Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.18
      1 /*	$NetBSD: pmap.c,v 1.18 2003/03/11 10:40:16 hannken 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/NBPG)+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] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
    211 	}
    212 	pm->pm_ptbl[seg][ptn] = pte;
    213 
    214 	/* Flush entry. */
    215 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    216 	return (1);
    217 }
    218 
    219 /*
    220  * Get a pointer to a PTE in a page table.
    221  */
    222 volatile u_int *
    223 pte_find(struct pmap *pm, vaddr_t va)
    224 {
    225 	int seg = STIDX(va);
    226 	int ptn = PTIDX(va);
    227 
    228 	if (pm->pm_ptbl[seg])
    229 		return (&pm->pm_ptbl[seg][ptn]);
    230 
    231 	return (NULL);
    232 }
    233 
    234 /*
    235  * This is called during initppc, before the system is really initialized.
    236  */
    237 void
    238 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    239 {
    240 	struct mem_region *mp, *mp1;
    241 	int cnt, i;
    242 	u_int s, e, sz;
    243 
    244 	/*
    245 	 * Allocate the kernel page table at the end of
    246 	 * kernel space so it's in the locked TTE.
    247 	 */
    248 	kernmap = (caddr_t)kernelend;
    249 
    250 	/*
    251 	 * Initialize kernel page table.
    252 	 */
    253 	for (i = 0; i < STSZ; i++) {
    254 		pmap_kernel()->pm_ptbl[i] = 0;
    255 	}
    256 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    257 
    258 	/*
    259 	 * Announce page-size to the VM-system
    260 	 */
    261 	uvmexp.pagesize = NBPG;
    262 	uvm_setpagesize();
    263 
    264 	/*
    265 	 * Get memory.
    266 	 */
    267 	mem_regions(&mem, &avail);
    268 	for (mp = mem; mp->size; mp++) {
    269 		physmem += btoc(mp->size);
    270 		printf("+%lx,",mp->size);
    271 	}
    272 	printf("\n");
    273 	ppc4xx_tlb_init();
    274 	/*
    275 	 * Count the number of available entries.
    276 	 */
    277 	for (cnt = 0, mp = avail; mp->size; mp++)
    278 		cnt++;
    279 
    280 	/*
    281 	 * Page align all regions.
    282 	 * Non-page aligned memory isn't very interesting to us.
    283 	 * Also, sort the entries for ascending addresses.
    284 	 */
    285 	kernelstart &= ~PGOFSET;
    286 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    287 	for (mp = avail; mp->size; mp++) {
    288 		s = mp->start;
    289 		e = mp->start + mp->size;
    290 		printf("%08x-%08x -> ",s,e);
    291 		/*
    292 		 * Check whether this region holds all of the kernel.
    293 		 */
    294 		if (s < kernelstart && e > kernelend) {
    295 			avail[cnt].start = kernelend;
    296 			avail[cnt++].size = e - kernelend;
    297 			e = kernelstart;
    298 		}
    299 		/*
    300 		 * Look whether this regions starts within the kernel.
    301 		 */
    302 		if (s >= kernelstart && s < kernelend) {
    303 			if (e <= kernelend)
    304 				goto empty;
    305 			s = kernelend;
    306 		}
    307 		/*
    308 		 * Now look whether this region ends within the kernel.
    309 		 */
    310 		if (e > kernelstart && e <= kernelend) {
    311 			if (s >= kernelstart)
    312 				goto empty;
    313 			e = kernelstart;
    314 		}
    315 		/*
    316 		 * Now page align the start and size of the region.
    317 		 */
    318 		s = round_page(s);
    319 		e = trunc_page(e);
    320 		if (e < s)
    321 			e = s;
    322 		sz = e - s;
    323 		printf("%08x-%08x = %x\n",s,e,sz);
    324 		/*
    325 		 * Check whether some memory is left here.
    326 		 */
    327 		if (sz == 0) {
    328 		empty:
    329 			memmove(mp, mp + 1,
    330 				(cnt - (mp - avail)) * sizeof *mp);
    331 			cnt--;
    332 			mp--;
    333 			continue;
    334 		}
    335 		/*
    336 		 * Do an insertion sort.
    337 		 */
    338 		npgs += btoc(sz);
    339 		for (mp1 = avail; mp1 < mp; mp1++)
    340 			if (s < mp1->start)
    341 				break;
    342 		if (mp1 < mp) {
    343 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    344 			mp1->start = s;
    345 			mp1->size = sz;
    346 		} else {
    347 			mp->start = s;
    348 			mp->size = sz;
    349 		}
    350 	}
    351 
    352 	/*
    353 	 * We cannot do pmap_steal_memory here,
    354 	 * since we don't run with translation enabled yet.
    355 	 */
    356 #ifndef MSGBUFADDR
    357 	/*
    358 	 * allow for msgbuf
    359 	 */
    360 	sz = round_page(MSGBUFSIZE);
    361 	mp = NULL;
    362 	for (mp1 = avail; mp1->size; mp1++)
    363 		if (mp1->size >= sz)
    364 			mp = mp1;
    365 	if (mp == NULL)
    366 		panic("not enough memory?");
    367 
    368 	npgs -= btoc(sz);
    369 	msgbuf_paddr = mp->start + mp->size - sz;
    370 	mp->size -= sz;
    371 	if (mp->size <= 0)
    372 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    373 #endif
    374 
    375 	printf("Loading pages\n");
    376 	for (mp = avail; mp->size; mp++)
    377 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    378 			atop(mp->start), atop(mp->start + mp->size),
    379 			VM_FREELIST_DEFAULT);
    380 
    381 	/*
    382 	 * Initialize kernel pmap and hardware.
    383 	 */
    384 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    385 	pmap_kernel()->pm_ctx = KERNEL_PID;
    386 	nextavail = avail->start;
    387 
    388 
    389 	evcnt_attach_static(&tlbhit_ev);
    390 	evcnt_attach_static(&tlbmiss_ev);
    391 	evcnt_attach_static(&tlbflush_ev);
    392 	evcnt_attach_static(&tlbenter_ev);
    393 	printf("Done\n");
    394 }
    395 
    396 /*
    397  * Restrict given range to physical memory
    398  *
    399  * (Used by /dev/mem)
    400  */
    401 void
    402 pmap_real_memory(paddr_t *start, psize_t *size)
    403 {
    404 	struct mem_region *mp;
    405 
    406 	for (mp = mem; mp->size; mp++) {
    407 		if (*start + *size > mp->start &&
    408 		    *start < mp->start + mp->size) {
    409 			if (*start < mp->start) {
    410 				*size -= mp->start - *start;
    411 				*start = mp->start;
    412 			}
    413 			if (*start + *size > mp->start + mp->size)
    414 				*size = mp->start + mp->size - *start;
    415 			return;
    416 		}
    417 	}
    418 	*size = 0;
    419 }
    420 
    421 /*
    422  * Initialize anything else for pmap handling.
