Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.1.2.5
      1 /*	$NetBSD: pmap.c,v 1.1.2.5 2002/06/23 17:39:39 jdolecek 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 #undef PPC_4XX_NOCACHE
     70 
     71 #include <sys/param.h>
     72 #include <sys/malloc.h>
     73 #include <sys/proc.h>
     74 #include <sys/user.h>
     75 #include <sys/queue.h>
     76 #include <sys/systm.h>
     77 #include <sys/pool.h>
     78 #include <sys/device.h>
     79 
     80 #include <uvm/uvm.h>
     81 
     82 #include <machine/cpu.h>
     83 #include <machine/pcb.h>
     84 #include <machine/powerpc.h>
     85 
     86 #include <powerpc/spr.h>
     87 #include <machine/tlb.h>
     88 
     89 /*
     90  * kernmap is an array of PTEs large enough to map in
     91  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     92  */
     93 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
     94 caddr_t kernmap;
     95 
     96 #define MINCTX		2
     97 #define NUMCTX		256
     98 volatile struct pmap *ctxbusy[NUMCTX];
     99 
    100 #define TLBF_USED	0x1
    101 #define	TLBF_REF	0x2
    102 #define	TLBF_LOCKED	0x4
    103 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    104 typedef struct tlb_info_s {
    105 	char	ti_flags;
    106 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    107 	u_int	ti_va;
    108 } tlb_info_t;
    109 
    110 volatile tlb_info_t tlb_info[NTLB];
    111 /* We'll use a modified FIFO replacement policy cause it's cheap */
    112 volatile int tlbnext = TLB_NRESERVED;
    113 
    114 u_long dtlb_miss_count = 0;
    115 u_long itlb_miss_count = 0;
    116 u_long ktlb_miss_count = 0;
    117 u_long utlb_miss_count = 0;
    118 
    119 /* Event counters -- XXX type `INTR' so we can see them with vmstat -i */
    120 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    121 	NULL, "cpu", "tlbmiss");
    122 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    123 	NULL, "cpu", "tlbhit");
    124 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    125 	NULL, "cpu", "tlbflush");
    126 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    127 	NULL, "cpu", "tlbenter");
    128 
    129 struct pmap kernel_pmap_;
    130 
    131 int physmem;
    132 static int npgs;
    133 static u_int nextavail;
    134 #ifndef MSGBUFADDR
    135 extern paddr_t msgbuf_paddr;
    136 #endif
    137 
    138 static struct mem_region *mem, *avail;
    139 
    140 /*
    141  * This is a cache of referenced/modified bits.
    142  * Bits herein are shifted by ATTRSHFT.
    143  */
    144 static char *pmap_attrib;
    145 
    146 #define PV_WIRED	0x1
    147 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    148 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    149 
    150 struct pv_entry {
    151 	struct pv_entry *pv_next;	/* Linked list of mappings */
    152 	vaddr_t pv_va;			/* virtual address of mapping */
    153 	struct pmap *pv_pm;
    154 };
    155 
    156 struct pv_entry *pv_table;
    157 static struct pool pv_pool;
    158 
    159 static int pmap_initialized;
    160 
    161 static int ctx_flush(int);
    162 
    163 inline struct pv_entry *pa_to_pv(paddr_t);
    164 static inline char *pa_to_attr(paddr_t);
    165 
    166 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    167 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    168 
    169 static void pmap_pinit(pmap_t);
    170 static void pmap_release(pmap_t);
    171 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    172 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    173 
    174 
    175 inline struct pv_entry *
    176 pa_to_pv(paddr_t pa)
    177 {
    178 	int bank, pg;
    179 
    180 	bank = vm_physseg_find(atop(pa), &pg);
    181 	if (bank == -1)
    182 		return NULL;
    183 	return &vm_physmem[bank].pmseg.pvent[pg];
    184 }
    185 
    186 static inline char *
    187 pa_to_attr(paddr_t pa)
    188 {
    189 	int bank, pg;
    190 
    191 	bank = vm_physseg_find(atop(pa), &pg);
    192 	if (bank == -1)
    193 		return NULL;
    194 	return &vm_physmem[bank].pmseg.attrs[pg];
    195 }
    196 
    197 /*
    198  * Insert PTE into page table.
    199  */
    200 int
    201 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    202 {
    203 	int seg = STIDX(va);
    204 	int ptn = PTIDX(va);
    205 	paddr_t pa;
    206 
    207 	if (!pm->pm_ptbl[seg]) {
    208 		/* Don't allocate a page to clear a non-existent mapping. */
    209 		if (!pte) return (1);
    210 		/* Allocate a page XXXX this will sleep! */
    211 		pa = 0;
    212 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
    213 	}
    214 	pm->pm_ptbl[seg][ptn] = pte;
    215 
    216 	/* Flush entry. */
    217 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    218 	return (1);
    219 }
    220 
    221 /*
    222  * Get a pointer to a PTE in a page table.
    223  */
    224 volatile u_int *
    225 pte_find(struct pmap *pm, vaddr_t va)
    226 {
    227 	int seg = STIDX(va);
    228 	int ptn = PTIDX(va);
    229 
    230 	if (pm->pm_ptbl[seg])
    231 		return (&pm->pm_ptbl[seg][ptn]);
    232 
    233 	return (NULL);
    234 }
    235 
    236 /*
    237  * This is called during initppc, before the system is really initialized.
    238  */
    239 void
    240 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    241 {
    242 	struct mem_region *mp, *mp1;
    243 	int cnt, i;
    244 	u_int s, e, sz;
    245 
    246 	/*
    247 	 * Allocate the kernel page table at the end of
    248 	 * kernel space so it's in the locked TTE.
    249 	 */
    250 	kernmap = (caddr_t)kernelend;
    251 
    252 	/*
    253 	 * Initialize kernel page table.
    254 	 */
    255 	for (i = 0; i < STSZ; i++) {
    256 		pmap_kernel()->pm_ptbl[i] = 0;
    257 	}
    258 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    259 
    260 	/*
    261 	 * Announce page-size to the VM-system
    262 	 */
    263 	uvmexp.pagesize = NBPG;
    264 	uvm_setpagesize();
    265 
    266 	/*
    267 	 * Get memory.
    268 	 */
    269 	mem_regions(&mem, &avail);
    270 	for (mp = mem; mp->size; mp++) {
    271 		physmem += btoc(mp->size);
    272 		printf("+%lx,",mp->size);
    273 	}
    274 	printf("\n");
    275 	ppc4xx_tlb_init();
    276 	/*
    277 	 * Count the number of available entries.
    278 	 */
    279 	for (cnt = 0, mp = avail; mp->size; mp++)
    280 		cnt++;
    281 
    282 	/*
    283 	 * Page align all regions.
    284 	 * Non-page aligned memory isn't very interesting to us.
    285 	 * Also, sort the entries for ascending addresses.
