Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.21.2.2
      1 /*	$NetBSD: pmap.c,v 1.21.2.2 2004/09/18 14:39:11 skrll 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/cdefs.h>
     70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.21.2.2 2004/09/18 14:39:11 skrll Exp $");
     71 
     72 #include <sys/param.h>
     73 #include <sys/malloc.h>
     74 #include <sys/proc.h>
     75 #include <sys/user.h>
     76 #include <sys/queue.h>
     77 #include <sys/systm.h>
     78 #include <sys/pool.h>
     79 #include <sys/device.h>
     80 
     81 #include <uvm/uvm.h>
     82 
     83 #include <machine/cpu.h>
     84 #include <machine/pcb.h>
     85 #include <machine/powerpc.h>
     86 
     87 #include <powerpc/spr.h>
     88 #include <machine/tlb.h>
     89 
     90 /*
     91  * kernmap is an array of PTEs large enough to map in
     92  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     93  */
     94 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
     95 caddr_t kernmap;
     96 
     97 #define MINCTX		2
     98 #define NUMCTX		256
     99 volatile struct pmap *ctxbusy[NUMCTX];
    100 
    101 #define TLBF_USED	0x1
    102 #define	TLBF_REF	0x2
    103 #define	TLBF_LOCKED	0x4
    104 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    105 typedef struct tlb_info_s {
    106 	char	ti_flags;
    107 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    108 	u_int	ti_va;
    109 } tlb_info_t;
    110 
    111 volatile tlb_info_t tlb_info[NTLB];
    112 /* We'll use a modified FIFO replacement policy cause it's cheap */
    113 volatile int tlbnext = TLB_NRESERVED;
    114 
    115 u_long dtlb_miss_count = 0;
    116 u_long itlb_miss_count = 0;
    117 u_long ktlb_miss_count = 0;
    118 u_long utlb_miss_count = 0;
    119 
    120 /* Event counters */
    121 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    122 	NULL, "cpu", "tlbmiss");
    123 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    124 	NULL, "cpu", "tlbhit");
    125 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    126 	NULL, "cpu", "tlbflush");
    127 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
    128 	NULL, "cpu", "tlbenter");
    129 
    130 struct pmap kernel_pmap_;
    131 
    132 int physmem;
    133 static int npgs;
    134 static u_int nextavail;
    135 #ifndef MSGBUFADDR
    136 extern paddr_t msgbuf_paddr;
    137 #endif
    138 
    139 static struct mem_region *mem, *avail;
    140 
    141 /*
    142  * This is a cache of referenced/modified bits.
    143  * Bits herein are shifted by ATTRSHFT.
    144  */
    145 static char *pmap_attrib;
    146 
    147 #define PV_WIRED	0x1
    148 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    149 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    150 
    151 struct pv_entry {
    152 	struct pv_entry *pv_next;	/* Linked list of mappings */
    153 	vaddr_t pv_va;			/* virtual address of mapping */
    154 	struct pmap *pv_pm;
    155 };
    156 
    157 struct pv_entry *pv_table;
    158 static struct pool pv_pool;
    159 
    160 static int pmap_initialized;
    161 
    162 static int ctx_flush(int);
    163 
    164 inline struct pv_entry *pa_to_pv(paddr_t);
    165 static inline char *pa_to_attr(paddr_t);
    166 
    167 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    168 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    169 
    170 static void pmap_pinit(pmap_t);
    171 static void pmap_release(pmap_t);
    172 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    173 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    174 
    175 
    176 inline struct pv_entry *
    177 pa_to_pv(paddr_t pa)
    178 {
    179 	int bank, pg;
    180 
    181 	bank = vm_physseg_find(atop(pa), &pg);
    182 	if (bank == -1)
    183 		return NULL;
    184 	return &vm_physmem[bank].pmseg.pvent[pg];
    185 }
    186 
    187 static inline char *
    188 pa_to_attr(paddr_t pa)
    189 {
    190 	int bank, pg;
    191 
    192 	bank = vm_physseg_find(atop(pa), &pg);
    193 	if (bank == -1)
    194 		return NULL;
    195 	return &vm_physmem[bank].pmseg.attrs[pg];
    196 }
    197 
    198 /*
    199  * Insert PTE into page table.
    200  */
    201 int
    202 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    203 {
    204 	int seg = STIDX(va);
    205 	int ptn = PTIDX(va);
    206 	u_int oldpte;
    207 
    208 	if (!pm->pm_ptbl[seg]) {
    209 		/* Don't allocate a page to clear a non-existent mapping. */
    210 		if (!pte) return (0);
    211 		/* Allocate a page XXXX this will sleep! */
    212 		pm->pm_ptbl[seg] =
    213 		    (uint *)uvm_km_zalloc(kernel_map, PAGE_SIZE);
    214 	}
    215 	oldpte = pm->pm_ptbl[seg][ptn];
    216 	pm->pm_ptbl[seg][ptn] = pte;
    217 
    218 	/* Flush entry. */
    219 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    220 	if (oldpte != pte) {
    221 		if (pte == 0)
    222 			pm->pm_stats.resident_count--;
    223 		else
    224 			pm->pm_stats.resident_count++;
    225 	}
    226 	return (1);
    227 }
    228 
    229 /*
    230  * Get a pointer to a PTE in a page table.
    231  */
    232 volatile u_int *
    233 pte_find(struct pmap *pm, vaddr_t va)
    234 {
    235 	int seg = STIDX(va);
    236 	int ptn = PTIDX(va);
    237 
    238 	if (pm->pm_ptbl[seg])
    239 		return (&pm->pm_ptbl[seg][ptn]);
    240 
    241 	return (NULL);
    242 }
    243 
    244 /*
    245  * This is called during initppc, before the system is really initialized.
    246  */
    247 void
    248 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    249 {
    250 	struct mem_region *mp, *mp1;
    251 	int cnt, i;
    252 	u_int s, e, sz;
    253 
    254 	/*
    255 	 * Allocate the kernel page table at the end of
    256 	 * kernel space so it's in the locked TTE.
    257 	 */
    258 	kernmap = (caddr_t)kernelend;
    259 
    260 	/*
    261 	 * Initialize kernel page table.
    262 	 */
    263 	for (i = 0; i < STSZ; i++) {
    264 		pmap_kernel()->pm_ptbl[i] = 0;
    265 	}
    266 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    267 
    268 	/*
    269 	 * Announce page-size to the VM-system
    270 	 */
    271 	uvmexp.pagesize = NBPG;
    272 	uvm_setpagesize();
    273 
    274 	/*
    275 	 * Get memory.
    276 	 */
    277 	mem_regions(&mem, &avail);
    278 	for (mp = mem; mp->size; mp++) {
    279 		physmem += btoc(mp->size);
    280 		printf("+%lx,",mp->size);
    281 	}
    282 	printf("\n");
    283 	ppc4xx_tlb_init();
    284 	/*
    285 	 * Count the number of available entries.
