Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.19
      1 /*	$NetBSD: pmap.c,v 1.19 2003/04/02 04:22:03 thorpej Exp $	*/
      2 
      3 /*
      4  * Copyright 2001 Wasabi Systems, Inc.
      5  * All rights reserved.
      6  *
      7  * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, Inc.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed for the NetBSD Project by
     20  *      Wasabi Systems, Inc.
     21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     22  *    or promote products derived from this software without specific prior
     23  *    written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     35  * POSSIBILITY OF SUCH DAMAGE.
     36  */
     37 
     38 /*
     39  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
     40  * Copyright (C) 1995, 1996 TooLs GmbH.
     41  * All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by TooLs GmbH.
     54  * 4. The name of TooLs GmbH may not be used to endorse or promote products
     55  *    derived from this software without specific prior written permission.
     56  *
     57  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
     58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     60  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     61  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
     62  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
     63  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     64  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
     65  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
     66  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     67  */
     68 
     69 #include <sys/param.h>
     70 #include <sys/malloc.h>
     71 #include <sys/proc.h>
     72 #include <sys/user.h>
     73 #include <sys/queue.h>
     74 #include <sys/systm.h>
     75 #include <sys/pool.h>
     76 #include <sys/device.h>
     77 
     78 #include <uvm/uvm.h>
     79 
     80 #include <machine/cpu.h>
     81 #include <machine/pcb.h>
     82 #include <machine/powerpc.h>
     83 
     84 #include <powerpc/spr.h>
     85 #include <machine/tlb.h>
     86 
     87 /*
     88  * kernmap is an array of PTEs large enough to map in
     89  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     90  */
     91 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
     92 caddr_t kernmap;
     93 
     94 #define MINCTX		2
     95 #define NUMCTX		256
     96 volatile struct pmap *ctxbusy[NUMCTX];
     97 
     98 #define TLBF_USED	0x1
     99 #define	TLBF_REF	0x2
    100 #define	TLBF_LOCKED	0x4
    101 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    102 typedef struct tlb_info_s {
    103 	char	ti_flags;
    104 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    105 	u_int	ti_va;
    106 } tlb_info_t;
    107 
    108 volatile tlb_info_t tlb_info[NTLB];
    109 /* We'll use a modified FIFO replacement policy cause it's cheap */
    110 volatile int tlbnext = TLB_NRESERVED;
    111 
    112 u_long dtlb_miss_count = 0;
    113 u_long itlb_miss_count = 0;
    114 u_long ktlb_miss_count = 0;
    115 u_long utlb_miss_count = 0;
    116 
    117 /* Event counters */
    118 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    119 	NULL, "cpu", "tlbmiss");
    120 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    121 	NULL, "cpu", "tlbhit");
    122 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    123 	NULL, "cpu", "tlbflush");
    124 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    125 	NULL, "cpu", "tlbenter");
    126 
    127 struct pmap kernel_pmap_;
    128 
    129 int physmem;
    130 static int npgs;
    131 static u_int nextavail;
    132 #ifndef MSGBUFADDR
    133 extern paddr_t msgbuf_paddr;
    134 #endif
    135 
    136 static struct mem_region *mem, *avail;
    137 
    138 /*
    139  * This is a cache of referenced/modified bits.
    140  * Bits herein are shifted by ATTRSHFT.
    141  */
    142 static char *pmap_attrib;
    143 
    144 #define PV_WIRED	0x1
    145 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    146 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    147 
    148 struct pv_entry {
    149 	struct pv_entry *pv_next;	/* Linked list of mappings */
    150 	vaddr_t pv_va;			/* virtual address of mapping */
    151 	struct pmap *pv_pm;
    152 };
    153 
    154 struct pv_entry *pv_table;
    155 static struct pool pv_pool;
    156 
    157 static int pmap_initialized;
    158 
    159 static int ctx_flush(int);
    160 
    161 inline struct pv_entry *pa_to_pv(paddr_t);
    162 static inline char *pa_to_attr(paddr_t);
    163 
    164 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    165 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    166 
    167 static void pmap_pinit(pmap_t);
    168 static void pmap_release(pmap_t);
    169 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    170 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    171 
    172 
    173 inline struct pv_entry *
    174 pa_to_pv(paddr_t pa)
    175 {
    176 	int bank, pg;
    177 
    178 	bank = vm_physseg_find(atop(pa), &pg);
    179 	if (bank == -1)
    180 		return NULL;
    181 	return &vm_physmem[bank].pmseg.pvent[pg];
    182 }
    183 
    184 static inline char *
    185 pa_to_attr(paddr_t pa)
    186 {
    187 	int bank, pg;
    188 
    189 	bank = vm_physseg_find(atop(pa), &pg);
    190 	if (bank == -1)
    191 		return NULL;
    192 	return &vm_physmem[bank].pmseg.attrs[pg];
    193 }
    194 
    195 /*
    196  * Insert PTE into page table.
    197  */
    198 int
    199 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    200 {
    201 	int seg = STIDX(va);
    202 	int ptn = PTIDX(va);
    203 	paddr_t pa;
    204 
    205 	if (!pm->pm_ptbl[seg]) {
    206 		/* Don't allocate a page to clear a non-existent mapping. */
    207 		if (!pte) return (1);
    208 		/* Allocate a page XXXX this will sleep! */
    209 		pa = 0;
    210 		pm->pm_ptbl[seg] =
    211 		    (uint *)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
    212 	}
    213 	pm->pm_ptbl[seg][ptn] = pte;
    214 
    215 	/* Flush entry. */
    216 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    217 	return (1);
    218 }
    219 
    220 /*
    221  * Get a pointer to a PTE in a page table.
    222  */
    223 volatile u_int *
    224 pte_find(struct pmap *pm, vaddr_t va)
    225 {
    226 	int seg = STIDX(va);
    227 	int ptn = PTIDX(va);
    228 
    229 	if (pm->pm_ptbl[seg])
    230 		return (&pm->pm_ptbl[seg][ptn]);
    231 
    232 	return (NULL);
    233 }
    234 
    235 /*
    236  * This is called during initppc, before the system is really initialized.
    237  */
    238 void
    239 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    240 {
    241 	struct mem_region *mp, *mp1;
    242 	int cnt, i;
    243 	u_int s, e, sz;
    244 
    245 	/*
    246 	 * Allocate the kernel page table at the end of
    247 	 * kernel space so it's in the locked TTE.
    248 	 */
    249 	kernmap = (caddr_t)kernelend;
    250 
    251 	/*
    252 	 * Initialize kernel page table.
    253 	 */
    254 	for (i = 0; i < STSZ; i++) {
    255 		pmap_kernel()->pm_ptbl[i] = 0;
    256 	}
    257 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    258 
    259 	/*
    260 	 * Announce page-size to the VM-system
    261 	 */
    262 	uvmexp.pagesize = NBPG;
    263 	uvm_setpagesize();
    264 
    265 	/*
    266 	 * Get memory.