    423  * Called during vm_init().
    424  */
    425 void
    426 pmap_init(void)
    427 {
    428 	struct pv_entry *pv;
    429 	vsize_t sz;
    430 	vaddr_t addr;
    431 	int i, s;
    432 	int bank;
    433 	char *attr;
    434 
    435 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    436 	sz = round_page(sz);
    437 	addr = uvm_km_zalloc(kernel_map, sz);
    438 	s = splvm();
    439 	pv = pv_table = (struct pv_entry *)addr;
    440 	for (i = npgs; --i >= 0;)
    441 		pv++->pv_pm = NULL;
    442 	pmap_attrib = (char *)pv;
    443 	memset(pv, 0, npgs);
    444 
    445 	pv = pv_table;
    446 	attr = pmap_attrib;
    447 	for (bank = 0; bank < vm_nphysseg; bank++) {
    448 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    449 		vm_physmem[bank].pmseg.pvent = pv;
    450 		vm_physmem[bank].pmseg.attrs = attr;
    451 		pv += sz;
    452 		attr += sz;
    453 	}
    454 
    455 	pmap_initialized = 1;
    456 	splx(s);
    457 
    458 	/* Setup a pool for additional pvlist structures */
    459 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    460 }
    461 
    462 /*
    463  * How much virtual space is available to the kernel?
    464  */
    465 void
    466 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    467 {
    468 
    469 #if 0
    470 	/*
    471 	 * Reserve one segment for kernel virtual memory
    472 	 */
    473 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    474 	*end = *start + SEGMENT_LENGTH;
    475 #else
    476 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    477 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    478 #endif
    479 }
    480 
    481 #ifdef PMAP_GROWKERNEL
    482 /*
    483  * Preallocate kernel page tables to a specified VA.
    484  * This simply loops through the first TTE for each
    485  * page table from the beginning of the kernel pmap,
    486  * reads the entry, and if the result is
    487  * zero (either invalid entry or no page table) it stores
    488  * a zero there, populating page tables in the process.
    489  * This is not the most efficient technique but i don't
    490  * expect it to be called that often.
    491  */
    492 extern struct vm_page *vm_page_alloc1 __P((void));
    493 extern void vm_page_free1 __P((struct vm_page *));
    494 
    495 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    496 
    497 vaddr_t
    498 pmap_growkernel(maxkvaddr)
    499 	vaddr_t maxkvaddr;
    500 {
    501 	int s;
    502 	int seg;
    503 	paddr_t pg;
    504 	struct pmap *pm = pmap_kernel();
    505 
    506 	s = splvm();
    507 
    508 	/* Align with the start of a page table */
    509 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    510 	     kbreak += PTMAP) {
    511 		seg = STIDX(kbreak);
    512 
    513 		if (pte_find(pm, kbreak)) continue;
    514 
    515 		if (uvm.page_init_done) {
    516 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    517 		} else {
    518 			if (!uvm_page_physget(&pg))
    519 				panic("pmap_growkernel: no memory");
    520 		}
    521 		if (!pg) panic("pmap_growkernel: no pages");
    522 		pmap_zero_page((paddr_t)pg);
    523 
    524 		/* XXX This is based on all phymem being addressable */
    525 		pm->pm_ptbl[seg] = (u_int *)pg;
    526 	}
    527 	splx(s);
    528 	return (kbreak);
    529 }
    530 
    531 /*
    532  *	vm_page_alloc1:
    533  *
    534  *	Allocate and return a memory cell with no associated object.
    535  */
    536 struct vm_page *
    537 vm_page_alloc1()
    538 {
    539 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    540 	if (pg) {
    541 		pg->wire_count = 1;	/* no mappings yet */
    542 		pg->flags &= ~PG_BUSY;	/* never busy */
    543 	}
    544 	return pg;
    545 }
    546 
    547 /*
    548  *	vm_page_free1:
    549  *
    550  *	Returns the given page to the free list,
    551  *	disassociating it with any VM object.
    552  *
    553  *	Object and page must be locked prior to entry.
    554  */
    555 void
    556 vm_page_free1(mem)
    557 	struct vm_page *mem;
    558 {
    559 #ifdef DIAGNOSTIC
    560 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    561 		printf("Freeing invalid page %p\n", mem);
    562 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    563 #ifdef DDB
    564 		Debugger();
    565 #endif
    566 		return;
    567 	}
    568 #endif
    569 	mem->flags |= PG_BUSY;
    570 	mem->wire_count = 0;
    571 	uvm_pagefree(mem);
    572 }
    573 #endif
    574 
    575 /*
    576  * Create and return a physical map.
    577  */
    578 struct pmap *
    579 pmap_create(void)
    580 {
    581 	struct pmap *pm;
    582 
    583 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    584 	memset((caddr_t)pm, 0, sizeof *pm);
    585 	pmap_pinit(pm);
    586 	return pm;
    587 }
    588 
    589 /*
    590  * Initialize a preallocated and zeroed pmap structure.
    591  */
    592 void
    593 pmap_pinit(struct pmap *pm)
    594 {
    595 	int i;
    596 
    597 	/*
    598 	 * Allocate some segment registers for this pmap.
    599 	 */
    600 	pm->pm_refs = 1;
    601 	for (i = 0; i < STSZ; i++)
    602 		pm->pm_ptbl[i] = NULL;
    603 }
    604 
    605 /*
    606  * Add a reference to the given pmap.
    607  */
    608 void
    609 pmap_reference(struct pmap *pm)
    610 {
    611 
    612 	pm->pm_refs++;
    613 }
    614 
    615 /*
    616  * Retire the given pmap from service.
    617  * Should only be called if the map contains no valid mappings.