    286 	 */
    287 	for (cnt = 0, mp = avail; mp->size; mp++)
    288 		cnt++;
    289 
    290 	/*
    291 	 * Page align all regions.
    292 	 * Non-page aligned memory isn't very interesting to us.
    293 	 * Also, sort the entries for ascending addresses.
    294 	 */
    295 	kernelstart &= ~PGOFSET;
    296 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    297 	for (mp = avail; mp->size; mp++) {
    298 		s = mp->start;
    299 		e = mp->start + mp->size;
    300 		printf("%08x-%08x -> ",s,e);
    301 		/*
    302 		 * Check whether this region holds all of the kernel.
    303 		 */
    304 		if (s < kernelstart && e > kernelend) {
    305 			avail[cnt].start = kernelend;
    306 			avail[cnt++].size = e - kernelend;
    307 			e = kernelstart;
    308 		}
    309 		/*
    310 		 * Look whether this regions starts within the kernel.
    311 		 */
    312 		if (s >= kernelstart && s < kernelend) {
    313 			if (e <= kernelend)
    314 				goto empty;
    315 			s = kernelend;
    316 		}
    317 		/*
    318 		 * Now look whether this region ends within the kernel.
    319 		 */
    320 		if (e > kernelstart && e <= kernelend) {
    321 			if (s >= kernelstart)
    322 				goto empty;
    323 			e = kernelstart;
    324 		}
    325 		/*
    326 		 * Now page align the start and size of the region.
    327 		 */
    328 		s = round_page(s);
    329 		e = trunc_page(e);
    330 		if (e < s)
    331 			e = s;
    332 		sz = e - s;
    333 		printf("%08x-%08x = %x\n",s,e,sz);
    334 		/*
    335 		 * Check whether some memory is left here.
    336 		 */
    337 		if (sz == 0) {
    338 		empty:
    339 			memmove(mp, mp + 1,
    340 				(cnt - (mp - avail)) * sizeof *mp);
    341 			cnt--;
    342 			mp--;
    343 			continue;
    344 		}
    345 		/*
    346 		 * Do an insertion sort.
    347 		 */
    348 		npgs += btoc(sz);
    349 		for (mp1 = avail; mp1 < mp; mp1++)
    350 			if (s < mp1->start)
    351 				break;
    352 		if (mp1 < mp) {
    353 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    354 			mp1->start = s;
    355 			mp1->size = sz;
    356 		} else {
    357 			mp->start = s;
    358 			mp->size = sz;
    359 		}
    360 	}
    361 
    362 	/*
    363 	 * We cannot do pmap_steal_memory here,
    364 	 * since we don't run with translation enabled yet.
    365 	 */
    366 #ifndef MSGBUFADDR
    367 	/*
    368 	 * allow for msgbuf
    369 	 */
    370 	sz = round_page(MSGBUFSIZE);
    371 	mp = NULL;
    372 	for (mp1 = avail; mp1->size; mp1++)
    373 		if (mp1->size >= sz)
    374 			mp = mp1;
    375 	if (mp == NULL)
    376 		panic("not enough memory?");
    377 
    378 	npgs -= btoc(sz);
    379 	msgbuf_paddr = mp->start + mp->size - sz;
    380 	mp->size -= sz;
    381 	if (mp->size <= 0)
    382 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    383 #endif
    384 
    385 	for (mp = avail; mp->size; mp++)
    386 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    387 			atop(mp->start), atop(mp->start + mp->size),
    388 			VM_FREELIST_DEFAULT);
    389 
    390 	/*
    391 	 * Initialize kernel pmap and hardware.
    392 	 */
    393 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    394 	pmap_kernel()->pm_ctx = KERNEL_PID;
    395 	nextavail = avail->start;
    396 
    397 
    398 	evcnt_attach_static(&tlbmiss_ev);
    399 	evcnt_attach_static(&tlbhit_ev);
    400 	evcnt_attach_static(&tlbflush_ev);
    401 	evcnt_attach_static(&tlbenter_ev);
    402 }
    403 
    404 /*
    405  * Restrict given range to physical memory
    406  *
    407  * (Used by /dev/mem)
    408  */
    409 void
    410 pmap_real_memory(paddr_t *start, psize_t *size)
    411 {
    412 	struct mem_region *mp;
    413 
    414 	for (mp = mem; mp->size; mp++) {
    415 		if (*start + *size > mp->start &&
    416 		    *start < mp->start + mp->size) {
    417 			if (*start < mp->start) {
    418 				*size -= mp->start - *start;
    419 				*start = mp->start;
    420 			}
    421 			if (*start + *size > mp->start + mp->size)
    422 				*size = mp->start + mp->size - *start;
    423 			return;
    424 		}
    425 	}
    426 	*size = 0;
    427 }
    428 
    429 /*
    430  * Initialize anything else for pmap handling.
    431  * Called during vm_init().
    432  */
    433 void
    434 pmap_init(void)
    435 {
    436 	struct pv_entry *pv;
    437 	vsize_t sz;
    438 	vaddr_t addr;
    439 	int i, s;
    440 	int bank;
    441 	char *attr;
    442 
    443 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    444 	sz = round_page(sz);
    445 	addr = uvm_km_zalloc(kernel_map, sz);
    446 	s = splvm();
    447 	pv = pv_table = (struct pv_entry *)addr;
    448 	for (i = npgs; --i >= 0;)
    449 		pv++->pv_pm = NULL;
    450 	pmap_attrib = (char *)pv;
    451 	memset(pv, 0, npgs);
    452 
    453 	pv = pv_table;
    454 	attr = pmap_attrib;
    455 	for (bank = 0; bank < vm_nphysseg; bank++) {
    456 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    457 		vm_physmem[bank].pmseg.pvent = pv;
    458 		vm_physmem[bank].pmseg.attrs = attr;
    459 		pv += sz;
    460 		attr += sz;
    461 	}
    462 
    463 	pmap_initialized = 1;
    464 	splx(s);
    465 
    466 	/* Setup a pool for additional pvlist structures */
    467 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
    468 }
    469 
    470 /*
    471  * How much virtual space is available to the kernel?
    472  */
    473 void
    474 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    475 {
    476 
    477 #if 0
    478 	/*
    479 	 * Reserve one segment for kernel virtual memory
    480 	 */
    481 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    482 	*end = *start + SEGMENT_LENGTH;
    483 #else
    484 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    485 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    486 #endif
    487 }
    488 
    489 #ifdef PMAP_GROWKERNEL
    490 /*
    491  * Preallocate kernel page tables to a specified VA.
    492  * This simply loops through the first TTE for each
    493  * page table from the beginning of the kernel pmap,
    494  * reads the entry, and if the result is
    495  * zero (either invalid entry or no page table) it stores
    496  * a zero there, populating page tables in the process.
    497  * This is not the most efficient technique but i don't
    498  * expect it to be called that often.
