707 	int s;
    708 
    709 	if (!pmap_initialized)
    710 		return 0;
    711 
    712 	s = splvm();
    713 
    714 	pv = pa_to_pv(pa);
    715 for (npv = pv; npv; npv = npv->pv_next)
    716 if (npv->pv_va == va && npv->pv_pm == pm) {
    717 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    718 #ifdef DDB
    719 Debugger();
    720 #endif
    721 return (1);
    722 }
    723 
    724 	if (!pv->pv_pm) {
    725 		/*
    726 		 * No entries yet, use header as the first entry.
    727 		 */
    728 		pv->pv_va = va;
    729 		pv->pv_pm = pm;
    730 		pv->pv_next = NULL;
    731 	} else {
    732 		/*
    733 		 * There is at least one other VA mapping this page.
    734 		 * Place this entry after the header.
    735 		 */
    736 		npv = pool_get(&pv_pool, PR_WAITOK);
    737 		if (!npv) return (0);
    738 		npv->pv_va = va;
    739 		npv->pv_pm = pm;
    740 		npv->pv_next = pv->pv_next;
    741 		pv->pv_next = npv;
    742 	}
    743 	splx(s);
    744 	return (1);
    745 }
    746 
    747 static void
    748 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    749 {
    750 	struct pv_entry *pv, *npv;
    751 
    752 	/*
    753 	 * Remove from the PV table.
    754 	 */
    755 	pv = pa_to_pv(pa);
    756 	if (!pv) return;
    757 
    758 	/*
    759 	 * If it is the first entry on the list, it is actually
    760 	 * in the header and we must copy the following entry up
    761 	 * to the header.  Otherwise we must search the list for
    762 	 * the entry.  In either case we free the now unused entry.
    763 	 */
    764 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    765 		if ((npv = pv->pv_next)) {
    766 			*pv = *npv;
    767 			pool_put(&pv_pool, npv);
    768 		} else
    769 			pv->pv_pm = NULL;
    770 	} else {
    771 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    772 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    773 				break;
    774 		if (npv) {
    775 			pv->pv_next = npv->pv_next;
    776 			pool_put(&pv_pool, npv);
    777 		}
    778 	}
    779 }
    780 
    781 /*
    782  * Insert physical page at pa into the given pmap at virtual address va.
    783  */
    784 int
    785 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    786 {
    787 	int s;
    788 	u_int tte;
    789 	int managed;
    790 
    791 	/*
    792 	 * Have to remove any existing mapping first.
    793 	 */
    794 	pmap_remove(pm, va, va + PAGE_SIZE);
    795 
    796 	if (flags & PMAP_WIRED) flags |= prot;
    797 
    798 	/* If it has no protections don't bother w/the rest */
    799 	if (!(flags & VM_PROT_ALL))
    800 		return (0);
    801 
    802 	managed = 0;
    803 	if (vm_physseg_find(atop(pa), NULL) != -1)
    804 		managed = 1;
    805 
    806 	/*
    807 	 * Generate TTE.
    808 	 *
    809 	 * XXXX
    810 	 *
    811 	 * Since the kernel does not handle execution privileges properly,
    812 	 * we will handle read and execute permissions together.
    813 	 */
    814 	tte = TTE_PA(pa) | TTE_EX;
    815 	/* XXXX -- need to support multiple page sizes. */
    816 	tte |= TTE_SZ_16K;
    817 #ifdef	DIAGNOSTIC
    818 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    819 		(PME_NOCACHE | PME_WRITETHROUG))
    820 		panic("pmap_enter: uncached & writethrough");
    821 #endif
    822 	if (flags & PME_NOCACHE)
    823 		/* Must be I/O mapping */
    824 		tte |= TTE_I | TTE_G;
    825 #ifdef PPC_4XX_NOCACHE
    826 	tte |= TTE_I;
    827 #else
    828 	else if (flags & PME_WRITETHROUG)
    829 		/* Uncached and writethrough are not compatible */
    830 		tte |= TTE_W;
    831 #endif
    832 	if (pm == pmap_kernel())
    833 		tte |= TTE_ZONE(ZONE_PRIV);
    834 	else
    835 		tte |= TTE_ZONE(ZONE_USER);
    836 
    837 	if (flags & VM_PROT_WRITE)
    838 		tte |= TTE_WR;
    839 
    840 	/*
    841 	 * Now record mapping for later back-translation.
