Home | History | Annotate | Line # | Download | only in ibm4xx
pmap.c revision 1.6
      1 /*	$NetBSD: pmap.c,v 1.6 2001/11/06 04:49:49 simonb 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 
     80 #include <uvm/uvm.h>
     81 
     82 #include <machine/pcb.h>
     83 #include <machine/powerpc.h>
     84 
     85 #include <powerpc/spr.h>
     86 #include <powerpc/ibm4xx/tlb.h>
     87 
     88 
     89 #define	CACHE_LINE	32
     90 
     91 /*
     92  * kernmap is an array of PTEs large enough to map in
     93  * 4GB.  At 16KB/page it is 256K entries or 2MB.
     94  */
     95 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
     96 caddr_t kernmap;
     97 
     98 #define MINCTX		2
     99 #define NUMCTX		256
    100 volatile struct pmap *ctxbusy[NUMCTX];
    101 
    102 #define TLBF_USED	0x1
    103 #define	TLBF_REF	0x2
    104 #define	TLBF_LOCKED	0x4
    105 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
    106 typedef struct tlb_info_s {
    107 	char	ti_flags;
    108 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
    109 	u_int	ti_va;
    110 } tlb_info_t;
    111 
    112 volatile tlb_info_t tlb_info[NTLB];
    113 /* We'll use a modified FIFO replacement policy cause it's cheap */
    114 volatile int tlbnext = TLB_NRESERVED;
    115 
    116 u_long dtlb_miss_count = 0;
    117 u_long itlb_miss_count = 0;
    118 u_long ktlb_miss_count = 0;
    119 u_long utlb_miss_count = 0;
    120 
    121 /* Event counters -- XXX type `INTR' so we can see them with vmstat -i */
    122 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    123 	NULL, "cpu", "tlbmiss");
    124 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    125 	NULL, "cpu", "tlbhit");
    126 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    127 	NULL, "cpu", "tlbflush");
    128 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
    129 	NULL, "cpu", "tlbenter");
    130 
    131 struct pmap kernel_pmap_;
    132 
    133 int physmem;
    134 static int npgs;
    135 static u_int nextavail;
    136 #ifndef MSGBUFADDR
    137 extern paddr_t msgbuf_paddr;
    138 #endif
    139 
    140 static struct mem_region *mem, *avail;
    141 
    142 /*
    143  * This is a cache of referenced/modified bits.
    144  * Bits herein are shifted by ATTRSHFT.
    145  */
    146 static char *pmap_attrib;
    147 
    148 #define PV_WIRED	0x1
    149 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
    150 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
    151 
    152 struct pv_entry {
    153 	struct pv_entry *pv_next;	/* Linked list of mappings */
    154 	vaddr_t pv_va;			/* virtual address of mapping */
    155 	struct pmap *pv_pm;
    156 };
    157 
    158 struct pv_entry *pv_table;
    159 static struct pool pv_pool;
    160 
    161 static int pmap_initialized;
    162 
    163 static int ctx_flush(int);
    164 
    165 static inline void dcache_flush_page(vaddr_t);
    166 static inline void icache_flush_page(vaddr_t);
    167 static inline void dcache_flush(vaddr_t, vsize_t);
    168 static inline void icache_flush(vaddr_t, vsize_t);
    169 
    170 inline struct pv_entry *pa_to_pv(paddr_t);
    171 static inline char *pa_to_attr(paddr_t);
    172 
    173 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
    174 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
    175 
    176 static void pmap_pinit(pmap_t);
    177 static void pmap_release(pmap_t);
    178 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
    179 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
    180 
    181 /*
    182  * These small routines may have to be replaced,
    183  * if/when we support processors other that the 604.
    184  */
    185 
    186 static inline void
    187 dcache_flush_page(vaddr_t va)
    188 {
    189 	int i;
    190 
    191 	for (i = 0; i < NBPG; i += CACHE_LINE)
    192 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    193 	asm volatile("sync;isync" : : );
    194 }
    195 
    196 static inline void
    197 icache_flush_page(vaddr_t va)
    198 {
    199 	int i;
    200 
    201 	for (i = 0; i < NBPG; i += CACHE_LINE)
    202 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    203 	asm volatile("sync;isync" : : );
    204 }
    205 
    206 static inline void
    207 dcache_flush(vaddr_t va, vsize_t len)
    208 {
    209 	int i;
    210 
    211 	if (len == 0)
    212 		return;
    213 
    214 	/* Make sure we flush all cache lines */
    215 	len += va & (CACHE_LINE-1);
    216 	for (i = 0; i < len; i += CACHE_LINE)
    217 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
    218 	asm volatile("sync;isync" : : );
    219 }
    220 
    221 static inline void
    222 icache_flush(vaddr_t va, vsize_t len)
    223 {
    224 	int i;
    225 
    226 	if (len == 0)
    227 		return;
    228 
    229 	/* Make sure we flush all cache lines */
    230 	len += va & (CACHE_LINE-1);
    231 	for (i = 0; i < len; i += CACHE_LINE)
    232 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
    233 	asm volatile("sync;isync" : : );
    234 }
    235 
    236 inline struct pv_entry *
    237 pa_to_pv(paddr_t pa)
    238 {
    239 	int bank, pg;
    240 
    241 	bank = vm_physseg_find(atop(pa), &pg);
    242 	if (bank == -1)
    243 		return NULL;
    244 	return &vm_physmem[bank].pmseg.pvent[pg];
    245 }
    246 
    247 static inline char *
    248 pa_to_attr(paddr_t pa)
    249 {
    250 	int bank, pg;
    251 
    252 	bank = vm_physseg_find(atop(pa), &pg);
    253 	if (bank == -1)
    254 		return NULL;
    255 	return &vm_physmem[bank].pmseg.attrs[pg];
    256 }
    257 
    258 /*
    259  * Insert PTE into page table.