    499  */
    500 extern struct vm_page *vm_page_alloc1 __P((void));
    501 extern void vm_page_free1 __P((struct vm_page *));
    502 
    503 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    504 
    505 vaddr_t
    506 pmap_growkernel(maxkvaddr)
    507 	vaddr_t maxkvaddr;
    508 {
    509 	int s;
    510 	int seg;
    511 	paddr_t pg;
    512 	struct pmap *pm = pmap_kernel();
    513 
    514 	s = splvm();
    515 
    516 	/* Align with the start of a page table */
    517 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    518 	     kbreak += PTMAP) {
    519 		seg = STIDX(kbreak);
    520 
    521 		if (pte_find(pm, kbreak)) continue;
    522 
    523 		if (uvm.page_init_done) {
    524 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    525 		} else {
    526 			if (!uvm_page_physget(&pg))
    527 				panic("pmap_growkernel: no memory");
    528 		}
    529 		if (!pg) panic("pmap_growkernel: no pages");
    530 		pmap_zero_page((paddr_t)pg);
    531 
    532 		/* XXX This is based on all phymem being addressable */
    533 		pm->pm_ptbl[seg] = (u_int *)pg;
    534 	}
    535 	splx(s);
    536 	return (kbreak);
    537 }
    538 
    539 /*
    540  *	vm_page_alloc1:
    541  *
    542  *	Allocate and return a memory cell with no associated object.
    543  */
    544 struct vm_page *
    545 vm_page_alloc1()
    546 {
    547 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    548 	if (pg) {
    549 		pg->wire_count = 1;	/* no mappings yet */
    550 		pg->flags &= ~PG_BUSY;	/* never busy */
    551 	}
    552 	return pg;
    553 }
    554 
    555 /*
    556  *	vm_page_free1:
    557  *
    558  *	Returns the given page to the free list,
    559  *	disassociating it with any VM object.
    560  *
    561  *	Object and page must be locked prior to entry.
    562  */
    563 void
    564 vm_page_free1(mem)
    565 	struct vm_page *mem;
    566 {
    567 #ifdef DIAGNOSTIC
    568 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    569 		printf("Freeing invalid page %p\n", mem);
    570 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    571 #ifdef DDB
    572 		Debugger();
    573 #endif
    574 		return;
    575 	}
    576 #endif
    577 	mem->flags |= PG_BUSY;
    578 	mem->wire_count = 0;
    579 	uvm_pagefree(mem);
    580 }
    581 #endif
    582 
    583 /*
    584  * Create and return a physical map.
    585  */
    586 struct pmap *
    587 pmap_create(void)
    588 {
    589 	struct pmap *pm;
    590 
    591 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    592 	memset((caddr_t)pm, 0, sizeof *pm);
    593 	pmap_pinit(pm);
    594 	return pm;
    595 }
    596 
    597 /*
    598  * Initialize a preallocated and zeroed pmap structure.
    599  */
    600 void
    601 pmap_pinit(struct pmap *pm)
    602 {
    603 	int i;
    604 
    605 	/*
    606 	 * Allocate some segment registers for this pmap.
    607 	 */
    608 	pm->pm_refs = 1;
    609 	for (i = 0; i < STSZ; i++)
    610 		pm->pm_ptbl[i] = NULL;
    611 }
    612 
    613 /*
    614  * Add a reference to the given pmap.
    615  */
    616 void
    617 pmap_reference(struct pmap *pm)
    618 {
    619 
    620 	pm->pm_refs++;
    621 }
    622 
    623 /*
    624  * Retire the given pmap from service.
    625  * Should only be called if the map contains no valid mappings.
    626  */
    627 void
    628 pmap_destroy(struct pmap *pm)
    629 {
    630 
    631 	if (--pm->pm_refs == 0) {
    632 		pmap_release(pm);
    633 		free((caddr_t)pm, M_VMPMAP);
    634 	}
    635 }
    636 
    637 /*
    638  * Release any resources held by the given physical map.
    639  * Called when a pmap initialized by pmap_pinit is being released.
    640  */
    641 static void
    642 pmap_release(struct pmap *pm)
    643 {
    644 	int i;
    645 
    646 	for (i = 0; i < STSZ; i++)
    647 		if (pm->pm_ptbl[i]) {
    648 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
    649 			    PAGE_SIZE);
    650 			pm->pm_ptbl[i] = NULL;
    651 		}
    652 	if (pm->pm_ctx) ctx_free(pm);
    653 }
    654 
    655 /*
    656  * Copy the range specified by src_addr/len
    657  * from the source map to the range dst_addr/len
    658  * in the destination map.
    659  *
    660  * This routine is only advisory and need not do anything.
    661  */
    662 void
    663 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    664 	  vsize_t len, vaddr_t src_addr)
    665 {
    666 }
    667 
    668 /*
    669  * Require that all active physical maps contain no
    670  * incorrect entries NOW.
    671  */
    672 void
    673 pmap_update(struct pmap *pmap)
    674 {
    675 }
    676 
    677 /*
    678  * Garbage collects the physical map system for
    679  * pages which are no longer used.
    680  * Success need not be guaranteed -- that is, there
    681  * may well be pages which are not referenced, but
    682  * others may be collected.
    683  * Called by the pageout daemon when pages are scarce.
    684  */
    685 void
    686 pmap_collect(struct pmap *pm)
    687 {
    688 }
    689 
    690 /*
    691  * Fill the given physical page with zeroes.
    692  */
    693 void
    694 pmap_zero_page(paddr_t pa)
    695 {
    696 
    697 #ifdef PPC_4XX_NOCACHE
    698 	memset((caddr_t)pa, 0, PAGE_SIZE);
    699 #else
    700 	int i;
    701 
    702 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
    703 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    704 		pa += CACHELINESIZE;
    705 	}
    706 #endif
    707 }
    708 
    709 /*
    710  * Copy the given physical source page to its destination.
    711  */
    712 void
    713 pmap_copy_page(paddr_t src, paddr_t dst)
    714 {
    715 
    716 	memcpy((caddr_t)dst, (caddr_t)src, PAGE_SIZE);
    717 	dcache_flush_page(dst);
    718 }
    719 
    720 /*
    721  * This returns whether this is the first mapping of a page.
    722  */
    723 static inline int
    724 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    725 {
    726 	struct pv_entry *pv, *npv = NULL;
    727 	int s;
    728 
    729 	if (!pmap_initialized)
    730 		return 0;
    731 
    732 	s = splvm();
    733 
    734 	pv = pa_to_pv(pa);
    735 for (npv = pv; npv; npv = npv->pv_next)
    736 if (npv->pv_va == va && npv->pv_pm == pm) {
    737 printf("Duplicate pv: va %lx pm %p\n", va, pm);
    738 #ifdef DDB
    739 Debugger();
    740 #endif
    741 return (1);
    742 }
    743 
    744 	if (!pv->pv_pm) {
    745 		/*
    746 		 * No entries yet, use header as the first entry.
    747 		 */
    748 		pv->pv_va = va;
    749 		pv->pv_pm = pm;
    750 		pv->pv_next = NULL;
    751 	} else {
    752 		/*
    753 		 * There is at least one other VA mapping this page.
