Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.5.4.2
      1 /*	$NetBSD: pmap.c,v 1.5.4.2 2001/11/05 19:46:16 briggs Exp $	*/
      2 
      3 /*
      4  * Copyright 2001 Wasabi Systems, Inc.
      5  * All rights reserved.
      6  *
      7  * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, Inc.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed for the NetBSD Project by
     20  *      Wasabi Systems, Inc.
     21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     22  *    or promote products derived from this software without specific prior
     23  *    written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     35  * POSSIBILITY OF SUCH DAMAGE.
     36  */
     37 
     38 /*
     39  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
     40  * Copyright (C) 1995, 1996 TooLs GmbH.
     41  * All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed by TooLs GmbH.
     54  * 4. The name of TooLs GmbH may not be used to endorse or promote products
     55  *    derived from this software without specific prior written permission.
     56  *
     57  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
     58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     60  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     61  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
     62  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
     63  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     64  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
     65  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
     66  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     67  */
     68 
     69 #undef NOCACHE
     70 
     71 #include <sys/param.h>
     72 #include <sys/malloc.h>
     73 #include <sys/proc.h>
     74 #include <sys/user.h>
     75 #include <sys/queue.h>
     76 #include <sys/systm.h>
     77 #include <sys/pool.h>
     78 #include <sys/device.h>
     79 #include <sys/lwp.h>
     80 
     81 #include <uvm/uvm.h>
     82 
     83 #include <machine/pcb.h>
     84 #include <machine/powerpc.h>
     85 
     86 #include <powerpc/spr.h>
     87 #include <powerpc/ibm4xx/tlb.h>
     88 
     89 
     90 #define	CACHE_LINE	32
     91 
     92 /*
     93  * kernmap is an array of PTEs large enough to map in
     94  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     95  */
     96 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
     97 caddr_t kernmap;
     98 
     99 #define MINCTX		2
    100 #define NUMCTX		256
    101 volatile struct pmap *ctxbusy[NUMCTX];
    102 
    103 #define TLBF_USED	0x1
    104 #define	TLBF_REF	0x2
    105 #define	TLBF_LOCKED	0x4
    106 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    107 typedef struct tlb_info_s {
    108 	char	ti_flags;
    109 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    110 	u_int	ti_va;
    111 } tlb_info_t;
    112 
    113 volatile tlb_info_t tlb_info[NTLB];
    114 /* We'll use a modified FIFO replacement policy cause it's cheap */
    115 volatile int tlbnext = TLB_NRESERVED;
    116 
    117 u_long dtlb_miss_count = 0;
    118 u_long itlb_miss_count = 0;
    119 u_long ktlb_miss_count = 0;
    120 u_long utlb_miss_count = 0;
    121 
    122 /* Event counters -- XXX type `INTR' so we can see them with vmstat -i */
    123 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    124 	NULL, "cpu", "tlbmiss");
    125 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    126 	NULL, "cpu", "tlbhit");
    127 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    128 	NULL, "cpu", "tlbflush");
    129 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    130 	NULL, "cpu", "tlbenter");
    131 
    132 struct pmap kernel_pmap_;
    133 
    134 int physmem;
    135 static int npgs;
    136 static u_int nextavail;
    137 #ifndef MSGBUFADDR
    138 extern paddr_t msgbuf_paddr;
    139 #endif
    140 
    141 static struct mem_region *mem, *avail;
    142 
    143 /*
    144  * This is a cache of referenced/modified bits.
    145  * Bits herein are shifted by ATTRSHFT.
    146  */
    147 static char *pmap_attrib;
    148 
    149 #define PV_WIRED	0x1
    150 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    151 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    152 
    153 struct pv_entry {
    154 	struct pv_entry *pv_next;	/* Linked list of mappings */
    155 	vaddr_t pv_va;			/* virtual address of mapping */
    156 	struct pmap *pv_pm;
    157 };
    158 
    159 struct pv_entry *pv_table;
    160 static struct pool pv_pool;
    161 
    162 static int pmap_initialized;
    163 
    164 static int ctx_flush(int);
    165 
    166 static inline void dcache_flush_page(vaddr_t);
    167 static inline void icache_flush_page(vaddr_t);
    168 static inline void dcache_flush(vaddr_t, vsize_t);
    169 static inline void icache_flush(vaddr_t, vsize_t);
    170 
    171 inline struct pv_entry *pa_to_pv(paddr_t);
    172 static inline char *pa_to_attr(paddr_t);
    173 
    174 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    175 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    176 
    177 static void pmap_pinit(pmap_t);
    178 static void pmap_release(pmap_t);
    179 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    180 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    181 
    182 /*
    183  * These small routines may have to be replaced,
    184  * if/when we support processors other that the 604.
    185  */
    186 
    187 static inline void
    188 dcache_flush_page(vaddr_t va)
    189 {
    190 	int i;
    191 
    192 	for (i = 0; i < NBPG; i += CACHE_LINE)
    193 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    194 	asm volatile("sync;isync" : : );
    195 }
    196 
    197 static inline void
    198 icache_flush_page(vaddr_t va)
    199 {
    200 	int i;
    201 
    202 	for (i = 0; i < NBPG; i += CACHE_LINE)
    203 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    204 	asm volatile("sync;isync" : : );
    205 }
    206 
    207 static inline void
    208 dcache_flush(vaddr_t va, vsize_t len)
    209 {
    210 	int i;
    211 
    212 	if (len == 0)
    213 		return;
    214 
    215 	/* Make sure we flush all cache lines */
    216 	len += va & (CACHE_LINE-1);
    217 	for (i = 0; i < len; i += CACHE_LINE)
    218 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    219 	asm volatile("sync;isync" : : );
    220 }
    221 
    222 static inline void
    223 icache_flush(vaddr_t va, vsize_t len)
    224 {
    225 	int i;
    226 
    227 	if (len == 0)
    228 		return;
    229 
    230 	/* Make sure we flush all cache lines */
    231 	len += va & (CACHE_LINE-1);
    232 	for (i = 0; i < len; i += CACHE_LINE)
    233 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    234 	asm volatile("sync;isync" : : );
    235 }
    236 
    237 inline struct pv_entry *
    238 pa_to_pv(paddr_t pa)
    239 {
    240 	int bank, pg;
    241 
    242 	bank = vm_physseg_find(atop(pa), &pg);
    243 	if (bank == -1)
    244 		return NULL;
    245 	return &vm_physmem[bank].pmseg.pvent[pg];
    246 }
    247 
    248 static inline char *
    249 pa_to_attr(paddr_t pa)
    250 {
    251 	int bank, pg;
    252 
    253 	bank = vm_physseg_find(atop(pa), &pg);
    254 	if (bank == -1)
    255 		return NULL;
    256 	return &vm_physmem[bank].pmseg.attrs[pg];
    257 }
    258 
    259 /*
    260  * Insert PTE into page table.
