Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.9
      1 /*	$NetBSD: pmap.c,v 1.9 2002/03/08 20:48:33 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 #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/pcb.h>
     83 #include <machine/powerpc.h>
     84 
     85 #include <powerpc/spr.h>
     86 #include <powerpc/ibm4xx/tlb.h>
     87 
     88 
     89 #define	CACHE_LINE	32
     90 
     91 /*
     92  * kernmap is an array of PTEs large enough to map in
     93  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     94  */
     95 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
     96 caddr_t kernmap;
     97 
     98 #define MINCTX		2
     99 #define NUMCTX		256
    100 volatile struct pmap *ctxbusy[NUMCTX];
    101 
    102 #define TLBF_USED	0x1
    103 #define	TLBF_REF	0x2
    104 #define	TLBF_LOCKED	0x4
    105 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    106 typedef struct tlb_info_s {
    107 	char	ti_flags;
    108 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    109 	u_int	ti_va;
    110 } tlb_info_t;
    111 
    112 volatile tlb_info_t tlb_info[NTLB];
    113 /* We'll use a modified FIFO replacement policy cause it's cheap */
    114 volatile int tlbnext = TLB_NRESERVED;
    115 
    116 u_long dtlb_miss_count = 0;
    117 u_long itlb_miss_count = 0;
    118 u_long ktlb_miss_count = 0;
    119 u_long utlb_miss_count = 0;
    120 
    121 /* Event counters -- XXX type `INTR' so we can see them with vmstat -i */
    122 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    123 	NULL, "cpu", "tlbmiss");
    124 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    125 	NULL, "cpu", "tlbhit");
    126 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    127 	NULL, "cpu", "tlbflush");
    128 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    129 	NULL, "cpu", "tlbenter");
    130 
    131 struct pmap kernel_pmap_;
    132 
    133 int physmem;
    134 static int npgs;
    135 static u_int nextavail;
    136 #ifndef MSGBUFADDR
    137 extern paddr_t msgbuf_paddr;
    138 #endif
    139 
    140 static struct mem_region *mem, *avail;
    141 
    142 /*
    143  * This is a cache of referenced/modified bits.
    144  * Bits herein are shifted by ATTRSHFT.
    145  */
    146 static char *pmap_attrib;
    147 
    148 #define PV_WIRED	0x1
    149 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    150 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    151 
    152 struct pv_entry {
    153 	struct pv_entry *pv_next;	/* Linked list of mappings */
    154 	vaddr_t pv_va;			/* virtual address of mapping */
    155 	struct pmap *pv_pm;
    156 };
    157 
    158 struct pv_entry *pv_table;
    159 static struct pool pv_pool;
    160 
    161 static int pmap_initialized;
    162 
    163 static int ctx_flush(int);
    164 
    165 static inline void dcache_flush_page(vaddr_t);
    166 static inline void icache_flush_page(vaddr_t);
    167 static inline void dcache_flush(vaddr_t, vsize_t);
    168 static inline void icache_flush(vaddr_t, vsize_t);
    169 
    170 inline struct pv_entry *pa_to_pv(paddr_t);
    171 static inline char *pa_to_attr(paddr_t);
    172 
    173 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    174 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    175 
    176 static void pmap_pinit(pmap_t);
    177 static void pmap_release(pmap_t);
    178 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    179 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    180 
    181 /*
    182  * These small routines may have to be replaced,
    183  * if/when we support processors other that the 604.
    184  */
    185 
    186 static inline void
    187 dcache_flush_page(vaddr_t va)
    188 {
    189 	int i;
    190 
    191 	for (i = 0; i < NBPG; i += CACHE_LINE)
    192 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    193 	asm volatile("sync;isync" : : );
    194 }
    195 
    196 static inline void
    197 icache_flush_page(vaddr_t va)
    198 {
    199 	int i;
    200 
    201 	for (i = 0; i < NBPG; i += CACHE_LINE)
    202 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    203 	asm volatile("sync;isync" : : );
    204 }
    205 
    206 static inline void
    207 dcache_flush(vaddr_t va, vsize_t len)
    208 {
    209 	int i;
    210 
    211 	if (len == 0)
    212 		return;
    213 
    214 	/* Make sure we flush all cache lines */
    215 	len += va & (CACHE_LINE-1);
    216 	for (i = 0; i < len; i += CACHE_LINE)
    217 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    218 	asm volatile("sync;isync" : : );
    219 }
    220 
    221 static inline void
    222 icache_flush(vaddr_t va, vsize_t len)
    223 {
    224 	int i;
    225 
    226 	if (len == 0)
    227 		return;
    228 
    229 	/* Make sure we flush all cache lines */
    230 	len += va & (CACHE_LINE-1);
    231 	for (i = 0; i < len; i += CACHE_LINE)
    232 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    233 	asm volatile("sync;isync" : : );
    234 }
    235 
    236 inline struct pv_entry *
    237 pa_to_pv(paddr_t pa)
    238 {
    239 	int bank, pg;
    240 
    241 	bank = vm_physseg_find(atop(pa), &pg);
    242 	if (bank == -1)
    243 		return NULL;
    244 	return &vm_physmem[bank].pmseg.pvent[pg];
    245 }
    246 
    247 static inline char *
    248 pa_to_attr(paddr_t pa)
    249 {
    250 	int bank, pg;
    251 
    252 	bank = vm_physseg_find(atop(pa), &pg);
    253 	if (bank == -1)
    254 		return NULL;
    255 	return &vm_physmem[bank].pmseg.attrs[pg];
    256 }
    257 
    258 /*
    259  * Insert PTE into page table.
    260  */
    261 int
    262 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    263 {
    264 	int seg = STIDX(va);
    265 	int ptn = PTIDX(va);
    266 	paddr_t pa;
    267 
    268 	if (!pm->pm_ptbl[seg]) {
    269 		/* Don't allocate a page to clear a non-existent mapping. */
    270 		if (!pte) return (1);
    271 		/* Allocate a page XXXX this will sleep! */
    272 		pa = 0;
    273 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
    274 	}
    275 	pm->pm_ptbl[seg][ptn] = pte;
    276 
    277 	/* Flush entry. */
    278 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    279 	return (1);
    280 }
    281 
    282 /*
    283  * Get a pointer to a PTE in a page table.
    284  */
    285 volatile u_int *
    286 pte_find(struct pmap *pm, vaddr_t va)
    287 {
    288 	int seg = STIDX(va);
    289 	int ptn = PTIDX(va);
    290 
    291 	if (pm->pm_ptbl[seg])
    292 		return (&pm->pm_ptbl[seg][ptn]);
    293 
    294 	return (NULL);
    295 }
    296 
    297 /*
    298  * This is called during initppc, before the system is really initialized.
    299  */
    300 void
    301 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    302 {
    303 	struct mem_region *mp, *mp1;
    304 	int cnt, i;
    305 	u_int s, e, sz;
    306 
    307 	/*
    308 	 * Allocate the kernel page table at the end of
    309 	 * kernel space so it's in the locked TTE.