    842 	 */
    843 	if (pmap_initialized && managed) {
    844 		char *attr;
    845 
    846 		if (!pmap_enter_pv(pm, va, pa)) {
    847 			/* Could not enter pv on a managed page */
    848 			return 1;
    849 		}
    850 
    851 		/* Now set attributes. */
    852 		attr = pa_to_attr(pa);
    853 #ifdef DIAGNOSTIC
    854 		if (!attr)
    855 			panic("managed but no attr");
    856 #endif
    857 		if (flags & VM_PROT_ALL)
    858 			*attr |= PTE_HI_REF;
    859 		if (flags & VM_PROT_WRITE)
    860 			*attr |= PTE_HI_CHG;
    861 	}
    862 
    863 	s = splvm();
    864 	pm->pm_stats.resident_count++;
    865 
    866 	/* Insert page into page table. */
    867 	pte_enter(pm, va, tte);
    868 
    869 	/* If this is a real fault, enter it in the tlb */
    870 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    871 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    872 	}
    873 	splx(s);
    874 
    875 	/* Flush the real memory from the instruction cache. */
    876 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    877 		__syncicache((void *)pa, PAGE_SIZE);
    878 
    879 	return 0;
    880 }
    881 
    882 void
    883 pmap_unwire(struct pmap *pm, vaddr_t va)
    884 {
    885 	struct pv_entry *pv, *npv;
    886 	paddr_t pa;
    887 	int s = splvm();
    888 
    889 	if (pm == NULL) {
    890 		return;
    891 	}
    892 
    893 	if (!pmap_extract(pm, va, &pa)) {
    894 		return;
    895 	}
    896 
    897 	va |= PV_WIRED;
    898 
    899 	pv = pa_to_pv(pa);
    900 	if (!pv) return;
    901 
    902 	/*
    903 	 * If it is the first entry on the list, it is actually
    904 	 * in the header and we must copy the following entry up
    905 	 * to the header.  Otherwise we must search the list for
    906 	 * the entry.  In either case we free the now unused entry.
    907 	 */
    908 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    909 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    910 			npv->pv_va &= ~PV_WIRED;
    911 			break;
    912 		}
    913 	}
    914 	splx(s);
    915 }
    916 
    917 void
    918 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    919 {
    920 	int s;
    921 	u_int tte;
    922 	struct pmap *pm = pmap_kernel();
    923 
    924 	/*
    925 	 * Have to remove any existing mapping first.
    926 	 */
    927 
    928 	/*
    929 	 * Generate TTE.
    930 	 *
    931 	 * XXXX
    932 	 *
    933 	 * Since the kernel does not handle execution privileges properly,
    934 	 * we will handle read and execute permissions together.
    935 	 */
    936 	tte = 0;
    937 	if (prot & VM_PROT_ALL) {
    938 
    939 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    940 		/* XXXX -- need to support multiple page sizes. */
    941 		tte |= TTE_SZ_16K;
    942 #ifdef DIAGNOSTIC
    943 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    944 			(PME_NOCACHE | PME_WRITETHROUG))
    945 			panic("pmap_kenter_pa: uncached & writethrough");
    946 #endif
    947 		if (prot & PME_NOCACHE)
    948 			/* Must be I/O mapping */
    949 			tte |= TTE_I | TTE_G;
    950 #ifdef PPC_4XX_NOCACHE
    951 		tte |= TTE_I;
    952 #else
    953 		else if (prot & PME_WRITETHROUG)
    954 			/* Uncached and writethrough are not compatible */
    955 			tte |= TTE_W;
    956 #endif
    957 		if (prot & VM_PROT_WRITE)
    958 			tte |= TTE_WR;
    959 	}
    960 
    961 	s = splvm();
    962 	pm->pm_stats.resident_count++;
    963 
    964 	/* Insert page into page table. */
    965 	pte_enter(pm, va, tte);
    966 	splx(s);
    967 }
    968 
    969 void
    970 pmap_kremove(vaddr_t va, vsize_t len)
    971 {
    972 
    973 	while (len > 0) {
    974 		pte_enter(pmap_kernel(), va, 0);
    975 		va += PAGE_SIZE;
    976 		len -= PAGE_SIZE;
    977 	}
    978 }
    979 
    980 /*
    981  * Remove the given range of mapping entries.