    754 		 * Place this entry after the header.
    755 		 */
    756 		npv = pool_get(&pv_pool, PR_WAITOK);
    757 		if (!npv) return (0);
    758 		npv->pv_va = va;
    759 		npv->pv_pm = pm;
    760 		npv->pv_next = pv->pv_next;
    761 		pv->pv_next = npv;
    762 	}
    763 	splx(s);
    764 	return (1);
    765 }
    766 
    767 static void
    768 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    769 {
    770 	struct pv_entry *pv, *npv;
    771 
    772 	/*
    773 	 * Remove from the PV table.
    774 	 */
    775 	pv = pa_to_pv(pa);
    776 	if (!pv) return;
    777 
    778 	/*
    779 	 * If it is the first entry on the list, it is actually
    780 	 * in the header and we must copy the following entry up
    781 	 * to the header.  Otherwise we must search the list for
    782 	 * the entry.  In either case we free the now unused entry.
    783 	 */
    784 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    785 		if ((npv = pv->pv_next)) {
    786 			*pv = *npv;
    787 			pool_put(&pv_pool, npv);
    788 		} else
    789 			pv->pv_pm = NULL;
    790 	} else {
    791 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    792 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    793 				break;
    794 		if (npv) {
    795 			pv->pv_next = npv->pv_next;
    796 			pool_put(&pv_pool, npv);
    797 		}
    798 	}
    799 }
    800 
    801 /*
    802  * Insert physical page at pa into the given pmap at virtual address va.
    803  */
    804 int
    805 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    806 {
    807 	int s;
    808 	u_int tte;
    809 	int managed;
    810 
    811 	/*
    812 	 * Have to remove any existing mapping first.
    813 	 */
    814 	pmap_remove(pm, va, va + PAGE_SIZE);
    815 
    816 	if (flags & PMAP_WIRED) flags |= prot;
    817 
    818 	/* If it has no protections don't bother w/the rest */
    819 	if (!(flags & VM_PROT_ALL))
    820 		return (0);
    821 
    822 	managed = 0;
    823 	if (vm_physseg_find(atop(pa), NULL) != -1)
    824 		managed = 1;
    825 
    826 	/*
    827 	 * Generate TTE.
    828 	 */
    829 	tte = TTE_PA(pa);
    830 	/* XXXX -- need to support multiple page sizes. */
    831 	tte |= TTE_SZ_16K;
    832 #ifdef	DIAGNOSTIC
    833 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    834 		(PME_NOCACHE | PME_WRITETHROUG))
    835 		panic("pmap_enter: uncached & writethrough");
    836 #endif
    837 	if (flags & PME_NOCACHE)
    838 		/* Must be I/O mapping */
    839 		tte |= TTE_I | TTE_G;
    840 #ifdef PPC_4XX_NOCACHE
    841 	tte |= TTE_I;
    842 #else
    843 	else if (flags & PME_WRITETHROUG)
    844 		/* Uncached and writethrough are not compatible */
    845 		tte |= TTE_W;
    846 #endif
    847 	if (pm == pmap_kernel())
    848 		tte |= TTE_ZONE(ZONE_PRIV);
    849 	else
    850 		tte |= TTE_ZONE(ZONE_USER);
    851 
    852 	if (flags & VM_PROT_WRITE)
    853 		tte |= TTE_WR;
    854 
    855 	if (flags & VM_PROT_EXECUTE)
    856 		tte |= TTE_EX;
    857 
    858 	/*
    859 	 * Now record mapping for later back-translation.
    860 	 */
    861 	if (pmap_initialized && managed) {
    862 		char *attr;
    863 
    864 		if (!pmap_enter_pv(pm, va, pa)) {
    865 			/* Could not enter pv on a managed page */
    866 			return 1;
    867 		}
    868 
    869 		/* Now set attributes. */
    870 		attr = pa_to_attr(pa);
    871 #ifdef DIAGNOSTIC
    872 		if (!attr)
    873 			panic("managed but no attr");
    874 #endif
    875 		if (flags & VM_PROT_ALL)
    876 			*attr |= PTE_HI_REF;
    877 		if (flags & VM_PROT_WRITE)
    878 			*attr |= PTE_HI_CHG;
    879 	}
    880 
    881 	s = splvm();
    882 
    883 	/* Insert page into page table. */
    884 	pte_enter(pm, va, tte);
    885 
    886 	/* If this is a real fault, enter it in the tlb */
    887 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    888 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    889 	}
    890 	splx(s);
    891 
    892 	/* Flush the real memory from the instruction cache. */
    893 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    894 		__syncicache((void *)pa, PAGE_SIZE);
    895 
    896 	return 0;
    897 }
    898 
    899 void
    900 pmap_unwire(struct pmap *pm, vaddr_t va)
    901 {
    902 	struct pv_entry *pv, *npv;
    903 	paddr_t pa;
    904 	int s = splvm();
    905 
    906 	if (pm == NULL) {
    907 		return;
    908 	}
    909 
    910 	if (!pmap_extract(pm, va, &pa)) {
    911 		return;
    912 	}
    913 
    914 	va |= PV_WIRED;
    915 
    916 	pv = pa_to_pv(pa);
    917 	if (!pv) return;
    918 
    919 	/*
    920 	 * If it is the first entry on the list, it is actually
    921 	 * in the header and we must copy the following entry up
    922 	 * to the header.  Otherwise we must search the list for
    923 	 * the entry.  In either case we free the now unused entry.
    924 	 */
    925 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    926 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    927 			npv->pv_va &= ~PV_WIRED;
    928 			break;
    929 		}
    930 	}
    931 	splx(s);
    932 }
    933 
    934 void
    935 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    936 {
    937 	int s;
    938 	u_int tte;
    939 	struct pmap *pm = pmap_kernel();
    940 
    941 	/*
    942 	 * Have to remove any existing mapping first.
    943 	 */
    944 
    945 	/*
    946 	 * Generate TTE.
    947 	 *
    948 	 * XXXX
    949 	 *
    950 	 * Since the kernel does not handle execution privileges properly,
    951 	 * we will handle read and execute permissions together.
    952 	 */
    953 	tte = 0;
    954 	if (prot & VM_PROT_ALL) {
    955 
    956 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
    957 		/* XXXX -- need to support multiple page sizes. */
    958 		tte |= TTE_SZ_16K;
    959 #ifdef DIAGNOSTIC
    960 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
    961 			(PME_NOCACHE | PME_WRITETHROUG))
    962 			panic("pmap_kenter_pa: uncached & writethrough");
    963 #endif
    964 		if (prot & PME_NOCACHE)
    965 			/* Must be I/O mapping */
    966 			tte |= TTE_I | TTE_G;
    967 #ifdef PPC_4XX_NOCACHE
    968 		tte |= TTE_I;
    969 #else
    970 		else if (prot & PME_WRITETHROUG)
    971 			/* Uncached and writethrough are not compatible */
    972 			tte |= TTE_W;
    973 #endif
    974 		if (prot & VM_PROT_WRITE)
    975 			tte |= TTE_WR;
    976 	}
    977 
    978 	s = splvm();
    979 
    980 	/* Insert page into page table. */
    981 	pte_enter(pm, va, tte);
    982 	splx(s);
    983 }
    984 
    985 void
    986 pmap_kremove(vaddr_t va, vsize_t len)
    987 {
    988 
    989 	while (len > 0) {
    990 		pte_enter(pmap_kernel(), va, 0);
    991 		va += PAGE_SIZE;
    992 		len -= PAGE_SIZE;
    993 	}
    994 }
    995 
    996 /*
    997  * Remove the given range of mapping entries.