    286 	 */
    287 	kernelstart &= ~PGOFSET;
    288 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    289 	for (mp = avail; mp->size; mp++) {
    290 		s = mp->start;
    291 		e = mp->start + mp->size;
    292 		printf("%08x-%08x -> ",s,e);
    293 		/*
    294 		 * Check whether this region holds all of the kernel.
    295 		 */
    296 		if (s < kernelstart && e > kernelend) {
    297 			avail[cnt].start = kernelend;
    298 			avail[cnt++].size = e - kernelend;
    299 			e = kernelstart;
    300 		}
    301 		/*
    302 		 * Look whether this regions starts within the kernel.
    303 		 */
    304 		if (s >= kernelstart && s < kernelend) {
    305 			if (e <= kernelend)
    306 				goto empty;
    307 			s = kernelend;
    308 		}
    309 		/*
    310 		 * Now look whether this region ends within the kernel.
    311 		 */
    312 		if (e > kernelstart && e <= kernelend) {
    313 			if (s >= kernelstart)
    314 				goto empty;
    315 			e = kernelstart;
    316 		}
    317 		/*
    318 		 * Now page align the start and size of the region.
    319 		 */
    320 		s = round_page(s);
    321 		e = trunc_page(e);
    322 		if (e < s)
    323 			e = s;
    324 		sz = e - s;
    325 		printf("%08x-%08x = %x\n",s,e,sz);
    326 		/*
    327 		 * Check whether some memory is left here.
    328 		 */
    329 		if (sz == 0) {
    330 		empty:
    331 			memmove(mp, mp + 1,
    332 				(cnt - (mp - avail)) * sizeof *mp);
    333 			cnt--;
    334 			mp--;
    335 			continue;
    336 		}
    337 		/*
    338 		 * Do an insertion sort.
    339 		 */
    340 		npgs += btoc(sz);
    341 		for (mp1 = avail; mp1 < mp; mp1++)
    342 			if (s < mp1->start)
    343 				break;
    344 		if (mp1 < mp) {
    345 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    346 			mp1->start = s;
    347 			mp1->size = sz;
    348 		} else {
    349 			mp->start = s;
    350 			mp->size = sz;
    351 		}
    352 	}
    353 
    354 	/*
    355 	 * We cannot do pmap_steal_memory here,
    356 	 * since we don't run with translation enabled yet.
    357 	 */
    358 #ifndef MSGBUFADDR
    359 	/*
    360 	 * allow for msgbuf
    361 	 */
    362 	sz = round_page(MSGBUFSIZE);
    363 	mp = NULL;
    364 	for (mp1 = avail; mp1->size; mp1++)
    365 		if (mp1->size >= sz)
    366 			mp = mp1;
    367 	if (mp == NULL)
    368 		panic("not enough memory?");
    369 
    370 	npgs -= btoc(sz);
    371 	msgbuf_paddr = mp->start + mp->size - sz;
    372 	mp->size -= sz;
    373 	if (mp->size <= 0)
    374 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    375 #endif
    376 
    377 	printf("Loading pages\n");
    378 	for (mp = avail; mp->size; mp++)
    379 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    380 			atop(mp->start), atop(mp->start + mp->size),
    381 			VM_FREELIST_DEFAULT);
    382 
    383 	/*
    384 	 * Initialize kernel pmap and hardware.
    385 	 */
    386 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    387 	pmap_kernel()->pm_ctx = KERNEL_PID;
    388 	nextavail = avail->start;
    389 
    390 
    391 	evcnt_attach_static(&tlbhit_ev);
    392 	evcnt_attach_static(&tlbmiss_ev);
    393 	evcnt_attach_static(&tlbflush_ev);
    394 	evcnt_attach_static(&tlbenter_ev);
    395 	printf("Done\n");
    396 }
    397 
    398 /*
    399  * Restrict given range to physical memory
    400  *
    401  * (Used by /dev/mem)
    402  */
    403 void
    404 pmap_real_memory(paddr_t *start, psize_t *size)
    405 {
    406 	struct mem_region *mp;
    407 
    408 	for (mp = mem; mp->size; mp++) {
    409 		if (*start + *size > mp->start &&
    410 		    *start < mp->start + mp->size) {
    411 			if (*start < mp->start) {
    412 				*size -= mp->start - *start;
    413 				*start = mp->start;
    414 			}
    415 			if (*start + *size > mp->start + mp->size)
    416 				*size = mp->start + mp->size - *start;
    417 			return;
    418 		}
    419 	}
    420 	*size = 0;
    421 }
    422 
    423 /*
    424  * Initialize anything else for pmap handling.
    425  * Called during vm_init().
    426  */
    427 void
    428 pmap_init(void)
    429 {
    430 	struct pv_entry *pv;
    431 	vsize_t sz;
    432 	vaddr_t addr;
    433 	int i, s;
    434 	int bank;
    435 	char *attr;
    436 
    437 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    438 	sz = round_page(sz);
    439 	addr = uvm_km_zalloc(kernel_map, sz);
    440 	s = splvm();
    441 	pv = pv_table = (struct pv_entry *)addr;
    442 	for (i = npgs; --i >= 0;)
    443 		pv++->pv_pm = NULL;
    444 	pmap_attrib = (char *)pv;
    445 	memset(pv, 0, npgs);
    446 
    447 	pv = pv_table;
    448 	attr = pmap_attrib;
    449 	for (bank = 0; bank < vm_nphysseg; bank++) {
    450 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    451 		vm_physmem[bank].pmseg.pvent = pv;
    452 		vm_physmem[bank].pmseg.attrs = attr;
    453 		pv += sz;
    454 		attr += sz;
    455 	}
    456 
    457 	pmap_initialized = 1;
    458 	splx(s);
    459 
    460 	/* Setup a pool for additional pvlist structures */
    461 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    462 }
    463 
    464 /*
    465  * How much virtual space is available to the kernel?
    466  */
    467 void
    468 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    469 {
    470 
    471 #if 0
    472 	/*
    473 	 * Reserve one segment for kernel virtual memory
    474 	 */
    475 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    476 	*end = *start + SEGMENT_LENGTH;
    477 #else
    478 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    479 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    480 #endif
    481 }
    482 
    483 #ifdef PMAP_GROWKERNEL
    484 /*
    485  * Preallocate kernel page tables to a specified VA.
    486  * This simply loops through the first TTE for each
    487  * page table from the beginning of the kernel pmap,
    488  * reads the entry, and if the result is
    489  * zero (either invalid entry or no page table) it stores
    490  * a zero there, populating page tables in the process.
    491  * This is not the most efficient technique but i don't
    492  * expect it to be called that often.