    618  */
    619 void
    620 pmap_destroy(struct pmap *pm)
    621 {
    622 
    623 	if (--pm->pm_refs == 0) {
    624 		pmap_release(pm);
    625 		free((caddr_t)pm, M_VMPMAP);
    626 	}
    627 }
    628 
    629 /*
    630  * Release any resources held by the given physical map.
    631  * Called when a pmap initialized by pmap_pinit is being released.
    632  */
    633 static void
    634 pmap_release(struct pmap *pm)
    635 {
    636 	int i;
    637 
    638 	for (i = 0; i < STSZ; i++)
    639 		if (pm->pm_ptbl[i]) {
    640 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
    641 			pm->pm_ptbl[i] = NULL;
    642 		}
    643 	if (pm->pm_ctx) ctx_free(pm);
    644 }
    645 
    646 /*
    647  * Copy the range specified by src_addr/len
    648  * from the source map to the range dst_addr/len
    649  * in the destination map.
    650  *
    651  * This routine is only advisory and need not do anything.
    652  */
    653 void
    654 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    655 	  vsize_t len, vaddr_t src_addr)
    656 {
    657 }
    658 
    659 /*
    660  * Require that all active physical maps contain no
    661  * incorrect entries NOW.
    662  */
    663 void
    664 pmap_update(struct pmap *pmap)
    665 {
    666 }
    667 
    668 /*
    669  * Garbage collects the physical map system for
    670  * pages which are no longer used.
    671  * Success need not be guaranteed -- that is, there
    672  * may well be pages which are not referenced, but
    673  * others may be collected.
    674  * Called by the pageout daemon when pages are scarce.
    675  */
    676 void
    677 pmap_collect(struct pmap *pm)
    678 {
    679 }
    680 
    681 /*
    682  * Fill the given physical page with zeroes.
    683  */
    684 void
    685 pmap_zero_page(paddr_t pa)
    686 {
    687 
    688 #ifdef PPC_4XX_NOCACHE
    689 	memset((caddr_t)pa, 0, NBPG);
    690 #else
    691 	int i;
    692 
    693 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
    694 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    695 		pa += CACHELINESIZE;
    696 	}
    697 #endif
    698 }
    699 
    700 /*
    701  * Copy the given physical source page to its destination.
    702  */
    703 void
    704 pmap_copy_page(paddr_t src, paddr_t dst)
    705 {
    706 
    707 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
    708 	dcache_flush_page(dst);
    709 }
    710 
    711 /*
    712  * This returns whether this is the first mapping of a page.
    713  */
    714 static inline int
    715 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    716 {
    717 	struct pv_entry *pv, *npv = NULL;
    718 	int s;
    719 
    720 	if (!pmap_initialized)
    721 		return 0;
    722 
    723 	s = splvm();
    724 
    725 	pv = pa_to_pv(pa);
    726 for (npv = pv; npv; npv = npv->pv_next)
    727 if (npv->pv_va == va && npv->pv_pm == pm) {
    728 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    729 #ifdef DDB
    730 Debugger();
    731 #endif
    732 return (1);
    733 }
    734 
    735 	if (!pv->pv_pm) {
    736 		/*
    737 		 * No entries yet, use header as the first entry.
    738 		 */
    739 		pv->pv_va = va;
    740 		pv->pv_pm = pm;
    741 		pv->pv_next = NULL;
    742 	} else {
    743 		/*
    744 		 * There is at least one other VA mapping this page.
    745 		 * Place this entry after the header.
    746 		 */
    747 		npv = pool_get(&pv_pool, PR_WAITOK);
    748 		if (!npv) return (0);
    749 		npv->pv_va = va;
    750 		npv->pv_pm = pm;
    751 		npv->pv_next = pv->pv_next;
    752 		pv->pv_next = npv;
    753 	}
    754 	splx(s);
    755 	return (1);
    756 }
    757 
    758 static void
    759 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    760 {
    761 	struct pv_entry *pv, *npv;
    762 
    763 	/*
    764 	 * Remove from the PV table.
    765 	 */
    766 	pv = pa_to_pv(pa);
    767 	if (!pv) return;
    768 
    769 	/*
    770 	 * If it is the first entry on the list, it is actually
    771 	 * in the header and we must copy the following entry up
    772 	 * to the header.  Otherwise we must search the list for
    773 	 * the entry.  In either case we free the now unused entry.
    774 	 */
    775 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    776 		if ((npv = pv->pv_next)) {
    777 			*pv = *npv;
    778 			pool_put(&pv_pool, npv);
    779 		} else
    780 			pv->pv_pm = NULL;
    781 	} else {
    782 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    783 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    784 				break;
    785 		if (npv) {
    786 			pv->pv_next = npv->pv_next;
    787 			pool_put(&pv_pool, npv);
    788 		}
    789 	}
    790 }
    791 
    792 /*
    793  * Insert physical page at pa into the given pmap at virtual address va.
    794  */
    795 int
    796 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    797 {
    798 	int s;
    799 	u_int tte;
    800 	int managed;
    801 
    802 	/*
    803 	 * Have to remove any existing mapping first.
    804 	 */
    805 	pmap_remove(pm, va, va + NBPG);
    806 
    807 	if (flags & PMAP_WIRED) flags |= prot;
    808 
    809 	/* If it has no protections don't bother w/the rest */
    810 	if (!(flags & VM_PROT_ALL))
    811 		return (0);
    812 
    813 	managed = 0;
    814 	if (vm_physseg_find(atop(pa), NULL) != -1)
    815 		managed = 1;
    816 
    817 	/*
    818 	 * Generate TTE.
    819 	 *
    820 	 * XXXX
    821 	 *
    822 	 * Since the kernel does not handle execution privileges properly,
    823 	 * we will handle read and execute permissions together.
    824 	 */
    825 	tte = TTE_PA(pa) | TTE_EX;
    826 	/* XXXX -- need to support multiple page sizes. */
    827 	tte |= TTE_SZ_16K;
    828 #ifdef	DIAGNOSTIC
    829 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    830 		(PME_NOCACHE | PME_WRITETHROUG))
    831 		panic("pmap_enter: uncached & writethrough");
    832 #endif
    833 	if (flags & PME_NOCACHE)
    834 		/* Must be I/O mapping */
    835 		tte |= TTE_I | TTE_G;
    836 #ifdef PPC_4XX_NOCACHE
    837 	tte |= TTE_I;
    838 #else
    839 	else if (flags & PME_WRITETHROUG)
    840 		/* Uncached and writethrough are not compatible */
    841 		tte |= TTE_W;
    842 #endif
    843 	if (pm == pmap_kernel())
    844 		tte |= TTE_ZONE(ZONE_PRIV);
    845 	else
    846 		tte |= TTE_ZONE(ZONE_USER);
    847 
    848 	if (flags & VM_PROT_WRITE)
    849 		tte |= TTE_WR;
    850 
    851 	/*
    852 	 * Now record mapping for later back-translation.