    260  */
    261 int
    262 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
    263 {
    264 	int seg = STIDX(va);
    265 	int ptn = PTIDX(va);
    266 	paddr_t pa;
    267 
    268 	if (!pm->pm_ptbl[seg]) {
    269 		/* Don't allocate a page to clear a non-existent mapping. */
    270 		if (!pte) return (1);
    271 		/* Allocate a page XXXX this will sleep! */
    272 		pa = 0;
    273 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
    274 	}
    275 	pm->pm_ptbl[seg][ptn] = pte;
    276 
    277 	/* Flush entry. */
    278 	ppc4xx_tlb_flush(va, pm->pm_ctx);
    279 	return (1);
    280 }
    281 
    282 /*
    283  * Get a pointer to a PTE in a page table.
    284  */
    285 volatile u_int *
    286 pte_find(struct pmap *pm, vaddr_t va)
    287 {
    288 	int seg = STIDX(va);
    289 	int ptn = PTIDX(va);
    290 
    291 	if (pm->pm_ptbl[seg])
    292 		return (&pm->pm_ptbl[seg][ptn]);
    293 
    294 	return (NULL);
    295 }
    296 
    297 /*
    298  * This is called during initppc, before the system is really initialized.
    299  */
    300 void
    301 pmap_bootstrap(u_int kernelstart, u_int kernelend)
    302 {
    303 	struct mem_region *mp, *mp1;
    304 	int cnt, i;
    305 	u_int s, e, sz;
    306 
    307 	/*
    308 	 * Allocate the kernel page table at the end of
    309 	 * kernel space so it's in the locked TTE.
    310 	 */
    311 	kernmap = (caddr_t)kernelend;
    312 //	kernelend += KERNMAP_SIZE*sizeof(struct pte);
    313 
    314 	/*
    315 	 * Initialize kernel page table.
    316 	 */
    317 //	memset(kernmap, 0, KERNMAP_SIZE*sizeof(struct pte));
    318 	for (i = 0; i < STSZ; i++) {
    319 		pmap_kernel()->pm_ptbl[i] = 0; // (u_int *)(kernmap + i*NBPG);
    320 	}
    321 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
    322 
    323 	/*
    324 	 * Announce page-size to the VM-system
    325 	 */
    326 	uvmexp.pagesize = NBPG;
    327 	uvm_setpagesize();
    328 
    329 	/*
    330 	 * Get memory.
    331 	 */
    332 	mem_regions(&mem, &avail);
    333 	for (mp = mem; mp->size; mp++) {
    334 		physmem += btoc(mp->size);
    335 		printf("+%lx,",mp->size);
    336 	}
    337 	printf("\n");
    338 	ppc4xx_tlb_init();
    339 	/*
    340 	 * Count the number of available entries.
    341 	 */
    342 	for (cnt = 0, mp = avail; mp->size; mp++)
    343 		cnt++;
    344 
    345 	/*
    346 	 * Page align all regions.
    347 	 * Non-page aligned memory isn't very interesting to us.
    348 	 * Also, sort the entries for ascending addresses.
    349 	 */
    350 	kernelstart &= ~PGOFSET;
    351 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
    352 	for (mp = avail; mp->size; mp++) {
    353 		s = mp->start;
    354 		e = mp->start + mp->size;
    355 		printf("%08x-%08x -> ",s,e);
    356 		/*
    357 		 * Check whether this region holds all of the kernel.
    358 		 */
    359 		if (s < kernelstart && e > kernelend) {
    360 			avail[cnt].start = kernelend;
    361 			avail[cnt++].size = e - kernelend;
    362 			e = kernelstart;
    363 		}
    364 		/*
    365 		 * Look whether this regions starts within the kernel.
    366 		 */
    367 		if (s >= kernelstart && s < kernelend) {
    368 			if (e <= kernelend)
    369 				goto empty;
    370 			s = kernelend;
    371 		}
    372 		/*
    373 		 * Now look whether this region ends within the kernel.
    374 		 */
    375 		if (e > kernelstart && e <= kernelend) {
    376 			if (s >= kernelstart)
    377 				goto empty;
    378 			e = kernelstart;
    379 		}
    380 		/*
    381 		 * Now page align the start and size of the region.
    382 		 */
    383 		s = round_page(s);
    384 		e = trunc_page(e);
    385 		if (e < s)
    386 			e = s;
    387 		sz = e - s;
    388 		printf("%08x-%08x = %x\n",s,e,sz);
    389 		/*
    390 		 * Check whether some memory is left here.
    391 		 */
    392 		if (sz == 0) {
    393 		empty:
    394 			memmove(mp, mp + 1,
    395 				(cnt - (mp - avail)) * sizeof *mp);
    396 			cnt--;
    397 			mp--;
    398 			continue;
    399 		}
    400 		/*
    401 		 * Do an insertion sort.