    998  */
    999 void
   1000 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
   1001 {
   1002 	int s;
   1003 	paddr_t pa;
   1004 	volatile u_int *ptp;
   1005 
   1006 	s = splvm();
   1007 	while (va < endva) {
   1008 
   1009 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1010 			pa = TTE_PA(pa);
   1011 			pmap_remove_pv(pm, va, pa);
   1012 			*ptp = 0;
   1013 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1014 			pm->pm_stats.resident_count--;
   1015 		}
   1016 		va += PAGE_SIZE;
   1017 	}
   1018 
   1019 	splx(s);
   1020 }
   1021 
   1022 /*
   1023  * Get the physical page address for the given pmap/virtual address.
   1024  */
   1025 boolean_t
   1026 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1027 {
   1028 	int seg = STIDX(va);
   1029 	int ptn = PTIDX(va);
   1030 	u_int pa = 0;
   1031 	int s = splvm();
   1032 
   1033 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1034 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1035 	}
   1036 	splx(s);
   1037 	return (pa != 0);
   1038 }
   1039 
   1040 /*
   1041  * Lower the protection on the specified range of this pmap.
   1042  *
   1043  * There are only two cases: either the protection is going to 0,
   1044  * or it is going to read-only.
   1045  */
   1046 void
   1047 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1048 {
   1049 	volatile u_int *ptp;
   1050 	int s, bic;
   1051 
   1052 	if ((prot & VM_PROT_READ) == 0) {
   1053 		pmap_remove(pm, sva, eva);
   1054 		return;
   1055 	}
   1056 	bic = 0;
   1057 	if ((prot & VM_PROT_WRITE) == 0) {
   1058 		bic |= TTE_WR;
   1059 	}
   1060 	if ((prot & VM_PROT_EXECUTE) == 0) {
   1061 		bic |= TTE_EX;
   1062 	}
   1063 	if (bic == 0) {
   1064 		return;
   1065 	}
   1066 	s = splvm();
   1067 	while (sva < eva) {
   1068 		if ((ptp = pte_find(pm, sva)) != NULL) {
   1069 			*ptp &= ~bic;
   1070 			ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1071 		}
   1072 		sva += PAGE_SIZE;
   1073 	}
   1074 	splx(s);
   1075 }
   1076 
   1077 boolean_t
   1078 check_attr(struct vm_page *pg, u_int mask, int clear)
   1079 {
   1080 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1081 	int s;
   1082 	char *attr;
   1083 	int rv;
   1084 
   1085 	/*
   1086 	 * First modify bits in cache.
   1087 	 */
   1088 	s = splvm();
   1089 	attr = pa_to_attr(pa);
   1090 	if (attr == NULL)
   1091 		return FALSE;
   1092 
   1093 	rv = ((*attr & mask) != 0);
   1094 	if (clear) {
   1095 		*attr &= ~mask;
   1096 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
   1097 	}
   1098 	splx(s);
   1099 	return rv;
   1100 }
   1101 
   1102 
   1103 /*
   1104  * Lower the protection on the specified physical page.
   1105  *
   1106  * There are only two cases: either the protection is going to 0,
   1107  * or it is going to read-only.
   1108  */
   1109 void
   1110 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1111 {
   1112 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1113 	vaddr_t va;
   1114 	struct pv_entry *pvh, *pv, *npv;
   1115 	struct pmap *pm;
   1116 
   1117 	pvh = pa_to_pv(pa);
   1118 	if (pvh == NULL)
   1119 		return;
   1120 
   1121 	/* Handle extra pvs which may be deleted in the operation */
   1122 	for (pv = pvh->pv_next; pv; pv = npv) {
   1123 		npv = pv->pv_next;
   1124 
   1125 		pm = pv->pv_pm;
   1126 		va = pv->pv_va;
   1127 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
   1128 	}
   1129 	/* Now check the head pv */
   1130 	if (pvh->pv_pm) {
   1131 		pv = pvh;
   1132 		pm = pv->pv_pm;
   1133 		va = pv->pv_va;
   1134 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
   1135 	}
   1136 }
   1137 
   1138 /*
   1139  * Activate the address space for the specified process.  If the process
   1140  * is the current process, load the new MMU context.
   1141  */
   1142 void
   1143 pmap_activate(struct lwp *l)
   1144 {
   1145 #if 0
   1146 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1147 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1148 
   1149 	/*
   1150 	 * XXX Normally performed in cpu_fork().
   1151 	 */
   1152 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1153 	pcb->pcb_pm = pmap;
   1154 #endif
   1155 }
   1156 
   1157 /*
   1158  * Deactivate the specified process's address space.
   1159  */
   1160 void
   1161 pmap_deactivate(struct lwp *l)
   1162 {
   1163 }
   1164 
   1165 /*
   1166  * Synchronize caches corresponding to [addr, addr+len) in p.
   1167  */
   1168 void
   1169 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1170 {
   1171 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1172 	int msr, ctx, opid, step;
   1173 
   1174 	step = CACHELINESIZE;
   1175 
   1176 	/*
   1177 	 * Need to turn off IMMU and switch to user context.
   1178 	 * (icbi uses DMMU).