    267 	 */
    268 	mem_regions(&mem, &avail);
    269 	for (mp = mem; mp->size; mp++) {
    270 		physmem += btoc(mp->size);
    271 		printf("+%lx,",mp->size);
    272 	}
    273 	printf("\n");
    274 	ppc4xx_tlb_init();
    275 	/*
    276 	 * Count the number of available entries.
    277 	 */
    278 	for (cnt = 0, mp = avail; mp->size; mp++)
    279 		cnt++;
    280 
    281 	/*
    282 	 * Page align all regions.
    283 	 * Non-page aligned memory isn't very interesting to us.
    284 	 * Also, sort the entries for ascending addresses.
    285 	 */
    286 	kernelstart &= ~PGOFSET;
    287 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    288 	for (mp = avail; mp->size; mp++) {
    289 		s = mp->start;
    290 		e = mp->start + mp->size;
    291 		printf("%08x-%08x -> ",s,e);
    292 		/*
    293 		 * Check whether this region holds all of the kernel.
    294 		 */
    295 		if (s < kernelstart && e > kernelend) {
    296 			avail[cnt].start = kernelend;
    297 			avail[cnt++].size = e - kernelend;
    298 			e = kernelstart;
    299 		}
    300 		/*
    301 		 * Look whether this regions starts within the kernel.
    302 		 */
    303 		if (s >= kernelstart && s < kernelend) {
    304 			if (e <= kernelend)
    305 				goto empty;
    306 			s = kernelend;
    307 		}
    308 		/*
    309 		 * Now look whether this region ends within the kernel.
    310 		 */
    311 		if (e > kernelstart && e <= kernelend) {
    312 			if (s >= kernelstart)
    313 				goto empty;
    314 			e = kernelstart;
    315 		}
    316 		/*
    317 		 * Now page align the start and size of the region.
    318 		 */
    319 		s = round_page(s);
    320 		e = trunc_page(e);
    321 		if (e < s)
    322 			e = s;
    323 		sz = e - s;
    324 		printf("%08x-%08x = %x\n",s,e,sz);
    325 		/*
    326 		 * Check whether some memory is left here.
    327 		 */
    328 		if (sz == 0) {
    329 		empty:
    330 			memmove(mp, mp + 1,
    331 				(cnt - (mp - avail)) * sizeof *mp);
    332 			cnt--;
    333 			mp--;
    334 			continue;
    335 		}
    336 		/*
    337 		 * Do an insertion sort.
    338 		 */
    339 		npgs += btoc(sz);
    340 		for (mp1 = avail; mp1 < mp; mp1++)
    341 			if (s < mp1->start)
    342 				break;
    343 		if (mp1 < mp) {
    344 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    345 			mp1->start = s;
    346 			mp1->size = sz;
    347 		} else {
    348 			mp->start = s;
    349 			mp->size = sz;
    350 		}
    351 	}
    352 
    353 	/*
    354 	 * We cannot do pmap_steal_memory here,
    355 	 * since we don't run with translation enabled yet.
    356 	 */
    357 #ifndef MSGBUFADDR
    358 	/*
    359 	 * allow for msgbuf
    360 	 */
    361 	sz = round_page(MSGBUFSIZE);
    362 	mp = NULL;
    363 	for (mp1 = avail; mp1->size; mp1++)
    364 		if (mp1->size >= sz)
    365 			mp = mp1;
    366 	if (mp == NULL)
    367 		panic("not enough memory?");
    368 
    369 	npgs -= btoc(sz);
    370 	msgbuf_paddr = mp->start + mp->size - sz;
    371 	mp->size -= sz;
    372 	if (mp->size <= 0)
    373 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    374 #endif
    375 
    376 	printf("Loading pages\n");
    377 	for (mp = avail; mp->size; mp++)
    378 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    379 			atop(mp->start), atop(mp->start + mp->size),
    380 			VM_FREELIST_DEFAULT);
    381 
    382 	/*
    383 	 * Initialize kernel pmap and hardware.
    384 	 */
    385 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    386 	pmap_kernel()->pm_ctx = KERNEL_PID;
    387 	nextavail = avail->start;
    388 
    389 
    390 	evcnt_attach_static(&tlbhit_ev);
    391 	evcnt_attach_static(&tlbmiss_ev);
    392 	evcnt_attach_static(&tlbflush_ev);
    393 	evcnt_attach_static(&tlbenter_ev);
    394 	printf("Done\n");
    395 }
    396 
    397 /*
    398  * Restrict given range to physical memory
    399  *
    400  * (Used by /dev/mem)
    401  */
    402 void
    403 pmap_real_memory(paddr_t *start, psize_t *size)
    404 {
    405 	struct mem_region *mp;
    406 
    407 	for (mp = mem; mp->size; mp++) {
    408 		if (*start + *size > mp->start &&
    409 		    *start < mp->start + mp->size) {
    410 			if (*start < mp->start) {
    411 				*size -= mp->start - *start;
    412 				*start = mp->start;
    413 			}
    414 			if (*start + *size > mp->start + mp->size)
    415 				*size = mp->start + mp->size - *start;
    416 			return;
    417 		}
    418 	}
    419 	*size = 0;
    420 }
    421 
    422 /*
    423  * Initialize anything else for pmap handling.
    424  * Called during vm_init().
    425  */
    426 void
    427 pmap_init(void)
    428 {
    429 	struct pv_entry *pv;
    430 	vsize_t sz;
    431 	vaddr_t addr;
    432 	int i, s;
    433 	int bank;
    434 	char *attr;
    435 
    436 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    437 	sz = round_page(sz);
    438 	addr = uvm_km_zalloc(kernel_map, sz);
    439 	s = splvm();
    440 	pv = pv_table = (struct pv_entry *)addr;
    441 	for (i = npgs; --i >= 0;)
    442 		pv++->pv_pm = NULL;
    443 	pmap_attrib = (char *)pv;
    444 	memset(pv, 0, npgs);
    445 
    446 	pv = pv_table;
    447 	attr = pmap_attrib;
    448 	for (bank = 0; bank < vm_nphysseg; bank++) {
    449 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    450 		vm_physmem[bank].pmseg.pvent = pv;
    451 		vm_physmem[bank].pmseg.attrs = attr;
    452 		pv += sz;
    453 		attr += sz;
    454 	}
    455 
    456 	pmap_initialized = 1;
    457 	splx(s);
    458 
    459 	/* Setup a pool for additional pvlist structures */
    460 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    461 }
    462 
    463 /*
    464  * How much virtual space is available to the kernel?