    982  */
    983 void
    984 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
    985 {
    986 	int s;
    987 	paddr_t pa;
    988 	volatile u_int *ptp;
    989 
    990 	s = splvm();
    991 	while (va < endva) {
    992 
    993 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
    994 			pa = TTE_PA(pa);
    995 			pmap_remove_pv(pm, va, pa);
    996 			*ptp = 0;
    997 			ppc4xx_tlb_flush(va, pm->pm_ctx);
    998 			pm->pm_stats.resident_count--;
    999 		}
   1000 		va += PAGE_SIZE;
   1001 	}
   1002 
   1003 	splx(s);
   1004 }
   1005 
   1006 /*
   1007  * Get the physical page address for the given pmap/virtual address.
   1008  */
   1009 boolean_t
   1010 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1011 {
   1012 	int seg = STIDX(va);
   1013 	int ptn = PTIDX(va);
   1014 	u_int pa = 0;
   1015 	int s = splvm();
   1016 
   1017 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1018 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1019 	}
   1020 	splx(s);
   1021 	return (pa != 0);
   1022 }
   1023 
   1024 /*
   1025  * Lower the protection on the specified range of this pmap.
   1026  *
   1027  * There are only two cases: either the protection is going to 0,
   1028  * or it is going to read-only.
   1029  */
   1030 void
   1031 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1032 {
   1033 	volatile u_int *ptp;
   1034 	int s;
   1035 
   1036 	if (prot & VM_PROT_READ) {
   1037 		s = splvm();
   1038 		while (sva < eva) {
   1039 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1040 				*ptp &= ~TTE_WR;
   1041 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1042 			}
   1043 			sva += PAGE_SIZE;
   1044 		}
   1045 		splx(s);
   1046 		return;
   1047 	}
   1048 	pmap_remove(pm, sva, eva);
   1049 }
   1050 
   1051 boolean_t
   1052 check_attr(struct vm_page *pg, u_int mask, int clear)
   1053 {
   1054 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1055 	int s;
   1056 	char *attr;
   1057 	int rv;
   1058 
   1059 	/*
   1060 	 * First modify bits in cache.
   1061 	 */
   1062 	s = splvm();
   1063 	attr = pa_to_attr(pa);
   1064 	if (attr == NULL)
   1065 		return FALSE;
   1066 
   1067 	rv = ((*attr & mask) != 0);
   1068 	if (clear) {
   1069 		*attr &= ~mask;
   1070 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1071 	}
   1072 	splx(s);
   1073 	return rv;
   1074 }
   1075 
   1076 
   1077 /*
   1078  * Lower the protection on the specified physical page.
   1079  *
   1080  * There are only two cases: either the protection is going to 0,
   1081  * or it is going to read-only.
   1082  */
   1083 void
   1084 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1085 {
   1086 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1087 	vaddr_t va;
   1088 	struct pv_entry *pvh, *pv, *npv;
   1089 	struct pmap *pm;
   1090 
   1091 	pvh = pa_to_pv(pa);
   1092 	if (pvh == NULL)
   1093 		return;
   1094 
   1095 	/* Handle extra pvs which may be deleted in the operation */
   1096 	for (pv = pvh->pv_next; pv; pv = npv) {
   1097 		npv = pv->pv_next;
   1098 
   1099 		pm = pv->pv_pm;
   1100 		va = pv->pv_va;
   1101 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1102 	}
   1103 	/* Now check the head pv */
   1104 	if (pvh->pv_pm) {
   1105 		pv = pvh;
   1106 		pm = pv->pv_pm;
   1107 		va = pv->pv_va;
   1108 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1109 	}
   1110 }
   1111 
   1112 /*
   1113  * Activate the address space for the specified process.  If the process
   1114  * is the current process, load the new MMU context.
   1115  */
   1116 void
   1117 pmap_activate(struct lwp *l)
   1118 {
   1119 #if 0
   1120 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1121 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1122 
   1123 	/*
   1124 	 * XXX Normally performed in cpu_fork().
   1125 	 */
   1126 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1127 	if (pcb->pcb_pm != pmap) {
   1128 		pcb->pcb_pm = pmap;
   1129 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1130 		    (paddr_t *)&pcb->pcb_pmreal);
   1131 	}
   1132 
   1133 	if (l == curlwp) {
   1134 		/* Store pointer to new current pmap. */
   1135 		curpm = pcb->pcb_pmreal;
   1136 	}
   1137 #endif
   1138 }
   1139 
   1140 /*
   1141  * Deactivate the specified process's address space.
   1142  */
   1143 void
   1144 pmap_deactivate(struct lwp *l)
   1145 {
   1146 }
   1147 
   1148 /*
   1149  * Synchronize caches corresponding to [addr, addr+len) in p.
