Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.22
      1 /*	$NetBSD: pmap.c,v 1.22 2003/07/03 13:18:42 scw Exp $	*/
      2 
      3 /*
      4  * Copyright 2001 Wasabi Systems, Inc.
      5  * All rights reserved.
      6  *
      7  * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, Inc.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed for the NetBSD Project by
     20  *      Wasabi Systems, Inc.
     21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     22  *    or promote products derived from this software without specific prior
     23  *    written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     35  * POSSIBILITY OF SUCH DAMAGE.
     36  */
     37 
     38 /*
     39  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
     40  * Copyright (C) 1995, 1996 TooLs GmbH.
     41  * All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by TooLs GmbH.
     54  * 4. The name of TooLs GmbH may not be used to endorse or promote products
     55  *    derived from this software without specific prior written permission.
     56  *
     57  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
     58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     60  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     61  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
     62  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
     63  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     64  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
     65  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
     66  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     67  */
     68 
     69 #include <sys/param.h>
     70 #include <sys/malloc.h>
     71 #include <sys/proc.h>
     72 #include <sys/user.h>
     73 #include <sys/queue.h>
     74 #include <sys/systm.h>
     75 #include <sys/pool.h>
     76 #include <sys/device.h>
     77 
     78 #include <uvm/uvm.h>
     79 
     80 #include <machine/cpu.h>
     81 #include <machine/pcb.h>
     82 #include <machine/powerpc.h>
     83 
     84 #include <powerpc/spr.h>
     85 #include <machine/tlb.h>
     86 
     87 /*
     88  * kernmap is an array of PTEs large enough to map in
     89  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     90  */
     91 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
     92 caddr_t kernmap;
     93 
     94 #define MINCTX		2
     95 #define NUMCTX		256
     96 volatile struct pmap *ctxbusy[NUMCTX];
     97 
     98 #define TLBF_USED	0x1
     99 #define	TLBF_REF	0x2
    100 #define	TLBF_LOCKED	0x4
    101 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    102 typedef struct tlb_info_s {
    103 	char	ti_flags;
    104 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    105 	u_int	ti_va;
    106 } tlb_info_t;
    107 
    108 volatile tlb_info_t tlb_info[NTLB];
    109 /* We'll use a modified FIFO replacement policy cause it's cheap */
    110 volatile int tlbnext = TLB_NRESERVED;
    111 
    112 u_long dtlb_miss_count = 0;
    113 u_long itlb_miss_count = 0;
    114 u_long ktlb_miss_count = 0;
    115 u_long utlb_miss_count = 0;
    116 
    117 /* Event counters */
    118 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    119 	NULL, "cpu", "tlbmiss");
    120 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    121 	NULL, "cpu", "tlbhit");
    122 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    123 	NULL, "cpu", "tlbflush");
    124 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    125 	NULL, "cpu", "tlbenter");
    126 
    127 struct pmap kernel_pmap_;
    128 
    129 int physmem;
    130 static int npgs;
    131 static u_int nextavail;
    132 #ifndef MSGBUFADDR
    133 extern paddr_t msgbuf_paddr;
    134 #endif
    135 
    136 static struct mem_region *mem, *avail;
    137 
    138 /*
    139  * This is a cache of referenced/modified bits.
    140  * Bits herein are shifted by ATTRSHFT.
    141  */
    142 static char *pmap_attrib;
    143 
    144 #define PV_WIRED	0x1
    145 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    146 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    147 
    148 struct pv_entry {
    149 	struct pv_entry *pv_next;	/* Linked list of mappings */
    150 	vaddr_t pv_va;			/* virtual address of mapping */
    151 	struct pmap *pv_pm;
    152 };
    153 
    154 struct pv_entry *pv_table;
    155 static struct pool pv_pool;
    156 
    157 static int pmap_initialized;
    158 
    159 static int ctx_flush(int);
    160 
    161 inline struct pv_entry *pa_to_pv(paddr_t);
    162 static inline char *pa_to_attr(paddr_t);
    163 
    164 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    165 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    166 
    167 static void pmap_pinit(pmap_t);
    168 static void pmap_release(pmap_t);
    169 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    170 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    171 
    172 
    173 inline struct pv_entry *
    174 pa_to_pv(paddr_t pa)
    175 {
    176 	int bank, pg;
    177 
    178 	bank = vm_physseg_find(atop(pa), &pg);
    179 	if (bank == -1)
    180 		return NULL;
    181 	return &vm_physmem[bank].pmseg.pvent[pg];
    182 }
    183 
    184 static inline char *
    185 pa_to_attr(paddr_t pa)
    186 {
    187 	int bank, pg;
    188 
    189 	bank = vm_physseg_find(atop(pa), &pg);
    190 	if (bank == -1)
    191 		return NULL;
    192 	return &vm_physmem[bank].pmseg.attrs[pg];
    193 }
    194 
    195 /*
    196  * Insert PTE into page table.
    197  */
    198 int
    199 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    200 {
    201 	int seg = STIDX(va);
    202 	int ptn = PTIDX(va);
    203 	u_int oldpte;
    204 	paddr_t pa;
    205 
    206 	if (!pm->pm_ptbl[seg]) {
    207 		/* Don't allocate a page to clear a non-existent mapping. */
    208 		if (!pte) return (0);
    209 		/* Allocate a page XXXX this will sleep! */
    210 		pa = 0;
    211 		pm->pm_ptbl[seg] =
    212 		    (uint *)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
    213 	}
    214 	oldpte = pm->pm_ptbl[seg][ptn];
    215 	pm->pm_ptbl[seg][ptn] = pte;
    216 
    217 	/* Flush entry. */
    218 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    219 	if (oldpte != pte) {
    220 		if (pte == 0)
    221 			pm->pm_stats.resident_count--;
    222 		else
    223 			pm->pm_stats.resident_count++;
    224 	}
    225 	return (1);
    226 }
    227 
    228 /*
    229  * Get a pointer to a PTE in a page table.
    230  */
    231 volatile u_int *
    232 pte_find(struct pmap *pm, vaddr_t va)
    233 {
    234 	int seg = STIDX(va);
    235 	int ptn = PTIDX(va);
    236 
    237 	if (pm->pm_ptbl[seg])
    238 		return (&pm->pm_ptbl[seg][ptn]);
    239 
    240 	return (NULL);
    241 }
    242 
    243 /*
    244  * This is called during initppc, before the system is really initialized.
    245  */
    246 void
    247 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    248 {
    249 	struct mem_region *mp, *mp1;
    250 	int cnt, i;
    251 	u_int s, e, sz;
    252 
    253 	/*
    254 	 * Allocate the kernel page table at the end of
    255 	 * kernel space so it's in the locked TTE.
    256 	 */
    257 	kernmap = (caddr_t)kernelend;
    258 
    259 	/*
    260 	 * Initialize kernel page table.