    402 		 */
    403 		npgs += btoc(sz);
    404 		for (mp1 = avail; mp1 < mp; mp1++)
    405 			if (s < mp1->start)
    406 				break;
    407 		if (mp1 < mp) {
    408 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
    409 			mp1->start = s;
    410 			mp1->size = sz;
    411 		} else {
    412 			mp->start = s;
    413 			mp->size = sz;
    414 		}
    415 	}
    416 
    417 	/*
    418 	 * We cannot do pmap_steal_memory here,
    419 	 * since we don't run with translation enabled yet.
    420 	 */
    421 #ifndef MSGBUFADDR
    422 	/*
    423 	 * allow for msgbuf
    424 	 */
    425 	sz = round_page(MSGBUFSIZE);
    426 	mp = NULL;
    427 	for (mp1 = avail; mp1->size; mp1++)
    428 		if (mp1->size >= sz)
    429 			mp = mp1;
    430 	if (mp == NULL)
    431 		panic("not enough memory?");
    432 
    433 	npgs -= btoc(sz);
    434 	msgbuf_paddr = mp->start + mp->size - sz;
    435 	mp->size -= sz;
    436 	if (mp->size <= 0)
    437 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
    438 #endif
    439 
    440 	printf("Loading pages\n");
    441 	for (mp = avail; mp->size; mp++)
    442 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
    443 			atop(mp->start), atop(mp->start + mp->size),
    444 			VM_FREELIST_DEFAULT);
    445 
    446 	/*
    447 	 * Initialize kernel pmap and hardware.
    448 	 */
    449 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
    450 	pmap_kernel()->pm_ctx = KERNEL_PID;
    451 	nextavail = avail->start;
    452 
    453 
    454 	evcnt_attach_static(&tlbhit_ev);
    455 	evcnt_attach_static(&tlbmiss_ev);
    456 	evcnt_attach_static(&tlbflush_ev);
    457 	evcnt_attach_static(&tlbenter_ev);
    458 	printf("Done\n");
    459 }
    460 
    461 /*
    462  * Restrict given range to physical memory
    463  *
    464  * (Used by /dev/mem)
    465  */
    466 void
    467 pmap_real_memory(paddr_t *start, psize_t *size)
    468 {
    469 	struct mem_region *mp;
    470 
    471 	for (mp = mem; mp->size; mp++) {
    472 		if (*start + *size > mp->start &&
    473 		    *start < mp->start + mp->size) {
    474 			if (*start < mp->start) {
    475 				*size -= mp->start - *start;
    476 				*start = mp->start;
    477 			}
    478 			if (*start + *size > mp->start + mp->size)
    479 				*size = mp->start + mp->size - *start;
    480 			return;
    481 		}
    482 	}
    483 	*size = 0;
    484 }
    485 
    486 /*
    487  * Initialize anything else for pmap handling.
    488  * Called during vm_init().
    489  */
    490 void
    491 pmap_init(void)
    492 {
    493 	struct pv_entry *pv;
    494 	vsize_t sz;
    495 	vaddr_t addr;
    496 	int i, s;
    497 	int bank;
    498 	char *attr;
    499 
    500 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
    501 	sz = round_page(sz);
    502 	addr = uvm_km_zalloc(kernel_map, sz);
    503 	s = splvm();
    504 	pv = pv_table = (struct pv_entry *)addr;
    505 	for (i = npgs; --i >= 0;)
    506 		pv++->pv_pm = NULL;
    507 	pmap_attrib = (char *)pv;
    508 	memset(pv, 0, npgs);
    509 
    510 	pv = pv_table;
    511 	attr = pmap_attrib;
    512 	for (bank = 0; bank < vm_nphysseg; bank++) {
    513 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
    514 		vm_physmem[bank].pmseg.pvent = pv;
    515 		vm_physmem[bank].pmseg.attrs = attr;
    516 		pv += sz;
    517 		attr += sz;
    518 	}
    519 
    520 	pmap_initialized = 1;
    521 	splx(s);
    522 
    523 	/* Setup a pool for additional pvlist structures */
    524 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", 0,
    525 		  NULL, NULL, 0);
    526 }
    527 
    528 /*
    529  * How much virtual space is available to the kernel?
    530  */
    531 void
    532 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
    533 {
    534 
    535 #if 0
    536 	/*
    537 	 * Reserve one segment for kernel virtual memory
    538 	 */
    539 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
    540 	*end = *start + SEGMENT_LENGTH;
    541 #else
    542 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
    543 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
    544 #endif
    545 }
    546 
    547 #ifdef PMAP_GROWKERNEL
    548 /*
    549  * Preallocate kernel page tables to a specified VA.
    550  * This simply loops through the first TTE for each
    551  * page table from the beginning of the kernel pmap,
    552  * reads the entry, and if the result is
    553  * zero (either invalid entry or no page table) it stores
    554  * a zero there, populating page tables in the process.
    555  * This is not the most efficient technique but i don't
    556  * expect it to be called that often.