    465  */
    466 void
    467 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    468 {
    469 
    470 #if 0
    471 	/*
    472 	 * Reserve one segment for kernel virtual memory
    473 	 */
    474 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    475 	*end = *start + SEGMENT_LENGTH;
    476 #else
    477 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    478 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    479 #endif
    480 }
    481 
    482 #ifdef PMAP_GROWKERNEL
    483 /*
    484  * Preallocate kernel page tables to a specified VA.
    485  * This simply loops through the first TTE for each
    486  * page table from the beginning of the kernel pmap,
    487  * reads the entry, and if the result is
    488  * zero (either invalid entry or no page table) it stores
    489  * a zero there, populating page tables in the process.
    490  * This is not the most efficient technique but i don't
    491  * expect it to be called that often.
    492  */
    493 extern struct vm_page *vm_page_alloc1 __P((void));
    494 extern void vm_page_free1 __P((struct vm_page *));
    495 
    496 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    497 
    498 vaddr_t
    499 pmap_growkernel(maxkvaddr)
    500 	vaddr_t maxkvaddr;
    501 {
    502 	int s;
    503 	int seg;
    504 	paddr_t pg;
    505 	struct pmap *pm = pmap_kernel();
    506 
    507 	s = splvm();
    508 
    509 	/* Align with the start of a page table */
    510 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    511 	     kbreak += PTMAP) {
    512 		seg = STIDX(kbreak);
    513 
    514 		if (pte_find(pm, kbreak)) continue;
    515 
    516 		if (uvm.page_init_done) {
    517 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    518 		} else {
    519 			if (!uvm_page_physget(&pg))
    520 				panic("pmap_growkernel: no memory");
    521 		}
    522 		if (!pg) panic("pmap_growkernel: no pages");
    523 		pmap_zero_page((paddr_t)pg);
    524 
    525 		/* XXX This is based on all phymem being addressable */
    526 		pm->pm_ptbl[seg] = (u_int *)pg;
    527 	}
    528 	splx(s);
    529 	return (kbreak);
    530 }
    531 
    532 /*
    533  *	vm_page_alloc1:
    534  *
    535  *	Allocate and return a memory cell with no associated object.
    536  */
    537 struct vm_page *
    538 vm_page_alloc1()
    539 {
    540 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    541 	if (pg) {
    542 		pg->wire_count = 1;	/* no mappings yet */
    543 		pg->flags &= ~PG_BUSY;	/* never busy */
    544 	}
    545 	return pg;
    546 }
    547 
    548 /*
    549  *	vm_page_free1:
    550  *
    551  *	Returns the given page to the free list,
    552  *	disassociating it with any VM object.
    553  *
    554  *	Object and page must be locked prior to entry.
    555  */
    556 void
    557 vm_page_free1(mem)
    558 	struct vm_page *mem;
    559 {
    560 #ifdef DIAGNOSTIC
    561 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    562 		printf("Freeing invalid page %p\n", mem);
    563 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    564 #ifdef DDB
    565 		Debugger();
    566 #endif
    567 		return;
    568 	}
    569 #endif
    570 	mem->flags |= PG_BUSY;
    571 	mem->wire_count = 0;
    572 	uvm_pagefree(mem);
    573 }
    574 #endif
    575 
    576 /*
    577  * Create and return a physical map.
    578  */
    579 struct pmap *
    580 pmap_create(void)
    581 {
    582 	struct pmap *pm;
    583 
    584 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    585 	memset((caddr_t)pm, 0, sizeof *pm);
    586 	pmap_pinit(pm);
    587 	return pm;
    588 }
    589 
    590 /*
    591  * Initialize a preallocated and zeroed pmap structure.
    592  */
    593 void
    594 pmap_pinit(struct pmap *pm)
    595 {
    596 	int i;
    597 
    598 	/*
    599 	 * Allocate some segment registers for this pmap.
    600 	 */
    601 	pm->pm_refs = 1;
    602 	for (i = 0; i < STSZ; i++)
    603 		pm->pm_ptbl[i] = NULL;
    604 }
    605 
    606 /*
    607  * Add a reference to the given pmap.
    608  */
    609 void
    610 pmap_reference(struct pmap *pm)
    611 {
    612 
    613 	pm->pm_refs++;
    614 }
    615 
    616 /*
    617  * Retire the given pmap from service.
    618  * Should only be called if the map contains no valid mappings.
    619  */
    620 void
    621 pmap_destroy(struct pmap *pm)
    622 {
    623 
    624 	if (--pm->pm_refs == 0) {
    625 		pmap_release(pm);
    626 		free((caddr_t)pm, M_VMPMAP);
    627 	}
    628 }
    629 
    630 /*
    631  * Release any resources held by the given physical map.
    632  * Called when a pmap initialized by pmap_pinit is being released.
    633  */
    634 static void
    635 pmap_release(struct pmap *pm)
    636 {
    637 	int i;
    638 
    639 	for (i = 0; i < STSZ; i++)
    640 		if (pm->pm_ptbl[i]) {
    641 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
    642 			    PAGE_SIZE);
    643 			pm->pm_ptbl[i] = NULL;
    644 		}
    645 	if (pm->pm_ctx) ctx_free(pm);
    646 }
    647 
    648 /*
    649  * Copy the range specified by src_addr/len
    650  * from the source map to the range dst_addr/len
    651  * in the destination map.
    652  *
    653  * This routine is only advisory and need not do anything.
    654  */
    655 void
    656 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    657 	  vsize_t len, vaddr_t src_addr)
    658 {
    659 }
    660 
    661 /*
    662  * Require that all active physical maps contain no
    663  * incorrect entries NOW.
    664  */
    665 void
    666 pmap_update(struct pmap *pmap)
    667 {
    668 }
    669 
    670 /*
    671  * Garbage collects the physical map system for
    672  * pages which are no longer used.
    673  * Success need not be guaranteed -- that is, there
    674  * may well be pages which are not referenced, but
    675  * others may be collected.
    676  * Called by the pageout daemon when pages are scarce.
    677  */
    678 void
    679 pmap_collect(struct pmap *pm)
    680 {
    681 }
    682 
    683 /*
    684  * Fill the given physical page with zeroes.
    685  */
    686 void
    687 pmap_zero_page(paddr_t pa)
    688 {
    689 
    690 #ifdef PPC_4XX_NOCACHE
    691 	memset((caddr_t)pa, 0, PAGE_SIZE);
    692 #else
    693 	int i;
    694 
    695 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
    696 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    697 		pa += CACHELINESIZE;
    698 	}
    699 #endif
    700 }
    701 
    702 /*
    703  * Copy the given physical source page to its destination.
    704  */
    705 void
    706 pmap_copy_page(paddr_t src, paddr_t dst)
    707 {
    708 
    709 	memcpy((caddr_t)dst, (caddr_t)src, PAGE_SIZE);
    710 	dcache_flush_page(dst);
    711 }
    712 
    713 /*
    714  * This returns whether this is the first mapping of a page.