    261 	 */
    262 	for (i = 0; i < STSZ; i++) {
    263 		pmap_kernel()->pm_ptbl[i] = 0;
    264 	}
    265 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    266 
    267 	/*
    268 	 * Announce page-size to the VM-system
    269 	 */
    270 	uvmexp.pagesize = NBPG;
    271 	uvm_setpagesize();
    272 
    273 	/*
    274 	 * Get memory.
    275 	 */
    276 	mem_regions(&mem, &avail);
    277 	for (mp = mem; mp->size; mp++) {
    278 		physmem += btoc(mp->size);
    279 		printf("+%lx,",mp->size);
    280 	}
    281 	printf("\n");
    282 	ppc4xx_tlb_init();
    283 	/*
    284 	 * Count the number of available entries.
    285 	 */
    286 	for (cnt = 0, mp = avail; mp->size; mp++)
    287 		cnt++;
    288 
    289 	/*
    290 	 * Page align all regions.
    291 	 * Non-page aligned memory isn't very interesting to us.
    292 	 * Also, sort the entries for ascending addresses.
    293 	 */
    294 	kernelstart &= ~PGOFSET;
    295 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    296 	for (mp = avail; mp->size; mp++) {
    297 		s = mp->start;
    298 		e = mp->start + mp->size;
    299 		printf("%08x-%08x -> ",s,e);
    300 		/*
    301 		 * Check whether this region holds all of the kernel.
    302 		 */
    303 		if (s < kernelstart && e > kernelend) {
    304 			avail[cnt].start = kernelend;
    305 			avail[cnt++].size = e - kernelend;
    306 			e = kernelstart;
    307 		}
    308 		/*
    309 		 * Look whether this regions starts within the kernel.
    310 		 */
    311 		if (s >= kernelstart && s < kernelend) {
    312 			if (e <= kernelend)
    313 				goto empty;
    314 			s = kernelend;
    315 		}
    316 		/*
    317 		 * Now look whether this region ends within the kernel.
    318 		 */
    319 		if (e > kernelstart && e <= kernelend) {
    320 			if (s >= kernelstart)
    321 				goto empty;
    322 			e = kernelstart;
    323 		}
    324 		/*
    325 		 * Now page align the start and size of the region.
    326 		 */
    327 		s = round_page(s);
    328 		e = trunc_page(e);
    329 		if (e < s)
    330 			e = s;
    331 		sz = e - s;
    332 		printf("%08x-%08x = %x\n",s,e,sz);
    333 		/*
    334 		 * Check whether some memory is left here.
    335 		 */
    336 		if (sz == 0) {
    337 		empty:
    338 			memmove(mp, mp + 1,
    339 				(cnt - (mp - avail)) * sizeof *mp);
    340 			cnt--;
    341 			mp--;
    342 			continue;
    343 		}
    344 		/*
    345 		 * Do an insertion sort.
    346 		 */
    347 		npgs += btoc(sz);
    348 		for (mp1 = avail; mp1 < mp; mp1++)
    349 			if (s < mp1->start)
    350 				break;
    351 		if (mp1 < mp) {
    352 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    353 			mp1->start = s;
    354 			mp1->size = sz;
    355 		} else {
    356 			mp->start = s;
    357 			mp->size = sz;
    358 		}
    359 	}
    360 
    361 	/*
    362 	 * We cannot do pmap_steal_memory here,
    363 	 * since we don't run with translation enabled yet.
    364 	 */
    365 #ifndef MSGBUFADDR
    366 	/*
    367 	 * allow for msgbuf
    368 	 */
    369 	sz = round_page(MSGBUFSIZE);
    370 	mp = NULL;
    371 	for (mp1 = avail; mp1->size; mp1++)
    372 		if (mp1->size >= sz)
    373 			mp = mp1;
    374 	if (mp == NULL)
    375 		panic("not enough memory?");
    376 
    377 	npgs -= btoc(sz);
    378 	msgbuf_paddr = mp->start + mp->size - sz;
    379 	mp->size -= sz;
    380 	if (mp->size <= 0)
    381 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    382 #endif
    383 
    384 	printf("Loading pages\n");
    385 	for (mp = avail; mp->size; mp++)
    386 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    387 			atop(mp->start), atop(mp->start + mp->size),
    388 			VM_FREELIST_DEFAULT);
    389 
    390 	/*
    391 	 * Initialize kernel pmap and hardware.
    392 	 */
    393 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    394 	pmap_kernel()->pm_ctx = KERNEL_PID;
    395 	nextavail = avail->start;
    396 
    397 
    398 	evcnt_attach_static(&tlbhit_ev);
    399 	evcnt_attach_static(&tlbmiss_ev);
    400 	evcnt_attach_static(&tlbflush_ev);
    401 	evcnt_attach_static(&tlbenter_ev);
    402 	printf("Done\n");
    403 }
    404 
    405 /*
    406  * Restrict given range to physical memory
    407  *
    408  * (Used by /dev/mem)
    409  */
    410 void
    411 pmap_real_memory(paddr_t *start, psize_t *size)
    412 {
    413 	struct mem_region *mp;
    414 
    415 	for (mp = mem; mp->size; mp++) {
    416 		if (*start + *size > mp->start &&
    417 		    *start < mp->start + mp->size) {
    418 			if (*start < mp->start) {
    419 				*size -= mp->start - *start;
    420 				*start = mp->start;
    421 			}
    422 			if (*start + *size > mp->start + mp->size)
    423 				*size = mp->start + mp->size - *start;
    424 			return;
    425 		}
    426 	}
    427 	*size = 0;
    428 }
    429 
    430 /*
    431  * Initialize anything else for pmap handling.
    432  * Called during vm_init().
    433  */
    434 void
    435 pmap_init(void)
    436 {
    437 	struct pv_entry *pv;
    438 	vsize_t sz;
    439 	vaddr_t addr;
    440 	int i, s;
    441 	int bank;
    442 	char *attr;
    443 
    444 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    445 	sz = round_page(sz);
    446 	addr = uvm_km_zalloc(kernel_map, sz);
    447 	s = splvm();
    448 	pv = pv_table = (struct pv_entry *)addr;
    449 	for (i = npgs; --i >= 0;)
    450 		pv++->pv_pm = NULL;
    451 	pmap_attrib = (char *)pv;
    452 	memset(pv, 0, npgs);
    453 
    454 	pv = pv_table;
    455 	attr = pmap_attrib;
    456 	for (bank = 0; bank < vm_nphysseg; bank++) {
    457 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    458 		vm_physmem[bank].pmseg.pvent = pv;
    459 		vm_physmem[bank].pmseg.attrs = attr;
    460 		pv += sz;
    461 		attr += sz;
    462 	}
    463 
    464 	pmap_initialized = 1;
    465 	splx(s);
    466 
    467 	/* Setup a pool for additional pvlist structures */
    468 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    469 }
    470 
    471 /*
    472  * How much virtual space is available to the kernel?
    473  */
    474 void
    475 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    476 {
    477 
    478 #if 0
    479 	/*
    480 	 * Reserve one segment for kernel virtual memory
    481 	 */
    482 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    483 	*end = *start + SEGMENT_LENGTH;
    484 #else
    485 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    486 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    487 #endif
    488 }
    489 
    490 #ifdef PMAP_GROWKERNEL
    491 /*
    492  * Preallocate kernel page tables to a specified VA.