    557  */
    558 extern struct vm_page *vm_page_alloc1 __P((void));
    559 extern void vm_page_free1 __P((struct vm_page *));
    560 
    561 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
    562 
    563 vaddr_t
    564 pmap_growkernel(maxkvaddr)
    565         vaddr_t maxkvaddr;
    566 {
    567 	int s;
    568 	int seg;
    569 	paddr_t pg;
    570 	struct pmap *pm = pmap_kernel();
    571 
    572 	s = splvm();
    573 
    574 	/* Align with the start of a page table */
    575 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
    576 	     kbreak += PTMAP) {
    577 		seg = STIDX(kbreak);
    578 
    579 		if (pte_find(pm, kbreak)) continue;
    580 
    581 		if (uvm.page_init_done) {
    582 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
    583 		} else {
    584 			if (!uvm_page_physget(&pg))
    585 				panic("pmap_growkernel: no memory");
    586 		}
    587 		if (!pg) panic("pmap_growkernel: no pages");
    588 		pmap_zero_page((paddr_t)pg);
    589 
    590 		/* XXX This is based on all phymem being addressable */
    591 		pm->pm_ptbl[seg] = (u_int *)pg;
    592 	}
    593 	splx(s);
    594 	return (kbreak);
    595 }
    596 
    597 /*
    598  *	vm_page_alloc1:
    599  *
    600  *	Allocate and return a memory cell with no associated object.
    601  */
    602 struct vm_page *
    603 vm_page_alloc1()
    604 {
    605 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
    606 	if (pg) {
    607 		pg->wire_count = 1;	/* no mappings yet */
    608 		pg->flags &= ~PG_BUSY;	/* never busy */
    609 	}
    610 	return pg;
    611 }
    612 
    613 /*
    614  *	vm_page_free1:
    615  *
    616  *	Returns the given page to the free list,
    617  *	disassociating it with any VM object.
    618  *
    619  *	Object and page must be locked prior to entry.
    620  */
    621 void
    622 vm_page_free1(mem)
    623 	struct vm_page *mem;
    624 {
    625 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
    626 		printf("Freeing invalid page %p\n", mem);
    627 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
    628 		Debugger();
    629 		return;
    630 	}
    631 	mem->flags |= PG_BUSY;
    632 	mem->wire_count = 0;
    633 	uvm_pagefree(mem);
    634 }
    635 #endif
    636 
    637 /*
    638  * Create and return a physical map.
    639  */
    640 struct pmap *
    641 pmap_create(void)
    642 {
    643 	struct pmap *pm;
    644 
    645 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
    646 	memset((caddr_t)pm, 0, sizeof *pm);
    647 	pmap_pinit(pm);
    648 	return pm;
    649 }
    650 
    651 /*
    652  * Initialize a preallocated and zeroed pmap structure.
    653  */
    654 void
    655 pmap_pinit(struct pmap *pm)
    656 {
    657 	int i;
    658 
    659 	/*
    660 	 * Allocate some segment registers for this pmap.
    661 	 */
    662 	pm->pm_refs = 1;
    663 	for (i = 0; i < STSZ; i++)
    664 		pm->pm_ptbl[i] = NULL;
    665 }
    666 
    667 /*
    668  * Add a reference to the given pmap.
    669  */
    670 void
    671 pmap_reference(struct pmap *pm)
    672 {
    673 
    674 	pm->pm_refs++;
    675 }
    676 
    677 /*
    678  * Retire the given pmap from service.
    679  * Should only be called if the map contains no valid mappings.
    680  */
    681 void
    682 pmap_destroy(struct pmap *pm)
    683 {
    684 
    685 	if (--pm->pm_refs == 0) {
    686 		pmap_release(pm);
    687 		free((caddr_t)pm, M_VMPMAP);
    688 	}
    689 }
    690 
    691 /*
    692  * Release any resources held by the given physical map.
    693  * Called when a pmap initialized by pmap_pinit is being released.
    694  */
    695 static void
    696 pmap_release(struct pmap *pm)
    697 {
    698 	int i;
    699 
    700 	for (i = 0; i < STSZ; i++)
    701 		if (pm->pm_ptbl[i]) {
    702 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
    703 			pm->pm_ptbl[i] = NULL;
    704 		}
    705 	if (pm->pm_ctx) ctx_free(pm);
    706 }
    707 
    708 /*
    709  * Copy the range specified by src_addr/len
    710  * from the source map to the range dst_addr/len
    711  * in the destination map.
    712  *
    713  * This routine is only advisory and need not do anything.
    714  */
    715 void
    716 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
    717 	  vsize_t len, vaddr_t src_addr)
    718 {
    719 }
    720 
    721 /*
    722  * Require that all active physical maps contain no
    723  * incorrect entries NOW.
    724  */
    725 void
    726 pmap_update(struct pmap *pmap)
    727 {
    728 }
    729 
    730 /*
    731  * Garbage collects the physical map system for
    732  * pages which are no longer used.
    733  * Success need not be guaranteed -- that is, there
    734  * may well be pages which are not referenced, but
    735  * others may be collected.
    736  * Called by the pageout daemon when pages are scarce.
    737  */
    738 void
    739 pmap_collect(struct pmap *pm)
    740 {
    741 }
    742 
    743 /*
    744  * Fill the given physical page with zeroes.
    745  */
    746 void
    747 pmap_zero_page(paddr_t pa)
    748 {
    749 
    750 #ifdef NOCACHE
    751 	memset((caddr_t)pa, 0, NBPG);
    752 #else
    753 	int i;
    754 
    755 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
    756 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
    757 		pa += CACHELINESIZE;
    758 	}
    759 #endif
    760 }
    761 
    762 /*
    763  * Copy the given physical source page to its destination.