    715  */
    716 static inline int
    717 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    718 {
    719 	struct pv_entry *pv, *npv = NULL;
    720 	int s;
    721 
    722 	if (!pmap_initialized)
    723 		return 0;
    724 
    725 	s = splvm();
    726 
    727 	pv = pa_to_pv(pa);
    728 for (npv = pv; npv; npv = npv->pv_next)
    729 if (npv->pv_va == va && npv->pv_pm == pm) {
    730 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    731 #ifdef DDB
    732 Debugger();
    733 #endif
    734 return (1);
    735 }
    736 
    737 	if (!pv->pv_pm) {
    738 		/*
    739 		 * No entries yet, use header as the first entry.
    740 		 */
    741 		pv->pv_va = va;
    742 		pv->pv_pm = pm;
    743 		pv->pv_next = NULL;
    744 	} else {
    745 		/*
    746 		 * There is at least one other VA mapping this page.
    747 		 * Place this entry after the header.
    748 		 */
    749 		npv = pool_get(&pv_pool, PR_WAITOK);
    750 		if (!npv) return (0);
    751 		npv->pv_va = va;
    752 		npv->pv_pm = pm;
    753 		npv->pv_next = pv->pv_next;
    754 		pv->pv_next = npv;
    755 	}
    756 	splx(s);
    757 	return (1);
    758 }
    759 
    760 static void
    761 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    762 {
    763 	struct pv_entry *pv, *npv;
    764 
    765 	/*
    766 	 * Remove from the PV table.
    767 	 */
    768 	pv = pa_to_pv(pa);
    769 	if (!pv) return;
    770 
    771 	/*
    772 	 * If it is the first entry on the list, it is actually
    773 	 * in the header and we must copy the following entry up
    774 	 * to the header.  Otherwise we must search the list for
    775 	 * the entry.  In either case we free the now unused entry.
    776 	 */
    777 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    778 		if ((npv = pv->pv_next)) {
    779 			*pv = *npv;
    780 			pool_put(&pv_pool, npv);
    781 		} else
    782 			pv->pv_pm = NULL;
    783 	} else {
    784 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    785 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    786 				break;
    787 		if (npv) {
    788 			pv->pv_next = npv->pv_next;
    789 			pool_put(&pv_pool, npv);
    790 		}
    791 	}
    792 }
    793 
    794 /*
    795  * Insert physical page at pa into the given pmap at virtual address va.
    796  */
    797 int
    798 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    799 {
    800 	int s;
    801 	u_int tte;
    802 	int managed;
    803 
    804 	/*
    805 	 * Have to remove any existing mapping first.
    806 	 */
    807 	pmap_remove(pm, va, va + PAGE_SIZE);
    808 
    809 	if (flags & PMAP_WIRED) flags |= prot;
    810 
    811 	/* If it has no protections don't bother w/the rest */
    812 	if (!(flags & VM_PROT_ALL))
    813 		return (0);
    814 
    815 	managed = 0;
    816 	if (vm_physseg_find(atop(pa), NULL) != -1)
    817 		managed = 1;
    818 
    819 	/*
    820 	 * Generate TTE.
    821 	 *
    822 	 * XXXX
    823 	 *
    824 	 * Since the kernel does not handle execution privileges properly,
    825 	 * we will handle read and execute permissions together.
    826 	 */
    827 	tte = TTE_PA(pa) | TTE_EX;
    828 	/* XXXX -- need to support multiple page sizes. */
    829 	tte |= TTE_SZ_16K;
    830 #ifdef	DIAGNOSTIC
    831 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    832 		(PME_NOCACHE | PME_WRITETHROUG))
    833 		panic("pmap_enter: uncached & writethrough");
    834 #endif
    835 	if (flags & PME_NOCACHE)
    836 		/* Must be I/O mapping */
    837 		tte |= TTE_I | TTE_G;
    838 #ifdef PPC_4XX_NOCACHE
    839 	tte |= TTE_I;
    840 #else
    841 	else if (flags & PME_WRITETHROUG)
    842 		/* Uncached and writethrough are not compatible */
    843 		tte |= TTE_W;
    844 #endif
    845 	if (pm == pmap_kernel())
    846 		tte |= TTE_ZONE(ZONE_PRIV);
    847 	else
    848 		tte |= TTE_ZONE(ZONE_USER);
    849 
    850 	if (flags & VM_PROT_WRITE)
    851 		tte |= TTE_WR;
    852 
    853 	/*
    854 	 * Now record mapping for later back-translation.
    855 	 */
    856 	if (pmap_initialized && managed) {
    857 		char *attr;
    858 
    859 		if (!pmap_enter_pv(pm, va, pa)) {
    860 			/* Could not enter pv on a managed page */
    861 			return 1;
    862 		}
    863 
    864 		/* Now set attributes. */
    865 		attr = pa_to_attr(pa);
    866 #ifdef DIAGNOSTIC
    867 		if (!attr)
    868 			panic("managed but no attr");
    869 #endif
    870 		if (flags & VM_PROT_ALL)
    871 			*attr |= PTE_HI_REF;
    872 		if (flags & VM_PROT_WRITE)
    873 			*attr |= PTE_HI_CHG;
    874 	}
    875 
    876 	s = splvm();
    877 	pm->pm_stats.resident_count++;
    878 
    879 	/* Insert page into page table. */
    880 	pte_enter(pm, va, tte);
    881 
    882 	/* If this is a real fault, enter it in the tlb */
    883 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    884 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    885 	}
    886 	splx(s);
    887 
    888 	/* Flush the real memory from the instruction cache. */
    889 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    890 		__syncicache((void *)pa, PAGE_SIZE);
    891 
    892 	return 0;
    893 }
    894 
    895 void
    896 pmap_unwire(struct pmap *pm, vaddr_t va)
    897 {
    898 	struct pv_entry *pv, *npv;
    899 	paddr_t pa;
    900 	int s = splvm();
    901 
    902 	if (pm == NULL) {
    903 		return;
    904 	}
    905 
    906 	if (!pmap_extract(pm, va, &pa)) {
    907 		return;
    908 	}
    909 
    910 	va |= PV_WIRED;
    911 
    912 	pv = pa_to_pv(pa);
    913 	if (!pv) return;
    914 
    915 	/*
    916 	 * If it is the first entry on the list, it is actually
    917 	 * in the header and we must copy the following entry up
    918 	 * to the header.  Otherwise we must search the list for
    919 	 * the entry.  In either case we free the now unused entry.
    920 	 */
    921 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    922 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    923 			npv->pv_va &= ~PV_WIRED;
    924 			break;
    925 		}
    926 	}
    927 	splx(s);
    928 }
    929 
    930 void
    931 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    932 {
    933 	int s;
    934 	u_int tte;
    935 	struct pmap *pm = pmap_kernel();
    936 
    937 	/*
    938 	 * Have to remove any existing mapping first.