    261  */
    262 int
    263 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    264 {
    265 	int seg = STIDX(va);
    266 	int ptn = PTIDX(va);
    267 	paddr_t pa;
    268 
    269 	if (!pm->pm_ptbl[seg]) {
    270 		/* Don't allocate a page to clear a non-existent mapping. */
    271 		if (!pte) return (1);
    272 		/* Allocate a page XXXX this will sleep! */
    273 		pa = 0;
    274 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
    275 	}
    276 	pm->pm_ptbl[seg][ptn] = pte;
    277 
    278 	/* Flush entry. */
    279 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    280 	return (1);
    281 }
    282 
    283 /*
    284  * Get a pointer to a PTE in a page table.
    285  */
    286 volatile u_int *
    287 pte_find(struct pmap *pm, vaddr_t va)
    288 {
    289 	int seg = STIDX(va);
    290 	int ptn = PTIDX(va);
    291 
    292 	if (pm->pm_ptbl[seg])
    293 		return (&pm->pm_ptbl[seg][ptn]);
    294 
    295 	return (NULL);
    296 }
    297 
    298 /*
    299  * This is called during initppc, before the system is really initialized.
    300  */
    301 void
    302 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    303 {
    304 	struct mem_region *mp, *mp1;
    305 	int cnt, i;
    306 	u_int s, e, sz;
    307 
    308 	/*
    309 	 * Allocate the kernel page table at the end of
    310 	 * kernel space so it's in the locked TTE.
    311 	 */
    312 	kernmap = (caddr_t)kernelend;
    313 //	kernelend += KERNMAP_SIZE*sizeof(struct pte);
    314 
    315 	/*
    316 	 * Initialize kernel page table.
    317 	 */
    318 //	memset(kernmap, 0, KERNMAP_SIZE*sizeof(struct pte));
    319 	for (i = 0; i < STSZ; i++) {
    320 		pmap_kernel()->pm_ptbl[i] = 0; // (u_int *)(kernmap + i*NBPG);
    321 	}
    322 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    323 
    324 	/*
    325 	 * Announce page-size to the VM-system
    326 	 */
    327 	uvmexp.pagesize = NBPG;
    328 	uvm_setpagesize();
    329 
    330 	/*
    331 	 * Get memory.
    332 	 */
    333 	mem_regions(&mem, &avail);
    334 	for (mp = mem; mp->size; mp++) {
    335 		physmem += btoc(mp->size);
    336 		printf("+%lx,",mp->size);
    337 	}
    338 	printf("\n");
    339 	ppc4xx_tlb_init();
    340 	/*
    341 	 * Count the number of available entries.
    342 	 */
    343 	for (cnt = 0, mp = avail; mp->size; mp++)
    344 		cnt++;
    345 
    346 	/*
    347 	 * Page align all regions.
    348 	 * Non-page aligned memory isn't very interesting to us.
    349 	 * Also, sort the entries for ascending addresses.
    350 	 */
    351 	kernelstart &= ~PGOFSET;
    352 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    353 	for (mp = avail; mp->size; mp++) {
    354 		s = mp->start;
    355 		e = mp->start + mp->size;
    356 		printf("%08x-%08x -> ",s,e);
    357 		/*
    358 		 * Check whether this region holds all of the kernel.
    359 		 */
    360 		if (s < kernelstart && e > kernelend) {
    361 			avail[cnt].start = kernelend;
    362 			avail[cnt++].size = e - kernelend;
    363 			e = kernelstart;
    364 		}
    365 		/*
    366 		 * Look whether this regions starts within the kernel.
    367 		 */
    368 		if (s >= kernelstart && s < kernelend) {
    369 			if (e <= kernelend)
    370 				goto empty;
    371 			s = kernelend;
    372 		}
    373 		/*
    374 		 * Now look whether this region ends within the kernel.
    375 		 */
    376 		if (e > kernelstart && e <= kernelend) {
    377 			if (s >= kernelstart)
    378 				goto empty;
    379 			e = kernelstart;
    380 		}
    381 		/*
    382 		 * Now page align the start and size of the region.
    383 		 */
    384 		s = round_page(s);
    385 		e = trunc_page(e);
    386 		if (e < s)
    387 			e = s;
    388 		sz = e - s;
    389 		printf("%08x-%08x = %x\n",s,e,sz);
    390 		/*
    391 		 * Check whether some memory is left here.
    392 		 */
    393 		if (sz == 0) {
    394 		empty:
    395 			memmove(mp, mp + 1,
    396 				(cnt - (mp - avail)) * sizeof *mp);
    397 			cnt--;
    398 			mp--;
    399 			continue;
    400 		}
    401 		/*
    402 		 * Do an insertion sort.
    403 		 */
    404 		npgs += btoc(sz);
    405 		for (mp1 = avail; mp1 < mp; mp1++)
    406 			if (s < mp1->start)
    407 				break;
    408 		if (mp1 < mp) {
    409 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    410 			mp1->start = s;
    411 			mp1->size = sz;
    412 		} else {
    413 			mp->start = s;
    414 			mp->size = sz;
    415 		}
    416 	}
    417 
    418 	/*
    419 	 * We cannot do pmap_steal_memory here,
    420 	 * since we don't run with translation enabled yet.
    421 	 */
    422 #ifndef MSGBUFADDR
    423 	/*
    424 	 * allow for msgbuf
    425 	 */
    426 	sz = round_page(MSGBUFSIZE);
    427 	mp = NULL;
    428 	for (mp1 = avail; mp1->size; mp1++)
    429 		if (mp1->size >= sz)
    430 			mp = mp1;
    431 	if (mp == NULL)
    432 		panic("not enough memory?");
    433 
    434 	npgs -= btoc(sz);
    435 	msgbuf_paddr = mp->start + mp->size - sz;
    436 	mp->size -= sz;
    437 	if (mp->size <= 0)
    438 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    439 #endif
    440 
    441 	printf("Loading pages\n");
    442 	for (mp = avail; mp->size; mp++)
    443 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    444 			atop(mp->start), atop(mp->start + mp->size),
    445 			VM_FREELIST_DEFAULT);
    446 
    447 	/*
    448 	 * Initialize kernel pmap and hardware.
    449 	 */
    450 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    451 	pmap_kernel()->pm_ctx = KERNEL_PID;
    452 	nextavail = avail->start;
    453 
    454 
    455 	evcnt_attach_static(&tlbhit_ev);
    456 	evcnt_attach_static(&tlbmiss_ev);
    457 	evcnt_attach_static(&tlbflush_ev);
    458 	evcnt_attach_static(&tlbenter_ev);
    459 	printf("Done\n");
    460 }
    461 
    462 /*
    463  * Restrict given range to physical memory
    464  *
    465  * (Used by /dev/mem)
    466  */
    467 void
    468 pmap_real_memory(paddr_t *start, psize_t *size)
    469 {
    470 	struct mem_region *mp;
    471 
    472 	for (mp = mem; mp->size; mp++) {
    473 		if (*start + *size > mp->start &&
    474 		    *start < mp->start + mp->size) {
    475 			if (*start < mp->start) {
    476 				*size -= mp->start - *start;
    477 				*start = mp->start;
    478 			}
    479 			if (*start + *size > mp->start + mp->size)
    480 				*size = mp->start + mp->size - *start;
    481 			return;
    482 		}
    483 	}
    484 	*size = 0;
    485 }
    486 
    487 /*
    488  * Initialize anything else for pmap handling.
    489  * Called during vm_init().