    764  */
    765 void
    766 pmap_copy_page(paddr_t src, paddr_t dst)
    767 {
    768 
    769 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
    770 	dcache_flush_page(dst);
    771 }
    772 
    773 /*
    774  * This returns whether this is the first mapping of a page.
    775  */
    776 static inline int
    777 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    778 {
    779 	struct pv_entry *pv, *npv = NULL;
    780 	int s;
    781 
    782 	if (!pmap_initialized)
    783 		return 0;
    784 
    785 	s = splvm();
    786 
    787 	pv = pa_to_pv(pa);
    788 	if (!pv->pv_pm) {
    789 		/*
    790 		 * No entries yet, use header as the first entry.
    791 		 */
    792 		pv->pv_va = va;
    793 		pv->pv_pm = pm;
    794 		pv->pv_next = NULL;
    795 	} else {
    796 		/*
    797 		 * There is at least one other VA mapping this page.
    798 		 * Place this entry after the header.
    799 		 */
    800 		npv = pool_get(&pv_pool, PR_WAITOK);
    801 		if (!npv) return (0);
    802 		npv->pv_va = va;
    803 		npv->pv_pm = pm;
    804 		npv->pv_next = pv->pv_next;
    805 		pv->pv_next = npv;
    806 	}
    807 	splx(s);
    808 	return (1);
    809 }
    810 
    811 static void
    812 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
    813 {
    814 	struct pv_entry *pv, *npv;
    815 
    816 	/*
    817 	 * Remove from the PV table.
    818 	 */
    819 	pv = pa_to_pv(pa);
    820 	if (!pv) return;
    821 
    822 	/*
    823 	 * If it is the first entry on the list, it is actually
    824 	 * in the header and we must copy the following entry up
    825 	 * to the header.  Otherwise we must search the list for
    826 	 * the entry.  In either case we free the now unused entry.
    827 	 */
    828 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
    829 		if ((npv = pv->pv_next)) {
    830 			*pv = *npv;
    831 			pool_put(&pv_pool, npv);
    832 		} else
    833 			pv->pv_pm = NULL;
    834 	} else {
    835 		for (; (npv = pv->pv_next) != NULL; pv = npv)
    836 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
    837 				break;
    838 		if (npv) {
    839 			pv->pv_next = npv->pv_next;
    840 			pool_put(&pv_pool, npv);
    841 		}
    842 	}
    843 }
    844 
    845 /*
    846  * Insert physical page at pa into the given pmap at virtual address va.
    847  */
    848 int
    849 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
    850 {
    851 	int s;
    852 	u_int tte;
    853 	int managed;
    854 
    855 	/*
    856 	 * Have to remove any existing mapping first.
    857 	 */
    858 	pmap_remove(pm, va, va + NBPG);
    859 
    860 	if (flags & PMAP_WIRED) flags |= prot;
    861 
    862 	/* If it has no protections don't bother w/the rest */
    863 	if (!(flags & VM_PROT_ALL))
    864 		return (0);
    865 
    866 	managed = 0;
    867 	if (vm_physseg_find(atop(pa), NULL) != -1)
    868 		managed = 1;
    869 
    870 	/*
    871 	 * Generate TTE.
    872 	 *
    873 	 * XXXX
    874 	 *
    875 	 * Since the kernel does not handle execution privileges properly,
    876 	 * we will handle read and execute permissions together.
    877 	 */
    878 	tte = TTE_PA(pa) | TTE_EX;
    879 	/* XXXX -- need to support multiple page sizes. */
    880 	tte |= TTE_SZ_16K;
    881 #ifdef	DIAGNOSTIC
    882 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
    883 		(PME_NOCACHE | PME_WRITETHROUG))
    884 		panic("pmap_enter: uncached & writethrough\n");
    885 #endif
    886 	if (flags & PME_NOCACHE)
    887 		/* Must be I/O mapping */
    888 		tte |= TTE_I | TTE_G;
    889 #ifdef NOCACHE
    890 	tte |= TTE_I;
    891 #else
    892 	else if (flags & PME_WRITETHROUG)
    893 		/* Uncached and writethrough are not compatible */
    894 		tte |= TTE_W;
    895 #endif
    896 	if (pm == pmap_kernel())
    897 		tte |= TTE_ZONE(ZONE_PRIV);
    898 	else
    899 		tte |= TTE_ZONE(ZONE_USER);
    900 
    901 	if (flags & VM_PROT_WRITE)
    902 		tte |= TTE_WR;
    903 
    904 	/*
    905 	 * Now record mapping for later back-translation.