   1179 	 */
   1180 	if (!(ctx = pm->pm_ctx)) {
   1181 		/* No context -- assign it one */
   1182 		ctx_alloc(pm);
   1183 		ctx = pm->pm_ctx;
   1184 	}
   1185 	__asm __volatile("mfmsr %0;"
   1186 		"li %1, %7;"
   1187 		"andc %1,%0,%1;"
   1188 		"mtmsr %1;"
   1189 		"sync;isync;"
   1190 		"mfpid %1;"
   1191 		"mtpid %2;"
   1192 		"sync; isync;"
   1193 		"1:"
   1194 		"dcbf 0,%3;"
   1195 		"icbi 0,%3;"
   1196 		"add %3,%3,%5;"
   1197 		"addc. %4,%4,%6;"
   1198 		"bge 1b;"
   1199 		"mtpid %1;"
   1200 		"mtmsr %0;"
   1201 		"sync; isync"
   1202 		: "=&r" (msr), "=&r" (opid)
   1203 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step),
   1204 		  "K" (PSL_IR | PSL_DR));
   1205 }
   1206 
   1207 
   1208 /* This has to be done in real mode !!! */
   1209 void
   1210 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1211 {
   1212 	u_long i, found;
   1213 	u_long msr;
   1214 
   1215 	/* If there's no context then it can't be mapped. */
   1216 	if (!pid)
   1217 		return;
   1218 
   1219 	asm("mfpid %1;"			/* Save PID */
   1220 		"mfmsr %2;"		/* Save MSR */
   1221 		"li %0,0;"		/* Now clear MSR */
   1222 		"mtmsr %0;"
   1223 		"mtpid %4;"		/* Set PID */
   1224 		"sync;"
   1225 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1226 		"sync;"
   1227 		"mtpid %1;"		/* Restore PID */
   1228 		"mtmsr %2;"		/* Restore MSR */
   1229 		"sync;isync;"
   1230 		"li %1,1;"
   1231 		"beq 1f;"
   1232 		"li %1,0;"
   1233 		"1:"
   1234 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1235 		: "r" (va), "r" (pid));
   1236 	if (found && !TLB_LOCKED(i)) {
   1237 
   1238 		/* Now flush translation */
   1239 		asm volatile(
   1240 			"tlbwe %0,%1,0;"
   1241 			"sync;isync;"
   1242 			: : "r" (0), "r" (i));
   1243 
   1244 		tlb_info[i].ti_ctx = 0;
   1245 		tlb_info[i].ti_flags = 0;
   1246 		tlbnext = i;
   1247 		/* Successful flushes */
   1248 		tlbflush_ev.ev_count++;
   1249 	}
   1250 }
   1251 
   1252 void
   1253 ppc4xx_tlb_flush_all(void)
   1254 {
   1255 	u_long i;
   1256 
   1257 	for (i = 0; i < NTLB; i++)
   1258 		if (!TLB_LOCKED(i)) {
   1259 			asm volatile(
   1260 				"tlbwe %0,%1,0;"
   1261 				"sync;isync;"
   1262 				: : "r" (0), "r" (i));
   1263 			tlb_info[i].ti_ctx = 0;
   1264 			tlb_info[i].ti_flags = 0;
   1265 		}
   1266 
   1267 	asm volatile("sync;isync");
   1268 }
   1269 
   1270 /* Find a TLB entry to evict. */
   1271 static int
   1272 ppc4xx_tlb_find_victim(void)
   1273 {
   1274 	int flags;
   1275 
   1276 	for (;;) {
   1277 		if (++tlbnext >= NTLB)
   1278 			tlbnext = TLB_NRESERVED;
   1279 		flags = tlb_info[tlbnext].ti_flags;
   1280 		if (!(flags & TLBF_USED) ||
   1281 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1282 			u_long va, stack = (u_long)&va;
   1283 
   1284 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1285 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1286 			     (flags & TLBF_USED)) {
   1287 				/* Kernel stack page */
   1288 				flags |= TLBF_USED;
   1289 				tlb_info[tlbnext].ti_flags = flags;
   1290 			} else {
   1291 				/* Found it! */
   1292 				return (tlbnext);
   1293 			}
   1294 		} else {
   1295 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1296 		}
   1297 	}
   1298 }
   1299 
   1300 void
   1301 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1302 {
   1303 	u_long th, tl, idx;
   1304 	tlbpid_t pid;
   1305 	u_short msr;
   1306 	paddr_t pa;
   1307 	int s, sz;
   1308 
   1309 	tlbenter_ev.ev_count++;
   1310 
   1311 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
   1312 	pa = (pte & TTE_RPN_MASK(sz));
   1313 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
   1314 	tl = (pte & ~TLB_RPN_MASK) | pa;
   1315 	tl |= ppc4xx_tlbflags(va, pa);
   1316 
   1317 	s = splhigh();
   1318 	idx = ppc4xx_tlb_find_victim();
   1319 
   1320 #ifdef DIAGNOSTIC
   1321 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1322 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
   1323 	}
   1324 #endif
   1325 
   1326 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1327 	tlb_info[idx].ti_ctx = ctx;
   1328 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1329 
   1330 	asm volatile(
   1331 		"mfmsr %0;"			/* Save MSR */
   1332 		"li %1,0;"
   1333 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1334 		"mtmsr %1;"			/* Clear MSR */
   1335 		"mfpid %1;"			/* Save old PID */
   1336 		"mtpid %2;"			/* Load translation ctx */
   1337 		"sync; isync;"
   1338 #ifdef DEBUG
   1339 		"andi. %3,%3,63;"
   1340 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1341 #endif
   1342 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1343 		"sync; isync;"
   1344 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1345 		"sync; isync;"
   1346 	: "=&r" (msr), "=&r" (pid)
   1347 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1348 	splx(s);
   1349 }
   1350 
   1351 void
   1352 ppc4xx_tlb_unpin(int i)
   1353 {
   1354 
   1355 	if (i == -1)
   1356 		for (i = 0; i < TLB_NRESERVED; i++)
   1357 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1358 	else
   1359 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1360 }
   1361 
   1362 void
   1363 ppc4xx_tlb_init(void)
   1364 {
   1365 	int i;
   1366 
   1367 	/* Mark reserved TLB entries */
   1368 	for (i = 0; i < TLB_NRESERVED; i++) {
   1369 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1370 		tlb_info[i].ti_ctx = KERNEL_PID;
   1371 	}
   1372 
   1373 	/* Setup security zones */
   1374 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1375 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1376 	 * Z3 - full access regardless of TLB entry permissions
   1377 	 */
   1378 
   1379 	asm volatile(
   1380 		"mtspr %0,%1;"
   1381 		"sync;"
   1382 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1383 }
   1384 
   1385 
   1386 /*
   1387  * We should pass the ctx in from trap code.
   1388  */
   1389 int
   1390 pmap_tlbmiss(vaddr_t va, int ctx)
   1391 {
   1392 	volatile u_int *pte;
   1393 	u_long tte;
   1394 
   1395 	tlbmiss_ev.ev_count++;
   1396 
   1397 	/*
   1398 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1399 	 */
   1400 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1401 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1402 		if (pte == NULL) {
   1403 			/* Map unmanaged addresses directly for kernel access */
   1404 			return 1;
   1405 		}
   1406 		tte = *pte;
   1407 		if (tte == 0) {
   1408 			return 1;
   1409 		}
   1410 	} else {
   1411 		/* Create a 16MB writable mapping. */
   1412 #ifdef PPC_4XX_NOCACHE
   1413 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1414 #else
   1415 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1416 #endif
   1417 	}
   1418 	tlbhit_ev.ev_count++;
   1419 	ppc4xx_tlb_enter(ctx, va, tte);
   1420 
   1421 	return 0;
   1422 }
   1423 
   1424 /*
   1425  * Flush all the entries matching a context from the TLB.