    853 	 */
    854 	if (pmap_initialized && managed) {
    855 		char *attr;
    856 
    857 		if (!pmap_enter_pv(pm, va, pa)) {
    858 			/* Could not enter pv on a managed page */
    859 			return 1;
    860 		}
    861 
    862 		/* Now set attributes. */
    863 		attr = pa_to_attr(pa);
    864 #ifdef DIAGNOSTIC
    865 		if (!attr)
    866 			panic("managed but no attr");
    867 #endif
    868 		if (flags & VM_PROT_ALL)
    869 			*attr |= PTE_HI_REF;
    870 		if (flags & VM_PROT_WRITE)
    871 			*attr |= PTE_HI_CHG;
    872 	}
    873 
    874 	s = splvm();
    875 	pm->pm_stats.resident_count++;
    876 
    877 	/* Insert page into page table. */
    878 	pte_enter(pm, va, tte);
    879 
    880 	/* If this is a real fault, enter it in the tlb */
    881 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    882 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    883 	}
    884 	splx(s);
    885 
    886 	/* Flush the real memory from the instruction cache. */
    887 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    888 		__syncicache((void *)pa, PAGE_SIZE);
    889 
    890 	return 0;
    891 }
    892 
    893 void
    894 pmap_unwire(struct pmap *pm, vaddr_t va)
    895 {
    896 	struct pv_entry *pv, *npv;
    897 	paddr_t pa;
    898 	int s = splvm();
    899 
    900 	if (pm == NULL) {
    901 		return;
    902 	}
    903 
    904 	if (!pmap_extract(pm, va, &pa)) {
    905 		return;
    906 	}
    907 
    908 	va |= PV_WIRED;
    909 
    910 	pv = pa_to_pv(pa);
    911 	if (!pv) return;
    912 
    913 	/*
    914 	 * If it is the first entry on the list, it is actually
    915 	 * in the header and we must copy the following entry up
    916 	 * to the header.  Otherwise we must search the list for
    917 	 * the entry.  In either case we free the now unused entry.
    918 	 */
    919 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    920 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    921 			npv->pv_va &= ~PV_WIRED;
    922 			break;
    923 		}
    924 	}
    925 	splx(s);
    926 }
    927 
    928 void
    929 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    930 {
    931 	int s;
    932 	u_int tte;
    933 	struct pmap *pm = pmap_kernel();
    934 
    935 	/*
    936 	 * Have to remove any existing mapping first.
    937 	 */
    938 
    939 	/*
    940 	 * Generate TTE.
    941 	 *
    942 	 * XXXX
    943 	 *
    944 	 * Since the kernel does not handle execution privileges properly,
    945 	 * we will handle read and execute permissions together.
    946 	 */
    947 	tte = 0;
    948 	if (prot & VM_PROT_ALL) {
    949 
    950 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    951 		/* XXXX -- need to support multiple page sizes. */
    952 		tte |= TTE_SZ_16K;
    953 #ifdef DIAGNOSTIC
    954 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    955 			(PME_NOCACHE | PME_WRITETHROUG))
    956 			panic("pmap_kenter_pa: uncached & writethrough");
    957 #endif
    958 		if (prot & PME_NOCACHE)
    959 			/* Must be I/O mapping */
    960 			tte |= TTE_I | TTE_G;
    961 #ifdef PPC_4XX_NOCACHE
    962 		tte |= TTE_I;
    963 #else
    964 		else if (prot & PME_WRITETHROUG)
    965 			/* Uncached and writethrough are not compatible */
    966 			tte |= TTE_W;
    967 #endif
    968 		if (prot & VM_PROT_WRITE)
    969 			tte |= TTE_WR;
    970 	}
    971 
    972 	s = splvm();
    973 	pm->pm_stats.resident_count++;
    974 
    975 	/* Insert page into page table. */
    976 	pte_enter(pm, va, tte);
    977 	splx(s);
    978 }
    979 
    980 void
    981 pmap_kremove(vaddr_t va, vsize_t len)
    982 {
    983 
    984 	while (len > 0) {
    985 		pte_enter(pmap_kernel(), va, 0);
    986 		va += PAGE_SIZE;
    987 		len -= PAGE_SIZE;
    988 	}
    989 }
    990 
    991 /*
    992  * Remove the given range of mapping entries.
    993  */
    994 void
    995 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
    996 {
    997 	int s;
    998 	paddr_t pa;
    999 	volatile u_int *ptp;
   1000 
   1001 	s = splvm();
   1002 	while (va < endva) {
   1003 
   1004 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1005 			pa = TTE_PA(pa);
   1006 			pmap_remove_pv(pm, va, pa);
   1007 			*ptp = 0;
   1008 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1009 			pm->pm_stats.resident_count--;
   1010 		}
   1011 		va += NBPG;
   1012 	}
   1013 
   1014 	splx(s);
   1015 }
   1016 
   1017 /*
   1018  * Get the physical page address for the given pmap/virtual address.
   1019  */
   1020 boolean_t
   1021 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1022 {
   1023 	int seg = STIDX(va);
   1024 	int ptn = PTIDX(va);
   1025 	u_int pa = 0;
   1026 	int s = splvm();
   1027 
   1028 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1029 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1030 	}
   1031 	splx(s);
   1032 	return (pa != 0);
   1033 }
   1034 
   1035 /*
   1036  * Lower the protection on the specified range of this pmap.
   1037  *
   1038  * There are only two cases: either the protection is going to 0,
   1039  * or it is going to read-only.