    493  */
    494 extern struct vm_page *vm_page_alloc1 __P((void));
    495 extern void vm_page_free1 __P((struct vm_page *));
    496 
    497 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    498 
    499 vaddr_t
    500 pmap_growkernel(maxkvaddr)
    501         vaddr_t maxkvaddr;
    502 {
    503 	int s;
    504 	int seg;
    505 	paddr_t pg;
    506 	struct pmap *pm = pmap_kernel();
    507 
    508 	s = splvm();
    509 
    510 	/* Align with the start of a page table */
    511 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    512 	     kbreak += PTMAP) {
    513 		seg = STIDX(kbreak);
    514 
    515 		if (pte_find(pm, kbreak)) continue;
    516 
    517 		if (uvm.page_init_done) {
    518 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    519 		} else {
    520 			if (!uvm_page_physget(&pg))
    521 				panic("pmap_growkernel: no memory");
    522 		}
    523 		if (!pg) panic("pmap_growkernel: no pages");
    524 		pmap_zero_page((paddr_t)pg);
    525 
    526 		/* XXX This is based on all phymem being addressable */
    527 		pm->pm_ptbl[seg] = (u_int *)pg;
    528 	}
    529 	splx(s);
    530 	return (kbreak);
    531 }
    532 
    533 /*
    534  *	vm_page_alloc1:
    535  *
    536  *	Allocate and return a memory cell with no associated object.
    537  */
    538 struct vm_page *
    539 vm_page_alloc1()
    540 {
    541 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    542 	if (pg) {
    543 		pg->wire_count = 1;	/* no mappings yet */
    544 		pg->flags &= ~PG_BUSY;	/* never busy */
    545 	}
    546 	return pg;
    547 }
    548 
    549 /*
    550  *	vm_page_free1:
    551  *
    552  *	Returns the given page to the free list,
    553  *	disassociating it with any VM object.
    554  *
    555  *	Object and page must be locked prior to entry.
    556  */
    557 void
    558 vm_page_free1(mem)
    559 	struct vm_page *mem;
    560 {
    561 #ifdef DIAGNOSTIC
    562 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    563 		printf("Freeing invalid page %p\n", mem);
    564 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    565 #ifdef DDB
    566 		Debugger();
    567 #endif
    568 		return;
    569 	}
    570 #endif
    571 	mem->flags |= PG_BUSY;
    572 	mem->wire_count = 0;
    573 	uvm_pagefree(mem);
    574 }
    575 #endif
    576 
    577 /*
    578  * Create and return a physical map.
    579  */
    580 struct pmap *
    581 pmap_create(void)
    582 {
    583 	struct pmap *pm;
    584 
    585 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    586 	memset((caddr_t)pm, 0, sizeof *pm);
    587 	pmap_pinit(pm);
    588 	return pm;
    589 }
    590 
    591 /*
    592  * Initialize a preallocated and zeroed pmap structure.
    593  */
    594 void
    595 pmap_pinit(struct pmap *pm)
    596 {
    597 	int i;
    598 
    599 	/*
    600 	 * Allocate some segment registers for this pmap.
    601 	 */
    602 	pm->pm_refs = 1;
    603 	for (i = 0; i < STSZ; i++)
    604 		pm->pm_ptbl[i] = NULL;
    605 }
    606 
    607 /*
    608  * Add a reference to the given pmap.
    609  */
    610 void
    611 pmap_reference(struct pmap *pm)
    612 {
    613 
    614 	pm->pm_refs++;
    615 }
    616 
    617 /*
    618  * Retire the given pmap from service.
    619  * Should only be called if the map contains no valid mappings.
    620  */
    621 void
    622 pmap_destroy(struct pmap *pm)
    623 {
    624 
    625 	if (--pm->pm_refs == 0) {
    626 		pmap_release(pm);
    627 		free((caddr_t)pm, M_VMPMAP);
    628 	}
    629 }
    630 
    631 /*
    632  * Release any resources held by the given physical map.
    633  * Called when a pmap initialized by pmap_pinit is being released.
    634  */
    635 static void
    636 pmap_release(struct pmap *pm)
    637 {
    638 	int i;
    639 
    640 	for (i = 0; i < STSZ; i++)
    641 		if (pm->pm_ptbl[i]) {
    642 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
    643 			pm->pm_ptbl[i] = NULL;
    644 		}
    645 	if (pm->pm_ctx) ctx_free(pm);
    646 }
    647 
    648 /*
    649  * Copy the range specified by src_addr/len
    650  * from the source map to the range dst_addr/len
    651  * in the destination map.
    652  *
    653  * This routine is only advisory and need not do anything.
    654  */
    655 void
    656 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    657 	  vsize_t len, vaddr_t src_addr)
    658 {
    659 }
    660 
    661 /*
    662  * Require that all active physical maps contain no
    663  * incorrect entries NOW.
    664  */
    665 void
    666 pmap_update(struct pmap *pmap)
    667 {
    668 }
    669 
    670 /*
    671  * Garbage collects the physical map system for
    672  * pages which are no longer used.
    673  * Success need not be guaranteed -- that is, there
    674  * may well be pages which are not referenced, but
    675  * others may be collected.
    676  * Called by the pageout daemon when pages are scarce.
    677  */
    678 void
    679 pmap_collect(struct pmap *pm)
    680 {
    681 }
    682 
    683 /*
    684  * Fill the given physical page with zeroes.
    685  */
    686 void
    687 pmap_zero_page(paddr_t pa)
    688 {
    689 
    690 #ifdef PPC_4XX_NOCACHE
    691 	memset((caddr_t)pa, 0, NBPG);
    692 #else
    693 	int i;
    694 
    695 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
    696 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    697 		pa += CACHELINESIZE;
    698 	}
    699 #endif
    700 }
    701 
    702 /*
    703  * Copy the given physical source page to its destination.
    704  */
    705 void
    706 pmap_copy_page(paddr_t src, paddr_t dst)
    707 {
    708 
    709 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
    710 	dcache_flush_page(dst);
    711 }
    712 
    713 /*
    714  * This returns whether this is the first mapping of a page.
    715  */
    716 static inline int
    717 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    718 {
    719 	struct pv_entry *pv, *npv = NULL;
    720 	int s;
    721 
    722 	if (!pmap_initialized)
    723 		return 0;
    724 
    725 	s = splvm();
    726 
    727 	pv = pa_to_pv(pa);
    728 for (npv = pv; npv; npv = npv->pv_next)
    729 if (npv->pv_va == va && npv->pv_pm == pm) {
    730 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    731 Debugger();
    732 return (1);
    733 }
    734 
    735 	if (!pv->pv_pm) {
    736 		/*
    737 		 * No entries yet, use header as the first entry.
    738 		 */
    739 		pv->pv_va = va;
    740 		pv->pv_pm = pm;
    741 		pv->pv_next = NULL;
    742 	} else {
    743 		/*
    744 		 * There is at least one other VA mapping this page.
    745 		 * Place this entry after the header.
    746 		 */
    747 		npv = pool_get(&pv_pool, PR_WAITOK);
    748 		if (!npv) return (0);
    749 		npv->pv_va = va;
    750 		npv->pv_pm = pm;
    751 		npv->pv_next = pv->pv_next;
    752 		pv->pv_next = npv;
    753 	}
    754 	splx(s);
    755 	return (1);
    756 }
    757 
    758 static void
    759 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    760 {
    761 	struct pv_entry *pv, *npv;
    762 
    763 	/*
    764 	 * Remove from the PV table.