   1150  */
   1151 void
   1152 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1153 {
   1154 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1155 	int msr, ctx, opid, step;
   1156 
   1157 
   1158 	step = CACHELINESIZE;
   1159 
   1160 	/*
   1161 	 * Need to turn off IMMU and switch to user context.
   1162 	 * (icbi uses DMMU).
   1163 	 */
   1164 	if (!(ctx = pm->pm_ctx)) {
   1165 		/* No context -- assign it one */
   1166 		ctx_alloc(pm);
   1167 		ctx = pm->pm_ctx;
   1168 	}
   1169 	__asm __volatile("mfmsr %0;"
   1170 		"li %1, 0x20;"
   1171 		"andc %1,%0,%1;"
   1172 		"mtmsr %1;"
   1173 		"sync;isync;"
   1174 		"mfpid %1;"
   1175 		"mtpid %2;"
   1176 		"sync; isync;"
   1177 		"1:"
   1178 		"dcbf 0,%3;"
   1179 		"icbi 0,%3;"
   1180 		"add %3,%3,%5;"
   1181 		"addc. %4,%4,%6;"
   1182 		"bge 1b;"
   1183 		"mtpid %1;"
   1184 		"mtmsr %0;"
   1185 		"sync; isync"
   1186 		: "=&r" (msr), "=&r" (opid)
   1187 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step));
   1188 }
   1189 
   1190 
   1191 /* This has to be done in real mode !!! */
   1192 void
   1193 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1194 {
   1195 	u_long i, found;
   1196 	u_long msr;
   1197 
   1198 	/* If there's no context then it can't be mapped. */
   1199 	if (!pid) return;
   1200 
   1201 	asm("mfpid %1;"			/* Save PID */
   1202 		"mfmsr %2;"		/* Save MSR */
   1203 		"li %0,0;"		/* Now clear MSR */
   1204 		"mtmsr %0;"
   1205 		"mtpid %4;"		/* Set PID */
   1206 		"sync;"
   1207 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1208 		"sync;"
   1209 		"mtpid %1;"		/* Restore PID */
   1210 		"mtmsr %2;"		/* Restore MSR */
   1211 		"sync;isync;"
   1212 		"li %1,1;"
   1213 		"beq 1f;"
   1214 		"li %1,0;"
   1215 		"1:"
   1216 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1217 		: "r" (va), "r" (pid));
   1218 	if (found && !TLB_LOCKED(i)) {
   1219 
   1220 		/* Now flush translation */
   1221 		asm volatile(
   1222 			"tlbwe %0,%1,0;"
   1223 			"sync;isync;"
   1224 			: : "r" (0), "r" (i));
   1225 
   1226 		tlb_info[i].ti_ctx = 0;
   1227 		tlb_info[i].ti_flags = 0;
   1228 		tlbnext = i;
   1229 		/* Successful flushes */
   1230 		tlbflush_ev.ev_count++;
   1231 	}
   1232 }
   1233 
   1234 void
   1235 ppc4xx_tlb_flush_all(void)
   1236 {
   1237 	u_long i;
   1238 
   1239 	for (i = 0; i < NTLB; i++)
   1240 		if (!TLB_LOCKED(i)) {
   1241 			asm volatile(
   1242 				"tlbwe %0,%1,0;"
   1243 				"sync;isync;"
   1244 				: : "r" (0), "r" (i));
   1245 			tlb_info[i].ti_ctx = 0;
   1246 			tlb_info[i].ti_flags = 0;
   1247 		}
   1248 
   1249 	asm volatile("sync;isync");
   1250 }
   1251 
   1252 /* Find a TLB entry to evict. */
   1253 static int
   1254 ppc4xx_tlb_find_victim(void)
   1255 {
   1256 	int flags;
   1257 
   1258 	for (;;) {
   1259 		if (++tlbnext >= NTLB)
   1260 			tlbnext = TLB_NRESERVED;
   1261 		flags = tlb_info[tlbnext].ti_flags;
   1262 		if (!(flags & TLBF_USED) ||
   1263 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1264 			u_long va, stack = (u_long)&va;
   1265 
   1266 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1267 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1268 			     (flags & TLBF_USED)) {
   1269 				/* Kernel stack page */
   1270 				flags |= TLBF_USED;
   1271 				tlb_info[tlbnext].ti_flags = flags;
   1272 			} else {
   1273 				/* Found it! */
   1274 				return (tlbnext);
   1275 			}
   1276 		} else {
   1277 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1278 		}
   1279 	}
   1280 }
   1281 
   1282 void
   1283 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1284 {
   1285 	u_long th, tl, idx;
   1286 	tlbpid_t pid;
   1287 	u_short msr;
   1288 	paddr_t pa;
   1289 	int s, sz;
   1290 
   1291 	tlbenter_ev.ev_count++;
   1292 
   1293 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1294 	pa = (pte & TTE_RPN_MASK(sz));
   1295 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1296 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1297 	tl |= ppc4xx_tlbflags(va, pa);
   1298 
   1299 	s = splhigh();
   1300 	idx = ppc4xx_tlb_find_victim();