    493  * This simply loops through the first TTE for each
    494  * page table from the beginning of the kernel pmap,
    495  * reads the entry, and if the result is
    496  * zero (either invalid entry or no page table) it stores
    497  * a zero there, populating page tables in the process.
    498  * This is not the most efficient technique but i don't
    499  * expect it to be called that often.
    500  */
    501 extern struct vm_page *vm_page_alloc1 __P((void));
    502 extern void vm_page_free1 __P((struct vm_page *));
    503 
    504 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    505 
    506 vaddr_t
    507 pmap_growkernel(maxkvaddr)
    508 	vaddr_t maxkvaddr;
    509 {
    510 	int s;
    511 	int seg;
    512 	paddr_t pg;
    513 	struct pmap *pm = pmap_kernel();
    514 
    515 	s = splvm();
    516 
    517 	/* Align with the start of a page table */
    518 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    519 	     kbreak += PTMAP) {
    520 		seg = STIDX(kbreak);
    521 
    522 		if (pte_find(pm, kbreak)) continue;
    523 
    524 		if (uvm.page_init_done) {
    525 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    526 		} else {
    527 			if (!uvm_page_physget(&pg))
    528 				panic("pmap_growkernel: no memory");
    529 		}
    530 		if (!pg) panic("pmap_growkernel: no pages");
    531 		pmap_zero_page((paddr_t)pg);
    532 
    533 		/* XXX This is based on all phymem being addressable */
    534 		pm->pm_ptbl[seg] = (u_int *)pg;
    535 	}
    536 	splx(s);
    537 	return (kbreak);
    538 }
    539 
    540 /*
    541  *	vm_page_alloc1:
    542  *
    543  *	Allocate and return a memory cell with no associated object.
    544  */
    545 struct vm_page *
    546 vm_page_alloc1()
    547 {
    548 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    549 	if (pg) {
    550 		pg->wire_count = 1;	/* no mappings yet */
    551 		pg->flags &= ~PG_BUSY;	/* never busy */
    552 	}
    553 	return pg;
    554 }
    555 
    556 /*
    557  *	vm_page_free1:
    558  *
    559  *	Returns the given page to the free list,
    560  *	disassociating it with any VM object.
    561  *
    562  *	Object and page must be locked prior to entry.
    563  */
    564 void
    565 vm_page_free1(mem)
    566 	struct vm_page *mem;
    567 {
    568 #ifdef DIAGNOSTIC
    569 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    570 		printf("Freeing invalid page %p\n", mem);
    571 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    572 #ifdef DDB
    573 		Debugger();
    574 #endif
    575 		return;
    576 	}
    577 #endif
    578 	mem->flags |= PG_BUSY;
    579 	mem->wire_count = 0;
    580 	uvm_pagefree(mem);
    581 }
    582 #endif
    583 
    584 /*
    585  * Create and return a physical map.
    586  */
    587 struct pmap *
    588 pmap_create(void)
    589 {
    590 	struct pmap *pm;
    591 
    592 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    593 	memset((caddr_t)pm, 0, sizeof *pm);
    594 	pmap_pinit(pm);
    595 	return pm;
    596 }
    597 
    598 /*
    599  * Initialize a preallocated and zeroed pmap structure.
    600  */
    601 void
    602 pmap_pinit(struct pmap *pm)
    603 {
    604 	int i;
    605 
    606 	/*
    607 	 * Allocate some segment registers for this pmap.
    608 	 */
    609 	pm->pm_refs = 1;
    610 	for (i = 0; i < STSZ; i++)
    611 		pm->pm_ptbl[i] = NULL;
    612 }
    613 
    614 /*
    615  * Add a reference to the given pmap.
    616  */
    617 void
    618 pmap_reference(struct pmap *pm)
    619 {
    620 
    621 	pm->pm_refs++;
    622 }
    623 
    624 /*
    625  * Retire the given pmap from service.
    626  * Should only be called if the map contains no valid mappings.
    627  */
    628 void
    629 pmap_destroy(struct pmap *pm)
    630 {
    631 
    632 	if (--pm->pm_refs == 0) {
    633 		pmap_release(pm);
    634 		free((caddr_t)pm, M_VMPMAP);
    635 	}
    636 }
    637 
    638 /*
    639  * Release any resources held by the given physical map.
    640  * Called when a pmap initialized by pmap_pinit is being released.
    641  */
    642 static void
    643 pmap_release(struct pmap *pm)
    644 {
    645 	int i;
    646 
    647 	for (i = 0; i < STSZ; i++)
    648 		if (pm->pm_ptbl[i]) {
    649 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
    650 			    PAGE_SIZE);
    651 			pm->pm_ptbl[i] = NULL;
    652 		}
    653 	if (pm->pm_ctx) ctx_free(pm);
    654 }
    655 
    656 /*
    657  * Copy the range specified by src_addr/len
    658  * from the source map to the range dst_addr/len
    659  * in the destination map.
    660  *
    661  * This routine is only advisory and need not do anything.
    662  */
    663 void
    664 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    665 	  vsize_t len, vaddr_t src_addr)
    666 {
    667 }
    668 
    669 /*
    670  * Require that all active physical maps contain no
    671  * incorrect entries NOW.
    672  */
    673 void
    674 pmap_update(struct pmap *pmap)
    675 {
    676 }
    677 
    678 /*
    679  * Garbage collects the physical map system for
    680  * pages which are no longer used.
    681  * Success need not be guaranteed -- that is, there
    682  * may well be pages which are not referenced, but
    683  * others may be collected.
    684  * Called by the pageout daemon when pages are scarce.
    685  */
    686 void
    687 pmap_collect(struct pmap *pm)
    688 {
    689 }
    690 
    691 /*
    692  * Fill the given physical page with zeroes.
    693  */
    694 void
    695 pmap_zero_page(paddr_t pa)
    696 {
    697 
    698 #ifdef PPC_4XX_NOCACHE
    699 	memset((caddr_t)pa, 0, PAGE_SIZE);
    700 #else
    701 	int i;
    702 
    703 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
    704 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    705 		pa += CACHELINESIZE;
    706 	}
    707 #endif
    708 }
    709 
    710 /*
    711  * Copy the given physical source page to its destination.
    712  */
    713 void
    714 pmap_copy_page(paddr_t src, paddr_t dst)
    715 {
    716 
    717 	memcpy((caddr_t)dst, (caddr_t)src, PAGE_SIZE);
    718 	dcache_flush_page(dst);
    719 }
    720 
    721 /*
    722  * This returns whether this is the first mapping of a page.