    490  */
    491 void
    492 pmap_init(void)
    493 {
    494 	struct pv_entry *pv;
    495 	vsize_t sz;
    496 	vaddr_t addr;
    497 	int i, s;
    498 	int bank;
    499 	char *attr;
    500 
    501 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    502 	sz = round_page(sz);
    503 	addr = uvm_km_zalloc(kernel_map, sz);
    504 	s = splvm();
    505 	pv = pv_table = (struct pv_entry *)addr;
    506 	for (i = npgs; --i >= 0;)
    507 		pv++->pv_pm = NULL;
    508 	pmap_attrib = (char *)pv;
    509 	memset(pv, 0, npgs);
    510 
    511 	pv = pv_table;
    512 	attr = pmap_attrib;
    513 	for (bank = 0; bank < vm_nphysseg; bank++) {
    514 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    515 		vm_physmem[bank].pmseg.pvent = pv;
    516 		vm_physmem[bank].pmseg.attrs = attr;
    517 		pv += sz;
    518 		attr += sz;
    519 	}
    520 
    521 	pmap_initialized = 1;
    522 	splx(s);
    523 
    524 	/* Setup a pool for additional pvlist structures */
    525 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", 0,
    526 		  NULL, NULL, 0);
    527 }
    528 
    529 /*
    530  * How much virtual space is available to the kernel?
    531  */
    532 void
    533 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    534 {
    535 
    536 #if 0
    537 	/*
    538 	 * Reserve one segment for kernel virtual memory
    539 	 */
    540 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    541 	*end = *start + SEGMENT_LENGTH;
    542 #else
    543 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    544 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    545 #endif
    546 }
    547 
    548 #ifdef PMAP_GROWKERNEL
    549 /*
    550  * Preallocate kernel page tables to a specified VA.
    551  * This simply loops through the first TTE for each
    552  * page table from the beginning of the kernel pmap,
    553  * reads the entry, and if the result is
    554  * zero (either invalid entry or no page table) it stores
    555  * a zero there, populating page tables in the process.
    556  * This is not the most efficient technique but i don't
    557  * expect it to be called that often.
    558  */
    559 extern struct vm_page *vm_page_alloc1 __P((void));
    560 extern void vm_page_free1 __P((struct vm_page *));
    561 
    562 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    563 
    564 vaddr_t
    565 pmap_growkernel(maxkvaddr)
    566         vaddr_t maxkvaddr;
    567 {
    568 	int s;
    569 	int seg;
    570 	paddr_t pg;
    571 	struct pmap *pm = pmap_kernel();
    572 
    573 	s = splvm();
    574 
    575 	/* Align with the start of a page table */
    576 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    577 	     kbreak += PTMAP) {
    578 		seg = STIDX(kbreak);
    579 
    580 		if (pte_find(pm, kbreak)) continue;
    581 
    582 		if (uvm.page_init_done) {
    583 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    584 		} else {
    585 			if (!uvm_page_physget(&pg))
    586 				panic("pmap_growkernel: no memory");
    587 		}
    588 		if (!pg) panic("pmap_growkernel: no pages");
    589 		pmap_zero_page((paddr_t)pg);
    590 
    591 		/* XXX This is based on all phymem being addressable */
    592 		pm->pm_ptbl[seg] = (u_int *)pg;
    593 	}
    594 	splx(s);
    595 	return (kbreak);
    596 }
    597 
    598 /*
    599  *	vm_page_alloc1:
    600  *
    601  *	Allocate and return a memory cell with no associated object.
    602  */
    603 struct vm_page *
    604 vm_page_alloc1()
    605 {
    606 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    607 	if (pg) {
    608 		pg->wire_count = 1;	/* no mappings yet */
    609 		pg->flags &= ~PG_BUSY;	/* never busy */
    610 	}
    611 	return pg;
    612 }
    613 
    614 /*
    615  *	vm_page_free1:
    616  *
    617  *	Returns the given page to the free list,
    618  *	disassociating it with any VM object.
    619  *
    620  *	Object and page must be locked prior to entry.
    621  */
    622 void
    623 vm_page_free1(mem)
    624 	struct vm_page *mem;
    625 {
    626 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    627 		printf("Freeing invalid page %p\n", mem);
    628 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    629 		Debugger();
    630 		return;
    631 	}
    632 	mem->flags |= PG_BUSY;
    633 	mem->wire_count = 0;
    634 	uvm_pagefree(mem);
    635 }
    636 #endif
    637 
    638 /*
    639  * Create and return a physical map.
    640  */
    641 struct pmap *
    642 pmap_create(void)
    643 {
    644 	struct pmap *pm;
    645 
    646 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    647 	memset((caddr_t)pm, 0, sizeof *pm);
    648 	pmap_pinit(pm);
    649 	return pm;
    650 }
    651 
    652 /*
    653  * Initialize a preallocated and zeroed pmap structure.
    654  */
    655 void
    656 pmap_pinit(struct pmap *pm)
    657 {
    658 	int i;
    659 
    660 	/*
    661 	 * Allocate some segment registers for this pmap.
    662 	 */
    663 	pm->pm_refs = 1;
    664 	for (i = 0; i < STSZ; i++)
    665 		pm->pm_ptbl[i] = NULL;
    666 }
    667 
    668 /*
    669  * Add a reference to the given pmap.
    670  */
    671 void
    672 pmap_reference(struct pmap *pm)
    673 {
    674 
    675 	pm->pm_refs++;
    676 }
    677 
    678 /*
    679  * Retire the given pmap from service.
    680  * Should only be called if the map contains no valid mappings.
    681  */
    682 void
    683 pmap_destroy(struct pmap *pm)
    684 {
    685 
    686 	if (--pm->pm_refs == 0) {
    687 		pmap_release(pm);
    688 		free((caddr_t)pm, M_VMPMAP);
    689 	}
    690 }
    691 
    692 /*
    693  * Release any resources held by the given physical map.
    694  * Called when a pmap initialized by pmap_pinit is being released.
    695  */
    696 static void
    697 pmap_release(struct pmap *pm)
    698 {
    699 	int i;
    700 
    701 	for (i = 0; i < STSZ; i++)
    702 		if (pm->pm_ptbl[i]) {
    703 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
    704 			pm->pm_ptbl[i] = NULL;
    705 		}
    706 	if (pm->pm_ctx) ctx_free(pm);
    707 }
    708 
    709 /*
    710  * Copy the range specified by src_addr/len
    711  * from the source map to the range dst_addr/len
    712  * in the destination map.
    713  *
    714  * This routine is only advisory and need not do anything.
    715  */
    716 void
    717 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    718 	  vsize_t len, vaddr_t src_addr)
    719 {
    720 }
    721 
    722 /*
    723  * Require that all active physical maps contain no
    724  * incorrect entries NOW.
    725  */
    726 void
    727 pmap_update(struct pmap *pmap)
    728 {
    729 }
    730 
    731 /*
    732  * Garbage collects the physical map system for
    733  * pages which are no longer used.
    734  * Success need not be guaranteed -- that is, there
    735  * may well be pages which are not referenced, but
    736  * others may be collected.
    737  * Called by the pageout daemon when pages are scarce.