   1301 
   1302 #ifdef DIAGNOSTIC
   1303 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1304 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
   1305 	}
   1306 #endif
   1307 
   1308 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1309 	tlb_info[idx].ti_ctx = ctx;
   1310 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1311 
   1312 	asm volatile(
   1313 		"mfmsr %0;"			/* Save MSR */
   1314 		"li %1,0;"
   1315 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1316 		"mtmsr %1;"			/* Clear MSR */
   1317 		"mfpid %1;"			/* Save old PID */
   1318 		"mtpid %2;"			/* Load translation ctx */
   1319 		"sync; isync;"
   1320 #ifdef DEBUG
   1321 		"andi. %3,%3,63;"
   1322 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1323 #endif
   1324 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1325 		"sync; isync;"
   1326 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1327 		"sync; isync;"
   1328 	: "=&r" (msr), "=&r" (pid)
   1329 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1330 	splx(s);
   1331 }
   1332 
   1333 void
   1334 ppc4xx_tlb_unpin(int i)
   1335 {
   1336 
   1337 	if (i == -1)
   1338 		for (i = 0; i < TLB_NRESERVED; i++)
   1339 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1340 	else
   1341 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1342 }
   1343 
   1344 void
   1345 ppc4xx_tlb_init(void)
   1346 {
   1347 	int i;
   1348 
   1349 	/* Mark reserved TLB entries */
   1350 	for (i = 0; i < TLB_NRESERVED; i++) {
   1351 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1352 		tlb_info[i].ti_ctx = KERNEL_PID;
   1353 	}
   1354 
   1355 	/* Setup security zones */
   1356 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1357 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1358 	 * Z3 - full access regardless of TLB entry permissions
   1359 	 */
   1360 
   1361 	asm volatile(
   1362 		"mtspr %0,%1;"
   1363 		"sync;"
   1364 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1365 }
   1366 
   1367 
   1368 /*
   1369  * We should pass the ctx in from trap code.
   1370  */
   1371 int
   1372 pmap_tlbmiss(vaddr_t va, int ctx)
   1373 {
   1374 	volatile u_int *pte;
   1375 	u_long tte;
   1376 
   1377 	tlbmiss_ev.ev_count++;
   1378 
   1379 	/*
   1380 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1381 	 */
   1382 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1383 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1384 		if (pte == NULL) {
   1385 			/* Map unmanaged addresses directly for kernel access */
   1386 			return 1;
   1387 		}
   1388 		tte = *pte;
   1389 		if (tte == 0) {
   1390 			return 1;
   1391 		}
   1392 	} else {
   1393 		/* Create a 16MB writable mapping. */
   1394 #ifdef PPC_4XX_NOCACHE
   1395 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1396 #else
   1397 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1398 #endif
   1399 	}
   1400 	tlbhit_ev.ev_count++;
   1401 	ppc4xx_tlb_enter(ctx, va, tte);
   1402 
   1403 	return 0;
   1404 }
   1405 
   1406 /*
   1407  * Flush all the entries matching a context from the TLB.
   1408  */
   1409 static int
   1410 ctx_flush(int cnum)
   1411 {
   1412 	int i;
   1413 
   1414 	/* We gotta steal this context */
   1415 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1416 		if (tlb_info[i].ti_ctx == cnum) {
   1417 			/* Can't steal ctx if it has a locked entry. */
   1418 			if (TLB_LOCKED(i)) {
   1419 #ifdef DIAGNOSTIC
   1420 				printf("ctx_flush: can't invalidate "
   1421 					"locked mapping %d "
   1422 					"for context %d\n", i, cnum);
   1423 #ifdef DDB
   1424 				Debugger();
   1425 #endif
   1426 #endif
   1427 				return (1);
   1428 			}
   1429 #ifdef DIAGNOSTIC
   1430 			if (i < TLB_NRESERVED)
   1431 				panic("TLB entry %d not locked", i);
   1432 #endif
   1433 			/* Invalidate particular TLB entry regardless of locked status */
   1434 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1435 			tlb_info[i].ti_flags = 0;
   1436 		}
   1437 	}
   1438 	return (0);
   1439 }
   1440 
   1441 /*
   1442  * Allocate a context.  If necessary, steal one from someone else.