    723  */
    724 static inline int
    725 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    726 {
    727 	struct pv_entry *pv, *npv = NULL;
    728 	int s;
    729 
    730 	if (!pmap_initialized)
    731 		return 0;
    732 
    733 	s = splvm();
    734 
    735 	pv = pa_to_pv(pa);
    736 for (npv = pv; npv; npv = npv->pv_next)
    737 if (npv->pv_va == va && npv->pv_pm == pm) {
    738 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    739 #ifdef DDB
    740 Debugger();
    741 #endif
    742 return (1);
    743 }
    744 
    745 	if (!pv->pv_pm) {
    746 		/*
    747 		 * No entries yet, use header as the first entry.
    748 		 */
    749 		pv->pv_va = va;
    750 		pv->pv_pm = pm;
    751 		pv->pv_next = NULL;
    752 	} else {
    753 		/*
    754 		 * There is at least one other VA mapping this page.
    755 		 * Place this entry after the header.
    756 		 */
    757 		npv = pool_get(&pv_pool, PR_WAITOK);
    758 		if (!npv) return (0);
    759 		npv->pv_va = va;
    760 		npv->pv_pm = pm;
    761 		npv->pv_next = pv->pv_next;
    762 		pv->pv_next = npv;
    763 	}
    764 	splx(s);
    765 	return (1);
    766 }
    767 
    768 static void
    769 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    770 {
    771 	struct pv_entry *pv, *npv;
    772 
    773 	/*
    774 	 * Remove from the PV table.
    775 	 */
    776 	pv = pa_to_pv(pa);
    777 	if (!pv) return;
    778 
    779 	/*
    780 	 * If it is the first entry on the list, it is actually
    781 	 * in the header and we must copy the following entry up
    782 	 * to the header.  Otherwise we must search the list for
    783 	 * the entry.  In either case we free the now unused entry.
    784 	 */
    785 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    786 		if ((npv = pv->pv_next)) {
    787 			*pv = *npv;
    788 			pool_put(&pv_pool, npv);
    789 		} else
    790 			pv->pv_pm = NULL;
    791 	} else {
    792 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    793 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    794 				break;
    795 		if (npv) {
    796 			pv->pv_next = npv->pv_next;
    797 			pool_put(&pv_pool, npv);
    798 		}
    799 	}
    800 }
    801 
    802 /*
    803  * Insert physical page at pa into the given pmap at virtual address va.
    804  */
    805 int
    806 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    807 {
    808 	int s;
    809 	u_int tte;
    810 	int managed;
    811 
    812 	/*
    813 	 * Have to remove any existing mapping first.
    814 	 */
    815 	pmap_remove(pm, va, va + PAGE_SIZE);
    816 
    817 	if (flags & PMAP_WIRED) flags |= prot;
    818 
    819 	/* If it has no protections don't bother w/the rest */
    820 	if (!(flags & VM_PROT_ALL))
    821 		return (0);
    822 
    823 	managed = 0;
    824 	if (vm_physseg_find(atop(pa), NULL) != -1)
    825 		managed = 1;
    826 
    827 	/*
    828 	 * Generate TTE.
    829 	 *
    830 	 * XXXX
    831 	 *
    832 	 * Since the kernel does not handle execution privileges properly,
    833 	 * we will handle read and execute permissions together.
    834 	 */
    835 	tte = TTE_PA(pa) | TTE_EX;
    836 	/* XXXX -- need to support multiple page sizes. */
    837 	tte |= TTE_SZ_16K;
    838 #ifdef	DIAGNOSTIC
    839 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    840 		(PME_NOCACHE | PME_WRITETHROUG))
    841 		panic("pmap_enter: uncached & writethrough");
    842 #endif
    843 	if (flags & PME_NOCACHE)
    844 		/* Must be I/O mapping */
    845 		tte |= TTE_I | TTE_G;
    846 #ifdef PPC_4XX_NOCACHE
    847 	tte |= TTE_I;
    848 #else
    849 	else if (flags & PME_WRITETHROUG)
    850 		/* Uncached and writethrough are not compatible */
    851 		tte |= TTE_W;
    852 #endif
    853 	if (pm == pmap_kernel())
    854 		tte |= TTE_ZONE(ZONE_PRIV);
    855 	else
    856 		tte |= TTE_ZONE(ZONE_USER);
    857 
    858 	if (flags & VM_PROT_WRITE)
    859 		tte |= TTE_WR;
    860 
    861 	/*
    862 	 * Now record mapping for later back-translation.
    863 	 */
    864 	if (pmap_initialized && managed) {
    865 		char *attr;
    866 
    867 		if (!pmap_enter_pv(pm, va, pa)) {
    868 			/* Could not enter pv on a managed page */
    869 			return 1;
    870 		}
    871 
    872 		/* Now set attributes. */
    873 		attr = pa_to_attr(pa);
    874 #ifdef DIAGNOSTIC
    875 		if (!attr)
    876 			panic("managed but no attr");
    877 #endif
    878 		if (flags & VM_PROT_ALL)
    879 			*attr |= PTE_HI_REF;
    880 		if (flags & VM_PROT_WRITE)
    881 			*attr |= PTE_HI_CHG;
    882 	}
    883 
    884 	s = splvm();
    885 
    886 	/* Insert page into page table. */
    887 	pte_enter(pm, va, tte);
    888 
    889 	/* If this is a real fault, enter it in the tlb */
    890 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    891 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    892 	}
    893 	splx(s);
    894 
    895 	/* Flush the real memory from the instruction cache. */
    896 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    897 		__syncicache((void *)pa, PAGE_SIZE);
    898 
    899 	return 0;
    900 }
    901 
    902 void
    903 pmap_unwire(struct pmap *pm, vaddr_t va)
    904 {
    905 	struct pv_entry *pv, *npv;
    906 	paddr_t pa;
    907 	int s = splvm();
    908 
    909 	if (pm == NULL) {
    910 		return;
    911 	}
    912 
    913 	if (!pmap_extract(pm, va, &pa)) {
    914 		return;
    915 	}
    916 
    917 	va |= PV_WIRED;
    918 
    919 	pv = pa_to_pv(pa);
    920 	if (!pv) return;
    921 
    922 	/*
    923 	 * If it is the first entry on the list, it is actually
    924 	 * in the header and we must copy the following entry up
    925 	 * to the header.  Otherwise we must search the list for
    926 	 * the entry.  In either case we free the now unused entry.
    927 	 */
    928 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    929 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    930 			npv->pv_va &= ~PV_WIRED;
    931 			break;
    932 		}
    933 	}
    934 	splx(s);
    935 }
    936 
    937 void
    938 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    939 {
    940 	int s;
    941 	u_int tte;
    942 	struct pmap *pm = pmap_kernel();
    943 
    944 	/*
    945 	 * Have to remove any existing mapping first.
    946 	 */
    947 
    948 	/*
    949 	 * Generate TTE.
    950 	 *
    951 	 * XXXX
    952 	 *
    953 	 * Since the kernel does not handle execution privileges properly,
    954 	 * we will handle read and execute permissions together.