    765 	 */
    766 	pv = pa_to_pv(pa);
    767 	if (!pv) return;
    768 
    769 	/*
    770 	 * If it is the first entry on the list, it is actually
    771 	 * in the header and we must copy the following entry up
    772 	 * to the header.  Otherwise we must search the list for
    773 	 * the entry.  In either case we free the now unused entry.
    774 	 */
    775 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    776 		if ((npv = pv->pv_next)) {
    777 			*pv = *npv;
    778 			pool_put(&pv_pool, npv);
    779 		} else
    780 			pv->pv_pm = NULL;
    781 	} else {
    782 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    783 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    784 				break;
    785 		if (npv) {
    786 			pv->pv_next = npv->pv_next;
    787 			pool_put(&pv_pool, npv);
    788 		}
    789 	}
    790 }
    791 
    792 /*
    793  * Insert physical page at pa into the given pmap at virtual address va.
    794  */
    795 int
    796 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    797 {
    798 	int s;
    799 	u_int tte;
    800 	int managed;
    801 
    802 	/*
    803 	 * Have to remove any existing mapping first.
    804 	 */
    805 	pmap_remove(pm, va, va + NBPG);
    806 
    807 	if (flags & PMAP_WIRED) flags |= prot;
    808 
    809 	/* If it has no protections don't bother w/the rest */
    810 	if (!(flags & VM_PROT_ALL))
    811 		return (0);
    812 
    813 	managed = 0;
    814 	if (vm_physseg_find(atop(pa), NULL) != -1)
    815 		managed = 1;
    816 
    817 	/*
    818 	 * Generate TTE.
    819 	 *
    820 	 * XXXX
    821 	 *
    822 	 * Since the kernel does not handle execution privileges properly,
    823 	 * we will handle read and execute permissions together.
    824 	 */
    825 	tte = TTE_PA(pa) | TTE_EX;
    826 	/* XXXX -- need to support multiple page sizes. */
    827 	tte |= TTE_SZ_16K;
    828 #ifdef	DIAGNOSTIC
    829 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    830 		(PME_NOCACHE | PME_WRITETHROUG))
    831 		panic("pmap_enter: uncached & writethrough\n");
    832 #endif
    833 	if (flags & PME_NOCACHE)
    834 		/* Must be I/O mapping */
    835 		tte |= TTE_I | TTE_G;
    836 #ifdef PPC_4XX_NOCACHE
    837 	tte |= TTE_I;
    838 #else
    839 	else if (flags & PME_WRITETHROUG)
    840 		/* Uncached and writethrough are not compatible */
    841 		tte |= TTE_W;
    842 #endif
    843 	if (pm == pmap_kernel())
    844 		tte |= TTE_ZONE(ZONE_PRIV);
    845 	else
    846 		tte |= TTE_ZONE(ZONE_USER);
    847 
    848 	if (flags & VM_PROT_WRITE)
    849 		tte |= TTE_WR;
    850 
    851 	/*
    852 	 * Now record mapping for later back-translation.
    853 	 */
    854 	if (pmap_initialized && managed) {
    855 		char *attr;
    856 
    857 		if (!pmap_enter_pv(pm, va, pa)) {
    858 			/* Could not enter pv on a managed page */
    859 			return 1;
    860 		}
    861 
    862 		/* Now set attributes. */
    863 		attr = pa_to_attr(pa);
    864 #ifdef DIAGNOSTIC
    865 		if (!attr)
    866 			panic("managed but no attr\n");
    867 #endif
    868 		if (flags & VM_PROT_ALL)
    869 			*attr |= PTE_HI_REF;
    870 		if (flags & VM_PROT_WRITE)
    871 			*attr |= PTE_HI_CHG;
    872 	}
    873 
    874 	s = splvm();
    875 	pm->pm_stats.resident_count++;
    876 
    877 	/* Insert page into page table. */
    878 	pte_enter(pm, va, tte);
    879 
    880 	/* If this is a real fault, enter it in the tlb */
    881 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    882 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    883 	}
    884 	splx(s);
    885 
    886 	/* Flush the real memory from the instruction cache. */
    887 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    888 		__syncicache((void *)pa, PAGE_SIZE);
    889 
    890 	return 0;
    891 }
    892 
    893 void
    894 pmap_unwire(struct pmap *pm, vaddr_t va)
    895 {
    896 	struct pv_entry *pv, *npv;
    897 	paddr_t pa;
    898 	int s = splvm();
    899 
    900         if (pm == NULL) {
    901                 return;
    902         }
    903 
    904 	if (!pmap_extract(pm, va, &pa)) {
    905 		return;
    906 	}
    907 
    908 	va |= PV_WIRED;
    909 
    910 	pv = pa_to_pv(pa);
    911 	if (!pv) return;
    912 
    913 	/*
    914 	 * If it is the first entry on the list, it is actually
    915 	 * in the header and we must copy the following entry up
    916 	 * to the header.  Otherwise we must search the list for
    917 	 * the entry.  In either case we free the now unused entry.
    918 	 */
    919 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    920 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    921 			npv->pv_va &= ~PV_WIRED;
    922 			break;
    923 		}
    924 	}
    925 	splx(s);
    926 }
    927 
    928 void
    929 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    930 {
    931 	int s;
    932 	u_int tte;
    933 	struct pmap *pm = pmap_kernel();
    934 
    935 	/*
    936 	 * Have to remove any existing mapping first.
    937 	 */
    938 
    939 	/*
    940 	 * Generate TTE.
    941 	 *
    942 	 * XXXX
    943 	 *
    944 	 * Since the kernel does not handle execution privileges properly,
    945 	 * we will handle read and execute permissions together.
    946 	 */
    947 	tte = 0;
    948 	if (prot & VM_PROT_ALL) {
    949 
    950 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    951 		/* XXXX -- need to support multiple page sizes. */
    952 		tte |= TTE_SZ_16K;
    953 #ifdef DIAGNOSTIC
    954 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    955 			(PME_NOCACHE | PME_WRITETHROUG))
    956 			panic("pmap_kenter_pa: uncached & writethrough\n");
    957 #endif
    958 		if (prot & PME_NOCACHE)
    959 			/* Must be I/O mapping */
    960 			tte |= TTE_I | TTE_G;
    961 #ifdef PPC_4XX_NOCACHE
    962 		tte |= TTE_I;
    963 #else
    964 		else if (prot & PME_WRITETHROUG)
    965 			/* Uncached and writethrough are not compatible */
    966 			tte |= TTE_W;
    967 #endif
    968 		if (prot & VM_PROT_WRITE)
    969 			tte |= TTE_WR;
    970 	}
    971 
    972 	s = splvm();
    973 	pm->pm_stats.resident_count++;
    974 
    975 	/* Insert page into page table. */
    976 	pte_enter(pm, va, tte);
    977 	splx(s);
    978 }
    979 
    980 void
    981 pmap_kremove(vaddr_t va, vsize_t len)
    982 {
    983 
    984 	while (len > 0) {
    985 		pte_enter(pmap_kernel(), va, 0);
    986 		va += PAGE_SIZE;
    987 		len -= PAGE_SIZE;
    988 	}
    989 }
    990 
    991 /*
    992  * Remove the given range of mapping entries.