    310 	 */
    311 	kernmap = (caddr_t)kernelend;
    312 //	kernelend += KERNMAP_SIZE*sizeof(struct pte);
    313 
    314 	/*
    315 	 * Initialize kernel page table.
    316 	 */
    317 //	memset(kernmap, 0, KERNMAP_SIZE*sizeof(struct pte));
    318 	for (i = 0; i < STSZ; i++) {
    319 		pmap_kernel()->pm_ptbl[i] = 0; // (u_int *)(kernmap + i*NBPG);
    320 	}
    321 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    322 
    323 	/*
    324 	 * Announce page-size to the VM-system
    325 	 */
    326 	uvmexp.pagesize = NBPG;
    327 	uvm_setpagesize();
    328 
    329 	/*
    330 	 * Get memory.
    331 	 */
    332 	mem_regions(&mem, &avail);
    333 	for (mp = mem; mp->size; mp++) {
    334 		physmem += btoc(mp->size);
    335 		printf("+%lx,",mp->size);
    336 	}
    337 	printf("\n");
    338 	ppc4xx_tlb_init();
    339 	/*
    340 	 * Count the number of available entries.
    341 	 */
    342 	for (cnt = 0, mp = avail; mp->size; mp++)
    343 		cnt++;
    344 
    345 	/*
    346 	 * Page align all regions.
    347 	 * Non-page aligned memory isn't very interesting to us.
    348 	 * Also, sort the entries for ascending addresses.
    349 	 */
    350 	kernelstart &= ~PGOFSET;
    351 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    352 	for (mp = avail; mp->size; mp++) {
    353 		s = mp->start;
    354 		e = mp->start + mp->size;
    355 		printf("%08x-%08x -> ",s,e);
    356 		/*
    357 		 * Check whether this region holds all of the kernel.
    358 		 */
    359 		if (s < kernelstart && e > kernelend) {
    360 			avail[cnt].start = kernelend;
    361 			avail[cnt++].size = e - kernelend;
    362 			e = kernelstart;
    363 		}
    364 		/*
    365 		 * Look whether this regions starts within the kernel.
    366 		 */
    367 		if (s >= kernelstart && s < kernelend) {
    368 			if (e <= kernelend)
    369 				goto empty;
    370 			s = kernelend;
    371 		}
    372 		/*
    373 		 * Now look whether this region ends within the kernel.
    374 		 */
    375 		if (e > kernelstart && e <= kernelend) {
    376 			if (s >= kernelstart)
    377 				goto empty;
    378 			e = kernelstart;
    379 		}
    380 		/*
    381 		 * Now page align the start and size of the region.
    382 		 */
    383 		s = round_page(s);
    384 		e = trunc_page(e);
    385 		if (e < s)
    386 			e = s;
    387 		sz = e - s;
    388 		printf("%08x-%08x = %x\n",s,e,sz);
    389 		/*
    390 		 * Check whether some memory is left here.
    391 		 */
    392 		if (sz == 0) {
    393 		empty:
    394 			memmove(mp, mp + 1,
    395 				(cnt - (mp - avail)) * sizeof *mp);
    396 			cnt--;
    397 			mp--;
    398 			continue;
    399 		}
    400 		/*
    401 		 * Do an insertion sort.
    402 		 */
    403 		npgs += btoc(sz);
    404 		for (mp1 = avail; mp1 < mp; mp1++)
    405 			if (s < mp1->start)
    406 				break;
    407 		if (mp1 < mp) {
    408 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    409 			mp1->start = s;
    410 			mp1->size = sz;
    411 		} else {
    412 			mp->start = s;
    413 			mp->size = sz;
    414 		}
    415 	}
    416 
    417 	/*
    418 	 * We cannot do pmap_steal_memory here,
    419 	 * since we don't run with translation enabled yet.
    420 	 */
    421 #ifndef MSGBUFADDR
    422 	/*
    423 	 * allow for msgbuf
    424 	 */
    425 	sz = round_page(MSGBUFSIZE);
    426 	mp = NULL;
    427 	for (mp1 = avail; mp1->size; mp1++)
    428 		if (mp1->size >= sz)
    429 			mp = mp1;
    430 	if (mp == NULL)
    431 		panic("not enough memory?");
    432 
    433 	npgs -= btoc(sz);
    434 	msgbuf_paddr = mp->start + mp->size - sz;
    435 	mp->size -= sz;
    436 	if (mp->size <= 0)
    437 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    438 #endif
    439 
    440 	printf("Loading pages\n");
    441 	for (mp = avail; mp->size; mp++)
    442 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    443 			atop(mp->start), atop(mp->start + mp->size),
    444 			VM_FREELIST_DEFAULT);
    445 
    446 	/*
    447 	 * Initialize kernel pmap and hardware.
    448 	 */
    449 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    450 	pmap_kernel()->pm_ctx = KERNEL_PID;
    451 	nextavail = avail->start;
    452 
    453 
    454 	evcnt_attach_static(&tlbhit_ev);
    455 	evcnt_attach_static(&tlbmiss_ev);
    456 	evcnt_attach_static(&tlbflush_ev);
    457 	evcnt_attach_static(&tlbenter_ev);
    458 	printf("Done\n");
    459 }
    460 
    461 /*
    462  * Restrict given range to physical memory
    463  *
    464  * (Used by /dev/mem)
    465  */
    466 void
    467 pmap_real_memory(paddr_t *start, psize_t *size)
    468 {
    469 	struct mem_region *mp;
    470 
    471 	for (mp = mem; mp->size; mp++) {
    472 		if (*start + *size > mp->start &&
    473 		    *start < mp->start + mp->size) {
    474 			if (*start < mp->start) {
    475 				*size -= mp->start - *start;
    476 				*start = mp->start;
    477 			}
    478 			if (*start + *size > mp->start + mp->size)
    479 				*size = mp->start + mp->size - *start;
    480 			return;
    481 		}
    482 	}
    483 	*size = 0;
    484 }
    485 
    486 /*
    487  * Initialize anything else for pmap handling.
    488  * Called during vm_init().
    489  */
    490 void
    491 pmap_init(void)
    492 {
    493 	struct pv_entry *pv;
    494 	vsize_t sz;
    495 	vaddr_t addr;
    496 	int i, s;
    497 	int bank;
    498 	char *attr;
    499 
    500 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    501 	sz = round_page(sz);
    502 	addr = uvm_km_zalloc(kernel_map, sz);
    503 	s = splvm();
    504 	pv = pv_table = (struct pv_entry *)addr;
    505 	for (i = npgs; --i >= 0;)
    506 		pv++->pv_pm = NULL;
    507 	pmap_attrib = (char *)pv;
    508 	memset(pv, 0, npgs);
    509 
    510 	pv = pv_table;
    511 	attr = pmap_attrib;
    512 	for (bank = 0; bank < vm_nphysseg; bank++) {
    513 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    514 		vm_physmem[bank].pmseg.pvent = pv;
    515 		vm_physmem[bank].pmseg.attrs = attr;
    516 		pv += sz;
    517 		attr += sz;
    518 	}
    519 
    520 	pmap_initialized = 1;
    521 	splx(s);
    522 
    523 	/* Setup a pool for additional pvlist structures */
    524 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    525 }
    526 
    527 /*
    528  * How much virtual space is available to the kernel?