   1426  */
   1427 static int
   1428 ctx_flush(int cnum)
   1429 {
   1430 	int i;
   1431 
   1432 	/* We gotta steal this context */
   1433 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1434 		if (tlb_info[i].ti_ctx == cnum) {
   1435 			/* Can't steal ctx if it has a locked entry. */
   1436 			if (TLB_LOCKED(i)) {
   1437 #ifdef DIAGNOSTIC
   1438 				printf("ctx_flush: can't invalidate "
   1439 					"locked mapping %d "
   1440 					"for context %d\n", i, cnum);
   1441 #ifdef DDB
   1442 				Debugger();
   1443 #endif
   1444 #endif
   1445 				return (1);
   1446 			}
   1447 #ifdef DIAGNOSTIC
   1448 			if (i < TLB_NRESERVED)
   1449 				panic("TLB entry %d not locked", i);
   1450 #endif
   1451 			/* Invalidate particular TLB entry regardless of locked status */
   1452 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1453 			tlb_info[i].ti_flags = 0;
   1454 		}
   1455 	}
   1456 	return (0);
   1457 }
   1458 
   1459 /*
   1460  * Allocate a context.  If necessary, steal one from someone else.
   1461  *
   1462  * The new context is flushed from the TLB before returning.
   1463  */
   1464 int
   1465 ctx_alloc(struct pmap *pm)
   1466 {
   1467 	int s, cnum;
   1468 	static int next = MINCTX;
   1469 
   1470 	if (pm == pmap_kernel()) {
   1471 #ifdef DIAGNOSTIC
   1472 		printf("ctx_alloc: kernel pmap!\n");
   1473 #endif
   1474 		return (0);
   1475 	}
   1476 	s = splvm();
   1477 
   1478 	/* Find a likely context. */
   1479 	cnum = next;
   1480 	do {
   1481 		if ((++cnum) > NUMCTX)
   1482 			cnum = MINCTX;
   1483 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1484 
   1485 	/* Now clean it out */
   1486 oops:
   1487 	if (cnum < MINCTX)
   1488 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1489 	if (ctx_flush(cnum)) {
   1490 		/* oops -- something's wired. */
   1491 		if ((++cnum) > NUMCTX)
   1492 			cnum = MINCTX;
   1493 		goto oops;
   1494 	}
   1495 
   1496 	if (ctxbusy[cnum]) {
   1497 #ifdef DEBUG
   1498 		/* We should identify this pmap and clear it */
   1499 		printf("Warning: stealing context %d\n", cnum);
   1500 #endif
   1501 		ctxbusy[cnum]->pm_ctx = 0;
   1502 	}
   1503 	ctxbusy[cnum] = pm;
   1504 	next = cnum;
   1505 	splx(s);
   1506 	pm->pm_ctx = cnum;
   1507 
   1508 	return cnum;
   1509 }
   1510 
   1511 /*
   1512  * Give away a context.
   1513  */
   1514 void
   1515 ctx_free(struct pmap *pm)
   1516 {
   1517 	int oldctx;
   1518 
   1519 	oldctx = pm->pm_ctx;
   1520 
   1521 	if (oldctx == 0)
   1522 		panic("ctx_free: freeing kernel context");
   1523 #ifdef DIAGNOSTIC
   1524 	if (ctxbusy[oldctx] == 0)
   1525 		printf("ctx_free: freeing free context %d\n", oldctx);
   1526 	if (ctxbusy[oldctx] != pm) {
   1527 		printf("ctx_free: freeing someone esle's context\n "
   1528 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1529 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1530 #ifdef DDB
   1531 		Debugger();
   1532 #endif
   1533 	}
   1534 #endif
   1535 	/* We should verify it has not been stolen and reallocated... */
   1536 	ctxbusy[oldctx] = NULL;
   1537 	ctx_flush(oldctx);
   1538 }
   1539 
   1540 
   1541 #ifdef DEBUG
   1542 /*
   1543  * Test ref/modify handling.
   1544  */
   1545 void pmap_testout __P((void));
   1546 void
   1547 pmap_testout()
   1548 {
   1549 	vaddr_t va;
   1550 	volatile int *loc;
   1551 	int val = 0;
   1552 	paddr_t pa;
   1553 	struct vm_page *pg;
   1554 	int ref, mod;
   1555 
   1556 	/* Allocate a page */
   1557 	va = (vaddr_t)uvm_km_zalloc(kernel_map, PAGE_SIZE);
   1558 	loc = (int*)va;
   1559 
   1560 	pmap_extract(pmap_kernel(), va, &pa);
   1561 	pg = PHYS_TO_VM_PAGE(pa);
   1562 	pmap_unwire(pmap_kernel(), va);
   1563 
   1564 	pmap_remove(pmap_kernel(), va, va+1);
   1565 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1566 	pmap_update(pmap_kernel());
   1567 
   1568 	/* Now clear reference and modify */
   1569 	ref = pmap_clear_reference(pg);
   1570 	mod = pmap_clear_modify(pg);
   1571 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1572 	       (void *)(u_long)va, (long)pa,
   1573 	       ref, mod);
   1574 
   1575 	/* Check it's properly cleared */
   1576 	ref = pmap_is_referenced(pg);
   1577 	mod = pmap_is_modified(pg);
   1578 	printf("Checking cleared page: ref %d, mod %d\n",
   1579 	       ref, mod);
   1580 
   1581 	/* Reference page */
   1582 	val = *loc;
   1583 
   1584 	ref = pmap_is_referenced(pg);
   1585 	mod = pmap_is_modified(pg);
   1586 	printf("Referenced page: ref %d, mod %d val %x\n",
   1587 	       ref, mod, val);
   1588 
   1589 	/* Now clear reference and modify */
   1590 	ref = pmap_clear_reference(pg);
   1591 	mod = pmap_clear_modify(pg);
   1592 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1593 	       (void *)(u_long)va, (long)pa,
   1594 	       ref, mod);
   1595 
   1596 	/* Modify page */
   1597 	*loc = 1;
   1598 
   1599 	ref = pmap_is_referenced(pg);
   1600 	mod = pmap_is_modified(pg);
   1601 	printf("Modified page: ref %d, mod %d\n",
   1602 	       ref, mod);
   1603 
   1604 	/* Now clear reference and modify */
   1605 	ref = pmap_clear_reference(pg);
   1606 	mod = pmap_clear_modify(pg);
   1607 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1608 	       (void *)(u_long)va, (long)pa,
   1609 	       ref, mod);
   1610 
   1611 	/* Check it's properly cleared */
   1612 	ref = pmap_is_referenced(pg);
   1613 	mod = pmap_is_modified(pg);
   1614 	printf("Checking cleared page: ref %d, mod %d\n",
   1615 	       ref, mod);
   1616 
   1617 	/* Modify page */
   1618 	*loc = 1;
   1619 
   1620 	ref = pmap_is_referenced(pg);
   1621 	mod = pmap_is_modified(pg);
   1622 	printf("Modified page: ref %d, mod %d\n",
   1623 	       ref, mod);
   1624 
   1625 	/* Check pmap_protect() */
   1626 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1627 	pmap_update(pmap_kernel());
   1628 	ref = pmap_is_referenced(pg);
   1629 	mod = pmap_is_modified(pg);
   1630 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1631 	       ref, mod);
   1632 
   1633 	/* Now clear reference and modify */