    993  */
    994 void
    995 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
    996 {
    997 	int s;
    998 	paddr_t pa;
    999 	volatile u_int *ptp;
   1000 
   1001 	s = splvm();
   1002 	while (va < endva) {
   1003 
   1004 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1005 			pa = TTE_PA(pa);
   1006 			pmap_remove_pv(pm, va, pa);
   1007 			*ptp = 0;
   1008 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1009 			pm->pm_stats.resident_count--;
   1010 		}
   1011 		va += NBPG;
   1012 	}
   1013 
   1014 	splx(s);
   1015 }
   1016 
   1017 /*
   1018  * Get the physical page address for the given pmap/virtual address.
   1019  */
   1020 boolean_t
   1021 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1022 {
   1023 	int seg = STIDX(va);
   1024 	int ptn = PTIDX(va);
   1025 	u_int pa = 0;
   1026 	int s = splvm();
   1027 
   1028 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1029 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1030 	}
   1031 	splx(s);
   1032 	return (pa != 0);
   1033 }
   1034 
   1035 /*
   1036  * Lower the protection on the specified range of this pmap.
   1037  *
   1038  * There are only two cases: either the protection is going to 0,
   1039  * or it is going to read-only.
   1040  */
   1041 void
   1042 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1043 {
   1044 	volatile u_int *ptp;
   1045 	int s;
   1046 
   1047 	if (prot & VM_PROT_READ) {
   1048 		s = splvm();
   1049 		while (sva < eva) {
   1050 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1051 				*ptp &= ~TTE_WR;
   1052 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1053 			}
   1054 			sva += NBPG;
   1055 		}
   1056 		splx(s);
   1057 		return;
   1058 	}
   1059 	pmap_remove(pm, sva, eva);
   1060 }
   1061 
   1062 boolean_t
   1063 check_attr(struct vm_page *pg, u_int mask, int clear)
   1064 {
   1065 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1066 	int s;
   1067 	char *attr;
   1068 	int rv;
   1069 
   1070 	/*
   1071 	 * First modify bits in cache.
   1072 	 */
   1073 	s = splvm();
   1074 	attr = pa_to_attr(pa);
   1075 	if (attr == NULL)
   1076 		return FALSE;
   1077 
   1078 	rv = ((*attr & mask) != 0);
   1079 	if (clear) {
   1080 		*attr &= ~mask;
   1081 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1082 	}
   1083 	splx(s);
   1084 	return rv;
   1085 }
   1086 
   1087 
   1088 /*
   1089  * Lower the protection on the specified physical page.
   1090  *
   1091  * There are only two cases: either the protection is going to 0,
   1092  * or it is going to read-only.
   1093  */
   1094 void
   1095 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1096 {
   1097 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1098 	vaddr_t va;
   1099 	struct pv_entry *pvh, *pv, *npv;
   1100 	struct pmap *pm;
   1101 
   1102 	pvh = pa_to_pv(pa);
   1103 	if (pvh == NULL)
   1104 		return;
   1105 
   1106 	/* Handle extra pvs which may be deleted in the operation */
   1107 	for (pv = pvh->pv_next; pv; pv = npv) {
   1108 		npv = pv->pv_next;
   1109 
   1110 		pm = pv->pv_pm;
   1111 		va = pv->pv_va;
   1112 		pmap_protect(pm, va, va+NBPG, prot);
   1113 	}
   1114 	/* Now check the head pv */
   1115 	if (pvh->pv_pm) {
   1116 		pv = pvh;
   1117 		pm = pv->pv_pm;
   1118 		va = pv->pv_va;
   1119 		pmap_protect(pm, va, va+NBPG, prot);
   1120 	}
   1121 }
   1122 
   1123 /*
   1124  * Activate the address space for the specified process.  If the process
   1125  * is the current process, load the new MMU context.
   1126  */
   1127 void
   1128 pmap_activate(struct proc *p)
   1129 {
   1130 #if 0
   1131 	struct pcb *pcb = &p->p_addr->u_pcb;
   1132 	pmap_t pmap = p->p_vmspace->vm_map.pmap;
   1133 
   1134 	/*
   1135 	 * XXX Normally performed in cpu_fork().
   1136 	 */
   1137 	printf("pmap_activate(%p), pmap=%p\n",p,pmap);
   1138 	if (pcb->pcb_pm != pmap) {
   1139 		pcb->pcb_pm = pmap;
   1140 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1141 		    (paddr_t *)&pcb->pcb_pmreal);
   1142 	}
   1143 
   1144 	if (p == curproc) {
   1145 		/* Store pointer to new current pmap. */
   1146 		curpm = pcb->pcb_pmreal;
   1147 	}
   1148 #endif
   1149 }
   1150 
   1151 /*
   1152  * Deactivate the specified process's address space.
   1153  */
   1154 void
   1155 pmap_deactivate(struct proc *p)
   1156 {
   1157 }
   1158 
   1159 /*
   1160  * Synchronize caches corresponding to [addr, addr+len) in p.
   1161  */
   1162 void
   1163 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1164 {
   1165 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1166 	int msr, ctx, opid;
   1167 
   1168 
   1169 	/*
   1170 	 * Need to turn off IMMU and switch to user context.
   1171 	 * (icbi uses DMMU).