    955 	 */
    956 	tte = 0;
    957 	if (prot & VM_PROT_ALL) {
    958 
    959 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    960 		/* XXXX -- need to support multiple page sizes. */
    961 		tte |= TTE_SZ_16K;
    962 #ifdef DIAGNOSTIC
    963 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    964 			(PME_NOCACHE | PME_WRITETHROUG))
    965 			panic("pmap_kenter_pa: uncached & writethrough");
    966 #endif
    967 		if (prot & PME_NOCACHE)
    968 			/* Must be I/O mapping */
    969 			tte |= TTE_I | TTE_G;
    970 #ifdef PPC_4XX_NOCACHE
    971 		tte |= TTE_I;
    972 #else
    973 		else if (prot & PME_WRITETHROUG)
    974 			/* Uncached and writethrough are not compatible */
    975 			tte |= TTE_W;
    976 #endif
    977 		if (prot & VM_PROT_WRITE)
    978 			tte |= TTE_WR;
    979 	}
    980 
    981 	s = splvm();
    982 
    983 	/* Insert page into page table. */
    984 	pte_enter(pm, va, tte);
    985 	splx(s);
    986 }
    987 
    988 void
    989 pmap_kremove(vaddr_t va, vsize_t len)
    990 {
    991 
    992 	while (len > 0) {
    993 		pte_enter(pmap_kernel(), va, 0);
    994 		va += PAGE_SIZE;
    995 		len -= PAGE_SIZE;
    996 	}
    997 }
    998 
    999 /*
   1000  * Remove the given range of mapping entries.
   1001  */
   1002 void
   1003 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
   1004 {
   1005 	int s;
   1006 	paddr_t pa;
   1007 	volatile u_int *ptp;
   1008 
   1009 	s = splvm();
   1010 	while (va < endva) {
   1011 
   1012 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1013 			pa = TTE_PA(pa);
   1014 			pmap_remove_pv(pm, va, pa);
   1015 			*ptp = 0;
   1016 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1017 			pm->pm_stats.resident_count--;
   1018 		}
   1019 		va += PAGE_SIZE;
   1020 	}
   1021 
   1022 	splx(s);
   1023 }
   1024 
   1025 /*
   1026  * Get the physical page address for the given pmap/virtual address.
   1027  */
   1028 boolean_t
   1029 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1030 {
   1031 	int seg = STIDX(va);
   1032 	int ptn = PTIDX(va);
   1033 	u_int pa = 0;
   1034 	int s = splvm();
   1035 
   1036 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1037 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1038 	}
   1039 	splx(s);
   1040 	return (pa != 0);
   1041 }
   1042 
   1043 /*
   1044  * Lower the protection on the specified range of this pmap.
   1045  *
   1046  * There are only two cases: either the protection is going to 0,
   1047  * or it is going to read-only.
   1048  */
   1049 void
   1050 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1051 {
   1052 	volatile u_int *ptp;
   1053 	int s;
   1054 
   1055 	if (prot & VM_PROT_READ) {
   1056 		s = splvm();
   1057 		while (sva < eva) {
   1058 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1059 				*ptp &= ~TTE_WR;
   1060 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1061 			}
   1062 			sva += PAGE_SIZE;
   1063 		}
   1064 		splx(s);
   1065 		return;
   1066 	}
   1067 	pmap_remove(pm, sva, eva);
   1068 }
   1069 
   1070 boolean_t
   1071 check_attr(struct vm_page *pg, u_int mask, int clear)
   1072 {
   1073 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1074 	int s;
   1075 	char *attr;
   1076 	int rv;
   1077 
   1078 	/*
   1079 	 * First modify bits in cache.
   1080 	 */
   1081 	s = splvm();
   1082 	attr = pa_to_attr(pa);
   1083 	if (attr == NULL)
   1084 		return FALSE;
   1085 
   1086 	rv = ((*attr & mask) != 0);
   1087 	if (clear) {
   1088 		*attr &= ~mask;
   1089 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1090 	}
   1091 	splx(s);
   1092 	return rv;
   1093 }
   1094 
   1095 
   1096 /*
   1097  * Lower the protection on the specified physical page.
   1098  *
   1099  * There are only two cases: either the protection is going to 0,
   1100  * or it is going to read-only.
   1101  */
   1102 void
   1103 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1104 {
   1105 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1106 	vaddr_t va;
   1107 	struct pv_entry *pvh, *pv, *npv;
   1108 	struct pmap *pm;
   1109 
   1110 	pvh = pa_to_pv(pa);
   1111 	if (pvh == NULL)
   1112 		return;
   1113 
   1114 	/* Handle extra pvs which may be deleted in the operation */
   1115 	for (pv = pvh->pv_next; pv; pv = npv) {
   1116 		npv = pv->pv_next;
   1117 
   1118 		pm = pv->pv_pm;
   1119 		va = pv->pv_va;
   1120 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1121 	}
   1122 	/* Now check the head pv */
   1123 	if (pvh->pv_pm) {
   1124 		pv = pvh;
   1125 		pm = pv->pv_pm;
   1126 		va = pv->pv_va;
   1127 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1128 	}
   1129 }
   1130 
   1131 /*
   1132  * Activate the address space for the specified process.  If the process
   1133  * is the current process, load the new MMU context.
   1134  */
   1135 void
   1136 pmap_activate(struct lwp *l)
   1137 {
   1138 #if 0
   1139 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1140 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1141 
   1142 	/*
   1143 	 * XXX Normally performed in cpu_fork().
   1144 	 */
   1145 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1146 	if (pcb->pcb_pm != pmap) {
   1147 		pcb->pcb_pm = pmap;
   1148 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1149 		    (paddr_t *)&pcb->pcb_pmreal);
   1150 	}
   1151 
   1152 	if (l == curlwp) {
   1153 		/* Store pointer to new current pmap. */
   1154 		curpm = pcb->pcb_pmreal;
   1155 	}
   1156 #endif
   1157 }
   1158 
   1159 /*
   1160  * Deactivate the specified process's address space.
   1161  */
   1162 void
   1163 pmap_deactivate(struct lwp *l)
   1164 {
   1165 }
   1166 
   1167 /*
   1168  * Synchronize caches corresponding to [addr, addr+len) in p.
   1169  */
   1170 void
   1171 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1172 {
   1173 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1174 	int msr, ctx, opid, step;
   1175 
   1176 
   1177 	step = CACHELINESIZE;
   1178 
   1179 	/*
   1180 	 * Need to turn off IMMU and switch to user context.
   1181 	 * (icbi uses DMMU).