    738  */
    739 void
    740 pmap_collect(struct pmap *pm)
    741 {
    742 }
    743 
    744 /*
    745  * Fill the given physical page with zeroes.
    746  */
    747 void
    748 pmap_zero_page(paddr_t pa)
    749 {
    750 
    751 #ifdef NOCACHE
    752 	memset((caddr_t)pa, 0, NBPG);
    753 #else
    754 	int i;
    755 
    756 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
    757 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    758 		pa += CACHELINESIZE;
    759 	}
    760 #endif
    761 }
    762 
    763 /*
    764  * Copy the given physical source page to its destination.
    765  */
    766 void
    767 pmap_copy_page(paddr_t src, paddr_t dst)
    768 {
    769 
    770 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
    771 	dcache_flush_page(dst);
    772 }
    773 
    774 /*
    775  * This returns whether this is the first mapping of a page.
    776  */
    777 static inline int
    778 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    779 {
    780 	struct pv_entry *pv, *npv = NULL;
    781 	int s;
    782 
    783 	if (!pmap_initialized)
    784 		return 0;
    785 
    786 	s = splvm();
    787 
    788 	pv = pa_to_pv(pa);
    789 	if (!pv->pv_pm) {
    790 		/*
    791 		 * No entries yet, use header as the first entry.
    792 		 */
    793 		pv->pv_va = va;
    794 		pv->pv_pm = pm;
    795 		pv->pv_next = NULL;
    796 	} else {
    797 		/*
    798 		 * There is at least one other VA mapping this page.
    799 		 * Place this entry after the header.
    800 		 */
    801 		npv = pool_get(&pv_pool, PR_WAITOK);
    802 		if (!npv) return (0);
    803 		npv->pv_va = va;
    804 		npv->pv_pm = pm;
    805 		npv->pv_next = pv->pv_next;
    806 		pv->pv_next = npv;
    807 	}
    808 	splx(s);
    809 	return (1);
    810 }
    811 
    812 static void
    813 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    814 {
    815 	struct pv_entry *pv, *npv;
    816 
    817 	/*
    818 	 * Remove from the PV table.
    819 	 */
    820 	pv = pa_to_pv(pa);
    821 	if (!pv) return;
    822 
    823 	/*
    824 	 * If it is the first entry on the list, it is actually
    825 	 * in the header and we must copy the following entry up
    826 	 * to the header.  Otherwise we must search the list for
    827 	 * the entry.  In either case we free the now unused entry.
    828 	 */
    829 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    830 		if ((npv = pv->pv_next)) {
    831 			*pv = *npv;
    832 			pool_put(&pv_pool, npv);
    833 		} else
    834 			pv->pv_pm = NULL;
    835 	} else {
    836 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    837 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    838 				break;
    839 		if (npv) {
    840 			pv->pv_next = npv->pv_next;
    841 			pool_put(&pv_pool, npv);
    842 		}
    843 	}
    844 }
    845 
    846 /*
    847  * Insert physical page at pa into the given pmap at virtual address va.
    848  */
    849 int
    850 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    851 {
    852 	int s;
    853 	u_int tte;
    854 	int managed;
    855 
    856 	/*
    857 	 * Have to remove any existing mapping first.
    858 	 */
    859 	pmap_remove(pm, va, va + NBPG);
    860 
    861 	if (flags & PMAP_WIRED) flags |= prot;
    862 
    863 	/* If it has no protections don't bother w/the rest */
    864 	if (!(flags & VM_PROT_ALL))
    865 		return (0);
    866 
    867 	managed = 0;
    868 	if (vm_physseg_find(atop(pa), NULL) != -1)
    869 		managed = 1;
    870 
    871 	/*
    872 	 * Generate TTE.
    873 	 *
    874 	 * XXXX
    875 	 *
    876 	 * Since the kernel does not handle execution privileges properly,
    877 	 * we will handle read and execute permissions together.
    878 	 */
    879 	tte = TTE_PA(pa) | TTE_EX;
    880 	/* XXXX -- need to support multiple page sizes. */
    881 	tte |= TTE_SZ_16K;
    882 #ifdef	DIAGNOSTIC
    883 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    884 		(PME_NOCACHE | PME_WRITETHROUG))
    885 		panic("pmap_enter: uncached & writethrough\n");
    886 #endif
    887 	if (flags & PME_NOCACHE)
    888 		/* Must be I/O mapping */
    889 		tte |= TTE_I | TTE_G;
    890 #ifdef NOCACHE
    891 	tte |= TTE_I;
    892 #else
    893 	else if (flags & PME_WRITETHROUG)
    894 		/* Uncached and writethrough are not compatible */
    895 		tte |= TTE_W;
    896 #endif
    897 	if (pm == pmap_kernel())
    898 		tte |= TTE_ZONE(ZONE_PRIV);
    899 	else
    900 		tte |= TTE_ZONE(ZONE_USER);
    901 
    902 	if (flags & VM_PROT_WRITE)
    903 		tte |= TTE_WR;
    904 
    905 	/*
    906 	 * Now record mapping for later back-translation.
    907 	 */
    908 	if (pmap_initialized && managed) {
    909 		char *attr;
    910 
    911 		if (!pmap_enter_pv(pm, va, pa)) {
    912 			/* Could not enter pv on a managed page */
    913 			return 1;
    914 		}
    915 
    916 		/* Now set attributes. */
    917 		attr = pa_to_attr(pa);
    918 #ifdef DIAGNOSTIC
    919 		if (!attr)
    920 			panic("managed but no attr\n");
    921 #endif
    922 		if (flags & VM_PROT_ALL)
    923 			*attr |= PTE_HI_REF;
    924 		if (flags & VM_PROT_WRITE)
    925 			*attr |= PTE_HI_CHG;
    926 	}
    927 
    928 	s = splvm();
    929 	pm->pm_stats.resident_count++;
    930 
    931 	/* Insert page into page table. */
    932 	pte_enter(pm, va, tte);
    933 
    934 	/* If this is a real fault, enter it in the tlb */
    935 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    936 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    937 	}
    938 	splx(s);
    939 	return 0;
    940 }
    941 
    942 void
    943 pmap_unwire(struct pmap *pm, vaddr_t va)
    944 {
    945 	struct pv_entry *pv, *npv;
    946 	paddr_t pa;
    947 	int s = splvm();
    948 
    949         if (pm == NULL) {
    950                 return;
    951         }
    952 
    953 	if (!pmap_extract(pm, va, &pa)) {
    954 		return;
    955 	}
    956 
    957 	va |= PV_WIRED;
    958 
    959 	pv = pa_to_pv(pa);
    960 	if (!pv) return;
    961 
    962 	/*
    963 	 * If it is the first entry on the list, it is actually
    964 	 * in the header and we must copy the following entry up
    965 	 * to the header.  Otherwise we must search the list for
    966 	 * the entry.  In either case we free the now unused entry.
    967 	 */
    968 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    969 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    970 			npv->pv_va &= ~PV_WIRED;
    971 			break;
    972 		}
    973 	}
    974 	splx(s);
    975 }
    976 
    977 void
    978 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    979 {
    980 	int s;
    981 	u_int tte;
    982 	struct pmap *pm = pmap_kernel();
    983 
    984 	/*
    985 	 * Have to remove any existing mapping first.