   1172 	 */
   1173 	if (!(ctx = pm->pm_ctx)) {
   1174 		/* No context -- assign it one */
   1175 		ctx_alloc(pm);
   1176 		ctx = pm->pm_ctx;
   1177 	}
   1178 	__asm __volatile("mfmsr %0;"
   1179 		"li %1, 0x20;"
   1180 		"andc %1,%0,%1;"
   1181 		"mtmsr %1;"
   1182 		"sync;isync;"
   1183 		"mfpid %1;"
   1184 		"mtpid %2;"
   1185 		"sync; isync;"
   1186 		"1:"
   1187 		"dcbf 0,%3;"
   1188 		"icbi 0,%3;"
   1189 		"addi %3,%3,32;"
   1190 		"addic. %4,%4,-32;"
   1191 		"bge 1b;"
   1192 		"mtpid %1;"
   1193 		"mtmsr %0;"
   1194 		"sync; isync"
   1195 		: "=&r" (msr), "=&r" (opid)
   1196 		: "r" (ctx), "r" (va), "r" (len));
   1197 }
   1198 
   1199 
   1200 /* This has to be done in real mode !!! */
   1201 void
   1202 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1203 {
   1204 	u_long i, found;
   1205 	u_long msr;
   1206 
   1207 	/* If there's no context then it can't be mapped. */
   1208 	if (!pid) return;
   1209 
   1210 	asm("mfpid %1;"			/* Save PID */
   1211 		"mfmsr %2;"		/* Save MSR */
   1212 		"li %0,0;"		/* Now clear MSR */
   1213 		"mtmsr %0;"
   1214 		"mtpid %4;"		/* Set PID */
   1215 		"sync;"
   1216 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1217 		"sync;"
   1218 		"mtpid %1;"		/* Restore PID */
   1219 		"mtmsr %2;"		/* Restore MSR */
   1220 		"sync;isync;"
   1221 		"li %1,1;"
   1222 		"beq 1f;"
   1223 		"li %1,0;"
   1224 		"1:"
   1225 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1226 		: "r" (va), "r" (pid));
   1227 	if (found && !TLB_LOCKED(i)) {
   1228 
   1229 		/* Now flush translation */
   1230 		asm volatile(
   1231 			"tlbwe %0,%1,0;"
   1232 			"sync;isync;"
   1233 			: : "r" (0), "r" (i));
   1234 
   1235 		tlb_info[i].ti_ctx = 0;
   1236 		tlb_info[i].ti_flags = 0;
   1237 		tlbnext = i;
   1238 		/* Successful flushes */
   1239 		tlbflush_ev.ev_count++;
   1240 	}
   1241 }
   1242 
   1243 void
   1244 ppc4xx_tlb_flush_all(void)
   1245 {
   1246 	u_long i;
   1247 
   1248 	for (i = 0; i < NTLB; i++)
   1249 		if (!TLB_LOCKED(i)) {
   1250 			asm volatile(
   1251 				"tlbwe %0,%1,0;"
   1252 				"sync;isync;"
   1253 				: : "r" (0), "r" (i));
   1254 			tlb_info[i].ti_ctx = 0;
   1255 			tlb_info[i].ti_flags = 0;
   1256 		}
   1257 
   1258 	asm volatile("sync;isync");
   1259 }
   1260 
   1261 /* Find a TLB entry to evict. */
   1262 static int
   1263 ppc4xx_tlb_find_victim(void)
   1264 {
   1265 	int flags;
   1266 
   1267 	for (;;) {
   1268 		if (++tlbnext >= NTLB)
   1269 			tlbnext = TLB_NRESERVED;
   1270 		flags = tlb_info[tlbnext].ti_flags;
   1271 		if (!(flags & TLBF_USED) ||
   1272 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1273 			u_long va, stack = (u_long)&va;
   1274 
   1275 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1276 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1277 			     (flags & TLBF_USED)) {
   1278 				/* Kernel stack page */
   1279 				flags |= TLBF_USED;
   1280 				tlb_info[tlbnext].ti_flags = flags;
   1281 			} else {
   1282 				/* Found it! */
   1283 				return (tlbnext);
   1284 			}
   1285 		} else {
   1286 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1287 		}
   1288 	}
   1289 }
   1290 
   1291 void
   1292 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1293 {
   1294 	u_long th, tl, idx;
   1295 	tlbpid_t pid;
   1296 	u_short msr;
   1297 	paddr_t pa;
   1298 	int s, sz;
   1299 
   1300 
   1301 	tlbenter_ev.ev_count++;
   1302 
   1303 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1304 	pa = (pte & TTE_RPN_MASK(sz));
   1305 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1306 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1307 	tl |= ppc4xx_tlbflags(va, pa);
   1308 
   1309 	s = splhigh();
   1310 	idx = ppc4xx_tlb_find_victim();
   1311 
   1312 #ifdef DIAGNOSTIC
   1313 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1314 		panic("ppc4xx_tlb_enter: repacing entry %ld\n", idx);
   1315 	}
   1316 #endif
   1317 
   1318 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1319 	tlb_info[idx].ti_ctx = ctx;
   1320 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1321 
   1322 	asm volatile(
   1323 		"mfmsr %0;"			/* Save MSR */
   1324 		"li %1,0;"
   1325 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1326 		"mtmsr %1;"			/* Clear MSR */
   1327 		"mfpid %1;"			/* Save old PID */
   1328 		"mtpid %2;"			/* Load translation ctx */
   1329 		"sync; isync;"
   1330 #ifdef DEBUG
   1331 		"andi. %3,%3,63;"
   1332 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1333 #endif
   1334 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1335 		"sync; isync;"
   1336 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1337 		"sync; isync;"
   1338 	: "=&r" (msr), "=&r" (pid)
   1339 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1340 	splx(s);
   1341 }
   1342 
   1343 void
   1344 ppc4xx_tlb_unpin(int i)
   1345 {
   1346 
   1347 	if (i == -1)
   1348 		for (i = 0; i < TLB_NRESERVED; i++)
   1349 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1350 	else
   1351 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1352 }
   1353 
   1354 void
   1355 ppc4xx_tlb_init(void)
   1356 {
   1357 	int i;
   1358 
   1359 	/* Mark reserved TLB entries */
   1360 	for (i = 0; i < TLB_NRESERVED; i++) {
   1361 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1362 		tlb_info[i].ti_ctx = KERNEL_PID;
   1363 	}
   1364 
   1365 	/* Setup security zones */
   1366 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1367 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1368 	 * Z3 - full access regardless of TLB entry permissions
   1369 	 */
   1370 
   1371 	asm volatile(
   1372 		"mtspr %0,%1;"
   1373 		"sync;"
   1374 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1375 }
   1376 
   1377 
   1378 /*
   1379  * We should pass the ctx in from trap code.
   1380  */
   1381 int
   1382 pmap_tlbmiss(vaddr_t va, int ctx)
   1383 {
   1384 	volatile u_int *pte;
   1385 	u_long tte;
   1386 
   1387 	tlbmiss_ev.ev_count++;
   1388 
   1389 	/*
   1390 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1391 	 */
   1392 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1393 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1394 		if (pte == NULL) {
   1395 			/* Map unmanaged addresses directly for kernel access */
   1396 			return 1;
   1397 		}
   1398 		tte = *pte;
   1399 		if (tte == 0) {
   1400 			return 1;
   1401 		}
   1402 	} else {
   1403 		/* Create a 16MB writeable mapping. */
   1404 #ifdef PPC_4XX_NOCACHE
   1405 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1406 #else
   1407 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1408 #endif
   1409 	}
   1410 	tlbhit_ev.ev_count++;
   1411 	ppc4xx_tlb_enter(ctx, va, tte);
   1412 
   1413 	return 0;
   1414 }
   1415 
   1416 /*
   1417  * Flush all the entries matching a context from the TLB.