   1040  */
   1041 void
   1042 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1043 {
   1044 	volatile u_int *ptp;
   1045 	int s;
   1046 
   1047 	if (prot & VM_PROT_READ) {
   1048 		s = splvm();
   1049 		while (sva < eva) {
   1050 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1051 				*ptp &= ~TTE_WR;
   1052 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1053 			}
   1054 			sva += NBPG;
   1055 		}
   1056 		splx(s);
   1057 		return;
   1058 	}
   1059 	pmap_remove(pm, sva, eva);
   1060 }
   1061 
   1062 boolean_t
   1063 check_attr(struct vm_page *pg, u_int mask, int clear)
   1064 {
   1065 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1066 	int s;
   1067 	char *attr;
   1068 	int rv;
   1069 
   1070 	/*
   1071 	 * First modify bits in cache.
   1072 	 */
   1073 	s = splvm();
   1074 	attr = pa_to_attr(pa);
   1075 	if (attr == NULL)
   1076 		return FALSE;
   1077 
   1078 	rv = ((*attr & mask) != 0);
   1079 	if (clear) {
   1080 		*attr &= ~mask;
   1081 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1082 	}
   1083 	splx(s);
   1084 	return rv;
   1085 }
   1086 
   1087 
   1088 /*
   1089  * Lower the protection on the specified physical page.
   1090  *
   1091  * There are only two cases: either the protection is going to 0,
   1092  * or it is going to read-only.
   1093  */
   1094 void
   1095 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1096 {
   1097 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1098 	vaddr_t va;
   1099 	struct pv_entry *pvh, *pv, *npv;
   1100 	struct pmap *pm;
   1101 
   1102 	pvh = pa_to_pv(pa);
   1103 	if (pvh == NULL)
   1104 		return;
   1105 
   1106 	/* Handle extra pvs which may be deleted in the operation */
   1107 	for (pv = pvh->pv_next; pv; pv = npv) {
   1108 		npv = pv->pv_next;
   1109 
   1110 		pm = pv->pv_pm;
   1111 		va = pv->pv_va;
   1112 		pmap_protect(pm, va, va+NBPG, prot);
   1113 	}
   1114 	/* Now check the head pv */
   1115 	if (pvh->pv_pm) {
   1116 		pv = pvh;
   1117 		pm = pv->pv_pm;
   1118 		va = pv->pv_va;
   1119 		pmap_protect(pm, va, va+NBPG, prot);
   1120 	}
   1121 }
   1122 
   1123 /*
   1124  * Activate the address space for the specified process.  If the process
   1125  * is the current process, load the new MMU context.
   1126  */
   1127 void
   1128 pmap_activate(struct lwp *l)
   1129 {
   1130 #if 0
   1131 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1132 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1133 
   1134 	/*
   1135 	 * XXX Normally performed in cpu_fork().
   1136 	 */
   1137 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1138 	if (pcb->pcb_pm != pmap) {
   1139 		pcb->pcb_pm = pmap;
   1140 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1141 		    (paddr_t *)&pcb->pcb_pmreal);
   1142 	}
   1143 
   1144 	if (l == curlwp) {
   1145 		/* Store pointer to new current pmap. */
   1146 		curpm = pcb->pcb_pmreal;
   1147 	}
   1148 #endif
   1149 }
   1150 
   1151 /*
   1152  * Deactivate the specified process's address space.
   1153  */
   1154 void
   1155 pmap_deactivate(struct lwp *l)
   1156 {
   1157 }
   1158 
   1159 /*
   1160  * Synchronize caches corresponding to [addr, addr+len) in p.
   1161  */
   1162 void
   1163 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1164 {
   1165 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1166 	int msr, ctx, opid, step;
   1167 
   1168 
   1169 	step = CACHELINESIZE;
   1170 
   1171 	/*
   1172 	 * Need to turn off IMMU and switch to user context.
   1173 	 * (icbi uses DMMU).
   1174 	 */
   1175 	if (!(ctx = pm->pm_ctx)) {
   1176 		/* No context -- assign it one */
   1177 		ctx_alloc(pm);
   1178 		ctx = pm->pm_ctx;
   1179 	}
   1180 	__asm __volatile("mfmsr %0;"
   1181 		"li %1, 0x20;"
   1182 		"andc %1,%0,%1;"
   1183 		"mtmsr %1;"
   1184 		"sync;isync;"
   1185 		"mfpid %1;"
   1186 		"mtpid %2;"
   1187 		"sync; isync;"
   1188 		"1:"
   1189 		"dcbf 0,%3;"
   1190 		"icbi 0,%3;"
   1191 		"add %3,%3,%5;"
   1192 		"addc. %4,%4,%6;"
   1193 		"bge 1b;"
   1194 		"mtpid %1;"
   1195 		"mtmsr %0;"
   1196 		"sync; isync"
   1197 		: "=&r" (msr), "=&r" (opid)
   1198 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step));
   1199 }
   1200 
   1201 
   1202 /* This has to be done in real mode !!! */
   1203 void
   1204 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1205 {
   1206 	u_long i, found;
   1207 	u_long msr;
   1208 
   1209 	/* If there's no context then it can't be mapped. */
   1210 	if (!pid) return;
   1211 
   1212 	asm("mfpid %1;"			/* Save PID */
   1213 		"mfmsr %2;"		/* Save MSR */
   1214 		"li %0,0;"		/* Now clear MSR */
   1215 		"mtmsr %0;"
   1216 		"mtpid %4;"		/* Set PID */
   1217 		"sync;"
   1218 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1219 		"sync;"
   1220 		"mtpid %1;"		/* Restore PID */
   1221 		"mtmsr %2;"		/* Restore MSR */
   1222 		"sync;isync;"
   1223 		"li %1,1;"
   1224 		"beq 1f;"
   1225 		"li %1,0;"
   1226 		"1:"
   1227 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1228 		: "r" (va), "r" (pid));
   1229 	if (found && !TLB_LOCKED(i)) {
   1230 
   1231 		/* Now flush translation */
   1232 		asm volatile(
   1233 			"tlbwe %0,%1,0;"
   1234 			"sync;isync;"
   1235 			: : "r" (0), "r" (i));
   1236 
   1237 		tlb_info[i].ti_ctx = 0;
   1238 		tlb_info[i].ti_flags = 0;
   1239 		tlbnext = i;
   1240 		/* Successful flushes */
   1241 		tlbflush_ev.ev_count++;
   1242 	}
   1243 }
   1244 
   1245 void
   1246 ppc4xx_tlb_flush_all(void)
   1247 {
   1248 	u_long i;
   1249 
   1250 	for (i = 0; i < NTLB; i++)
   1251 		if (!TLB_LOCKED(i)) {
   1252 			asm volatile(
   1253 				"tlbwe %0,%1,0;"
   1254 				"sync;isync;"
   1255 				: : "r" (0), "r" (i));
   1256 			tlb_info[i].ti_ctx = 0;