    529  */
    530 void
    531 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    532 {
    533 
    534 #if 0
    535 	/*
    536 	 * Reserve one segment for kernel virtual memory
    537 	 */
    538 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    539 	*end = *start + SEGMENT_LENGTH;
    540 #else
    541 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    542 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    543 #endif
    544 }
    545 
    546 #ifdef PMAP_GROWKERNEL
    547 /*
    548  * Preallocate kernel page tables to a specified VA.
    549  * This simply loops through the first TTE for each
    550  * page table from the beginning of the kernel pmap,
    551  * reads the entry, and if the result is
    552  * zero (either invalid entry or no page table) it stores
    553  * a zero there, populating page tables in the process.
    554  * This is not the most efficient technique but i don't
    555  * expect it to be called that often.
    556  */
    557 extern struct vm_page *vm_page_alloc1 __P((void));
    558 extern void vm_page_free1 __P((struct vm_page *));
    559 
    560 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    561 
    562 vaddr_t
    563 pmap_growkernel(maxkvaddr)
    564         vaddr_t maxkvaddr;
    565 {
    566 	int s;
    567 	int seg;
    568 	paddr_t pg;
    569 	struct pmap *pm = pmap_kernel();
    570 
    571 	s = splvm();
    572 
    573 	/* Align with the start of a page table */
    574 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    575 	     kbreak += PTMAP) {
    576 		seg = STIDX(kbreak);
    577 
    578 		if (pte_find(pm, kbreak)) continue;
    579 
    580 		if (uvm.page_init_done) {
    581 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    582 		} else {
    583 			if (!uvm_page_physget(&pg))
    584 				panic("pmap_growkernel: no memory");
    585 		}
    586 		if (!pg) panic("pmap_growkernel: no pages");
    587 		pmap_zero_page((paddr_t)pg);
    588 
    589 		/* XXX This is based on all phymem being addressable */
    590 		pm->pm_ptbl[seg] = (u_int *)pg;
    591 	}
    592 	splx(s);
    593 	return (kbreak);
    594 }
    595 
    596 /*
    597  *	vm_page_alloc1:
    598  *
    599  *	Allocate and return a memory cell with no associated object.
    600  */
    601 struct vm_page *
    602 vm_page_alloc1()
    603 {
    604 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    605 	if (pg) {
    606 		pg->wire_count = 1;	/* no mappings yet */
    607 		pg->flags &= ~PG_BUSY;	/* never busy */
    608 	}
    609 	return pg;
    610 }
    611 
    612 /*
    613  *	vm_page_free1:
    614  *
    615  *	Returns the given page to the free list,
    616  *	disassociating it with any VM object.
    617  *
    618  *	Object and page must be locked prior to entry.
    619  */
    620 void
    621 vm_page_free1(mem)
    622 	struct vm_page *mem;
    623 {
    624 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    625 		printf("Freeing invalid page %p\n", mem);
    626 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    627 		Debugger();
    628 		return;
    629 	}
    630 	mem->flags |= PG_BUSY;
    631 	mem->wire_count = 0;
    632 	uvm_pagefree(mem);
    633 }
    634 #endif
    635 
    636 /*
    637  * Create and return a physical map.
    638  */
    639 struct pmap *
    640 pmap_create(void)
    641 {
    642 	struct pmap *pm;
    643 
    644 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    645 	memset((caddr_t)pm, 0, sizeof *pm);
    646 	pmap_pinit(pm);
    647 	return pm;
    648 }
    649 
    650 /*
    651  * Initialize a preallocated and zeroed pmap structure.
    652  */
    653 void
    654 pmap_pinit(struct pmap *pm)
    655 {
    656 	int i;
    657 
    658 	/*
    659 	 * Allocate some segment registers for this pmap.
    660 	 */
    661 	pm->pm_refs = 1;
    662 	for (i = 0; i < STSZ; i++)
    663 		pm->pm_ptbl[i] = NULL;
    664 }
    665 
    666 /*
    667  * Add a reference to the given pmap.
    668  */
    669 void
    670 pmap_reference(struct pmap *pm)
    671 {
    672 
    673 	pm->pm_refs++;
    674 }
    675 
    676 /*
    677  * Retire the given pmap from service.
    678  * Should only be called if the map contains no valid mappings.
    679  */
    680 void
    681 pmap_destroy(struct pmap *pm)
    682 {
    683 
    684 	if (--pm->pm_refs == 0) {
    685 		pmap_release(pm);
    686 		free((caddr_t)pm, M_VMPMAP);
    687 	}
    688 }
    689 
    690 /*
    691  * Release any resources held by the given physical map.
    692  * Called when a pmap initialized by pmap_pinit is being released.
    693  */
    694 static void
    695 pmap_release(struct pmap *pm)
    696 {
    697 	int i;
    698 
    699 	for (i = 0; i < STSZ; i++)
    700 		if (pm->pm_ptbl[i]) {
    701 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
    702 			pm->pm_ptbl[i] = NULL;
    703 		}
    704 	if (pm->pm_ctx) ctx_free(pm);
    705 }
    706 
    707 /*
    708  * Copy the range specified by src_addr/len
    709  * from the source map to the range dst_addr/len
    710  * in the destination map.
    711  *
    712  * This routine is only advisory and need not do anything.
    713  */
    714 void
    715 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    716 	  vsize_t len, vaddr_t src_addr)
    717 {
    718 }
    719 
    720 /*
    721  * Require that all active physical maps contain no
    722  * incorrect entries NOW.
    723  */
    724 void
    725 pmap_update(struct pmap *pmap)
    726 {
    727 }
    728 
    729 /*
    730  * Garbage collects the physical map system for
    731  * pages which are no longer used.
    732  * Success need not be guaranteed -- that is, there
    733  * may well be pages which are not referenced, but
    734  * others may be collected.
    735  * Called by the pageout daemon when pages are scarce.
    736  */
    737 void
    738 pmap_collect(struct pmap *pm)
    739 {
    740 }
    741 
    742 /*
    743  * Fill the given physical page with zeroes.
    744  */
    745 void
    746 pmap_zero_page(paddr_t pa)
    747 {
    748 
    749 #ifdef PPC_4XX_NOCACHE
    750 	memset((caddr_t)pa, 0, NBPG);
    751 #else
    752 	int i;
    753 
    754 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
    755 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    756 		pa += CACHELINESIZE;
    757 	}
    758 #endif
    759 }
    760 
    761 /*
    762  * Copy the given physical source page to its destination.
    763  */
    764 void
    765 pmap_copy_page(paddr_t src, paddr_t dst)
    766 {
    767 
    768 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
    769 	dcache_flush_page(dst);
    770 }
    771 
    772 /*
    773  * This returns whether this is the first mapping of a page.