   1443  *
   1444  * The new context is flushed from the TLB before returning.
   1445  */
   1446 int
   1447 ctx_alloc(struct pmap *pm)
   1448 {
   1449 	int s, cnum;
   1450 	static int next = MINCTX;
   1451 
   1452 	if (pm == pmap_kernel()) {
   1453 #ifdef DIAGNOSTIC
   1454 		printf("ctx_alloc: kernel pmap!\n");
   1455 #endif
   1456 		return (0);
   1457 	}
   1458 	s = splvm();
   1459 
   1460 	/* Find a likely context. */
   1461 	cnum = next;
   1462 	do {
   1463 		if ((++cnum) > NUMCTX)
   1464 			cnum = MINCTX;
   1465 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1466 
   1467 	/* Now clean it out */
   1468 oops:
   1469 	if (cnum < MINCTX)
   1470 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1471 	if (ctx_flush(cnum)) {
   1472 		/* oops -- something's wired. */
   1473 		if ((++cnum) > NUMCTX)
   1474 			cnum = MINCTX;
   1475 		goto oops;
   1476 	}
   1477 
   1478 	if (ctxbusy[cnum]) {
   1479 #ifdef DEBUG
   1480 		/* We should identify this pmap and clear it */
   1481 		printf("Warning: stealing context %d\n", cnum);
   1482 #endif
   1483 		ctxbusy[cnum]->pm_ctx = 0;
   1484 	}
   1485 	ctxbusy[cnum] = pm;
   1486 	next = cnum;
   1487 	splx(s);
   1488 	pm->pm_ctx = cnum;
   1489 
   1490 	return cnum;
   1491 }
   1492 
   1493 /*
   1494  * Give away a context.
   1495  */
   1496 void
   1497 ctx_free(struct pmap *pm)
   1498 {
   1499 	int oldctx;
   1500 
   1501 	oldctx = pm->pm_ctx;
   1502 
   1503 	if (oldctx == 0)
   1504 		panic("ctx_free: freeing kernel context");
   1505 #ifdef DIAGNOSTIC
   1506 	if (ctxbusy[oldctx] == 0)
   1507 		printf("ctx_free: freeing free context %d\n", oldctx);
   1508 	if (ctxbusy[oldctx] != pm) {
   1509 		printf("ctx_free: freeing someone esle's context\n "
   1510 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1511 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1512 #ifdef DDB
   1513 		Debugger();
   1514 #endif
   1515 	}
   1516 #endif
   1517 	/* We should verify it has not been stolen and reallocated... */
   1518 	ctxbusy[oldctx] = NULL;
   1519 	ctx_flush(oldctx);
   1520 }
   1521 
   1522 
   1523 #ifdef DEBUG
   1524 /*
   1525  * Test ref/modify handling.