   1257 			tlb_info[i].ti_flags = 0;
   1258 		}
   1259 
   1260 	asm volatile("sync;isync");
   1261 }
   1262 
   1263 /* Find a TLB entry to evict. */
   1264 static int
   1265 ppc4xx_tlb_find_victim(void)
   1266 {
   1267 	int flags;
   1268 
   1269 	for (;;) {
   1270 		if (++tlbnext >= NTLB)
   1271 			tlbnext = TLB_NRESERVED;
   1272 		flags = tlb_info[tlbnext].ti_flags;
   1273 		if (!(flags & TLBF_USED) ||
   1274 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1275 			u_long va, stack = (u_long)&va;
   1276 
   1277 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1278 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1279 			     (flags & TLBF_USED)) {
   1280 				/* Kernel stack page */
   1281 				flags |= TLBF_USED;
   1282 				tlb_info[tlbnext].ti_flags = flags;
   1283 			} else {
   1284 				/* Found it! */
   1285 				return (tlbnext);
   1286 			}
   1287 		} else {
   1288 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1289 		}
   1290 	}
   1291 }
   1292 
   1293 void
   1294 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1295 {
   1296 	u_long th, tl, idx;
   1297 	tlbpid_t pid;
   1298 	u_short msr;
   1299 	paddr_t pa;
   1300 	int s, sz;
   1301 
   1302 	tlbenter_ev.ev_count++;
   1303 
   1304 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1305 	pa = (pte & TTE_RPN_MASK(sz));
   1306 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1307 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1308 	tl |= ppc4xx_tlbflags(va, pa);
   1309 
   1310 	s = splhigh();
   1311 	idx = ppc4xx_tlb_find_victim();
   1312 
   1313 #ifdef DIAGNOSTIC
   1314 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1315 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
   1316 	}
   1317 #endif
   1318 
   1319 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1320 	tlb_info[idx].ti_ctx = ctx;
   1321 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1322 
   1323 	asm volatile(
   1324 		"mfmsr %0;"			/* Save MSR */
   1325 		"li %1,0;"
   1326 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1327 		"mtmsr %1;"			/* Clear MSR */
   1328 		"mfpid %1;"			/* Save old PID */
   1329 		"mtpid %2;"			/* Load translation ctx */
   1330 		"sync; isync;"
   1331 #ifdef DEBUG
   1332 		"andi. %3,%3,63;"
   1333 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1334 #endif
   1335 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1336 		"sync; isync;"
   1337 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1338 		"sync; isync;"
   1339 	: "=&r" (msr), "=&r" (pid)
   1340 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1341 	splx(s);
   1342 }
   1343 
   1344 void
   1345 ppc4xx_tlb_unpin(int i)
   1346 {
   1347 
   1348 	if (i == -1)
   1349 		for (i = 0; i < TLB_NRESERVED; i++)
   1350 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1351 	else
   1352 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1353 }
   1354 
   1355 void
   1356 ppc4xx_tlb_init(void)
   1357 {
   1358 	int i;
   1359 
   1360 	/* Mark reserved TLB entries */
   1361 	for (i = 0; i < TLB_NRESERVED; i++) {
   1362 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1363 		tlb_info[i].ti_ctx = KERNEL_PID;
   1364 	}
   1365 
   1366 	/* Setup security zones */
   1367 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1368 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1369 	 * Z3 - full access regardless of TLB entry permissions
   1370 	 */
   1371 
   1372 	asm volatile(
   1373 		"mtspr %0,%1;"
   1374 		"sync;"
   1375 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1376 }
   1377 
   1378 
   1379 /*
   1380  * We should pass the ctx in from trap code.
   1381  */
   1382 int
   1383 pmap_tlbmiss(vaddr_t va, int ctx)
   1384 {
   1385 	volatile u_int *pte;
   1386 	u_long tte;
   1387 
   1388 	tlbmiss_ev.ev_count++;
   1389 
   1390 	/*
   1391 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1392 	 */
   1393 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1394 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1395 		if (pte == NULL) {
   1396 			/* Map unmanaged addresses directly for kernel access */
   1397 			return 1;
   1398 		}
   1399 		tte = *pte;
   1400 		if (tte == 0) {
   1401 			return 1;
   1402 		}
   1403 	} else {
   1404 		/* Create a 16MB writable mapping. */
   1405 #ifdef PPC_4XX_NOCACHE
   1406 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1407 #else
   1408 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1409 #endif
   1410 	}
   1411 	tlbhit_ev.ev_count++;
   1412 	ppc4xx_tlb_enter(ctx, va, tte);
   1413 
   1414 	return 0;
   1415 }
   1416 
   1417 /*
   1418  * Flush all the entries matching a context from the TLB.
   1419  */
   1420 static int
   1421 ctx_flush(int cnum)
   1422 {
   1423 	int i;
   1424 
   1425 	/* We gotta steal this context */
   1426 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1427 		if (tlb_info[i].ti_ctx == cnum) {
   1428 			/* Can't steal ctx if it has a locked entry. */
   1429 			if (TLB_LOCKED(i)) {
   1430 #ifdef DIAGNOSTIC
   1431 				printf("ctx_flush: can't invalidate "
   1432 					"locked mapping %d "
   1433 					"for context %d\n", i, cnum);
   1434 #ifdef DDB
   1435 				Debugger();
   1436 #endif
   1437 #endif
   1438 				return (1);
   1439 			}
   1440 #ifdef DIAGNOSTIC
   1441 			if (i < TLB_NRESERVED)
   1442 				panic("TLB entry %d not locked", i);
   1443 #endif
   1444 			/* Invalidate particular TLB entry regardless of locked status */
   1445 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1446 			tlb_info[i].ti_flags = 0;
   1447 		}
   1448 	}
   1449 	return (0);
   1450 }
   1451 
   1452 /*
   1453  * Allocate a context.  If necessary, steal one from someone else.
   1454  *
   1455  * The new context is flushed from the TLB before returning.