   1418  */
   1419 static int
   1420 ctx_flush(int cnum)
   1421 {
   1422 	int i;
   1423 
   1424 	/* We gotta steal this context */
   1425 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1426 		if (tlb_info[i].ti_ctx == cnum) {
   1427 			/* Can't steal ctx if it has a locked entry. */
   1428 			if (TLB_LOCKED(i)) {
   1429 #ifdef DIAGNOSTIC
   1430 				printf("ctx_flush: can't invalidate "
   1431 					"locked mapping %d "
   1432 					"for context %d\n", i, cnum);
   1433 #ifdef DDB
   1434 				Debugger();
   1435 #endif
   1436 #endif
   1437 				return (1);
   1438 			}
   1439 #ifdef DIAGNOSTIC
   1440 			if (i < TLB_NRESERVED)
   1441 				panic("TLB entry %d not locked\n", i);
   1442 #endif
   1443 			/* Invalidate particular TLB entry regardless of locked status */
   1444 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1445 			tlb_info[i].ti_flags = 0;
   1446 		}
   1447 	}
   1448 	return (0);
   1449 }
   1450 
   1451 /*
   1452  * Allocate a context.  If necessary, steal one from someone else.
   1453  *
   1454  * The new context is flushed from the TLB before returning.
   1455  */
   1456 int
   1457 ctx_alloc(struct pmap *pm)
   1458 {
   1459 	int s, cnum;
   1460 	static int next = MINCTX;
   1461 
   1462 	if (pm == pmap_kernel()) {
   1463 #ifdef DIAGNOSTIC
   1464 		printf("ctx_alloc: kernel pmap!\n");
   1465 #endif
   1466 		return (0);
   1467 	}
   1468 	s = splvm();
   1469 
   1470 	/* Find a likely context. */
   1471 	cnum = next;
   1472 	do {
   1473 		if ((++cnum) > NUMCTX)
   1474 			cnum = MINCTX;
   1475 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1476 
   1477 	/* Now clean it out */
   1478 oops:
   1479 	if (cnum < MINCTX)
   1480 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1481 	if (ctx_flush(cnum)) {
   1482 		/* oops -- something's wired. */
   1483 		if ((++cnum) > NUMCTX)
   1484 			cnum = MINCTX;
   1485 		goto oops;
   1486 	}
   1487 
   1488 	if (ctxbusy[cnum]) {
   1489 #ifdef DEBUG
   1490 		/* We should identify this pmap and clear it */
   1491 		printf("Warning: stealing context %d\n", cnum);
   1492 #endif
   1493 		ctxbusy[cnum]->pm_ctx = 0;
   1494 	}
   1495 	ctxbusy[cnum] = pm;
   1496 	next = cnum;
   1497 	splx(s);
   1498 	pm->pm_ctx = cnum;
   1499 
   1500 	return cnum;
   1501 }
   1502 
   1503 /*
   1504  * Give away a context.
   1505  */
   1506 void
   1507 ctx_free(struct pmap *pm)
   1508 {
   1509 	int oldctx;
   1510 
   1511 	oldctx = pm->pm_ctx;
   1512 
   1513 	if (oldctx == 0)
   1514 		panic("ctx_free: freeing kernel context");
   1515 #ifdef DIAGNOSTIC
   1516 	if (ctxbusy[oldctx] == 0)
   1517 		printf("ctx_free: freeing free context %d\n", oldctx);
   1518 	if (ctxbusy[oldctx] != pm) {
   1519 		printf("ctx_free: freeing someone esle's context\n "
   1520 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1521 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1522 #ifdef DDB
   1523 		Debugger();
   1524 #endif
   1525 	}
   1526 #endif
   1527 	/* We should verify it has not been stolen and reallocated... */
   1528 	ctxbusy[oldctx] = NULL;
   1529 	ctx_flush(oldctx);
   1530 }
   1531 
   1532 
   1533 #ifdef DEBUG
   1534 /*
   1535  * Test ref/modify handling.
   1536  */
   1537 void pmap_testout __P((void));
   1538 void
   1539 pmap_testout()
   1540 {
   1541 	vaddr_t va;
   1542 	volatile int *loc;
   1543 	int val = 0;
   1544 	paddr_t pa;
   1545 	struct vm_page *pg;
   1546 	int ref, mod;
   1547 
   1548 	/* Allocate a page */
   1549 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
   1550 	loc = (int*)va;
   1551 
   1552 	pmap_extract(pmap_kernel(), va, &pa);
   1553 	pg = PHYS_TO_VM_PAGE(pa);
   1554 	pmap_unwire(pmap_kernel(), va);
   1555 
   1556 	pmap_remove(pmap_kernel(), va, va+1);
   1557 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1558 	pmap_update(pmap_kernel());
   1559 
   1560 	/* Now clear reference and modify */
   1561 	ref = pmap_clear_reference(pg);
   1562 	mod = pmap_clear_modify(pg);
   1563 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1564 	       (void *)(u_long)va, (long)pa,
   1565 	       ref, mod);
   1566 
   1567 	/* Check it's properly cleared */
   1568 	ref = pmap_is_referenced(pg);
   1569 	mod = pmap_is_modified(pg);
   1570 	printf("Checking cleared page: ref %d, mod %d\n",
   1571 	       ref, mod);
   1572 
   1573 	/* Reference page */
   1574 	val = *loc;
   1575 
   1576 	ref = pmap_is_referenced(pg);
   1577 	mod = pmap_is_modified(pg);
   1578 	printf("Referenced page: ref %d, mod %d val %x\n",
   1579 	       ref, mod, val);
   1580 
   1581 	/* Now clear reference and modify */
   1582 	ref = pmap_clear_reference(pg);
   1583 	mod = pmap_clear_modify(pg);
   1584 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1585 	       (void *)(u_long)va, (long)pa,
   1586 	       ref, mod);
   1587 
   1588 	/* Modify page */
   1589 	*loc = 1;
   1590 
   1591 	ref = pmap_is_referenced(pg);
   1592 	mod = pmap_is_modified(pg);
   1593 	printf("Modified page: ref %d, mod %d\n",
   1594 	       ref, mod);
   1595 
   1596 	/* Now clear reference and modify */
   1597 	ref = pmap_clear_reference(pg);
   1598 	mod = pmap_clear_modify(pg);
   1599 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1600 	       (void *)(u_long)va, (long)pa,
   1601 	       ref, mod);
   1602 
   1603 	/* Check it's properly cleared */
   1604 	ref = pmap_is_referenced(pg);
   1605 	mod = pmap_is_modified(pg);
   1606 	printf("Checking cleared page: ref %d, mod %d\n",
   1607 	       ref, mod);
   1608 
   1609 	/* Modify page */
   1610 	*loc = 1;
   1611 
   1612 	ref = pmap_is_referenced(pg);
   1613 	mod = pmap_is_modified(pg);
   1614 	printf("Modified page: ref %d, mod %d\n",
   1615 	       ref, mod);
   1616 
   1617 	/* Check pmap_protect() */
   1618 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1619 	pmap_update(pmap_kernel());
   1620 	ref = pmap_is_referenced(pg);
   1621 	mod = pmap_is_modified(pg);
   1622 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1623 	       ref, mod);
   1624 
   1625 	/* Now clear reference and modify */