   1526  */
   1527 void pmap_testout __P((void));
   1528 void
   1529 pmap_testout()
   1530 {
   1531 	vaddr_t va;
   1532 	volatile int *loc;
   1533 	int val = 0;
   1534 	paddr_t pa;
   1535 	struct vm_page *pg;
   1536 	int ref, mod;
   1537 
   1538 	/* Allocate a page */
   1539 	va = (vaddr_t)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
   1540 	loc = (int*)va;
   1541 
   1542 	pmap_extract(pmap_kernel(), va, &pa);
   1543 	pg = PHYS_TO_VM_PAGE(pa);
   1544 	pmap_unwire(pmap_kernel(), va);
   1545 
   1546 	pmap_remove(pmap_kernel(), va, va+1);
   1547 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1548 	pmap_update(pmap_kernel());
   1549 
   1550 	/* Now clear reference and modify */
   1551 	ref = pmap_clear_reference(pg);
   1552 	mod = pmap_clear_modify(pg);
   1553 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1554 	       (void *)(u_long)va, (long)pa,
   1555 	       ref, mod);
   1556 
   1557 	/* Check it's properly cleared */
   1558 	ref = pmap_is_referenced(pg);
   1559 	mod = pmap_is_modified(pg);
   1560 	printf("Checking cleared page: ref %d, mod %d\n",
   1561 	       ref, mod);
   1562 
   1563 	/* Reference page */
   1564 	val = *loc;
   1565 
   1566 	ref = pmap_is_referenced(pg);
   1567 	mod = pmap_is_modified(pg);
   1568 	printf("Referenced page: ref %d, mod %d val %x\n",
   1569 	       ref, mod, val);
   1570 
   1571 	/* Now clear reference and modify */
   1572 	ref = pmap_clear_reference(pg);
   1573 	mod = pmap_clear_modify(pg);
   1574 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1575 	       (void *)(u_long)va, (long)pa,
   1576 	       ref, mod);
   1577 
   1578 	/* Modify page */
   1579 	*loc = 1;
   1580 
   1581 	ref = pmap_is_referenced(pg);
   1582 	mod = pmap_is_modified(pg);
   1583 	printf("Modified page: ref %d, mod %d\n",
   1584 	       ref, mod);
   1585 
   1586 	/* Now clear reference and modify */
   1587 	ref = pmap_clear_reference(pg);
   1588 	mod = pmap_clear_modify(pg);
   1589 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1590 	       (void *)(u_long)va, (long)pa,
   1591 	       ref, mod);
   1592 
   1593 	/* Check it's properly cleared */
   1594 	ref = pmap_is_referenced(pg);
   1595 	mod = pmap_is_modified(pg);
   1596 	printf("Checking cleared page: ref %d, mod %d\n",
   1597 	       ref, mod);
   1598 
   1599 	/* Modify page */
   1600 	*loc = 1;
   1601 
   1602 	ref = pmap_is_referenced(pg);
   1603 	mod = pmap_is_modified(pg);
   1604 	printf("Modified page: ref %d, mod %d\n",
   1605 	       ref, mod);
   1606 
   1607 	/* Check pmap_protect() */
   1608 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1609 	pmap_update(pmap_kernel());
   1610 	ref = pmap_is_referenced(pg);
   1611 	mod = pmap_is_modified(pg);
   1612 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1613 	       ref, mod);
   1614 
   1615 	/* Now clear reference and modify */
   1616 	ref = pmap_clear_reference(pg);
   1617 	mod = pmap_clear_modify(pg);
   1618 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1619 	       (void *)(u_long)va, (long)pa,
   1620 	       ref, mod);
   1621 
   1622 	/* Reference page */
   1623 	val = *loc;
   1624 
   1625 	ref = pmap_is_referenced(pg);
   1626 	mod = pmap_is_modified(pg);
   1627 	printf("Referenced page: ref %d, mod %d val %x\n",
   1628 	       ref, mod, val);
   1629 
   1630 	/* Now clear reference and modify */
   1631 	ref = pmap_clear_reference(pg);
   1632 	mod = pmap_clear_modify(pg);
   1633 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1634 	       (void *)(u_long)va, (long)pa,
   1635 	       ref, mod);
   1636 
   1637 	/* Modify page */
   1638 #if 0
   1639 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1640 	pmap_update(pmap_kernel());
   1641 #endif
   1642 	*loc = 1;
   1643 
   1644 	ref = pmap_is_referenced(pg);
   1645 	mod = pmap_is_modified(pg);
   1646 	printf("Modified page: ref %d, mod %d\n",
   1647 	       ref, mod);
   1648 
   1649 	/* Check pmap_protect() */
   1650 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1651 	pmap_update(pmap_kernel());
   1652 	ref = pmap_is_referenced(pg);
   1653 	mod = pmap_is_modified(pg);
   1654 	printf("pmap_protect(): ref %d, mod %d\n",
   1655 	       ref, mod);
   1656 
   1657 	/* Now clear reference and modify */
   1658 	ref = pmap_clear_reference(pg);
   1659 	mod = pmap_clear_modify(pg);
   1660 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1661 	       (void *)(u_long)va, (long)pa,