   1456  */
   1457 int
   1458 ctx_alloc(struct pmap *pm)
   1459 {
   1460 	int s, cnum;
   1461 	static int next = MINCTX;
   1462 
   1463 	if (pm == pmap_kernel()) {
   1464 #ifdef DIAGNOSTIC
   1465 		printf("ctx_alloc: kernel pmap!\n");
   1466 #endif
   1467 		return (0);
   1468 	}
   1469 	s = splvm();
   1470 
   1471 	/* Find a likely context. */
   1472 	cnum = next;
   1473 	do {
   1474 		if ((++cnum) > NUMCTX)
   1475 			cnum = MINCTX;
   1476 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1477 
   1478 	/* Now clean it out */
   1479 oops:
   1480 	if (cnum < MINCTX)
   1481 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1482 	if (ctx_flush(cnum)) {
   1483 		/* oops -- something's wired. */
   1484 		if ((++cnum) > NUMCTX)
   1485 			cnum = MINCTX;
   1486 		goto oops;
   1487 	}
   1488 
   1489 	if (ctxbusy[cnum]) {
   1490 #ifdef DEBUG
   1491 		/* We should identify this pmap and clear it */
   1492 		printf("Warning: stealing context %d\n", cnum);
   1493 #endif
   1494 		ctxbusy[cnum]->pm_ctx = 0;
   1495 	}
   1496 	ctxbusy[cnum] = pm;
   1497 	next = cnum;
   1498 	splx(s);
   1499 	pm->pm_ctx = cnum;
   1500 
   1501 	return cnum;
   1502 }
   1503 
   1504 /*
   1505  * Give away a context.
   1506  */
   1507 void
   1508 ctx_free(struct pmap *pm)
   1509 {
   1510 	int oldctx;
   1511 
   1512 	oldctx = pm->pm_ctx;
   1513 
   1514 	if (oldctx == 0)
   1515 		panic("ctx_free: freeing kernel context");
   1516 #ifdef DIAGNOSTIC
   1517 	if (ctxbusy[oldctx] == 0)
   1518 		printf("ctx_free: freeing free context %d\n", oldctx);
   1519 	if (ctxbusy[oldctx] != pm) {
   1520 		printf("ctx_free: freeing someone esle's context\n "
   1521 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1522 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1523 #ifdef DDB
   1524 		Debugger();
   1525 #endif
   1526 	}
   1527 #endif
   1528 	/* We should verify it has not been stolen and reallocated... */
   1529 	ctxbusy[oldctx] = NULL;
   1530 	ctx_flush(oldctx);
   1531 }
   1532 
   1533 
   1534 #ifdef DEBUG
   1535 /*
   1536  * Test ref/modify handling.
   1537  */
   1538 void pmap_testout __P((void));
   1539 void
   1540 pmap_testout()
   1541 {
   1542 	vaddr_t va;
   1543 	volatile int *loc;
   1544 	int val = 0;
   1545 	paddr_t pa;
   1546 	struct vm_page *pg;
   1547 	int ref, mod;
   1548 
   1549 	/* Allocate a page */
   1550 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
   1551 	loc = (int*)va;
   1552 
   1553 	pmap_extract(pmap_kernel(), va, &pa);
   1554 	pg = PHYS_TO_VM_PAGE(pa);
   1555 	pmap_unwire(pmap_kernel(), va);
   1556 
   1557 	pmap_remove(pmap_kernel(), va, va+1);
   1558 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1559 	pmap_update(pmap_kernel());
   1560 
   1561 	/* Now clear reference and modify */
   1562 	ref = pmap_clear_reference(pg);
   1563 	mod = pmap_clear_modify(pg);
   1564 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1565 	       (void *)(u_long)va, (long)pa,
   1566 	       ref, mod);
   1567 
   1568 	/* Check it's properly cleared */
   1569 	ref = pmap_is_referenced(pg);
   1570 	mod = pmap_is_modified(pg);
   1571 	printf("Checking cleared page: ref %d, mod %d\n",
   1572 	       ref, mod);
   1573 
   1574 	/* Reference page */
   1575 	val = *loc;
   1576 
   1577 	ref = pmap_is_referenced(pg);
   1578 	mod = pmap_is_modified(pg);
   1579 	printf("Referenced page: ref %d, mod %d val %x\n",
   1580 	       ref, mod, val);
   1581 
   1582 	/* Now clear reference and modify */
   1583 	ref = pmap_clear_reference(pg);
   1584 	mod = pmap_clear_modify(pg);
   1585 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1586 	       (void *)(u_long)va, (long)pa,
   1587 	       ref, mod);
   1588 
   1589 	/* Modify page */
   1590 	*loc = 1;
   1591 
   1592 	ref = pmap_is_referenced(pg);
   1593 	mod = pmap_is_modified(pg);
   1594 	printf("Modified page: ref %d, mod %d\n",
   1595 	       ref, mod);
   1596 
   1597 	/* Now clear reference and modify */
   1598 	ref = pmap_clear_reference(pg);
   1599 	mod = pmap_clear_modify(pg);
   1600 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1601 	       (void *)(u_long)va, (long)pa,
   1602 	       ref, mod);
   1603 
   1604 	/* Check it's properly cleared */
   1605 	ref = pmap_is_referenced(pg);
   1606 	mod = pmap_is_modified(pg);
   1607 	printf("Checking cleared page: ref %d, mod %d\n",
   1608 	       ref, mod);
   1609 
   1610 	/* Modify page */
   1611 	*loc = 1;
   1612 
   1613 	ref = pmap_is_referenced(pg);
   1614 	mod = pmap_is_modified(pg);
   1615 	printf("Modified page: ref %d, mod %d\n",
   1616 	       ref, mod);
   1617 
   1618 	/* Check pmap_protect() */
   1619 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1620 	pmap_update(pmap_kernel());
   1621 	ref = pmap_is_referenced(pg);
   1622 	mod = pmap_is_modified(pg);
   1623 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1624 	       ref, mod);
   1625 
   1626 	/* Now clear reference and modify */
   1627 	ref = pmap_clear_reference(pg);
   1628 	mod = pmap_clear_modify(pg);
   1629 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1630 	       (void *)(u_long)va, (long)pa,
   1631 	       ref, mod);
   1632 
   1633 	/* Reference page */
   1634 	val = *loc;
   1635 
   1636 	ref = pmap_is_referenced(pg);
   1637 	mod = pmap_is_modified(pg);
   1638 	printf("Referenced page: ref %d, mod %d val %x\n",
   1639 	       ref, mod, val);
   1640 
   1641 	/* Now clear reference and modify */
   1642 	ref = pmap_clear_reference(pg);