    774  */
    775 static inline int
    776 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    777 {
    778 	struct pv_entry *pv, *npv = NULL;
    779 	int s;
    780 
    781 	if (!pmap_initialized)
    782 		return 0;
    783 
    784 	s = splvm();
    785 
    786 	pv = pa_to_pv(pa);
    787 	if (!pv->pv_pm) {
    788 		/*
    789 		 * No entries yet, use header as the first entry.
    790 		 */
    791 		pv->pv_va = va;
    792 		pv->pv_pm = pm;
    793 		pv->pv_next = NULL;
    794 	} else {
    795 		/*
    796 		 * There is at least one other VA mapping this page.
    797 		 * Place this entry after the header.
    798 		 */
    799 		npv = pool_get(&pv_pool, PR_WAITOK);
    800 		if (!npv) return (0);
    801 		npv->pv_va = va;
    802 		npv->pv_pm = pm;
    803 		npv->pv_next = pv->pv_next;
    804 		pv->pv_next = npv;
    805 	}
    806 	splx(s);
    807 	return (1);
    808 }
    809 
    810 static void
    811 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    812 {
    813 	struct pv_entry *pv, *npv;
    814 
    815 	/*
    816 	 * Remove from the PV table.
    817 	 */
    818 	pv = pa_to_pv(pa);
    819 	if (!pv) return;
    820 
    821 	/*
    822 	 * If it is the first entry on the list, it is actually
    823 	 * in the header and we must copy the following entry up
    824 	 * to the header.  Otherwise we must search the list for
    825 	 * the entry.  In either case we free the now unused entry.
    826 	 */
    827 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    828 		if ((npv = pv->pv_next)) {
    829 			*pv = *npv;
    830 			pool_put(&pv_pool, npv);
    831 		} else
    832 			pv->pv_pm = NULL;
    833 	} else {
    834 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    835 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    836 				break;
    837 		if (npv) {
    838 			pv->pv_next = npv->pv_next;
    839 			pool_put(&pv_pool, npv);
    840 		}
    841 	}
    842 }
    843 
    844 /*
    845  * Insert physical page at pa into the given pmap at virtual address va.
    846  */
    847 int
    848 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    849 {
    850 	int s;
    851 	u_int tte;
    852 	int managed;
    853 
    854 	/*
    855 	 * Have to remove any existing mapping first.
    856 	 */
    857 	pmap_remove(pm, va, va + NBPG);
    858 
    859 	if (flags & PMAP_WIRED) flags |= prot;
    860 
    861 	/* If it has no protections don't bother w/the rest */
    862 	if (!(flags & VM_PROT_ALL))
    863 		return (0);
    864 
    865 	managed = 0;
    866 	if (vm_physseg_find(atop(pa), NULL) != -1)
    867 		managed = 1;
    868 
    869 	/*
    870 	 * Generate TTE.
    871 	 *
    872 	 * XXXX
    873 	 *
    874 	 * Since the kernel does not handle execution privileges properly,
    875 	 * we will handle read and execute permissions together.
    876 	 */
    877 	tte = TTE_PA(pa) | TTE_EX;
    878 	/* XXXX -- need to support multiple page sizes. */
    879 	tte |= TTE_SZ_16K;
    880 #ifdef	DIAGNOSTIC
    881 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    882 		(PME_NOCACHE | PME_WRITETHROUG))
    883 		panic("pmap_enter: uncached & writethrough\n");
    884 #endif
    885 	if (flags & PME_NOCACHE)
    886 		/* Must be I/O mapping */
    887 		tte |= TTE_I | TTE_G;
    888 #ifdef PPC_4XX_NOCACHE
    889 	tte |= TTE_I;
    890 #else
    891 	else if (flags & PME_WRITETHROUG)
    892 		/* Uncached and writethrough are not compatible */
    893 		tte |= TTE_W;
    894 #endif
    895 	if (pm == pmap_kernel())
    896 		tte |= TTE_ZONE(ZONE_PRIV);
    897 	else
    898 		tte |= TTE_ZONE(ZONE_USER);
    899 
    900 	if (flags & VM_PROT_WRITE)
    901 		tte |= TTE_WR;
    902 
    903 	/*
    904 	 * Now record mapping for later back-translation.
    905 	 */
    906 	if (pmap_initialized && managed) {
    907 		char *attr;
    908 
    909 		if (!pmap_enter_pv(pm, va, pa)) {
    910 			/* Could not enter pv on a managed page */
    911 			return 1;
    912 		}
    913 
    914 		/* Now set attributes. */
    915 		attr = pa_to_attr(pa);
    916 #ifdef DIAGNOSTIC
    917 		if (!attr)
    918 			panic("managed but no attr\n");
    919 #endif
    920 		if (flags & VM_PROT_ALL)
    921 			*attr |= PTE_HI_REF;
    922 		if (flags & VM_PROT_WRITE)
    923 			*attr |= PTE_HI_CHG;
    924 	}
    925 
    926 	s = splvm();
    927 	pm->pm_stats.resident_count++;
    928 
    929 	/* Insert page into page table. */
    930 	pte_enter(pm, va, tte);
    931 
    932 	/* If this is a real fault, enter it in the tlb */
    933 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    934 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    935 	}
    936 	splx(s);
    937 
    938 	/* Flush the real memory from the instruction cache. */
    939 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    940 		__syncicache((void *)pa, PAGE_SIZE);
    941 
    942 	return 0;
    943 }
    944 
    945 void
    946 pmap_unwire(struct pmap *pm, vaddr_t va)
    947 {
    948 	struct pv_entry *pv, *npv;
    949 	paddr_t pa;
    950 	int s = splvm();
    951 
    952         if (pm == NULL) {
    953                 return;
    954         }
    955 
    956 	if (!pmap_extract(pm, va, &pa)) {
    957 		return;
    958 	}
    959 
    960 	va |= PV_WIRED;
    961 
    962 	pv = pa_to_pv(pa);
    963 	if (!pv) return;
    964 
    965 	/*
    966 	 * If it is the first entry on the list, it is actually
    967 	 * in the header and we must copy the following entry up
    968 	 * to the header.  Otherwise we must search the list for
    969 	 * the entry.  In either case we free the now unused entry.
    970 	 */
    971 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    972 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    973 			npv->pv_va &= ~PV_WIRED;
    974 			break;
    975 		}
    976 	}
    977 	splx(s);
    978 }
    979 
    980 void
    981 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    982 {
    983 	int s;
    984 	u_int tte;
    985 	struct pmap *pm = pmap_kernel();
    986 
    987 	/*
    988 	 * Have to remove any existing mapping first.
    989 	 */
    990 
    991 	/*
    992 	 * Generate TTE.
    993 	 *
    994 	 * XXXX
    995 	 *
    996 	 * Since the kernel does not handle execution privileges properly,
    997 	 * we will handle read and execute permissions together.