    906 	 */
    907 	if (pmap_initialized && managed) {
    908 		char *attr;
    909 
    910 		if (!pmap_enter_pv(pm, va, pa)) {
    911 			/* Could not enter pv on a managed page */
    912 			return 1;
    913 		}
    914 
    915 		/* Now set attributes. */
    916 		attr = pa_to_attr(pa);
    917 #ifdef DIAGNOSTIC
    918 		if (!attr)
    919 			panic("managed but no attr\n");
    920 #endif
    921 		if (flags & VM_PROT_ALL)
    922 			*attr |= PTE_HI_REF;
    923 		if (flags & VM_PROT_WRITE)
    924 			*attr |= PTE_HI_CHG;
    925 	}
    926 
    927 	s = splvm();
    928 	pm->pm_stats.resident_count++;
    929 
    930 	/* Insert page into page table. */
    931 	pte_enter(pm, va, tte);
    932 
    933 	/* If this is a real fault, enter it in the tlb */
    934 	if (tte && ((flags & PMAP_WIRED) == 0)) {
    935 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
    936 	}
    937 	splx(s);
    938 
    939 	/* Flush the real memory from the instruction cache. */
    940 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
    941 		__syncicache((void *)pa, PAGE_SIZE);
    942 
    943 	return 0;
    944 }
    945 
    946 void
    947 pmap_unwire(struct pmap *pm, vaddr_t va)
    948 {
    949 	struct pv_entry *pv, *npv;
    950 	paddr_t pa;
    951 	int s = splvm();
    952 
    953         if (pm == NULL) {
    954                 return;
    955         }
    956 
    957 	if (!pmap_extract(pm, va, &pa)) {
    958 		return;
    959 	}
    960 
    961 	va |= PV_WIRED;
    962 
    963 	pv = pa_to_pv(pa);
    964 	if (!pv) return;
    965 
    966 	/*
    967 	 * If it is the first entry on the list, it is actually
    968 	 * in the header and we must copy the following entry up
    969 	 * to the header.  Otherwise we must search the list for
    970 	 * the entry.  In either case we free the now unused entry.
    971 	 */
    972 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
    973 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
    974 			npv->pv_va &= ~PV_WIRED;
    975 			break;
    976 		}
    977 	}
    978 	splx(s);
    979 }
    980 
    981 void
    982 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
    983 {
    984 	int s;
    985 	u_int tte;
    986 	struct pmap *pm = pmap_kernel();
    987 
    988 	/*
    989 	 * Have to remove any existing mapping first.
    990 	 */
    991 
    992 	/*
    993 	 * Generate TTE.
    994 	 *
    995 	 * XXXX
    996 	 *
    997 	 * Since the kernel does not handle execution privileges properly,
    998 	 * we will handle read and execute permissions together.
    999 	 */
   1000 	tte = 0;
   1001 	if (prot & VM_PROT_ALL) {
   1002 
   1003 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
   1004 		/* XXXX -- need to support multiple page sizes. */
   1005 		tte |= TTE_SZ_16K;
   1006 #ifdef DIAGNOSTIC
   1007 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
   1008 			(PME_NOCACHE | PME_WRITETHROUG))
   1009 			panic("pmap_kenter_pa: uncached & writethrough\n");
   1010 #endif
   1011 		if (prot & PME_NOCACHE)
   1012 			/* Must be I/O mapping */
   1013 			tte |= TTE_I | TTE_G;
   1014 #ifdef NOCACHE
   1015 		tte |= TTE_I;
   1016 #else
   1017 		else if (prot & PME_WRITETHROUG)
   1018 			/* Uncached and writethrough are not compatible */
   1019 			tte |= TTE_W;
   1020 #endif
   1021 		if (prot & VM_PROT_WRITE)
   1022 			tte |= TTE_WR;
   1023 	}
   1024 
   1025 	s = splvm();
   1026 	pm->pm_stats.resident_count++;
   1027 
   1028 	/* Insert page into page table. */
   1029 	pte_enter(pm, va, tte);
   1030 	splx(s);
   1031 }
   1032 
   1033 void
   1034 pmap_kremove(vaddr_t va, vsize_t len)
   1035 {
   1036 
   1037 	while (len > 0) {
   1038 		pte_enter(pmap_kernel(), va, 0);
   1039 		va += PAGE_SIZE;
   1040 		len -= PAGE_SIZE;
   1041 	}
   1042 }
   1043 
   1044 /*
   1045  * Remove the given range of mapping entries.
   1046  */
   1047 void
   1048 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
   1049 {
   1050 	int s;
   1051 	paddr_t pa;
   1052 	volatile u_int *ptp;
   1053 
   1054 	s = splvm();
   1055 	while (va < endva) {
   1056 
   1057 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
   1058 			pa = TTE_PA(pa);
   1059 			pmap_remove_pv(pm, va, pa);
   1060 			*ptp = 0;
   1061 			ppc4xx_tlb_flush(va, pm->pm_ctx);
   1062 			pm->pm_stats.resident_count--;
   1063 		}
   1064 		va += NBPG;
   1065 	}
   1066 
   1067 	splx(s);
   1068 }
   1069 
   1070 /*
   1071  * Get the physical page address for the given pmap/virtual address.
   1072  */
   1073 boolean_t
   1074 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
   1075 {
   1076 	int seg = STIDX(va);
   1077 	int ptn = PTIDX(va);
   1078 	u_int pa = 0;
   1079 	int s = splvm();
   1080 
   1081 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
   1082 		*pap = TTE_PA(pa) | (va & PGOFSET);
   1083 	}
   1084 	splx(s);
   1085 	return (pa != 0);
   1086 }
   1087 
   1088 /*
   1089  * Lower the protection on the specified range of this pmap.
   1090  *
   1091  * There are only two cases: either the protection is going to 0,
   1092  * or it is going to read-only.