   1182 	 */
   1183 	if (!(ctx = pm->pm_ctx)) {
   1184 		/* No context -- assign it one */
   1185 		ctx_alloc(pm);
   1186 		ctx = pm->pm_ctx;
   1187 	}
   1188 	__asm __volatile("mfmsr %0;"
   1189 		"li %1, 0x20;"
   1190 		"andc %1,%0,%1;"
   1191 		"mtmsr %1;"
   1192 		"sync;isync;"
   1193 		"mfpid %1;"
   1194 		"mtpid %2;"
   1195 		"sync; isync;"
   1196 		"1:"
   1197 		"dcbf 0,%3;"
   1198 		"icbi 0,%3;"
   1199 		"add %3,%3,%5;"
   1200 		"addc. %4,%4,%6;"
   1201 		"bge 1b;"
   1202 		"mtpid %1;"
   1203 		"mtmsr %0;"
   1204 		"sync; isync"
   1205 		: "=&r" (msr), "=&r" (opid)
   1206 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step));
   1207 }
   1208 
   1209 
   1210 /* This has to be done in real mode !!! */
   1211 void
   1212 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1213 {
   1214 	u_long i, found;
   1215 	u_long msr;
   1216 
   1217 	/* If there's no context then it can't be mapped. */
   1218 	if (!pid) return;
   1219 
   1220 	asm("mfpid %1;"			/* Save PID */
   1221 		"mfmsr %2;"		/* Save MSR */
   1222 		"li %0,0;"		/* Now clear MSR */
   1223 		"mtmsr %0;"
   1224 		"mtpid %4;"		/* Set PID */
   1225 		"sync;"
   1226 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1227 		"sync;"
   1228 		"mtpid %1;"		/* Restore PID */
   1229 		"mtmsr %2;"		/* Restore MSR */
   1230 		"sync;isync;"
   1231 		"li %1,1;"
   1232 		"beq 1f;"
   1233 		"li %1,0;"
   1234 		"1:"
   1235 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1236 		: "r" (va), "r" (pid));
   1237 	if (found && !TLB_LOCKED(i)) {
   1238 
   1239 		/* Now flush translation */
   1240 		asm volatile(
   1241 			"tlbwe %0,%1,0;"
   1242 			"sync;isync;"
   1243 			: : "r" (0), "r" (i));
   1244 
   1245 		tlb_info[i].ti_ctx = 0;
   1246 		tlb_info[i].ti_flags = 0;
   1247 		tlbnext = i;
   1248 		/* Successful flushes */
   1249 		tlbflush_ev.ev_count++;
   1250 	}
   1251 }
   1252 
   1253 void
   1254 ppc4xx_tlb_flush_all(void)
   1255 {
   1256 	u_long i;
   1257 
   1258 	for (i = 0; i < NTLB; i++)
   1259 		if (!TLB_LOCKED(i)) {
   1260 			asm volatile(
   1261 				"tlbwe %0,%1,0;"
   1262 				"sync;isync;"
   1263 				: : "r" (0), "r" (i));
   1264 			tlb_info[i].ti_ctx = 0;
   1265 			tlb_info[i].ti_flags = 0;
   1266 		}
   1267 
   1268 	asm volatile("sync;isync");
   1269 }
   1270 
   1271 /* Find a TLB entry to evict. */
   1272 static int
   1273 ppc4xx_tlb_find_victim(void)
   1274 {
   1275 	int flags;
   1276 
   1277 	for (;;) {
   1278 		if (++tlbnext >= NTLB)
   1279 			tlbnext = TLB_NRESERVED;
   1280 		flags = tlb_info[tlbnext].ti_flags;
   1281 		if (!(flags & TLBF_USED) ||
   1282 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1283 			u_long va, stack = (u_long)&va;
   1284 
   1285 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1286 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1287 			     (flags & TLBF_USED)) {
   1288 				/* Kernel stack page */
   1289 				flags |= TLBF_USED;
   1290 				tlb_info[tlbnext].ti_flags = flags;
   1291 			} else {
   1292 				/* Found it! */
   1293 				return (tlbnext);
   1294 			}
   1295 		} else {
   1296 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1297 		}
   1298 	}
   1299 }
   1300 
   1301 void
   1302 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1303 {
   1304 	u_long th, tl, idx;
   1305 	tlbpid_t pid;
   1306 	u_short msr;
   1307 	paddr_t pa;
   1308 	int s, sz;
   1309 
   1310 	tlbenter_ev.ev_count++;
   1311 
   1312 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1313 	pa = (pte & TTE_RPN_MASK(sz));
   1314 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1315 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1316 	tl |= ppc4xx_tlbflags(va, pa);
   1317 
   1318 	s = splhigh();
   1319 	idx = ppc4xx_tlb_find_victim();
   1320 
   1321 #ifdef DIAGNOSTIC
   1322 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1323 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
   1324 	}
   1325 #endif
   1326 
   1327 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1328 	tlb_info[idx].ti_ctx = ctx;
   1329 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1330 
   1331 	asm volatile(
   1332 		"mfmsr %0;"			/* Save MSR */
   1333 		"li %1,0;"
   1334 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1335 		"mtmsr %1;"			/* Clear MSR */
   1336 		"mfpid %1;"			/* Save old PID */
   1337 		"mtpid %2;"			/* Load translation ctx */
   1338 		"sync; isync;"
   1339 #ifdef DEBUG
   1340 		"andi. %3,%3,63;"
   1341 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1342 #endif
   1343 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1344 		"sync; isync;"
   1345 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1346 		"sync; isync;"
   1347 	: "=&r" (msr), "=&r" (pid)
   1348 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1349 	splx(s);
   1350 }
   1351 
   1352 void
   1353 ppc4xx_tlb_unpin(int i)
   1354 {
   1355 
   1356 	if (i == -1)
   1357 		for (i = 0; i < TLB_NRESERVED; i++)
   1358 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1359 	else
   1360 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1361 }
   1362 
   1363 void
   1364 ppc4xx_tlb_init(void)
   1365 {
   1366 	int i;
   1367 
   1368 	/* Mark reserved TLB entries */
   1369 	for (i = 0; i < TLB_NRESERVED; i++) {
   1370 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1371 		tlb_info[i].ti_ctx = KERNEL_PID;
   1372 	}
   1373 
   1374 	/* Setup security zones */
   1375 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1376 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1377 	 * Z3 - full access regardless of TLB entry permissions
   1378 	 */
   1379 
   1380 	asm volatile(
   1381 		"mtspr %0,%1;"
   1382 		"sync;"
   1383 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1384 }
   1385 
   1386 
   1387 /*
   1388  * We should pass the ctx in from trap code.
   1389  */
   1390 int
   1391 pmap_tlbmiss(vaddr_t va, int ctx)
   1392 {
   1393 	volatile u_int *pte;
   1394 	u_long tte;
   1395 
   1396 	tlbmiss_ev.ev_count++;
   1397 
   1398 	/*
   1399 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1400 	 */
   1401 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1402 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1403 		if (pte == NULL) {
   1404 			/* Map unmanaged addresses directly for kernel access */
   1405 			return 1;
   1406 		}
   1407 		tte = *pte;
   1408 		if (tte == 0) {
   1409 			return 1;
   1410 		}
   1411 	} else {
   1412 		/* Create a 16MB writable mapping. */
   1413 #ifdef PPC_4XX_NOCACHE
   1414 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1415 #else
   1416 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1417 #endif
   1418 	}
   1419 	tlbhit_ev.ev_count++;
   1420 	ppc4xx_tlb_enter(ctx, va, tte);
   1421 
   1422 	return 0;
   1423 }
   1424 
   1425 /*
   1426  * Flush all the entries matching a context from the TLB.