    998 	 */
    999 	tte = 0;
   1000 	if (prot & VM_PROT_ALL) {
   1001 
   1002 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
   1003 		/* XXXX -- need to support multiple page sizes. */
   1004 		tte |= TTE_SZ_16K;
   1005 #ifdef DIAGNOSTIC
   1006 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
   1007 			(PME_NOCACHE | PME_WRITETHROUG))
   1008 			panic("pmap_kenter_pa: uncached & writethrough\n");
   1009 #endif
   1010 		if (prot & PME_NOCACHE)
   1011 			/* Must be I/O mapping */
   1012 			tte |= TTE_I | TTE_G;
   1013 #ifdef PPC_4XX_NOCACHE
   1014 		tte |= TTE_I;
   1015 #else
   1016 		else if (prot & PME_WRITETHROUG)
   1017 			/* Uncached and writethrough are not compatible */
   1018 			tte |= TTE_W;
   1019 #endif
   1020 		if (prot & VM_PROT_WRITE)
   1021 			tte |= TTE_WR;
   1022 	}
   1023 
   1024 	s = splvm();
   1025 	pm->pm_stats.resident_count++;
   1026 
   1027 	/* Insert page into page table. */
   1028 	pte_enter(pm, va, tte);
   1029 	splx(s);
   1030 }
   1031 
   1032 void
   1033 pmap_kremove(vaddr_t va, vsize_t len)
   1034 {
   1035 
   1036 	while (len > 0) {
   1037 		pte_enter(pmap_kernel(), va, 0);
   1038 		va += PAGE_SIZE;
   1039 		len -= PAGE_SIZE;
   1040 	}
   1041 }
   1042 
   1043 /*
   1044  * Remove the given range of mapping entries.
   1045  */
   1046 void
   1047 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
   1048 {
   1049 	int s;
   1050 	paddr_t pa;
   1051 	volatile u_int *ptp;
   1052 
   1053 	s = splvm();
   1054 	while (va < endva) {
   1055 
   1056 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1057 			pa = TTE_PA(pa);
   1058 			pmap_remove_pv(pm, va, pa);
   1059 			*ptp = 0;
   1060 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1061 			pm->pm_stats.resident_count--;
   1062 		}
   1063 		va += NBPG;
   1064 	}
   1065 
   1066 	splx(s);
   1067 }
   1068 
   1069 /*
   1070  * Get the physical page address for the given pmap/virtual address.
   1071  */
   1072 boolean_t
   1073 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1074 {
   1075 	int seg = STIDX(va);
   1076 	int ptn = PTIDX(va);
   1077 	u_int pa = 0;
   1078 	int s = splvm();
   1079 
   1080 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1081 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1082 	}
   1083 	splx(s);
   1084 	return (pa != 0);
   1085 }
   1086 
   1087 /*
   1088  * Lower the protection on the specified range of this pmap.
   1089  *
   1090  * There are only two cases: either the protection is going to 0,
   1091  * or it is going to read-only.
   1092  */
   1093 void
   1094 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1095 {
   1096 	volatile u_int *ptp;
   1097 	int s;
   1098 
   1099 	if (prot & VM_PROT_READ) {
   1100 		s = splvm();
   1101 		while (sva < eva) {
   1102 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1103 				*ptp &= ~TTE_WR;
   1104 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1105 			}
   1106 			sva += NBPG;
   1107 		}
   1108 		splx(s);
   1109 		return;
   1110 	}
   1111 	pmap_remove(pm, sva, eva);
   1112 }
   1113 
   1114 boolean_t
   1115 check_attr(struct vm_page *pg, u_int mask, int clear)
   1116 {
   1117 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1118 	int s;
   1119 	char *attr;
   1120 	int rv;
   1121 
   1122 	/*
   1123 	 * First modify bits in cache.
   1124 	 */
   1125 	s = splvm();
   1126 	attr = pa_to_attr(pa);
   1127 	if (attr == NULL)
   1128 		return FALSE;
   1129 
   1130 	rv = ((*attr & mask) != 0);
   1131 	if (clear)
   1132 		*attr &= ~mask;
   1133 
   1134 	splx(s);
   1135 	return rv;
   1136 }
   1137 
   1138 
   1139 /*
   1140  * Lower the protection on the specified physical page.
   1141  *
   1142  * There are only two cases: either the protection is going to 0,
   1143  * or it is going to read-only.
   1144  */
   1145 void
   1146 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1147 {
   1148 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1149 	vaddr_t va;
   1150 	struct pv_entry *pvh, *pv, *npv;
   1151 	struct pmap *pm;
   1152 
   1153 	pvh = pa_to_pv(pa);
   1154 	if (pvh == NULL)
   1155 		return;
   1156 
   1157 	/* Handle extra pvs which may be deleted in the operation */
   1158 	for (pv = pvh->pv_next; pv; pv = npv) {
   1159 		npv = pv->pv_next;
   1160 
   1161 		pm = pv->pv_pm;
   1162 		va = pv->pv_va;
   1163 		pmap_protect(pm, va, va+NBPG, prot);
   1164 	}
   1165 	/* Now check the head pv */
   1166 	if (pvh->pv_pm) {
   1167 		pv = pvh;
   1168 		pm = pv->pv_pm;
   1169 		va = pv->pv_va;
   1170 		pmap_protect(pm, va, va+NBPG, prot);
   1171 	}
   1172 }
   1173 
   1174 /*
   1175  * Activate the address space for the specified process.  If the process
   1176  * is the current process, load the new MMU context.
   1177  */
   1178 void
   1179 pmap_activate(struct proc *p)
   1180 {
   1181 #if 0
   1182 	struct pcb *pcb = &p->p_addr->u_pcb;
   1183 	pmap_t pmap = p->p_vmspace->vm_map.pmap;
   1184 
   1185 	/*
   1186 	 * XXX Normally performed in cpu_fork().
   1187 	 */
   1188 	printf("pmap_activate(%p), pmap=%p\n",p,pmap);
   1189 	if (pcb->pcb_pm != pmap) {
   1190 		pcb->pcb_pm = pmap;
   1191 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1192 		    (paddr_t *)&pcb->pcb_pmreal);
   1193 	}
   1194 
   1195 	if (p == curproc) {
   1196 		/* Store pointer to new current pmap. */
   1197 		curpm = pcb->pcb_pmreal;
   1198 	}
   1199 #endif
   1200 }
   1201 
   1202 /*
   1203  * Deactivate the specified process's address space.
   1204  */
   1205 void
   1206 pmap_deactivate(struct proc *p)
   1207 {
   1208 }
   1209 
   1210 /*
   1211  * Synchronize caches corresponding to [addr, addr+len) in p.
   1212  */
   1213 void
   1214 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1215 {
   1216 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1217 	int msr, ctx, opid;
   1218 
   1219 
   1220 	/*
   1221 	 * Need to turn off IMMU and switch to user context.
   1222 	 * (icbi uses DMMU).