   1093  */
   1094 void
   1095 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
   1096 {
   1097 	volatile u_int *ptp;
   1098 	int s;
   1099 
   1100 	if (prot & VM_PROT_READ) {
   1101 		s = splvm();
   1102 		while (sva < eva) {
   1103 			if ((ptp = pte_find(pm, sva)) != NULL) {
   1104 				*ptp &= ~TTE_WR;
   1105 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
   1106 			}
   1107 			sva += NBPG;
   1108 		}
   1109 		splx(s);
   1110 		return;
   1111 	}
   1112 	pmap_remove(pm, sva, eva);
   1113 }
   1114 
   1115 boolean_t
   1116 check_attr(struct vm_page *pg, u_int mask, int clear)
   1117 {
   1118 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1119 	int s;
   1120 	char *attr;
   1121 	int rv;
   1122 
   1123 	/*
   1124 	 * First modify bits in cache.
   1125 	 */
   1126 	s = splvm();
   1127 	attr = pa_to_attr(pa);
   1128 	if (attr == NULL)
   1129 		return FALSE;
   1130 
   1131 	rv = ((*attr & mask) != 0);
   1132 	if (clear)
   1133 		*attr &= ~mask;
   1134 
   1135 	splx(s);
   1136 	return rv;
   1137 }
   1138 
   1139 
   1140 /*
   1141  * Lower the protection on the specified physical page.
   1142  *
   1143  * There are only two cases: either the protection is going to 0,
   1144  * or it is going to read-only.
   1145  */
   1146 void
   1147 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
   1148 {
   1149 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
   1150 	vaddr_t va;
   1151 	struct pv_entry *pvh, *pv, *npv;
   1152 	struct pmap *pm;
   1153 
   1154 	pvh = pa_to_pv(pa);
   1155 	if (pvh == NULL)
   1156 		return;
   1157 
   1158 	/* Handle extra pvs which may be deleted in the operation */
   1159 	for (pv = pvh->pv_next; pv; pv = npv) {
   1160 		npv = pv->pv_next;
   1161 
   1162 		pm = pv->pv_pm;
   1163 		va = pv->pv_va;
   1164 		pmap_protect(pm, va, va+NBPG, prot);
   1165 	}
   1166 	/* Now check the head pv */
   1167 	if (pvh->pv_pm) {
   1168 		pv = pvh;
   1169 		pm = pv->pv_pm;
   1170 		va = pv->pv_va;
   1171 		pmap_protect(pm, va, va+NBPG, prot);
   1172 	}
   1173 }
   1174 
   1175 /*
   1176  * Activate the address space for the specified process.  If the process
   1177  * is the current process, load the new MMU context.
   1178  */
   1179 void
   1180 pmap_activate(struct proc *p)
   1181 {
   1182 #if 0
   1183 	struct pcb *pcb = &p->p_addr->u_pcb;
   1184 	pmap_t pmap = p->p_vmspace->vm_map.pmap;
   1185 
   1186 	/*
   1187 	 * XXX Normally performed in cpu_fork().
   1188 	 */
   1189 	printf("pmap_activate(%p), pmap=%p\n",p,pmap);
   1190 	if (pcb->pcb_pm != pmap) {
   1191 		pcb->pcb_pm = pmap;
   1192 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
   1193 		    (paddr_t *)&pcb->pcb_pmreal);
   1194 	}
   1195 
   1196 	if (p == curproc) {
   1197 		/* Store pointer to new current pmap. */
   1198 		curpm = pcb->pcb_pmreal;
   1199 	}
   1200 #endif
   1201 }
   1202 
   1203 /*
   1204  * Deactivate the specified process's address space.
   1205  */
   1206 void
   1207 pmap_deactivate(struct proc *p)
   1208 {
   1209 }
   1210 
   1211 /*
   1212  * Synchronize caches corresponding to [addr, addr+len) in p.
   1213  */
   1214 void
   1215 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
   1216 {
   1217 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
   1218 	int msr, ctx, opid;
   1219 
   1220 
   1221 	/*
   1222 	 * Need to turn off IMMU and switch to user context.
   1223 	 * (icbi uses DMMU).
   1224 	 */
   1225 	if (!(ctx = pm->pm_ctx)) {
   1226 		/* No context -- assign it one */
   1227 		ctx_alloc(pm);
   1228 		ctx = pm->pm_ctx;
   1229 	}
   1230 	__asm __volatile("mfmsr %0;"
   1231 		"li %1, 0x20;"
   1232 		"andc %1,%0,%1;"
   1233 		"mtmsr %1;"
   1234 		"sync;isync;"
   1235 		"mfpid %1;"
   1236 		"mtpid %2;"
   1237 		"sync; isync;"
   1238 		"1:"
   1239 		"dcbf 0,%3;"
   1240 		"icbi 0,%3;"
   1241 		"addi %3,%3,32;"
   1242 		"addic. %4,%4,-32;"
   1243 		"bge 1b;"
   1244 		"mtpid %1;"
   1245 		"mtmsr %0;"
   1246 		"sync; isync"
   1247 		: "=&r" (msr), "=&r" (opid)
   1248 		: "r" (ctx), "r" (va), "r" (len));
   1249 }
   1250 
   1251 
   1252 /* This has to be done in real mode !!! */
   1253 void
   1254 ppc4xx_tlb_flush(vaddr_t va, int pid)
   1255 {
   1256 	u_long i, found;
   1257 	u_long msr;
   1258 
   1259 	/* If there's no context then it can't be mapped. */
   1260 	if (!pid) return;
   1261 
   1262 	asm("mfpid %1;"			/* Save PID */
   1263 		"mfmsr %2;"		/* Save MSR */
   1264 		"li %0,0;"		/* Now clear MSR */