   1427  */
   1428 static int
   1429 ctx_flush(int cnum)
   1430 {
   1431 	int i;
   1432 
   1433 	/* We gotta steal this context */
   1434 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1435 		if (tlb_info[i].ti_ctx == cnum) {
   1436 			/* Can't steal ctx if it has a locked entry. */
   1437 			if (TLB_LOCKED(i)) {
   1438 #ifdef DIAGNOSTIC
   1439 				printf("ctx_flush: can't invalidate "
   1440 					"locked mapping %d "
   1441 					"for context %d\n", i, cnum);
   1442 #ifdef DDB
   1443 				Debugger();
   1444 #endif
   1445 #endif
   1446 				return (1);
   1447 			}
   1448 #ifdef DIAGNOSTIC
   1449 			if (i < TLB_NRESERVED)
   1450 				panic("TLB entry %d not locked", i);
   1451 #endif
   1452 			/* Invalidate particular TLB entry regardless of locked status */
   1453 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1454 			tlb_info[i].ti_flags = 0;
   1455 		}
   1456 	}
   1457 	return (0);
   1458 }
   1459 
   1460 /*
   1461  * Allocate a context.  If necessary, steal one from someone else.
   1462  *
   1463  * The new context is flushed from the TLB before returning.
   1464  */
   1465 int
   1466 ctx_alloc(struct pmap *pm)
   1467 {
   1468 	int s, cnum;
   1469 	static int next = MINCTX;
   1470 
   1471 	if (pm == pmap_kernel()) {
   1472 #ifdef DIAGNOSTIC
   1473 		printf("ctx_alloc: kernel pmap!\n");
   1474 #endif
   1475 		return (0);
   1476 	}
   1477 	s = splvm();
   1478 
   1479 	/* Find a likely context. */
   1480 	cnum = next;
   1481 	do {
   1482 		if ((++cnum) > NUMCTX)
   1483 			cnum = MINCTX;
   1484 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1485 
   1486 	/* Now clean it out */
   1487 oops:
   1488 	if (cnum < MINCTX)
   1489 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1490 	if (ctx_flush(cnum)) {
   1491 		/* oops -- something's wired. */
   1492 		if ((++cnum) > NUMCTX)
   1493 			cnum = MINCTX;
   1494 		goto oops;
   1495 	}
   1496 
   1497 	if (ctxbusy[cnum]) {
   1498 #ifdef DEBUG
   1499 		/* We should identify this pmap and clear it */
   1500 		printf("Warning: stealing context %d\n", cnum);
   1501 #endif
   1502 		ctxbusy[cnum]->pm_ctx = 0;
   1503 	}
   1504 	ctxbusy[cnum] = pm;
   1505 	next = cnum;
   1506 	splx(s);
   1507 	pm->pm_ctx = cnum;
   1508 
   1509 	return cnum;
   1510 }
   1511 
   1512 /*
   1513  * Give away a context.
   1514  */
   1515 void
   1516 ctx_free(struct pmap *pm)
   1517 {
   1518 	int oldctx;
   1519 
   1520 	oldctx = pm->pm_ctx;
   1521 
   1522 	if (oldctx == 0)
   1523 		panic("ctx_free: freeing kernel context");
   1524 #ifdef DIAGNOSTIC
   1525 	if (ctxbusy[oldctx] == 0)
   1526 		printf("ctx_free: freeing free context %d\n", oldctx);
   1527 	if (ctxbusy[oldctx] != pm) {
   1528 		printf("ctx_free: freeing someone esle's context\n "
   1529 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1530 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1531 #ifdef DDB
   1532 		Debugger();
   1533 #endif
   1534 	}
   1535 #endif
   1536 	/* We should verify it has not been stolen and reallocated... */
   1537 	ctxbusy[oldctx] = NULL;
   1538 	ctx_flush(oldctx);
   1539 }
   1540 
   1541 
   1542 #ifdef DEBUG
   1543 /*
   1544  * Test ref/modify handling.
   1545  */
   1546 void pmap_testout __P((void));
   1547 void
   1548 pmap_testout()
   1549 {
   1550 	vaddr_t va;
   1551 	volatile int *loc;
   1552 	int val = 0;
   1553 	paddr_t pa;
   1554 	struct vm_page *pg;
   1555 	int ref, mod;
   1556 
   1557 	/* Allocate a page */
   1558 	va = (vaddr_t)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
   1559 	loc = (int*)va;
   1560 
   1561 	pmap_extract(pmap_kernel(), va, &pa);
   1562 	pg = PHYS_TO_VM_PAGE(pa);
   1563 	pmap_unwire(pmap_kernel(), va);
   1564 
   1565 	pmap_remove(pmap_kernel(), va, va+1);
   1566 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1567 	pmap_update(pmap_kernel());
   1568 
   1569 	/* Now clear reference and modify */
   1570 	ref = pmap_clear_reference(pg);
   1571 	mod = pmap_clear_modify(pg);
   1572 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1573 	       (void *)(u_long)va, (long)pa,
   1574 	       ref, mod);
   1575 
   1576 	/* Check it's properly cleared */
   1577 	ref = pmap_is_referenced(pg);
   1578 	mod = pmap_is_modified(pg);
   1579 	printf("Checking cleared page: ref %d, mod %d\n",
   1580 	       ref, mod);
   1581 
   1582 	/* Reference page */
   1583 	val = *loc;
   1584 
   1585 	ref = pmap_is_referenced(pg);
   1586 	mod = pmap_is_modified(pg);
   1587 	printf("Referenced page: ref %d, mod %d val %x\n",
   1588 	       ref, mod, val);
   1589 
   1590 	/* Now clear reference and modify */
   1591 	ref = pmap_clear_reference(pg);
   1592 	mod = pmap_clear_modify(pg);
   1593 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1594 	       (void *)(u_long)va, (long)pa,
   1595 	       ref, mod);
   1596 
   1597 	/* Modify page */
   1598 	*loc = 1;
   1599 
   1600 	ref = pmap_is_referenced(pg);
   1601 	mod = pmap_is_modified(pg);
   1602 	printf("Modified page: ref %d, mod %d\n",
   1603 	       ref, mod);
   1604 
   1605 	/* Now clear reference and modify */
   1606 	ref = pmap_clear_reference(pg);
   1607 	mod = pmap_clear_modify(pg);
   1608 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1609 	       (void *)(u_long)va, (long)pa,
   1610 	       ref, mod);
   1611 
   1612 	/* Check it's properly cleared */
   1613 	ref = pmap_is_referenced(pg);
   1614 	mod = pmap_is_modified(pg);
   1615 	printf("Checking cleared page: ref %d, mod %d\n",
   1616 	       ref, mod);
   1617 
   1618 	/* Modify page */
   1619 	*loc = 1;
   1620 
   1621 	ref = pmap_is_referenced(pg);
   1622 	mod = pmap_is_modified(pg);
   1623 	printf("Modified page: ref %d, mod %d\n",
   1624 	       ref, mod);
   1625 
   1626 	/* Check pmap_protect() */
   1627 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1628 	pmap_update(pmap_kernel());
   1629 	ref = pmap_is_referenced(pg);
   1630 	mod = pmap_is_modified(pg);
   1631 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1632 	       ref, mod);
   1633 