   1223 	 */
   1224 	if (!(ctx = pm->pm_ctx)) {
   1225 		/* No context -- assign it one */
   1226 		ctx_alloc(pm);
   1227 		ctx = pm->pm_ctx;
   1228 	}
   1229 	__asm __volatile("mfmsr %0;"
   1230 		"li %1, 0x20;"
   1231 		"andc %1,%0,%1;"
   1232 		"mtmsr %1;"
   1233 		"sync;isync;"
   1234 		"mfpid %1;"
   1235 		"mtpid %2;"
   1236 		"sync; isync;"
   1237 		"1:"
   1238 		"dcbf 0,%3;"
   1239 		"icbi 0,%3;"
   1240 		"addi %3,%3,32;"
   1241 		"addic. %4,%4,-32;"
   1242 		"bge 1b;"
   1243 		"mtpid %1;"
   1244 		"mtmsr %0;"
   1245 		"sync; isync"
   1246 		: "=&r" (msr), "=&r" (opid)
   1247 		: "r" (ctx), "r" (va), "r" (len));
   1248 }
   1249 
   1250 
   1251 /* This has to be done in real mode !!! */
   1252 void
   1253 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1254 {
   1255 	u_long i, found;
   1256 	u_long msr;
   1257 
   1258 	/* If there's no context then it can't be mapped. */
   1259 	if (!pid) return;
   1260 
   1261 	asm("mfpid %1;"			/* Save PID */
   1262 		"mfmsr %2;"		/* Save MSR */
   1263 		"li %0,0;"		/* Now clear MSR */
   1264 		"mtmsr %0;"
   1265 		"mtpid %4;"		/* Set PID */
   1266 		"sync;"
   1267 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1268 		"sync;"
   1269 		"mtpid %1;"		/* Restore PID */
   1270 		"mtmsr %2;"		/* Restore MSR */
   1271 		"sync;isync;"
   1272 		"li %1,1;"
   1273 		"beq 1f;"
   1274 		"li %1,0;"
   1275 		"1:"
   1276 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1277 		: "r" (va), "r" (pid));
   1278 	if (found && !TLB_LOCKED(i)) {
   1279 
   1280 		/* Now flush translation */
   1281 		asm volatile(
   1282 			"tlbwe %0,%1,0;"
   1283 			"sync;isync;"
   1284 			: : "r" (0), "r" (i));
   1285 
   1286 		tlb_info[i].ti_ctx = 0;
   1287 		tlb_info[i].ti_flags = 0;
   1288 		tlbnext = i;
   1289 		/* Successful flushes */
   1290 		tlbflush_ev.ev_count++;
   1291 	}
   1292 }
   1293 
   1294 void
   1295 ppc4xx_tlb_flush_all(void)
   1296 {
   1297 	u_long i;
   1298 
   1299 	for (i = 0; i < NTLB; i++)
   1300 		if (!TLB_LOCKED(i)) {
   1301 			asm volatile(
   1302 				"tlbwe %0,%1,0;"
   1303 				"sync;isync;"
   1304 				: : "r" (0), "r" (i));
   1305 			tlb_info[i].ti_ctx = 0;
   1306 			tlb_info[i].ti_flags = 0;
   1307 		}
   1308 
   1309 	asm volatile("sync;isync");
   1310 }
   1311 
   1312 /* Find a TLB entry to evict. */
   1313 static int
   1314 ppc4xx_tlb_find_victim(void)
   1315 {
   1316 	int flags;
   1317 
   1318 	for (;;) {
   1319 		if (++tlbnext >= NTLB)
   1320 			tlbnext = TLB_NRESERVED;
   1321 		flags = tlb_info[tlbnext].ti_flags;
   1322 		if (!(flags & TLBF_USED) ||
   1323 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1324 			u_long va, stack = (u_long)&va;
   1325 
   1326 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1327 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1328 			     (flags & TLBF_USED)) {
   1329 				/* Kernel stack page */
   1330 				flags |= TLBF_USED;
   1331 				tlb_info[tlbnext].ti_flags = flags;
   1332 			} else {
   1333 				/* Found it! */
   1334 				return (tlbnext);
   1335 			}
   1336 		} else {
   1337 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1338 		}
   1339 	}
   1340 }
   1341 
   1342 void
   1343 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1344 {
   1345 	u_long th, tl, idx;
   1346 	tlbpid_t pid;
   1347 	u_short msr;
   1348 	int s;
   1349 
   1350 	tlbenter_ev.ev_count++;
   1351 
   1352 	th = (va & TLB_EPN_MASK) |
   1353 		(((pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT) << TLB_SIZE_SHFT) |
   1354 		TLB_VALID;
   1355 	tl = pte & ~(TTE_SZ_MASK|TTE_ENDIAN);
   1356 
   1357 	s = splhigh();
   1358 	idx = ppc4xx_tlb_find_victim();
   1359 
   1360 #ifdef DIAGNOSTIC
   1361 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1362 		panic("ppc4xx_tlb_enter: repacing entry %ld\n", idx);
   1363 	}
   1364 #endif
   1365 
   1366 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1367 	tlb_info[idx].ti_ctx = ctx;
   1368 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1369 
   1370 	asm volatile(
   1371 		"mfmsr %0;"			/* Save MSR */
   1372 		"li %1,0;"
   1373 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1374 		"mtmsr %1;"			/* Clear MSR */
   1375 		"mfpid %1;"			/* Save old PID */
   1376 		"mtpid %2;"			/* Load translation ctx */
   1377 		"sync; isync;"
   1378 #ifdef DEBUG
   1379 		"andi. %3,%3,63;"
   1380 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1381 #endif
   1382 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1383 		"sync; isync;"
   1384 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1385 		"sync; isync;"
   1386 	: "=&r" (msr), "=&r" (pid)
   1387 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1388 	splx(s);
   1389 }
   1390 
   1391 void
   1392 ppc4xx_tlb_unpin(int i)
   1393 {
   1394 
   1395 	if (i == -1)
   1396 		for (i = 0; i < TLB_NRESERVED; i++)
   1397 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1398 	else
   1399 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1400 }
   1401 
   1402 void
   1403 ppc4xx_tlb_init(void)
   1404 {
   1405 	int i;
   1406 
   1407 	/* Mark reserved TLB entries */
   1408 	for (i = 0; i < TLB_NRESERVED; i++) {
   1409 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1410 		tlb_info[i].ti_ctx = KERNEL_PID;
   1411 	}
   1412 
   1413 	/* Setup security zones */
   1414 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1415 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1416 	 * Z3 - full access regardless of TLB entry permissions
   1417 	 */
   1418 
   1419 	asm volatile(
   1420 		"mtspr %0,%1;"
   1421 		"sync;"
   1422 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1423 }
   1424 
   1425 
   1426 /*
   1427  * We should pass the ctx in from trap code.