   1643 	mod = pmap_clear_modify(pg);
   1644 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1645 	       (void *)(u_long)va, (long)pa,
   1646 	       ref, mod);
   1647 
   1648 	/* Modify page */
   1649 #if 0
   1650 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1651 	pmap_update(pmap_kernel());
   1652 #endif
   1653 	*loc = 1;
   1654 
   1655 	ref = pmap_is_referenced(pg);
   1656 	mod = pmap_is_modified(pg);
   1657 	printf("Modified page: ref %d, mod %d\n",
   1658 	       ref, mod);
   1659 
   1660 	/* Check pmap_protect() */
   1661 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1662 	pmap_update(pmap_kernel());
   1663 	ref = pmap_is_referenced(pg);
   1664 	mod = pmap_is_modified(pg);
   1665 	printf("pmap_protect(): ref %d, mod %d\n",
   1666 	       ref, mod);
   1667 
   1668 	/* Now clear reference and modify */
   1669 	ref = pmap_clear_reference(pg);
   1670 	mod = pmap_clear_modify(pg);
   1671 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1672 	       (void *)(u_long)va, (long)pa,
   1673 	       ref, mod);
   1674 
   1675 	/* Reference page */
   1676 	val = *loc;
   1677 
   1678 	ref = pmap_is_referenced(pg);
   1679 	mod = pmap_is_modified(pg);
   1680 	printf("Referenced page: ref %d, mod %d val %x\n",
   1681 	       ref, mod, val);
   1682 
   1683 	/* Now clear reference and modify */
   1684 	ref = pmap_clear_reference(pg);
   1685 	mod = pmap_clear_modify(pg);
   1686 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1687 	       (void *)(u_long)va, (long)pa,
   1688 	       ref, mod);
   1689 
   1690 	/* Modify page */
   1691 #if 0
   1692 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1693 	pmap_update(pmap_kernel());
   1694 #endif
   1695 	*loc = 1;
   1696 
   1697 	ref = pmap_is_referenced(pg);
   1698 	mod = pmap_is_modified(pg);
   1699 	printf("Modified page: ref %d, mod %d\n",
   1700 	       ref, mod);
   1701 
   1702 	/* Check pmap_pag_protect() */
   1703 	pmap_page_protect(pg, VM_PROT_READ);
   1704 	ref = pmap_is_referenced(pg);
   1705 	mod = pmap_is_modified(pg);
   1706 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1707 	       ref, mod);
   1708 
   1709 	/* Now clear reference and modify */
   1710 	ref = pmap_clear_reference(pg);
   1711 	mod = pmap_clear_modify(pg);
   1712 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1713 	       (void *)(u_long)va, (long)pa,
   1714 	       ref, mod);
   1715 
   1716 	/* Reference page */
   1717 	val = *loc;
   1718 
   1719 	ref = pmap_is_referenced(pg);
   1720 	mod = pmap_is_modified(pg);
   1721 	printf("Referenced page: ref %d, mod %d val %x\n",
   1722 	       ref, mod, val);
   1723 
   1724 	/* Now clear reference and modify */
   1725 	ref = pmap_clear_reference(pg);
   1726 	mod = pmap_clear_modify(pg);
   1727 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1728 	       (void *)(u_long)va, (long)pa,
   1729 	       ref, mod);
   1730 
   1731 	/* Modify page */
   1732 #if 0
   1733 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1734 	pmap_update(pmap_kernel());
   1735 #endif
   1736 	*loc = 1;
   1737 
   1738 	ref = pmap_is_referenced(pg);
   1739 	mod = pmap_is_modified(pg);
   1740 	printf("Modified page: ref %d, mod %d\n",
   1741 	       ref, mod);
   1742 
   1743 	/* Check pmap_pag_protect() */
   1744 	pmap_page_protect(pg, VM_PROT_NONE);
   1745 	ref = pmap_is_referenced(pg);
   1746 	mod = pmap_is_modified(pg);
   1747 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1748 	       ref, mod);
   1749 
   1750 	/* Now clear reference and modify */
   1751 	ref = pmap_clear_reference(pg);
   1752 	mod = pmap_clear_modify(pg);
   1753 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1754 	       (void *)(u_long)va, (long)pa,
   1755 	       ref, mod);
   1756 
   1757 
   1758 	/* Reference page */
   1759 	val = *loc;
   1760 
   1761 	ref = pmap_is_referenced(pg);
   1762 	mod = pmap_is_modified(pg);
   1763 	printf("Referenced page: ref %d, mod %d val %x\n",
   1764 	       ref, mod, val);
   1765 
   1766 	/* Now clear reference and modify */
   1767 	ref = pmap_clear_reference(pg);
   1768 	mod = pmap_clear_modify(pg);
   1769 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1770 	       (void *)(u_long)va, (long)pa,
   1771 	       ref, mod);
   1772 
   1773 	/* Modify page */
   1774 #if 0
   1775 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1776 	pmap_update(pmap_kernel());
   1777 #endif
   1778 	*loc = 1;
   1779 
   1780 	ref = pmap_is_referenced(pg);
   1781 	mod = pmap_is_modified(pg);
   1782 	printf("Modified page: ref %d, mod %d\n",
   1783 	       ref, mod);
   1784 
   1785 	/* Unmap page */
   1786 	pmap_remove(pmap_kernel(), va, va+1);
   1787 	pmap_update(pmap_kernel());
   1788 	ref = pmap_is_referenced(pg);
   1789 	mod = pmap_is_modified(pg);
   1790 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1791 
   1792 	/* Now clear reference and modify */
   1793 	ref = pmap_clear_reference(pg);
   1794 	mod = pmap_clear_modify(pg);
   1795 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1796 	       (void *)(u_long)va, (long)pa, ref, mod);
   1797 
   1798 	/* Check it's properly cleared */
   1799 	ref = pmap_is_referenced(pg);
   1800 	mod = pmap_is_modified(pg);
   1801 	printf("Checking cleared page: ref %d, mod %d\n",
   1802 	       ref, mod);
   1803 
   1804 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1805 		VM_PROT_ALL|PMAP_WIRED);
   1806 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
   1807 }
   1808 #endif
   1809