    939 	 */
    940 
    941 	/*
    942 	 * Generate TTE.
    943 	 *
    944 	 * XXXX
    945 	 *
    946 	 * Since the kernel does not handle execution privileges properly,
    947 	 * we will handle read and execute permissions together.
    948 	 */
    949 	tte = 0;
    950 	if (prot & VM_PROT_ALL) {
    951 
    952 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    953 		/* XXXX -- need to support multiple page sizes. */
    954 		tte |= TTE_SZ_16K;
    955 #ifdef DIAGNOSTIC
    956 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    957 			(PME_NOCACHE | PME_WRITETHROUG))
    958 			panic("pmap_kenter_pa: uncached & writethrough");
    959 #endif
    960 		if (prot & PME_NOCACHE)
    961 			/* Must be I/O mapping */
    962 			tte |= TTE_I | TTE_G;
    963 #ifdef PPC_4XX_NOCACHE
    964 		tte |= TTE_I;
    965 #else
    966 		else if (prot & PME_WRITETHROUG)
    967 			/* Uncached and writethrough are not compatible */
    968 			tte |= TTE_W;
    969 #endif
    970 		if (prot & VM_PROT_WRITE)
    971 			tte |= TTE_WR;
    972 	}
    973 
    974 	s = splvm();
    975 	pm->pm_stats.resident_count++;
    976 
    977 	/* Insert page into page table. */
    978 	pte_enter(pm, va, tte);
    979 	splx(s);
    980 }
    981 
    982 void
    983 pmap_kremove(vaddr_t va, vsize_t len)
    984 {
    985 
    986 	while (len > 0) {
    987 		pte_enter(pmap_kernel(), va, 0);
    988 		va += PAGE_SIZE;
    989 		len -= PAGE_SIZE;
    990 	}
    991 }
    992 
    993 /*
    994  * Remove the given range of mapping entries.
    995  */
    996 void
    997 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
    998 {
    999 	int s;
   1000 	paddr_t pa;
   1001 	volatile u_int *ptp;
   1002 
   1003 	s = splvm();
   1004 	while (va < endva) {
   1005 
   1006 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1007 			pa = TTE_PA(pa);
   1008 			pmap_remove_pv(pm, va, pa);
   1009 			*ptp = 0;
   1010 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1011 			pm->pm_stats.resident_count--;
   1012 		}
   1013 		va += PAGE_SIZE;
   1014 	}
   1015 
   1016 	splx(s);
   1017 }
   1018 
   1019 /*
   1020  * Get the physical page address for the given pmap/virtual address.
   1021  */
   1022 boolean_t
   1023 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1024 {
   1025 	int seg = STIDX(va);
   1026 	int ptn = PTIDX(va);
   1027 	u_int pa = 0;
   1028 	int s = splvm();
   1029 
   1030 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1031 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1032 	}
   1033 	splx(s);
   1034 	return (pa != 0);
   1035 }
   1036 
   1037 /*
   1038  * Lower the protection on the specified range of this pmap.
   1039  *
   1040  * There are only two cases: either the protection is going to 0,
   1041  * or it is going to read-only.
   1042  */
   1043 void
   1044 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1045 {
   1046 	volatile u_int *ptp;
   1047 	int s;
   1048 
   1049 	if (prot & VM_PROT_READ) {
   1050 		s = splvm();
   1051 		while (sva < eva) {
   1052 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1053 				*ptp &= ~TTE_WR;
   1054 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1055 			}
   1056 			sva += PAGE_SIZE;
   1057 		}
   1058 		splx(s);
   1059 		return;
   1060 	}
   1061 	pmap_remove(pm, sva, eva);
   1062 }
   1063 
   1064 boolean_t
   1065 check_attr(struct vm_page *pg, u_int mask, int clear)
   1066 {
   1067 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1068 	int s;
   1069 	char *attr;
   1070 	int rv;
   1071 
   1072 	/*
   1073 	 * First modify bits in cache.
   1074 	 */
   1075 	s = splvm();
   1076 	attr = pa_to_attr(pa);
   1077 	if (attr == NULL)
   1078 		return FALSE;
   1079 
   1080 	rv = ((*attr & mask) != 0);
   1081 	if (clear) {
   1082 		*attr &= ~mask;
   1083 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1084 	}
   1085 	splx(s);
   1086 	return rv;
   1087 }
   1088 
   1089 
   1090 /*
   1091  * Lower the protection on the specified physical page.
   1092  *
   1093  * There are only two cases: either the protection is going to 0,
   1094  * or it is going to read-only.
   1095  */
   1096 void
   1097 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1098 {
   1099 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1100 	vaddr_t va;
   1101 	struct pv_entry *pvh, *pv, *npv;
   1102 	struct pmap *pm;
   1103 
   1104 	pvh = pa_to_pv(pa);
   1105 	if (pvh == NULL)
   1106 		return;
   1107 
   1108 	/* Handle extra pvs which may be deleted in the operation */
   1109 	for (pv = pvh->pv_next; pv; pv = npv) {
   1110 		npv = pv->pv_next;
   1111 
   1112 		pm = pv->pv_pm;
   1113 		va = pv->pv_va;
   1114 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1115 	}
   1116 	/* Now check the head pv */
   1117 	if (pvh->pv_pm) {
   1118 		pv = pvh;
   1119 		pm = pv->pv_pm;
   1120 		va = pv->pv_va;
   1121 		pmap_protect(pm, va, va+PAGE_SIZE, prot);
   1122 	}
   1123 }
   1124 
   1125 /*
   1126  * Activate the address space for the specified process.  If the process
   1127  * is the current process, load the new MMU context.
   1128  */
   1129 void
   1130 pmap_activate(struct lwp *l)
   1131 {
   1132 #if 0
   1133 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1134 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1135 
   1136 	/*
   1137 	 * XXX Normally performed in cpu_fork().
   1138 	 */
   1139 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1140 	if (pcb->pcb_pm != pmap) {
   1141 		pcb->pcb_pm = pmap;
   1142 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1143 		    (paddr_t *)&pcb->pcb_pmreal);
   1144 	}
   1145 
   1146 	if (l == curlwp) {
   1147 		/* Store pointer to new current pmap. */
   1148 		curpm = pcb->pcb_pmreal;
   1149 	}
   1150 #endif
   1151 }
   1152 
   1153 /*
   1154  * Deactivate the specified process's address space.
   1155  */
   1156 void
   1157 pmap_deactivate(struct lwp *l)
   1158 {
   1159 }
   1160 
   1161 /*
   1162  * Synchronize caches corresponding to [addr, addr+len) in p.
   1163  */
   1164 void
   1165 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1166 {
   1167 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1168 	int msr, ctx, opid, step;
   1169 
   1170 
   1171 	step = CACHELINESIZE;
   1172 
   1173 	/*
   1174 	 * Need to turn off IMMU and switch to user context.