    986 	 */
    987 
    988 	/*
    989 	 * Generate TTE.
    990 	 *
    991 	 * XXXX
    992 	 *
    993 	 * Since the kernel does not handle execution privileges properly,
    994 	 * we will handle read and execute permissions together.
    995 	 */
    996 	tte = 0;
    997 	if (prot & VM_PROT_ALL) {
    998 
    999 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
   1000 		/* XXXX -- need to support multiple page sizes. */
   1001 		tte |= TTE_SZ_16K;
   1002 #ifdef DIAGNOSTIC
   1003 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
   1004 			(PME_NOCACHE | PME_WRITETHROUG))
   1005 			panic("pmap_kenter_pa: uncached & writethrough\n");
   1006 #endif
   1007 		if (prot & PME_NOCACHE)
   1008 			/* Must be I/O mapping */
   1009 			tte |= TTE_I | TTE_G;
   1010 #ifdef NOCACHE
   1011 		tte |= TTE_I;
   1012 #else
   1013 		else if (prot & PME_WRITETHROUG)
   1014 			/* Uncached and writethrough are not compatible */
   1015 			tte |= TTE_W;
   1016 #endif
   1017 		if (prot & VM_PROT_WRITE)
   1018 			tte |= TTE_WR;
   1019 	}
   1020 
   1021 	s = splvm();
   1022 	pm->pm_stats.resident_count++;
   1023 
   1024 	/* Insert page into page table. */
   1025 	pte_enter(pm, va, tte);
   1026 	splx(s);
   1027 }
   1028 
   1029 void
   1030 pmap_kremove(vaddr_t va, vsize_t len)
   1031 {
   1032 
   1033 	while (len > 0) {
   1034 		pte_enter(pmap_kernel(), va, 0);
   1035 		va += PAGE_SIZE;
   1036 		len -= PAGE_SIZE;
   1037 	}
   1038 }
   1039 
   1040 /*
   1041  * Remove the given range of mapping entries.
   1042  */
   1043 void
   1044 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
   1045 {
   1046 	int s;
   1047 	paddr_t pa;
   1048 	volatile u_int *ptp;
   1049 
   1050 	s = splvm();
   1051 	while (va < endva) {
   1052 
   1053 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1054 			pa = TTE_PA(pa);
   1055 			pmap_remove_pv(pm, va, pa);
   1056 			*ptp = 0;
   1057 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1058 			pm->pm_stats.resident_count--;
   1059 		}
   1060 		va += NBPG;
   1061 	}
   1062 
   1063 	splx(s);
   1064 }
   1065 
   1066 /*
   1067  * Get the physical page address for the given pmap/virtual address.
   1068  */
   1069 boolean_t
   1070 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1071 {
   1072 	int seg = STIDX(va);
   1073 	int ptn = PTIDX(va);
   1074 	u_int pa = 0;
   1075 	int s = splvm();
   1076 
   1077 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1078 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1079 	}
   1080 	splx(s);
   1081 	return (pa != 0);
   1082 }
   1083 
   1084 /*
   1085  * Lower the protection on the specified range of this pmap.
   1086  *
   1087  * There are only two cases: either the protection is going to 0,
   1088  * or it is going to read-only.
   1089  */
   1090 void
   1091 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1092 {
   1093 	volatile u_int *ptp;
   1094 	int s;
   1095 
   1096 	if (prot & VM_PROT_READ) {
   1097 		s = splvm();
   1098 		while (sva < eva) {
   1099 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1100 				*ptp &= ~TTE_WR;
   1101 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1102 			}
   1103 			sva += NBPG;
   1104 		}
   1105 		splx(s);
   1106 		return;
   1107 	}
   1108 	pmap_remove(pm, sva, eva);
   1109 }
   1110 
   1111 boolean_t
   1112 check_attr(struct vm_page *pg, u_int mask, int clear)
   1113 {
   1114 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1115 	int s;
   1116 	char *attr;
   1117 	int rv;
   1118 
   1119 	/*
   1120 	 * First modify bits in cache.
   1121 	 */
   1122 	s = splvm();
   1123 	attr = pa_to_attr(pa);
   1124 	if (attr == NULL)
   1125 		return FALSE;
   1126 
   1127 	rv = ((*attr & mask) != 0);
   1128 	if (clear)
   1129 		*attr &= ~mask;
   1130 
   1131 	splx(s);
   1132 	return rv;
   1133 }
   1134 
   1135 
   1136 /*
   1137  * Lower the protection on the specified physical page.
   1138  *
   1139  * There are only two cases: either the protection is going to 0,
   1140  * or it is going to read-only.
   1141  */
   1142 void
   1143 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1144 {
   1145 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1146 	vaddr_t va;
   1147 	struct pv_entry *pvh, *pv, *npv;
   1148 	struct pmap *pm;
   1149 
   1150 	pvh = pa_to_pv(pa);
   1151 	if (pvh == NULL)
   1152 		return;
   1153 
   1154 	/* Handle extra pvs which may be deleted in the operation */
   1155 	for (pv = pvh->pv_next; pv; pv = npv) {
   1156 		npv = pv->pv_next;
   1157 
   1158 		pm = pv->pv_pm;
   1159 		va = pv->pv_va;
   1160 		pmap_protect(pm, va, va+NBPG, prot);
   1161 	}
   1162 	/* Now check the head pv */
   1163 	if (pvh->pv_pm) {
   1164 		pv = pvh;
   1165 		pm = pv->pv_pm;
   1166 		va = pv->pv_va;
   1167 		pmap_protect(pm, va, va+NBPG, prot);
   1168 	}
   1169 }
   1170 
   1171 /*
   1172  * Activate the address space for the specified process.  If the process
   1173  * is the current process, load the new MMU context.
   1174  */
   1175 void
   1176 pmap_activate(struct lwp *l)
   1177 {
   1178 #if 0
   1179 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
   1180 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
   1181 
   1182 	/*
   1183 	 * XXX Normally performed in cpu_fork().
   1184 	 */
   1185 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
   1186 	if (pcb->pcb_pm != pmap) {
   1187 		pcb->pcb_pm = pmap;
   1188 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1189 		    (paddr_t *)&pcb->pcb_pmreal);
   1190 	}
   1191 
   1192 	if (l == curproc) {
   1193 		/* Store pointer to new current pmap. */
   1194 		curpm = pcb->pcb_pmreal;
   1195 	}
   1196 #endif
   1197 }
   1198 
   1199 /*
   1200  * Deactivate the specified process's address space.
   1201  */
   1202 void
   1203 pmap_deactivate(struct lwp *l)
   1204 {
   1205 }
   1206 
   1207 /*
   1208  * Synchronize caches corresponding to [addr, addr+len) in p.
   1209  */
   1210 void
   1211 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1212 {
   1213 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1214 	int msr, ctx, opid;
   1215 
   1216 
   1217 	/*
   1218 	 * Need to turn off IMMU and switch to user context.
   1219 	 * (icbi uses DMMU).