   1428  */
   1429 int
   1430 pmap_tlbmiss(vaddr_t va, int ctx)
   1431 {
   1432 	volatile u_int *pte;
   1433 	u_long tte;
   1434 
   1435 	tlbmiss_ev.ev_count++;
   1436 
   1437 	/*
   1438 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1439 	 */
   1440 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1441 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1442 		if (pte == NULL) {
   1443 			/* Map unmanaged addresses directly for kernel access */
   1444 			return 1;
   1445 		}
   1446 		tte = *pte;
   1447 		if (tte == 0) {
   1448 			return 1;
   1449 		}
   1450 	} else {
   1451 		/* Create a 16MB writeable mapping. */
   1452 #ifdef PPC_4XX_NOCACHE
   1453 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1454 #else
   1455 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1456 #endif
   1457 	}
   1458 	tlbhit_ev.ev_count++;
   1459 	ppc4xx_tlb_enter(ctx, va, tte);
   1460 
   1461 	return 0;
   1462 }
   1463 
   1464 /*
   1465  * Flush all the entries matching a context from the TLB.
   1466  */
   1467 static int
   1468 ctx_flush(int cnum)
   1469 {
   1470 	int i;
   1471 
   1472 	/* We gotta steal this context */
   1473 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1474 		if (tlb_info[i].ti_ctx == cnum) {
   1475 			/* Can't steal ctx if it has a locked entry. */
   1476 			if (TLB_LOCKED(i)) {
   1477 #ifdef DIAGNOSTIC
   1478 				printf("ctx_flush: can't invalidate "
   1479 					"locked mapping %d "
   1480 					"for context %d\n", i, cnum);
   1481 				Debugger();
   1482 #endif
   1483 				return (1);
   1484 			}
   1485 #ifdef DIAGNOSTIC
   1486 			if (i < TLB_NRESERVED)
   1487 				panic("TLB entry %d not locked\n", i);
   1488 #endif
   1489 			/* Invalidate particular TLB entry regardless of locked status */
   1490 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1491 			tlb_info[i].ti_flags = 0;
   1492 		}
   1493 	}
   1494 	return (0);
   1495 }
   1496 
   1497 /*
   1498  * Allocate a context.  If necessary, steal one from someone else.
   1499  *
   1500  * The new context is flushed from the TLB before returning.
   1501  */
   1502 int
   1503 ctx_alloc(struct pmap *pm)
   1504 {
   1505 	int s, cnum;
   1506 	static int next = MINCTX;
   1507 
   1508 	if (pm == pmap_kernel()) {
   1509 #ifdef DIAGNOSTIC
   1510 		printf("ctx_alloc: kernel pmap!\n");
   1511 #endif
   1512 		return (0);
   1513 	}
   1514 	s = splvm();
   1515 
   1516 	/* Find a likely context. */
   1517 	cnum = next;
   1518 	do {
   1519 		if ((++cnum) > NUMCTX)
   1520 			cnum = MINCTX;
   1521 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1522 
   1523 	/* Now clean it out */
   1524 oops:
   1525 	if (cnum < MINCTX)
   1526 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1527 	if (ctx_flush(cnum)) {
   1528 		/* oops -- something's wired. */
   1529 		if ((++cnum) > NUMCTX)
   1530 			cnum = MINCTX;
   1531 		goto oops;
   1532 	}
   1533 
   1534 	if (ctxbusy[cnum]) {
   1535 #ifdef DEBUG
   1536 		/* We should identify this pmap and clear it */
   1537 		printf("Warning: stealing context %d\n", cnum);
   1538 #endif
   1539 		ctxbusy[cnum]->pm_ctx = 0;
   1540 	}
   1541 	ctxbusy[cnum] = pm;
   1542 	next = cnum;
   1543 	splx(s);
   1544 	pm->pm_ctx = cnum;
   1545 
   1546 	return cnum;
   1547 }
   1548 
   1549 /*
   1550  * Give away a context.
   1551  */
   1552 void
   1553 ctx_free(struct pmap *pm)
   1554 {
   1555 	int oldctx;
   1556 
   1557 	oldctx = pm->pm_ctx;
   1558 
   1559 	if (oldctx == 0)
   1560 		panic("ctx_free: freeing kernel context");
   1561 #ifdef DIAGNOSTIC
   1562 	if (ctxbusy[oldctx] == 0)
   1563 		printf("ctx_free: freeing free context %d\n", oldctx);
   1564 	if (ctxbusy[oldctx] != pm) {
   1565 		printf("ctx_free: freeing someone esle's context\n "
   1566 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1567 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1568 		Debugger();
   1569 	}
   1570 #endif
   1571 	/* We should verify it has not been stolen and reallocated... */
   1572 	ctxbusy[oldctx] = NULL;
   1573 	ctx_flush(oldctx);
   1574 }
   1575 
   1576 
   1577 #ifdef DEBUG
   1578 /*
   1579  * Test ref/modify handling.
   1580  */
   1581 void pmap_testout __P((void));
   1582 void
   1583 pmap_testout()
   1584 {
   1585 	vaddr_t va;
   1586 	volatile int *loc;
   1587 	int val = 0;
   1588 	paddr_t pa;
   1589 	struct vm_page *pg;
   1590 	int ref, mod;
   1591 
   1592 	/* Allocate a page */
   1593 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
   1594 	loc = (int*)va;
   1595 
   1596 	pmap_extract(pmap_kernel(), va, &pa);
   1597 	pg = PHYS_TO_VM_PAGE(pa);
   1598 	pmap_unwire(pmap_kernel(), va);
   1599 
   1600 	pmap_remove(pmap_kernel(), va, va+1);
   1601 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1602 	pmap_update(pmap_kernel());
   1603 
   1604 	/* Now clear reference and modify */
   1605 	ref = pmap_clear_reference(pg);
   1606 	mod = pmap_clear_modify(pg);
   1607 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1608 	       (void *)(u_long)va, (long)pa,
   1609 	       ref, mod);
   1610 
   1611 	/* Check it's properly cleared */
   1612 	ref = pmap_is_referenced(pg);
   1613 	mod = pmap_is_modified(pg);
   1614 	printf("Checking cleared page: ref %d, mod %d\n",
   1615 	       ref, mod);
   1616 
   1617 	/* Reference page */
   1618 	val = *loc;
   1619 
   1620 	ref = pmap_is_referenced(pg);
   1621 	mod = pmap_is_modified(pg);
   1622 	printf("Referenced page: ref %d, mod %d val %x\n",
   1623 	       ref, mod, val);
   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 	/* Modify page */
   1633 	*loc = 1;
   1634 
   1635 	ref = pmap_is_referenced(pg);
   1636 	mod = pmap_is_modified(pg);
   1637 	printf("Modified page: ref %d, mod %d\n",
   1638 	       ref, mod);
   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 	/* Check it's properly cleared */
   1648 	ref = pmap_is_referenced(pg);
   1649 	mod = pmap_is_modified(pg);
   1650 	printf("Checking cleared page: ref %d, mod %d\n",
   1651 	       ref, mod);
   1652 
   1653 	/* Modify page */
   1654 	*loc = 1;
   1655 
   1656 	ref = pmap_is_referenced(pg);
   1657 	mod = pmap_is_modified(pg);
   1658 	printf("Modified page: ref %d, mod %d\n",
   1659 	       ref, mod);
   1660 
   1661 	/* Check pmap_protect() */