   1662 	       ref, mod);
   1663 
   1664 	/* Reference page */
   1665 	val = *loc;
   1666 
   1667 	ref = pmap_is_referenced(pg);
   1668 	mod = pmap_is_modified(pg);
   1669 	printf("Referenced page: ref %d, mod %d val %x\n",
   1670 	       ref, mod, val);
   1671 
   1672 	/* Now clear reference and modify */
   1673 	ref = pmap_clear_reference(pg);
   1674 	mod = pmap_clear_modify(pg);
   1675 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1676 	       (void *)(u_long)va, (long)pa,
   1677 	       ref, mod);
   1678 
   1679 	/* Modify page */
   1680 #if 0
   1681 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1682 	pmap_update(pmap_kernel());
   1683 #endif
   1684 	*loc = 1;
   1685 
   1686 	ref = pmap_is_referenced(pg);
   1687 	mod = pmap_is_modified(pg);
   1688 	printf("Modified page: ref %d, mod %d\n",
   1689 	       ref, mod);
   1690 
   1691 	/* Check pmap_pag_protect() */
   1692 	pmap_page_protect(pg, VM_PROT_READ);
   1693 	ref = pmap_is_referenced(pg);
   1694 	mod = pmap_is_modified(pg);
   1695 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1696 	       ref, mod);
   1697 
   1698 	/* Now clear reference and modify */
   1699 	ref = pmap_clear_reference(pg);
   1700 	mod = pmap_clear_modify(pg);
   1701 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1702 	       (void *)(u_long)va, (long)pa,
   1703 	       ref, mod);
   1704 
   1705 	/* Reference page */
   1706 	val = *loc;
   1707 
   1708 	ref = pmap_is_referenced(pg);
   1709 	mod = pmap_is_modified(pg);
   1710 	printf("Referenced page: ref %d, mod %d val %x\n",
   1711 	       ref, mod, val);
   1712 
   1713 	/* Now clear reference and modify */
   1714 	ref = pmap_clear_reference(pg);
   1715 	mod = pmap_clear_modify(pg);
   1716 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1717 	       (void *)(u_long)va, (long)pa,
   1718 	       ref, mod);
   1719 
   1720 	/* Modify page */
   1721 #if 0
   1722 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1723 	pmap_update(pmap_kernel());
   1724 #endif
   1725 	*loc = 1;
   1726 
   1727 	ref = pmap_is_referenced(pg);
   1728 	mod = pmap_is_modified(pg);
   1729 	printf("Modified page: ref %d, mod %d\n",
   1730 	       ref, mod);
   1731 
   1732 	/* Check pmap_pag_protect() */
   1733 	pmap_page_protect(pg, VM_PROT_NONE);
   1734 	ref = pmap_is_referenced(pg);
   1735 	mod = pmap_is_modified(pg);
   1736 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1737 	       ref, mod);
   1738 
   1739 	/* Now clear reference and modify */
   1740 	ref = pmap_clear_reference(pg);
   1741 	mod = pmap_clear_modify(pg);
   1742 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1743 	       (void *)(u_long)va, (long)pa,
   1744 	       ref, mod);
   1745 
   1746 
   1747 	/* Reference page */
   1748 	val = *loc;
   1749 
   1750 	ref = pmap_is_referenced(pg);
   1751 	mod = pmap_is_modified(pg);
   1752 	printf("Referenced page: ref %d, mod %d val %x\n",
   1753 	       ref, mod, val);
   1754 
   1755 	/* Now clear reference and modify */
   1756 	ref = pmap_clear_reference(pg);
   1757 	mod = pmap_clear_modify(pg);
   1758 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1759 	       (void *)(u_long)va, (long)pa,
   1760 	       ref, mod);
   1761 
   1762 	/* Modify page */
   1763 #if 0
   1764 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1765 	pmap_update(pmap_kernel());
   1766 #endif
   1767 	*loc = 1;
   1768 
   1769 	ref = pmap_is_referenced(pg);
   1770 	mod = pmap_is_modified(pg);
   1771 	printf("Modified page: ref %d, mod %d\n",
   1772 	       ref, mod);
   1773 
   1774 	/* Unmap page */
   1775 	pmap_remove(pmap_kernel(), va, va+1);
   1776 	pmap_update(pmap_kernel());
   1777 	ref = pmap_is_referenced(pg);
   1778 	mod = pmap_is_modified(pg);
   1779 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1780 
   1781 	/* Now clear reference and modify */
   1782 	ref = pmap_clear_reference(pg);
   1783 	mod = pmap_clear_modify(pg);
   1784 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1785 	       (void *)(u_long)va, (long)pa, ref, mod);
   1786 
   1787 	/* Check it's properly cleared */
   1788 	ref = pmap_is_referenced(pg);
   1789 	mod = pmap_is_modified(pg);
   1790 	printf("Checking cleared page: ref %d, mod %d\n",
   1791 	       ref, mod);
   1792 
   1793 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1794 		VM_PROT_ALL|PMAP_WIRED);
   1795 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE);
   1796 }
   1797 #endif
   1798