   1220 	 */
   1221 	if (!(ctx = pm->pm_ctx)) {
   1222 		/* No context -- assign it one */
   1223 		ctx_alloc(pm);
   1224 		ctx = pm->pm_ctx;
   1225 	}
   1226 	__asm __volatile("mfmsr %0;"
   1227 		"li %1, 0x20;"
   1228 		"andc %1,%0,%1;"
   1229 		"mtmsr %1;"
   1230 		"sync;isync;"
   1231 		"mfpid %1;"
   1232 		"mtpid %2;"
   1233 		"sync; isync;"
   1234 		"1:"
   1235 		"dcbf 0,%3;"
   1236 		"icbi 0,%3;"
   1237 		"addi %3,%3,32;"
   1238 		"addic. %4,%4,-32;"
   1239 		"bge 1b;"
   1240 		"mtpid %1;"
   1241 		"mtmsr %0;"
   1242 		"sync; isync"
   1243 		: "=&r" (msr), "=&r" (opid)
   1244 		: "r" (ctx), "r" (va), "r" (len));
   1245 }
   1246 
   1247 
   1248 /* This has to be done in real mode !!! */
   1249 void
   1250 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1251 {
   1252 	u_long i, found;
   1253 	u_long msr;
   1254 
   1255 	/* If there's no context then it can't be mapped. */
   1256 	if (!pid) return;
   1257 
   1258 	asm("mfpid %1;"			/* Save PID */
   1259 		"mfmsr %2;"		/* Save MSR */
   1260 		"li %0,0;"		/* Now clear MSR */
   1261 		"mtmsr %0;"
   1262 		"mtpid %4;"		/* Set PID */
   1263 		"sync;"
   1264 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1265 		"sync;"
   1266 		"mtpid %1;"		/* Restore PID */
   1267 		"mtmsr %2;"		/* Restore MSR */
   1268 		"sync;isync;"
   1269 		"li %1,1;"
   1270 		"beq 1f;"
   1271 		"li %1,0;"
   1272 		"1:"
   1273 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1274 		: "r" (va), "r" (pid));
   1275 	if (found && !TLB_LOCKED(i)) {
   1276 
   1277 		/* Now flush translation */
   1278 		asm volatile(
   1279 			"tlbwe %0,%1,0;"
   1280 			"sync;isync;"
   1281 			: : "r" (0), "r" (i));
   1282 
   1283 		tlb_info[i].ti_ctx = 0;
   1284 		tlb_info[i].ti_flags = 0;
   1285 		tlbnext = i;
   1286 		/* Successful flushes */
   1287 		tlbflush_ev.ev_count++;
   1288 	}
   1289 }
   1290 
   1291 void
   1292 ppc4xx_tlb_flush_all(void)
   1293 {
   1294 	u_long i;
   1295 
   1296 	for (i = 0; i < NTLB; i++)
   1297 		if (!TLB_LOCKED(i)) {
   1298 			asm volatile(
   1299 				"tlbwe %0,%1,0;"
   1300 				"sync;isync;"
   1301 				: : "r" (0), "r" (i));
   1302 			tlb_info[i].ti_ctx = 0;
   1303 			tlb_info[i].ti_flags = 0;
   1304 		}
   1305 
   1306 	asm volatile("sync;isync");
   1307 }
   1308 
   1309 /* Find a TLB entry to evict. */
   1310 static int
   1311 ppc4xx_tlb_find_victim(void)
   1312 {
   1313 	int flags;
   1314 
   1315 	for (;;) {
   1316 		if (++tlbnext >= NTLB)
   1317 			tlbnext = TLB_NRESERVED;
   1318 		flags = tlb_info[tlbnext].ti_flags;
   1319 		if (!(flags & TLBF_USED) ||
   1320 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1321 			u_long va, stack = (u_long)&va;
   1322 
   1323 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1324 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1325 			     (flags & TLBF_USED)) {
   1326 				/* Kernel stack page */
   1327 				flags |= TLBF_USED;
   1328 				tlb_info[tlbnext].ti_flags = flags;
   1329 			} else {
   1330 				/* Found it! */
   1331 				return (tlbnext);
   1332 			}
   1333 		} else {
   1334 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1335 		}
   1336 	}
   1337 }
   1338 
   1339 void
   1340 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1341 {
   1342 	u_long th, tl, idx;
   1343 	tlbpid_t pid;
   1344 	u_short msr;
   1345 	int s;
   1346 
   1347 	tlbenter_ev.ev_count++;
   1348 
   1349 	th = (va & TLB_EPN_MASK) |
   1350 		(((pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT) << TLB_SIZE_SHFT) |
   1351 		TLB_VALID;
   1352 	tl = pte;
   1353 
   1354 	s = splhigh();
   1355 	idx = ppc4xx_tlb_find_victim();
   1356 
   1357 #ifdef DIAGNOSTIC
   1358 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1359 		panic("ppc4xx_tlb_enter: repacing entry %ld\n", idx);
   1360 	}
   1361 #endif
   1362 
   1363 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1364 	tlb_info[idx].ti_ctx = ctx;
   1365 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1366 
   1367 	asm volatile(
   1368 		"mfmsr %0;"			/* Save MSR */
   1369 		"li %1,0;"
   1370 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1371 		"mtmsr %1;"			/* Clear MSR */
   1372 		"mfpid %1;"			/* Save old PID */
   1373 		"mtpid %2;"			/* Load translation ctx */
   1374 		"sync; isync;"
   1375 #ifdef DEBUG
   1376 		"andi. %3,%3,63;"
   1377 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1378 #endif
   1379 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1380 		"sync; isync;"
   1381 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1382 		"sync; isync;"
   1383 	: "=&r" (msr), "=&r" (pid)
   1384 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1385 	splx(s);
   1386 }
   1387 
   1388 void
   1389 ppc4xx_tlb_unpin(int i)
   1390 {
   1391 
   1392 	if (i == -1)
   1393 		for (i = 0; i < TLB_NRESERVED; i++)
   1394 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1395 	else
   1396 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1397 }
   1398 
   1399 void
   1400 ppc4xx_tlb_init(void)
   1401 {
   1402 	int i;
   1403 
   1404 	/* Mark reserved TLB entries */
   1405 	for (i = 0; i < TLB_NRESERVED; i++) {
   1406 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1407 		tlb_info[i].ti_ctx = KERNEL_PID;
   1408 	}
   1409 
   1410 	/* Setup security zones */
   1411 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1412 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1413 	 * Z3 - full access regardless of TLB entry permissions
   1414 	 */
   1415 
   1416 	asm volatile(
   1417 		"mtspr %0,%1;"
   1418 		"sync;"
   1419 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1420 }
   1421 
   1422 
   1423 /*
   1424  * We should pass the ctx in from trap code.
   1425  */
   1426 int
   1427 pmap_tlbmiss(vaddr_t va, int ctx)
   1428 {
   1429 	volatile u_int *pte;
   1430 	u_long tte;
   1431 
   1432 	tlbmiss_ev.ev_count++;
   1433 
   1434 	/*
   1435 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1436 	 */
   1437 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1438 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1439 		if (pte == NULL) {
   1440 			/* Map unmanaged addresses directly for kernel access */
   1441 			return 1;
   1442 		}
   1443 		tte = *pte;
   1444 		if (tte == 0) {
   1445 			return 1;
   1446 		}
   1447 	} else {
   1448 		/* Create a 16MB writeable mapping. */
   1449 #ifdef NOCACHE
   1450 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1451 #else
   1452 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1453 #endif
   1454 	}
   1455 	tlbhit_ev.ev_count++;
   1456 	ppc4xx_tlb_enter(ctx, va, tte);
   1457 
   1458 	return 0;
   1459 }
   1460 
   1461 /*
   1462  * Flush all the entries matching a context from the TLB.