   1175 	 * (icbi uses DMMU).
   1176 	 */
   1177 	if (!(ctx = pm->pm_ctx)) {
   1178 		/* No context -- assign it one */
   1179 		ctx_alloc(pm);
   1180 		ctx = pm->pm_ctx;
   1181 	}
   1182 	__asm __volatile("mfmsr %0;"
   1183 		"li %1, 0x20;"
   1184 		"andc %1,%0,%1;"
   1185 		"mtmsr %1;"
   1186 		"sync;isync;"
   1187 		"mfpid %1;"
   1188 		"mtpid %2;"
   1189 		"sync; isync;"
   1190 		"1:"
   1191 		"dcbf 0,%3;"
   1192 		"icbi 0,%3;"
   1193 		"add %3,%3,%5;"
   1194 		"addc. %4,%4,%6;"
   1195 		"bge 1b;"
   1196 		"mtpid %1;"
   1197 		"mtmsr %0;"
   1198 		"sync; isync"
   1199 		: "=&r" (msr), "=&r" (opid)
   1200 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step));
   1201 }
   1202 
   1203 
   1204 /* This has to be done in real mode !!! */
   1205 void
   1206 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1207 {
   1208 	u_long i, found;
   1209 	u_long msr;
   1210 
   1211 	/* If there's no context then it can't be mapped. */
   1212 	if (!pid) return;
   1213 
   1214 	asm("mfpid %1;"			/* Save PID */
   1215 		"mfmsr %2;"		/* Save MSR */
   1216 		"li %0,0;"		/* Now clear MSR */
   1217 		"mtmsr %0;"
   1218 		"mtpid %4;"		/* Set PID */
   1219 		"sync;"
   1220 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1221 		"sync;"
   1222 		"mtpid %1;"		/* Restore PID */
   1223 		"mtmsr %2;"		/* Restore MSR */
   1224 		"sync;isync;"
   1225 		"li %1,1;"
   1226 		"beq 1f;"
   1227 		"li %1,0;"
   1228 		"1:"
   1229 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1230 		: "r" (va), "r" (pid));
   1231 	if (found && !TLB_LOCKED(i)) {
   1232 
   1233 		/* Now flush translation */
   1234 		asm volatile(
   1235 			"tlbwe %0,%1,0;"
   1236 			"sync;isync;"
   1237 			: : "r" (0), "r" (i));
   1238 
   1239 		tlb_info[i].ti_ctx = 0;
   1240 		tlb_info[i].ti_flags = 0;
   1241 		tlbnext = i;
   1242 		/* Successful flushes */
   1243 		tlbflush_ev.ev_count++;
   1244 	}
   1245 }
   1246 
   1247 void
   1248 ppc4xx_tlb_flush_all(void)
   1249 {
   1250 	u_long i;
   1251 
   1252 	for (i = 0; i < NTLB; i++)
   1253 		if (!TLB_LOCKED(i)) {
   1254 			asm volatile(
   1255 				"tlbwe %0,%1,0;"
   1256 				"sync;isync;"
   1257 				: : "r" (0), "r" (i));
   1258 			tlb_info[i].ti_ctx = 0;
   1259 			tlb_info[i].ti_flags = 0;
   1260 		}
   1261 
   1262 	asm volatile("sync;isync");
   1263 }
   1264 
   1265 /* Find a TLB entry to evict. */
   1266 static int
   1267 ppc4xx_tlb_find_victim(void)
   1268 {
   1269 	int flags;
   1270 
   1271 	for (;;) {
   1272 		if (++tlbnext >= NTLB)
   1273 			tlbnext = TLB_NRESERVED;
   1274 		flags = tlb_info[tlbnext].ti_flags;
   1275 		if (!(flags & TLBF_USED) ||
   1276 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1277 			u_long va, stack = (u_long)&va;
   1278 
   1279 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1280 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1281 			     (flags & TLBF_USED)) {
   1282 				/* Kernel stack page */
   1283 				flags |= TLBF_USED;
   1284 				tlb_info[tlbnext].ti_flags = flags;
   1285 			} else {
   1286 				/* Found it! */
   1287 				return (tlbnext);
   1288 			}
   1289 		} else {
   1290 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1291 		}
   1292 	}
   1293 }
   1294 
   1295 void
   1296 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1297 {
   1298 	u_long th, tl, idx;
   1299 	tlbpid_t pid;
   1300 	u_short msr;
   1301 	paddr_t pa;
   1302 	int s, sz;
   1303 
   1304 	tlbenter_ev.ev_count++;
   1305 
   1306 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1307 	pa = (pte & TTE_RPN_MASK(sz));
   1308 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1309 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1310 	tl |= ppc4xx_tlbflags(va, pa);
   1311 
   1312 	s = splhigh();
   1313 	idx = ppc4xx_tlb_find_victim();
   1314 
   1315 #ifdef DIAGNOSTIC
   1316 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1317 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
   1318 	}
   1319 #endif
   1320 
   1321 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1322 	tlb_info[idx].ti_ctx = ctx;
   1323 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1324 
   1325 	asm volatile(
   1326 		"mfmsr %0;"			/* Save MSR */
   1327 		"li %1,0;"
   1328 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1329 		"mtmsr %1;"			/* Clear MSR */
   1330 		"mfpid %1;"			/* Save old PID */
   1331 		"mtpid %2;"			/* Load translation ctx */
   1332 		"sync; isync;"
   1333 #ifdef DEBUG
   1334 		"andi. %3,%3,63;"
   1335 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1336 #endif
   1337 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1338 		"sync; isync;"
   1339 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1340 		"sync; isync;"
   1341 	: "=&r" (msr), "=&r" (pid)
   1342 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1343 	splx(s);
   1344 }
   1345 
   1346 void
   1347 ppc4xx_tlb_unpin(int i)
   1348 {
   1349 
   1350 	if (i == -1)
   1351 		for (i = 0; i < TLB_NRESERVED; i++)
   1352 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1353 	else
   1354 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1355 }
   1356 
   1357 void
   1358 ppc4xx_tlb_init(void)
   1359 {
   1360 	int i;
   1361 
   1362 	/* Mark reserved TLB entries */
   1363 	for (i = 0; i < TLB_NRESERVED; i++) {
   1364 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1365 		tlb_info[i].ti_ctx = KERNEL_PID;
   1366 	}
   1367 
   1368 	/* Setup security zones */
   1369 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1370 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1371 	 * Z3 - full access regardless of TLB entry permissions
   1372 	 */
   1373 
   1374 	asm volatile(
   1375 		"mtspr %0,%1;"
   1376 		"sync;"
   1377 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1378 }
   1379 
   1380 
   1381 /*
   1382  * We should pass the ctx in from trap code.