   1634 	ref = pmap_clear_reference(pg);
   1635 	mod = pmap_clear_modify(pg);
   1636 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1637 	       (void *)(u_long)va, (long)pa,
   1638 	       ref, mod);
   1639 
   1640 	/* Reference page */
   1641 	val = *loc;
   1642 
   1643 	ref = pmap_is_referenced(pg);
   1644 	mod = pmap_is_modified(pg);
   1645 	printf("Referenced page: ref %d, mod %d val %x\n",
   1646 	       ref, mod, val);
   1647 
   1648 	/* Now clear reference and modify */
   1649 	ref = pmap_clear_reference(pg);
   1650 	mod = pmap_clear_modify(pg);
   1651 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1652 	       (void *)(u_long)va, (long)pa,
   1653 	       ref, mod);
   1654 
   1655 	/* Modify page */
   1656 #if 0
   1657 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1658 	pmap_update(pmap_kernel());
   1659 #endif
   1660 	*loc = 1;
   1661 
   1662 	ref = pmap_is_referenced(pg);
   1663 	mod = pmap_is_modified(pg);
   1664 	printf("Modified page: ref %d, mod %d\n",
   1665 	       ref, mod);
   1666 
   1667 	/* Check pmap_protect() */
   1668 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1669 	pmap_update(pmap_kernel());
   1670 	ref = pmap_is_referenced(pg);
   1671 	mod = pmap_is_modified(pg);
   1672 	printf("pmap_protect(): ref %d, mod %d\n",
   1673 	       ref, mod);
   1674 
   1675 	/* Now clear reference and modify */
   1676 	ref = pmap_clear_reference(pg);
   1677 	mod = pmap_clear_modify(pg);
   1678 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1679 	       (void *)(u_long)va, (long)pa,
   1680 	       ref, mod);
   1681 
   1682 	/* Reference page */
   1683 	val = *loc;
   1684 
   1685 	ref = pmap_is_referenced(pg);
   1686 	mod = pmap_is_modified(pg);
   1687 	printf("Referenced page: ref %d, mod %d val %x\n",
   1688 	       ref, mod, val);
   1689 
   1690 	/* Now clear reference and modify */
   1691 	ref = pmap_clear_reference(pg);
   1692 	mod = pmap_clear_modify(pg);
   1693 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1694 	       (void *)(u_long)va, (long)pa,
   1695 	       ref, mod);
   1696 
   1697 	/* Modify page */
   1698 #if 0
   1699 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1700 	pmap_update(pmap_kernel());
   1701 #endif
   1702 	*loc = 1;
   1703 
   1704 	ref = pmap_is_referenced(pg);
   1705 	mod = pmap_is_modified(pg);
   1706 	printf("Modified page: ref %d, mod %d\n",
   1707 	       ref, mod);
   1708 
   1709 	/* Check pmap_pag_protect() */
   1710 	pmap_page_protect(pg, VM_PROT_READ);
   1711 	ref = pmap_is_referenced(pg);
   1712 	mod = pmap_is_modified(pg);
   1713 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1714 	       ref, mod);
   1715 
   1716 	/* Now clear reference and modify */
   1717 	ref = pmap_clear_reference(pg);
   1718 	mod = pmap_clear_modify(pg);
   1719 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1720 	       (void *)(u_long)va, (long)pa,
   1721 	       ref, mod);
   1722 
   1723 	/* Reference page */
   1724 	val = *loc;
   1725 
   1726 	ref = pmap_is_referenced(pg);
   1727 	mod = pmap_is_modified(pg);
   1728 	printf("Referenced page: ref %d, mod %d val %x\n",
   1729 	       ref, mod, val);
   1730 
   1731 	/* Now clear reference and modify */
   1732 	ref = pmap_clear_reference(pg);
   1733 	mod = pmap_clear_modify(pg);
   1734 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1735 	       (void *)(u_long)va, (long)pa,
   1736 	       ref, mod);
   1737 
   1738 	/* Modify page */
   1739 #if 0
   1740 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1741 	pmap_update(pmap_kernel());
   1742 #endif
   1743 	*loc = 1;
   1744 
   1745 	ref = pmap_is_referenced(pg);
   1746 	mod = pmap_is_modified(pg);
   1747 	printf("Modified page: ref %d, mod %d\n",
   1748 	       ref, mod);
   1749 
   1750 	/* Check pmap_pag_protect() */
   1751 	pmap_page_protect(pg, VM_PROT_NONE);
   1752 	ref = pmap_is_referenced(pg);
   1753 	mod = pmap_is_modified(pg);
   1754 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1755 	       ref, mod);
   1756 
   1757 	/* Now clear reference and modify */
   1758 	ref = pmap_clear_reference(pg);
   1759 	mod = pmap_clear_modify(pg);
   1760 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1761 	       (void *)(u_long)va, (long)pa,
   1762 	       ref, mod);
   1763 
   1764 
   1765 	/* Reference page */
   1766 	val = *loc;
   1767 
   1768 	ref = pmap_is_referenced(pg);
   1769 	mod = pmap_is_modified(pg);
   1770 	printf("Referenced page: ref %d, mod %d val %x\n",
   1771 	       ref, mod, val);
   1772 
   1773 	/* Now clear reference and modify */
   1774 	ref = pmap_clear_reference(pg);
   1775 	mod = pmap_clear_modify(pg);
   1776 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1777 	       (void *)(u_long)va, (long)pa,
   1778 	       ref, mod);
   1779 
   1780 	/* Modify page */
   1781 #if 0
   1782 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1783 	pmap_update(pmap_kernel());
   1784 #endif
   1785 	*loc = 1;
   1786 
   1787 	ref = pmap_is_referenced(pg);
   1788 	mod = pmap_is_modified(pg);
   1789 	printf("Modified page: ref %d, mod %d\n",
   1790 	       ref, mod);
   1791 
   1792 	/* Unmap page */
   1793 	pmap_remove(pmap_kernel(), va, va+1);
   1794 	pmap_update(pmap_kernel());
   1795 	ref = pmap_is_referenced(pg);
   1796 	mod = pmap_is_modified(pg);
   1797 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1798 
   1799 	/* Now clear reference and modify */
   1800 	ref = pmap_clear_reference(pg);
   1801 	mod = pmap_clear_modify(pg);
   1802 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1803 	       (void *)(u_long)va, (long)pa, ref, mod);
   1804 
   1805 	/* Check it's properly cleared */
   1806 	ref = pmap_is_referenced(pg);
   1807 	mod = pmap_is_modified(pg);
   1808 	printf("Checking cleared page: ref %d, mod %d\n",
   1809 	       ref, mod);
   1810 
   1811 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1812 		VM_PROT_ALL|PMAP_WIRED);
   1813 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE);
   1814 }
   1815 #endif
   1816