   1626 	ref = pmap_clear_reference(pg);
   1627 	mod = pmap_clear_modify(pg);
   1628 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1629 	       (void *)(u_long)va, (long)pa,
   1630 	       ref, mod);
   1631 
   1632 	/* Reference page */
   1633 	val = *loc;
   1634 
   1635 	ref = pmap_is_referenced(pg);
   1636 	mod = pmap_is_modified(pg);
   1637 	printf("Referenced page: ref %d, mod %d val %x\n",
   1638 	       ref, mod, val);
   1639 
   1640 	/* Now clear reference and modify */
   1641 	ref = pmap_clear_reference(pg);
   1642 	mod = pmap_clear_modify(pg);
   1643 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1644 	       (void *)(u_long)va, (long)pa,
   1645 	       ref, mod);
   1646 
   1647 	/* Modify page */
   1648 #if 0
   1649 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1650 	pmap_update(pmap_kernel());
   1651 #endif
   1652 	*loc = 1;
   1653 
   1654 	ref = pmap_is_referenced(pg);
   1655 	mod = pmap_is_modified(pg);
   1656 	printf("Modified page: ref %d, mod %d\n",
   1657 	       ref, mod);
   1658 
   1659 	/* Check pmap_protect() */
   1660 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1661 	pmap_update(pmap_kernel());
   1662 	ref = pmap_is_referenced(pg);
   1663 	mod = pmap_is_modified(pg);
   1664 	printf("pmap_protect(): ref %d, mod %d\n",
   1665 	       ref, mod);
   1666 
   1667 	/* Now clear reference and modify */
   1668 	ref = pmap_clear_reference(pg);
   1669 	mod = pmap_clear_modify(pg);
   1670 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1671 	       (void *)(u_long)va, (long)pa,
   1672 	       ref, mod);
   1673 
   1674 	/* Reference page */
   1675 	val = *loc;
   1676 
   1677 	ref = pmap_is_referenced(pg);
   1678 	mod = pmap_is_modified(pg);
   1679 	printf("Referenced page: ref %d, mod %d val %x\n",
   1680 	       ref, mod, val);
   1681 
   1682 	/* Now clear reference and modify */
   1683 	ref = pmap_clear_reference(pg);
   1684 	mod = pmap_clear_modify(pg);
   1685 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1686 	       (void *)(u_long)va, (long)pa,
   1687 	       ref, mod);
   1688 
   1689 	/* Modify page */
   1690 #if 0
   1691 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1692 	pmap_update(pmap_kernel());
   1693 #endif
   1694 	*loc = 1;
   1695 
   1696 	ref = pmap_is_referenced(pg);
   1697 	mod = pmap_is_modified(pg);
   1698 	printf("Modified page: ref %d, mod %d\n",
   1699 	       ref, mod);
   1700 
   1701 	/* Check pmap_pag_protect() */
   1702 	pmap_page_protect(pg, VM_PROT_READ);
   1703 	ref = pmap_is_referenced(pg);
   1704 	mod = pmap_is_modified(pg);
   1705 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1706 	       ref, mod);
   1707 
   1708 	/* Now clear reference and modify */
   1709 	ref = pmap_clear_reference(pg);
   1710 	mod = pmap_clear_modify(pg);
   1711 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1712 	       (void *)(u_long)va, (long)pa,
   1713 	       ref, mod);
   1714 
   1715 	/* Reference page */
   1716 	val = *loc;
   1717 
   1718 	ref = pmap_is_referenced(pg);
   1719 	mod = pmap_is_modified(pg);
   1720 	printf("Referenced page: ref %d, mod %d val %x\n",
   1721 	       ref, mod, val);
   1722 
   1723 	/* Now clear reference and modify */
   1724 	ref = pmap_clear_reference(pg);
   1725 	mod = pmap_clear_modify(pg);
   1726 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1727 	       (void *)(u_long)va, (long)pa,
   1728 	       ref, mod);
   1729 
   1730 	/* Modify page */
   1731 #if 0
   1732 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1733 	pmap_update(pmap_kernel());
   1734 #endif
   1735 	*loc = 1;
   1736 
   1737 	ref = pmap_is_referenced(pg);
   1738 	mod = pmap_is_modified(pg);
   1739 	printf("Modified page: ref %d, mod %d\n",
   1740 	       ref, mod);
   1741 
   1742 	/* Check pmap_pag_protect() */
   1743 	pmap_page_protect(pg, VM_PROT_NONE);
   1744 	ref = pmap_is_referenced(pg);
   1745 	mod = pmap_is_modified(pg);
   1746 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1747 	       ref, mod);
   1748 
   1749 	/* Now clear reference and modify */
   1750 	ref = pmap_clear_reference(pg);
   1751 	mod = pmap_clear_modify(pg);
   1752 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1753 	       (void *)(u_long)va, (long)pa,
   1754 	       ref, mod);
   1755 
   1756 
   1757 	/* Reference page */
   1758 	val = *loc;
   1759 
   1760 	ref = pmap_is_referenced(pg);
   1761 	mod = pmap_is_modified(pg);
   1762 	printf("Referenced page: ref %d, mod %d val %x\n",
   1763 	       ref, mod, val);
   1764 
   1765 	/* Now clear reference and modify */
   1766 	ref = pmap_clear_reference(pg);
   1767 	mod = pmap_clear_modify(pg);
   1768 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1769 	       (void *)(u_long)va, (long)pa,
   1770 	       ref, mod);
   1771 
   1772 	/* Modify page */
   1773 #if 0
   1774 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1775 	pmap_update(pmap_kernel());
   1776 #endif
   1777 	*loc = 1;
   1778 
   1779 	ref = pmap_is_referenced(pg);
   1780 	mod = pmap_is_modified(pg);
   1781 	printf("Modified page: ref %d, mod %d\n",
   1782 	       ref, mod);
   1783 
   1784 	/* Unmap page */
   1785 	pmap_remove(pmap_kernel(), va, va+1);
   1786 	pmap_update(pmap_kernel());
   1787 	ref = pmap_is_referenced(pg);
   1788 	mod = pmap_is_modified(pg);
   1789 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1790 
   1791 	/* Now clear reference and modify */
   1792 	ref = pmap_clear_reference(pg);
   1793 	mod = pmap_clear_modify(pg);
   1794 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1795 	       (void *)(u_long)va, (long)pa, ref, mod);
   1796 
   1797 	/* Check it's properly cleared */
   1798 	ref = pmap_is_referenced(pg);
   1799 	mod = pmap_is_modified(pg);
   1800 	printf("Checking cleared page: ref %d, mod %d\n",
   1801 	       ref, mod);
   1802 
   1803 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1804 		VM_PROT_ALL|PMAP_WIRED);
   1805 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
   1806 }
   1807 #endif
   1808