   1383  */
   1384 int
   1385 pmap_tlbmiss(vaddr_t va, int ctx)
   1386 {
   1387 	volatile u_int *pte;
   1388 	u_long tte;
   1389 
   1390 	tlbmiss_ev.ev_count++;
   1391 
   1392 	/*
   1393 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1394 	 */
   1395 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1396 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1397 		if (pte == NULL) {
   1398 			/* Map unmanaged addresses directly for kernel access */
   1399 			return 1;
   1400 		}
   1401 		tte = *pte;
   1402 		if (tte == 0) {
   1403 			return 1;
   1404 		}
   1405 	} else {
   1406 		/* Create a 16MB writable mapping. */
   1407 #ifdef PPC_4XX_NOCACHE
   1408 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1409 #else
   1410 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1411 #endif
   1412 	}
   1413 	tlbhit_ev.ev_count++;
   1414 	ppc4xx_tlb_enter(ctx, va, tte);
   1415 
   1416 	return 0;
   1417 }
   1418 
   1419 /*
   1420  * Flush all the entries matching a context from the TLB.
   1421  */
   1422 static int
   1423 ctx_flush(int cnum)
   1424 {
   1425 	int i;
   1426 
   1427 	/* We gotta steal this context */
   1428 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1429 		if (tlb_info[i].ti_ctx == cnum) {
   1430 			/* Can't steal ctx if it has a locked entry. */
   1431 			if (TLB_LOCKED(i)) {
   1432 #ifdef DIAGNOSTIC
   1433 				printf("ctx_flush: can't invalidate "
   1434 					"locked mapping %d "
   1435 					"for context %d\n", i, cnum);
   1436 #ifdef DDB
   1437 				Debugger();
   1438 #endif
   1439 #endif
   1440 				return (1);
   1441 			}
   1442 #ifdef DIAGNOSTIC
   1443 			if (i < TLB_NRESERVED)
   1444 				panic("TLB entry %d not locked", i);
   1445 #endif
   1446 			/* Invalidate particular TLB entry regardless of locked status */
   1447 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1448 			tlb_info[i].ti_flags = 0;
   1449 		}
   1450 	}
   1451 	return (0);
   1452 }
   1453 
   1454 /*
   1455  * Allocate a context.  If necessary, steal one from someone else.
   1456  *
   1457  * The new context is flushed from the TLB before returning.
   1458  */
   1459 int
   1460 ctx_alloc(struct pmap *pm)
   1461 {
   1462 	int s, cnum;
   1463 	static int next = MINCTX;
   1464 
   1465 	if (pm == pmap_kernel()) {
   1466 #ifdef DIAGNOSTIC
   1467 		printf("ctx_alloc: kernel pmap!\n");
   1468 #endif
   1469 		return (0);
   1470 	}
   1471 	s = splvm();
   1472 
   1473 	/* Find a likely context. */
   1474 	cnum = next;
   1475 	do {
   1476 		if ((++cnum) > NUMCTX)
   1477 			cnum = MINCTX;
   1478 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1479 
   1480 	/* Now clean it out */
   1481 oops:
   1482 	if (cnum < MINCTX)
   1483 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1484 	if (ctx_flush(cnum)) {
   1485 		/* oops -- something's wired. */
   1486 		if ((++cnum) > NUMCTX)
   1487 			cnum = MINCTX;
   1488 		goto oops;
   1489 	}
   1490 
   1491 	if (ctxbusy[cnum]) {
   1492 #ifdef DEBUG
   1493 		/* We should identify this pmap and clear it */
   1494 		printf("Warning: stealing context %d\n", cnum);
   1495 #endif
   1496 		ctxbusy[cnum]->pm_ctx = 0;
   1497 	}
   1498 	ctxbusy[cnum] = pm;
   1499 	next = cnum;
   1500 	splx(s);
   1501 	pm->pm_ctx = cnum;
   1502 
   1503 	return cnum;
   1504 }
   1505 
   1506 /*
   1507  * Give away a context.
   1508  */
   1509 void
   1510 ctx_free(struct pmap *pm)
   1511 {
   1512 	int oldctx;
   1513 
   1514 	oldctx = pm->pm_ctx;
   1515 
   1516 	if (oldctx == 0)
   1517 		panic("ctx_free: freeing kernel context");
   1518 #ifdef DIAGNOSTIC
   1519 	if (ctxbusy[oldctx] == 0)
   1520 		printf("ctx_free: freeing free context %d\n", oldctx);
   1521 	if (ctxbusy[oldctx] != pm) {
   1522 		printf("ctx_free: freeing someone esle's context\n "
   1523 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1524 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1525 #ifdef DDB
   1526 		Debugger();
   1527 #endif
   1528 	}
   1529 #endif
   1530 	/* We should verify it has not been stolen and reallocated... */
   1531 	ctxbusy[oldctx] = NULL;
   1532 	ctx_flush(oldctx);
   1533 }
   1534 
   1535 
   1536 #ifdef DEBUG
   1537 /*
   1538  * Test ref/modify handling.