   1662 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1663 	pmap_update(pmap_kernel());
   1664 	ref = pmap_is_referenced(pg);
   1665 	mod = pmap_is_modified(pg);
   1666 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1667 	       ref, mod);
   1668 
   1669 	/* Now clear reference and modify */
   1670 	ref = pmap_clear_reference(pg);
   1671 	mod = pmap_clear_modify(pg);
   1672 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1673 	       (void *)(u_long)va, (long)pa,
   1674 	       ref, mod);
   1675 
   1676 	/* Reference page */
   1677 	val = *loc;
   1678 
   1679 	ref = pmap_is_referenced(pg);
   1680 	mod = pmap_is_modified(pg);
   1681 	printf("Referenced page: ref %d, mod %d val %x\n",
   1682 	       ref, mod, val);
   1683 
   1684 	/* Now clear reference and modify */
   1685 	ref = pmap_clear_reference(pg);
   1686 	mod = pmap_clear_modify(pg);
   1687 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1688 	       (void *)(u_long)va, (long)pa,
   1689 	       ref, mod);
   1690 
   1691 	/* Modify page */
   1692 #if 0
   1693 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1694 	pmap_update(pmap_kernel());
   1695 #endif
   1696 	*loc = 1;
   1697 
   1698 	ref = pmap_is_referenced(pg);
   1699 	mod = pmap_is_modified(pg);
   1700 	printf("Modified page: ref %d, mod %d\n",
   1701 	       ref, mod);
   1702 
   1703 	/* Check pmap_protect() */
   1704 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1705 	pmap_update(pmap_kernel());
   1706 	ref = pmap_is_referenced(pg);
   1707 	mod = pmap_is_modified(pg);
   1708 	printf("pmap_protect(): ref %d, mod %d\n",
   1709 	       ref, mod);
   1710 
   1711 	/* Now clear reference and modify */
   1712 	ref = pmap_clear_reference(pg);
   1713 	mod = pmap_clear_modify(pg);
   1714 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1715 	       (void *)(u_long)va, (long)pa,
   1716 	       ref, mod);
   1717 
   1718 	/* Reference page */
   1719 	val = *loc;
   1720 
   1721 	ref = pmap_is_referenced(pg);
   1722 	mod = pmap_is_modified(pg);
   1723 	printf("Referenced page: ref %d, mod %d val %x\n",
   1724 	       ref, mod, val);
   1725 
   1726 	/* Now clear reference and modify */
   1727 	ref = pmap_clear_reference(pg);
   1728 	mod = pmap_clear_modify(pg);
   1729 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1730 	       (void *)(u_long)va, (long)pa,
   1731 	       ref, mod);
   1732 
   1733 	/* Modify page */
   1734 #if 0
   1735 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1736 	pmap_update(pmap_kernel());
   1737 #endif
   1738 	*loc = 1;
   1739 
   1740 	ref = pmap_is_referenced(pg);
   1741 	mod = pmap_is_modified(pg);
   1742 	printf("Modified page: ref %d, mod %d\n",
   1743 	       ref, mod);
   1744 
   1745 	/* Check pmap_pag_protect() */
   1746 	pmap_page_protect(pg, VM_PROT_READ);
   1747 	ref = pmap_is_referenced(pg);
   1748 	mod = pmap_is_modified(pg);
   1749 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1750 	       ref, mod);
   1751 
   1752 	/* Now clear reference and modify */
   1753 	ref = pmap_clear_reference(pg);
   1754 	mod = pmap_clear_modify(pg);
   1755 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1756 	       (void *)(u_long)va, (long)pa,
   1757 	       ref, mod);
   1758 
   1759 	/* Reference page */
   1760 	val = *loc;
   1761 
   1762 	ref = pmap_is_referenced(pg);
   1763 	mod = pmap_is_modified(pg);
   1764 	printf("Referenced page: ref %d, mod %d val %x\n",
   1765 	       ref, mod, val);
   1766 
   1767 	/* Now clear reference and modify */
   1768 	ref = pmap_clear_reference(pg);
   1769 	mod = pmap_clear_modify(pg);
   1770 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1771 	       (void *)(u_long)va, (long)pa,
   1772 	       ref, mod);
   1773 
   1774 	/* Modify page */
   1775 #if 0
   1776 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1777 	pmap_update(pmap_kernel());
   1778 #endif
   1779 	*loc = 1;
   1780 
   1781 	ref = pmap_is_referenced(pg);
   1782 	mod = pmap_is_modified(pg);
   1783 	printf("Modified page: ref %d, mod %d\n",
   1784 	       ref, mod);
   1785 
   1786 	/* Check pmap_pag_protect() */
   1787 	pmap_page_protect(pg, VM_PROT_NONE);
   1788 	ref = pmap_is_referenced(pg);
   1789 	mod = pmap_is_modified(pg);
   1790 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1791 	       ref, mod);
   1792 
   1793 	/* Now clear reference and modify */
   1794 	ref = pmap_clear_reference(pg);
   1795 	mod = pmap_clear_modify(pg);
   1796 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1797 	       (void *)(u_long)va, (long)pa,
   1798 	       ref, mod);
   1799 
   1800 
   1801 	/* Reference page */
   1802 	val = *loc;
   1803 
   1804 	ref = pmap_is_referenced(pg);
   1805 	mod = pmap_is_modified(pg);
   1806 	printf("Referenced page: ref %d, mod %d val %x\n",
   1807 	       ref, mod, val);
   1808 
   1809 	/* Now clear reference and modify */
   1810 	ref = pmap_clear_reference(pg);
   1811 	mod = pmap_clear_modify(pg);
   1812 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1813 	       (void *)(u_long)va, (long)pa,
   1814 	       ref, mod);
   1815 
   1816 	/* Modify page */
   1817 #if 0
   1818 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1819 	pmap_update(pmap_kernel());
   1820 #endif
   1821 	*loc = 1;
   1822 
   1823 	ref = pmap_is_referenced(pg);
   1824 	mod = pmap_is_modified(pg);
   1825 	printf("Modified page: ref %d, mod %d\n",
   1826 	       ref, mod);
   1827 
   1828 	/* Unmap page */
   1829 	pmap_remove(pmap_kernel(), va, va+1);
   1830 	pmap_update(pmap_kernel());
   1831 	ref = pmap_is_referenced(pg);
   1832 	mod = pmap_is_modified(pg);
   1833 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1834 
   1835 	/* Now clear reference and modify */
   1836 	ref = pmap_clear_reference(pg);
   1837 	mod = pmap_clear_modify(pg);
   1838 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1839 	       (void *)(u_long)va, (long)pa, ref, mod);
   1840 
   1841 	/* Check it's properly cleared */
   1842 	ref = pmap_is_referenced(pg);
   1843 	mod = pmap_is_modified(pg);
   1844 	printf("Checking cleared page: ref %d, mod %d\n",
   1845 	       ref, mod);
   1846 
   1847 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1848 		VM_PROT_ALL|PMAP_WIRED);
   1849 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
   1850 }
   1851 #endif
   1852