   1634 	/* Now clear reference and modify */
   1635 	ref = pmap_clear_reference(pg);
   1636 	mod = pmap_clear_modify(pg);
   1637 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1638 	       (void *)(u_long)va, (long)pa,
   1639 	       ref, mod);
   1640 
   1641 	/* Reference page */
   1642 	val = *loc;
   1643 
   1644 	ref = pmap_is_referenced(pg);
   1645 	mod = pmap_is_modified(pg);
   1646 	printf("Referenced page: ref %d, mod %d val %x\n",
   1647 	       ref, mod, val);
   1648 
   1649 	/* Now clear reference and modify */
   1650 	ref = pmap_clear_reference(pg);
   1651 	mod = pmap_clear_modify(pg);
   1652 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1653 	       (void *)(u_long)va, (long)pa,
   1654 	       ref, mod);
   1655 
   1656 	/* Modify page */
   1657 #if 0
   1658 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1659 	pmap_update(pmap_kernel());
   1660 #endif
   1661 	*loc = 1;
   1662 
   1663 	ref = pmap_is_referenced(pg);
   1664 	mod = pmap_is_modified(pg);
   1665 	printf("Modified page: ref %d, mod %d\n",
   1666 	       ref, mod);
   1667 
   1668 	/* Check pmap_protect() */
   1669 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1670 	pmap_update(pmap_kernel());
   1671 	ref = pmap_is_referenced(pg);
   1672 	mod = pmap_is_modified(pg);
   1673 	printf("pmap_protect(): ref %d, mod %d\n",
   1674 	       ref, mod);
   1675 
   1676 	/* Now clear reference and modify */
   1677 	ref = pmap_clear_reference(pg);
   1678 	mod = pmap_clear_modify(pg);
   1679 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1680 	       (void *)(u_long)va, (long)pa,
   1681 	       ref, mod);
   1682 
   1683 	/* Reference page */
   1684 	val = *loc;
   1685 
   1686 	ref = pmap_is_referenced(pg);
   1687 	mod = pmap_is_modified(pg);
   1688 	printf("Referenced page: ref %d, mod %d val %x\n",
   1689 	       ref, mod, val);
   1690 
   1691 	/* Now clear reference and modify */
   1692 	ref = pmap_clear_reference(pg);
   1693 	mod = pmap_clear_modify(pg);
   1694 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1695 	       (void *)(u_long)va, (long)pa,
   1696 	       ref, mod);
   1697 
   1698 	/* Modify page */
   1699 #if 0
   1700 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1701 	pmap_update(pmap_kernel());
   1702 #endif
   1703 	*loc = 1;
   1704 
   1705 	ref = pmap_is_referenced(pg);
   1706 	mod = pmap_is_modified(pg);
   1707 	printf("Modified page: ref %d, mod %d\n",
   1708 	       ref, mod);
   1709 
   1710 	/* Check pmap_pag_protect() */
   1711 	pmap_page_protect(pg, VM_PROT_READ);
   1712 	ref = pmap_is_referenced(pg);
   1713 	mod = pmap_is_modified(pg);
   1714 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1715 	       ref, mod);
   1716 
   1717 	/* Now clear reference and modify */
   1718 	ref = pmap_clear_reference(pg);
   1719 	mod = pmap_clear_modify(pg);
   1720 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1721 	       (void *)(u_long)va, (long)pa,
   1722 	       ref, mod);
   1723 
   1724 	/* Reference page */
   1725 	val = *loc;
   1726 
   1727 	ref = pmap_is_referenced(pg);
   1728 	mod = pmap_is_modified(pg);
   1729 	printf("Referenced page: ref %d, mod %d val %x\n",
   1730 	       ref, mod, val);
   1731 
   1732 	/* Now clear reference and modify */
   1733 	ref = pmap_clear_reference(pg);
   1734 	mod = pmap_clear_modify(pg);
   1735 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1736 	       (void *)(u_long)va, (long)pa,
   1737 	       ref, mod);
   1738 
   1739 	/* Modify page */
   1740 #if 0
   1741 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1742 	pmap_update(pmap_kernel());
   1743 #endif
   1744 	*loc = 1;
   1745 
   1746 	ref = pmap_is_referenced(pg);
   1747 	mod = pmap_is_modified(pg);
   1748 	printf("Modified page: ref %d, mod %d\n",
   1749 	       ref, mod);
   1750 
   1751 	/* Check pmap_pag_protect() */
   1752 	pmap_page_protect(pg, VM_PROT_NONE);
   1753 	ref = pmap_is_referenced(pg);
   1754 	mod = pmap_is_modified(pg);
   1755 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1756 	       ref, mod);
   1757 
   1758 	/* Now clear reference and modify */
   1759 	ref = pmap_clear_reference(pg);
   1760 	mod = pmap_clear_modify(pg);
   1761 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1762 	       (void *)(u_long)va, (long)pa,
   1763 	       ref, mod);
   1764 
   1765 
   1766 	/* Reference page */
   1767 	val = *loc;
   1768 
   1769 	ref = pmap_is_referenced(pg);
   1770 	mod = pmap_is_modified(pg);
   1771 	printf("Referenced page: ref %d, mod %d val %x\n",
   1772 	       ref, mod, val);
   1773 
   1774 	/* Now clear reference and modify */
   1775 	ref = pmap_clear_reference(pg);
   1776 	mod = pmap_clear_modify(pg);
   1777 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1778 	       (void *)(u_long)va, (long)pa,
   1779 	       ref, mod);
   1780 
   1781 	/* Modify page */
   1782 #if 0
   1783 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1784 	pmap_update(pmap_kernel());
   1785 #endif
   1786 	*loc = 1;
   1787 
   1788 	ref = pmap_is_referenced(pg);
   1789 	mod = pmap_is_modified(pg);
   1790 	printf("Modified page: ref %d, mod %d\n",
   1791 	       ref, mod);
   1792 
   1793 	/* Unmap page */
   1794 	pmap_remove(pmap_kernel(), va, va+1);
   1795 	pmap_update(pmap_kernel());
   1796 	ref = pmap_is_referenced(pg);
   1797 	mod = pmap_is_modified(pg);
   1798 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1799 
   1800 	/* Now clear reference and modify */
   1801 	ref = pmap_clear_reference(pg);
   1802 	mod = pmap_clear_modify(pg);
   1803 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1804 	       (void *)(u_long)va, (long)pa, ref, mod);
   1805 
   1806 	/* Check it's properly cleared */
   1807 	ref = pmap_is_referenced(pg);
   1808 	mod = pmap_is_modified(pg);
   1809 	printf("Checking cleared page: ref %d, mod %d\n",
   1810 	       ref, mod);
   1811 
   1812 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1813 		VM_PROT_ALL|PMAP_WIRED);
   1814 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE);
   1815 }
   1816 #endif
   1817