   1463  */
   1464 static int
   1465 ctx_flush(int cnum)
   1466 {
   1467 	int i;
   1468 
   1469 	/* We gotta steal this context */
   1470 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1471 		if (tlb_info[i].ti_ctx == cnum) {
   1472 			/* Can't steal ctx if it has a locked entry. */
   1473 			if (TLB_LOCKED(i)) {
   1474 #ifdef DIAGNOSTIC
   1475 				printf("ctx_flush: can't invalidate "
   1476 					"locked mapping %d "
   1477 					"for context %d\n", i, cnum);
   1478 				Debugger();
   1479 #endif
   1480 				return (1);
   1481 			}
   1482 #ifdef DIAGNOSTIC
   1483 			if (i < TLB_NRESERVED)
   1484 				panic("TLB entry %d not locked\n", i);
   1485 #endif
   1486 			/* Invalidate particular TLB entry regardless of locked status */
   1487 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1488 			tlb_info[i].ti_flags = 0;
   1489 		}
   1490 	}
   1491 	return (0);
   1492 }
   1493 
   1494 /*
   1495  * Allocate a context.  If necessary, steal one from someone else.
   1496  *
   1497  * The new context is flushed from the TLB before returning.
   1498  */
   1499 int
   1500 ctx_alloc(struct pmap *pm)
   1501 {
   1502 	int s, cnum;
   1503 	static int next = MINCTX;
   1504 
   1505 	if (pm == pmap_kernel()) {
   1506 #ifdef DIAGNOSTIC
   1507 		printf("ctx_alloc: kernel pmap!\n");
   1508 #endif
   1509 		return (0);
   1510 	}
   1511 	s = splvm();
   1512 
   1513 	/* Find a likely context. */
   1514 	cnum = next;
   1515 	do {
   1516 		if ((++cnum) > NUMCTX)
   1517 			cnum = MINCTX;
   1518 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1519 
   1520 	/* Now clean it out */
   1521 oops:
   1522 	if (cnum < MINCTX)
   1523 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1524 	if (ctx_flush(cnum)) {
   1525 		/* oops -- something's wired. */
   1526 		if ((++cnum) > NUMCTX)
   1527 			cnum = MINCTX;
   1528 		goto oops;
   1529 	}
   1530 
   1531 	if (ctxbusy[cnum]) {
   1532 #ifdef DEBUG
   1533 		/* We should identify this pmap and clear it */
   1534 		printf("Warning: stealing context %d\n", cnum);
   1535 #endif
   1536 		ctxbusy[cnum]->pm_ctx = 0;
   1537 	}
   1538 	ctxbusy[cnum] = pm;
   1539 	next = cnum;
   1540 	splx(s);
   1541 	pm->pm_ctx = cnum;
   1542 
   1543 	return cnum;
   1544 }
   1545 
   1546 /*
   1547  * Give away a context.
   1548  */
   1549 void
   1550 ctx_free(struct pmap *pm)
   1551 {
   1552 	int oldctx;
   1553 
   1554 	oldctx = pm->pm_ctx;
   1555 
   1556 	if (oldctx == 0)
   1557 		panic("ctx_free: freeing kernel context");
   1558 #ifdef DIAGNOSTIC
   1559 	if (ctxbusy[oldctx] == 0)
   1560 		printf("ctx_free: freeing free context %d\n", oldctx);
   1561 	if (ctxbusy[oldctx] != pm) {
   1562 		printf("ctx_free: freeing someone esle's context\n "
   1563 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1564 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1565 		Debugger();
   1566 	}
   1567 #endif
   1568 	/* We should verify it has not been stolen and reallocated... */
   1569 	ctxbusy[oldctx] = NULL;
   1570 	ctx_flush(oldctx);
   1571 }
   1572 
   1573 
   1574 #ifdef DEBUG
   1575 /*
   1576  * Test ref/modify handling.
   1577  */
   1578 void pmap_testout __P((void));
   1579 void
   1580 pmap_testout()
   1581 {
   1582 	vaddr_t va;
   1583 	volatile int *loc;
   1584 	int val = 0;
   1585 	paddr_t pa;
   1586 	struct vm_page *pg;
   1587 	int ref, mod;
   1588 
   1589 	/* Allocate a page */
   1590 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
   1591 	loc = (int*)va;
   1592 
   1593 	pmap_extract(pmap_kernel(), va, &pa);
   1594 	pg = PHYS_TO_VM_PAGE(pa);
   1595 	pmap_unwire(pmap_kernel(), va);
   1596 
   1597 	pmap_remove(pmap_kernel(), va, va+1);
   1598 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1599 	pmap_update(pmap_kernel());
   1600 
   1601 	/* Now clear reference and modify */
   1602 	ref = pmap_clear_reference(pg);
   1603 	mod = pmap_clear_modify(pg);
   1604 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1605 	       (void *)(u_long)va, (long)pa,
   1606 	       ref, mod);
   1607 
   1608 	/* Check it's properly cleared */
   1609 	ref = pmap_is_referenced(pg);
   1610 	mod = pmap_is_modified(pg);
   1611 	printf("Checking cleared page: ref %d, mod %d\n",
   1612 	       ref, mod);
   1613 
   1614 	/* Reference page */
   1615 	val = *loc;
   1616 
   1617 	ref = pmap_is_referenced(pg);
   1618 	mod = pmap_is_modified(pg);
   1619 	printf("Referenced page: ref %d, mod %d val %x\n",
   1620 	       ref, mod, val);
   1621 
   1622 	/* Now clear reference and modify */
   1623 	ref = pmap_clear_reference(pg);
   1624 	mod = pmap_clear_modify(pg);
   1625 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1626 	       (void *)(u_long)va, (long)pa,
   1627 	       ref, mod);
   1628 
   1629 	/* Modify page */
   1630 	*loc = 1;
   1631 
   1632 	ref = pmap_is_referenced(pg);
   1633 	mod = pmap_is_modified(pg);
   1634 	printf("Modified page: ref %d, mod %d\n",
   1635 	       ref, mod);
   1636 
   1637 	/* Now clear reference and modify */
   1638 	ref = pmap_clear_reference(pg);
   1639 	mod = pmap_clear_modify(pg);
   1640 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1641 	       (void *)(u_long)va, (long)pa,
   1642 	       ref, mod);
   1643 
   1644 	/* Check it's properly cleared */
   1645 	ref = pmap_is_referenced(pg);
   1646 	mod = pmap_is_modified(pg);
   1647 	printf("Checking cleared page: ref %d, mod %d\n",
   1648 	       ref, mod);
   1649 
   1650 	/* Modify page */
   1651 	*loc = 1;
   1652 
   1653 	ref = pmap_is_referenced(pg);
   1654 	mod = pmap_is_modified(pg);
   1655 	printf("Modified page: ref %d, mod %d\n",
   1656 	       ref, mod);
   1657 
   1658 	/* Check pmap_protect() */
   1659 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1660 	pmap_update(pmap_kernel());
   1661 	ref = pmap_is_referenced(pg);
   1662 	mod = pmap_is_modified(pg);
   1663 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1664 	       ref, mod);
   1665 
   1666 	/* Now clear reference and modify */