   1539  */
   1540 void pmap_testout __P((void));
   1541 void
   1542 pmap_testout()
   1543 {
   1544 	vaddr_t va;
   1545 	volatile int *loc;
   1546 	int val = 0;
   1547 	paddr_t pa;
   1548 	struct vm_page *pg;
   1549 	int ref, mod;
   1550 
   1551 	/* Allocate a page */
   1552 	va = (vaddr_t)uvm_km_alloc1(kernel_map, PAGE_SIZE, 1);
   1553 	loc = (int*)va;
   1554 
   1555 	pmap_extract(pmap_kernel(), va, &pa);
   1556 	pg = PHYS_TO_VM_PAGE(pa);
   1557 	pmap_unwire(pmap_kernel(), va);
   1558 
   1559 	pmap_remove(pmap_kernel(), va, va+1);
   1560 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1561 	pmap_update(pmap_kernel());
   1562 
   1563 	/* Now clear reference and modify */
   1564 	ref = pmap_clear_reference(pg);
   1565 	mod = pmap_clear_modify(pg);
   1566 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1567 	       (void *)(u_long)va, (long)pa,
   1568 	       ref, mod);
   1569 
   1570 	/* Check it's properly cleared */
   1571 	ref = pmap_is_referenced(pg);
   1572 	mod = pmap_is_modified(pg);
   1573 	printf("Checking cleared page: ref %d, mod %d\n",
   1574 	       ref, mod);
   1575 
   1576 	/* Reference page */
   1577 	val = *loc;
   1578 
   1579 	ref = pmap_is_referenced(pg);
   1580 	mod = pmap_is_modified(pg);
   1581 	printf("Referenced page: ref %d, mod %d val %x\n",
   1582 	       ref, mod, val);
   1583 
   1584 	/* Now clear reference and modify */
   1585 	ref = pmap_clear_reference(pg);
   1586 	mod = pmap_clear_modify(pg);
   1587 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1588 	       (void *)(u_long)va, (long)pa,
   1589 	       ref, mod);
   1590 
   1591 	/* Modify page */
   1592 	*loc = 1;
   1593 
   1594 	ref = pmap_is_referenced(pg);
   1595 	mod = pmap_is_modified(pg);
   1596 	printf("Modified page: ref %d, mod %d\n",
   1597 	       ref, mod);
   1598 
   1599 	/* Now clear reference and modify */
   1600 	ref = pmap_clear_reference(pg);
   1601 	mod = pmap_clear_modify(pg);
   1602 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1603 	       (void *)(u_long)va, (long)pa,
   1604 	       ref, mod);
   1605 
   1606 	/* Check it's properly cleared */
   1607 	ref = pmap_is_referenced(pg);
   1608 	mod = pmap_is_modified(pg);
   1609 	printf("Checking cleared page: ref %d, mod %d\n",
   1610 	       ref, mod);
   1611 
   1612 	/* Modify page */
   1613 	*loc = 1;
   1614 
   1615 	ref = pmap_is_referenced(pg);
   1616 	mod = pmap_is_modified(pg);
   1617 	printf("Modified page: ref %d, mod %d\n",
   1618 	       ref, mod);
   1619 
   1620 	/* Check pmap_protect() */
   1621 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1622 	pmap_update(pmap_kernel());
   1623 	ref = pmap_is_referenced(pg);
   1624 	mod = pmap_is_modified(pg);
   1625 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1626 	       ref, mod);
   1627 
   1628 	/* Now clear reference and modify */
   1629 	ref = pmap_clear_reference(pg);
   1630 	mod = pmap_clear_modify(pg);
   1631 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1632 	       (void *)(u_long)va, (long)pa,
   1633 	       ref, mod);
   1634 
   1635 	/* Reference page */
   1636 	val = *loc;
   1637 
   1638 	ref = pmap_is_referenced(pg);
   1639 	mod = pmap_is_modified(pg);
   1640 	printf("Referenced page: ref %d, mod %d val %x\n",
   1641 	       ref, mod, val);
   1642 
   1643 	/* Now clear reference and modify */
   1644 	ref = pmap_clear_reference(pg);
   1645 	mod = pmap_clear_modify(pg);
   1646 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1647 	       (void *)(u_long)va, (long)pa,
   1648 	       ref, mod);
   1649 
   1650 	/* Modify page */
   1651 #if 0
   1652 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1653 	pmap_update(pmap_kernel());
   1654 #endif
   1655 	*loc = 1;
   1656 
   1657 	ref = pmap_is_referenced(pg);
   1658 	mod = pmap_is_modified(pg);
   1659 	printf("Modified page: ref %d, mod %d\n",
   1660 	       ref, mod);
   1661 
   1662 	/* Check pmap_protect() */
   1663 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1664 	pmap_update(pmap_kernel());
   1665 	ref = pmap_is_referenced(pg);
   1666 	mod = pmap_is_modified(pg);
   1667 	printf("pmap_protect(): ref %d, mod %d\n",
   1668 	       ref, mod);
   1669 
   1670 	/* Now clear reference and modify */
   1671 	ref = pmap_clear_reference(pg);
   1672 	mod = pmap_clear_modify(pg);
   1673 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1674 	       (void *)(u_long)va, (long)pa,
   1675 	       ref, mod);
   1676 
   1677 	/* Reference page */
   1678 	val = *loc;
   1679 
   1680 	ref = pmap_is_referenced(pg);
   1681 	mod = pmap_is_modified(pg);
   1682 	printf("Referenced page: ref %d, mod %d val %x\n",
   1683 	       ref, mod, val);
   1684 
   1685 	/* Now clear reference and modify */
   1686 	ref = pmap_clear_reference(pg);
   1687 	mod = pmap_clear_modify(pg);
   1688 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1689 	       (void *)(u_long)va, (long)pa,
   1690 	       ref, mod);
   1691 
   1692 	/* Modify page */
   1693 #if 0
   1694 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1695 	pmap_update(pmap_kernel());
   1696 #endif
   1697 	*loc = 1;
   1698 
   1699 	ref = pmap_is_referenced(pg);
   1700 	mod = pmap_is_modified(pg);
   1701 	printf("Modified page: ref %d, mod %d\n",
   1702 	       ref, mod);
   1703 
   1704 	/* Check pmap_pag_protect() */
   1705 	pmap_page_protect(pg, VM_PROT_READ);
   1706 	ref = pmap_is_referenced(pg);
   1707 	mod = pmap_is_modified(pg);
   1708 	printf("pmap_page_protect(VM_PROT_READ): 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_NONE);
   1747 	ref = pmap_is_referenced(pg);
   1748 	mod = pmap_is_modified(pg);
   1749 	printf("pmap_page_protect(): 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 
   1760 	/* Reference page */
   1761 	val = *loc;
   1762 
   1763 	ref = pmap_is_referenced(pg);
   1764 	mod = pmap_is_modified(pg);
   1765 	printf("Referenced page: ref %d, mod %d val %x\n",
   1766 	       ref, mod, val);
   1767 
   1768 	/* Now clear reference and modify */
   1769 	ref = pmap_clear_reference(pg);
   1770 	mod = pmap_clear_modify(pg);
   1771 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1772 	       (void *)(u_long)va, (long)pa,
   1773 	       ref, mod);
   1774 
   1775 	/* Modify page */
   1776 #if 0
   1777 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1778 	pmap_update(pmap_kernel());
   1779 #endif
   1780 	*loc = 1;
   1781 
   1782 	ref = pmap_is_referenced(pg);
   1783 	mod = pmap_is_modified(pg);
   1784 	printf("Modified page: ref %d, mod %d\n",
   1785 	       ref, mod);
   1786 
   1787 	/* Unmap page */
   1788 	pmap_remove(pmap_kernel(), va, va+1);
   1789 	pmap_update(pmap_kernel());
   1790 	ref = pmap_is_referenced(pg);
   1791 	mod = pmap_is_modified(pg);
   1792 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1793 
   1794 	/* Now clear reference and modify */
   1795 	ref = pmap_clear_reference(pg);
   1796 	mod = pmap_clear_modify(pg);
   1797 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1798 	       (void *)(u_long)va, (long)pa, ref, mod);
   1799 
   1800 	/* Check it's properly cleared */
   1801 	ref = pmap_is_referenced(pg);
   1802 	mod = pmap_is_modified(pg);
   1803 	printf("Checking cleared page: ref %d, mod %d\n",
   1804 	       ref, mod);
   1805 
   1806 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1807 		VM_PROT_ALL|PMAP_WIRED);
   1808 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE);
   1809 }
   1810 #endif
   1811