   1667 	ref = pmap_clear_reference(pg);
   1668 	mod = pmap_clear_modify(pg);
   1669 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1670 	       (void *)(u_long)va, (long)pa,
   1671 	       ref, mod);
   1672 
   1673 	/* Reference page */
   1674 	val = *loc;
   1675 
   1676 	ref = pmap_is_referenced(pg);
   1677 	mod = pmap_is_modified(pg);
   1678 	printf("Referenced page: ref %d, mod %d val %x\n",
   1679 	       ref, mod, val);
   1680 
   1681 	/* Now clear reference and modify */
   1682 	ref = pmap_clear_reference(pg);
   1683 	mod = pmap_clear_modify(pg);
   1684 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1685 	       (void *)(u_long)va, (long)pa,
   1686 	       ref, mod);
   1687 
   1688 	/* Modify page */
   1689 #if 0
   1690 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1691 	pmap_update(pmap_kernel());
   1692 #endif
   1693 	*loc = 1;
   1694 
   1695 	ref = pmap_is_referenced(pg);
   1696 	mod = pmap_is_modified(pg);
   1697 	printf("Modified page: ref %d, mod %d\n",
   1698 	       ref, mod);
   1699 
   1700 	/* Check pmap_protect() */
   1701 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1702 	pmap_update(pmap_kernel());
   1703 	ref = pmap_is_referenced(pg);
   1704 	mod = pmap_is_modified(pg);
   1705 	printf("pmap_protect(): ref %d, mod %d\n",
   1706 	       ref, mod);
   1707 
   1708 	/* Now clear reference and modify */
   1709 	ref = pmap_clear_reference(pg);
   1710 	mod = pmap_clear_modify(pg);
   1711 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1712 	       (void *)(u_long)va, (long)pa,
   1713 	       ref, mod);
   1714 
   1715 	/* Reference page */
   1716 	val = *loc;
   1717 
   1718 	ref = pmap_is_referenced(pg);
   1719 	mod = pmap_is_modified(pg);
   1720 	printf("Referenced page: ref %d, mod %d val %x\n",
   1721 	       ref, mod, val);
   1722 
   1723 	/* Now clear reference and modify */
   1724 	ref = pmap_clear_reference(pg);
   1725 	mod = pmap_clear_modify(pg);
   1726 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1727 	       (void *)(u_long)va, (long)pa,
   1728 	       ref, mod);
   1729 
   1730 	/* Modify page */
   1731 #if 0
   1732 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1733 	pmap_update(pmap_kernel());
   1734 #endif
   1735 	*loc = 1;
   1736 
   1737 	ref = pmap_is_referenced(pg);
   1738 	mod = pmap_is_modified(pg);
   1739 	printf("Modified page: ref %d, mod %d\n",
   1740 	       ref, mod);
   1741 
   1742 	/* Check pmap_pag_protect() */
   1743 	pmap_page_protect(pg, VM_PROT_READ);
   1744 	ref = pmap_is_referenced(pg);
   1745 	mod = pmap_is_modified(pg);
   1746 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1747 	       ref, mod);
   1748 
   1749 	/* Now clear reference and modify */
   1750 	ref = pmap_clear_reference(pg);
   1751 	mod = pmap_clear_modify(pg);
   1752 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1753 	       (void *)(u_long)va, (long)pa,
   1754 	       ref, mod);
   1755 
   1756 	/* Reference page */
   1757 	val = *loc;
   1758 
   1759 	ref = pmap_is_referenced(pg);
   1760 	mod = pmap_is_modified(pg);
   1761 	printf("Referenced page: ref %d, mod %d val %x\n",
   1762 	       ref, mod, val);
   1763 
   1764 	/* Now clear reference and modify */
   1765 	ref = pmap_clear_reference(pg);
   1766 	mod = pmap_clear_modify(pg);
   1767 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1768 	       (void *)(u_long)va, (long)pa,
   1769 	       ref, mod);
   1770 
   1771 	/* Modify page */
   1772 #if 0
   1773 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1774 	pmap_update(pmap_kernel());
   1775 #endif
   1776 	*loc = 1;
   1777 
   1778 	ref = pmap_is_referenced(pg);
   1779 	mod = pmap_is_modified(pg);
   1780 	printf("Modified page: ref %d, mod %d\n",
   1781 	       ref, mod);
   1782 
   1783 	/* Check pmap_pag_protect() */
   1784 	pmap_page_protect(pg, VM_PROT_NONE);
   1785 	ref = pmap_is_referenced(pg);
   1786 	mod = pmap_is_modified(pg);
   1787 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1788 	       ref, mod);
   1789 
   1790 	/* Now clear reference and modify */
   1791 	ref = pmap_clear_reference(pg);
   1792 	mod = pmap_clear_modify(pg);
   1793 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1794 	       (void *)(u_long)va, (long)pa,
   1795 	       ref, mod);
   1796 
   1797 
   1798 	/* Reference page */
   1799 	val = *loc;
   1800 
   1801 	ref = pmap_is_referenced(pg);
   1802 	mod = pmap_is_modified(pg);
   1803 	printf("Referenced page: ref %d, mod %d val %x\n",
   1804 	       ref, mod, val);
   1805 
   1806 	/* Now clear reference and modify */
   1807 	ref = pmap_clear_reference(pg);
   1808 	mod = pmap_clear_modify(pg);
   1809 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1810 	       (void *)(u_long)va, (long)pa,
   1811 	       ref, mod);
   1812 
   1813 	/* Modify page */
   1814 #if 0
   1815 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1816 	pmap_update(pmap_kernel());
   1817 #endif
   1818 	*loc = 1;
   1819 
   1820 	ref = pmap_is_referenced(pg);
   1821 	mod = pmap_is_modified(pg);
   1822 	printf("Modified page: ref %d, mod %d\n",
   1823 	       ref, mod);
   1824 
   1825 	/* Unmap page */
   1826 	pmap_remove(pmap_kernel(), va, va+1);
   1827 	pmap_update(pmap_kernel());
   1828 	ref = pmap_is_referenced(pg);
   1829 	mod = pmap_is_modified(pg);
   1830 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1831 
   1832 	/* Now clear reference and modify */
   1833 	ref = pmap_clear_reference(pg);
   1834 	mod = pmap_clear_modify(pg);
   1835 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1836 	       (void *)(u_long)va, (long)pa, ref, mod);
   1837 
   1838 	/* Check it's properly cleared */
   1839 	ref = pmap_is_referenced(pg);
   1840 	mod = pmap_is_modified(pg);
   1841 	printf("Checking cleared page: ref %d, mod %d\n",
   1842 	       ref, mod);
   1843 
   1844 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1845 		VM_PROT_ALL|PMAP_WIRED);
   1846 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
   1847 }
   1848 #endif
   1849