   1265 		"mtmsr %0;"
   1266 		"mtpid %4;"		/* Set PID */
   1267 		"sync;"
   1268 		"tlbsx. %0,0,%3;"	/* Search TLB */
   1269 		"sync;"
   1270 		"mtpid %1;"		/* Restore PID */
   1271 		"mtmsr %2;"		/* Restore MSR */
   1272 		"sync;isync;"
   1273 		"li %1,1;"
   1274 		"beq 1f;"
   1275 		"li %1,0;"
   1276 		"1:"
   1277 		: "=&r" (i), "=&r" (found), "=&r" (msr)
   1278 		: "r" (va), "r" (pid));
   1279 	if (found && !TLB_LOCKED(i)) {
   1280 
   1281 		/* Now flush translation */
   1282 		asm volatile(
   1283 			"tlbwe %0,%1,0;"
   1284 			"sync;isync;"
   1285 			: : "r" (0), "r" (i));
   1286 
   1287 		tlb_info[i].ti_ctx = 0;
   1288 		tlb_info[i].ti_flags = 0;
   1289 		tlbnext = i;
   1290 		/* Successful flushes */
   1291 		tlbflush_ev.ev_count++;
   1292 	}
   1293 }
   1294 
   1295 void
   1296 ppc4xx_tlb_flush_all(void)
   1297 {
   1298 	u_long i;
   1299 
   1300 	for (i = 0; i < NTLB; i++)
   1301 		if (!TLB_LOCKED(i)) {
   1302 			asm volatile(
   1303 				"tlbwe %0,%1,0;"
   1304 				"sync;isync;"
   1305 				: : "r" (0), "r" (i));
   1306 			tlb_info[i].ti_ctx = 0;
   1307 			tlb_info[i].ti_flags = 0;
   1308 		}
   1309 
   1310 	asm volatile("sync;isync");
   1311 }
   1312 
   1313 /* Find a TLB entry to evict. */
   1314 static int
   1315 ppc4xx_tlb_find_victim(void)
   1316 {
   1317 	int flags;
   1318 
   1319 	for (;;) {
   1320 		if (++tlbnext >= NTLB)
   1321 			tlbnext = TLB_NRESERVED;
   1322 		flags = tlb_info[tlbnext].ti_flags;
   1323 		if (!(flags & TLBF_USED) ||
   1324 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
   1325 			u_long va, stack = (u_long)&va;
   1326 
   1327 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
   1328 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
   1329 			     (flags & TLBF_USED)) {
   1330 				/* Kernel stack page */
   1331 				flags |= TLBF_USED;
   1332 				tlb_info[tlbnext].ti_flags = flags;
   1333 			} else {
   1334 				/* Found it! */
   1335 				return (tlbnext);
   1336 			}
   1337 		} else {
   1338 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
   1339 		}
   1340 	}
   1341 }
   1342 
   1343 void
   1344 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
   1345 {
   1346 	u_long th, tl, idx;
   1347 	tlbpid_t pid;
   1348 	u_short msr;
   1349 	int s;
   1350 
   1351 	tlbenter_ev.ev_count++;
   1352 
   1353 	th = (va & TLB_EPN_MASK) |
   1354 		(((pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT) << TLB_SIZE_SHFT) |
   1355 		TLB_VALID;
   1356 	tl = pte;
   1357 
   1358 	s = splhigh();
   1359 	idx = ppc4xx_tlb_find_victim();
   1360 
   1361 #ifdef DIAGNOSTIC
   1362 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
   1363 		panic("ppc4xx_tlb_enter: repacing entry %ld\n", idx);
   1364 	}
   1365 #endif
   1366 
   1367 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
   1368 	tlb_info[idx].ti_ctx = ctx;
   1369 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
   1370 
   1371 	asm volatile(
   1372 		"mfmsr %0;"			/* Save MSR */
   1373 		"li %1,0;"
   1374 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
   1375 		"mtmsr %1;"			/* Clear MSR */
   1376 		"mfpid %1;"			/* Save old PID */
   1377 		"mtpid %2;"			/* Load translation ctx */
   1378 		"sync; isync;"
   1379 #ifdef DEBUG
   1380 		"andi. %3,%3,63;"
   1381 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
   1382 #endif
   1383 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
   1384 		"sync; isync;"
   1385 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
   1386 		"sync; isync;"
   1387 	: "=&r" (msr), "=&r" (pid)
   1388 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
   1389 	splx(s);
   1390 }
   1391 
   1392 void
   1393 ppc4xx_tlb_unpin(int i)
   1394 {
   1395 
   1396 	if (i == -1)
   1397 		for (i = 0; i < TLB_NRESERVED; i++)
   1398 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1399 	else
   1400 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
   1401 }
   1402 
   1403 void
   1404 ppc4xx_tlb_init(void)
   1405 {
   1406 	int i;
   1407 
   1408 	/* Mark reserved TLB entries */
   1409 	for (i = 0; i < TLB_NRESERVED; i++) {
   1410 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
   1411 		tlb_info[i].ti_ctx = KERNEL_PID;
   1412 	}
   1413 
   1414 	/* Setup security zones */
   1415 	/* Z0 - accessible by kernel only if TLB entry permissions allow
   1416 	 * Z1,Z2 - access is controlled by TLB entry permissions
   1417 	 * Z3 - full access regardless of TLB entry permissions
   1418 	 */
   1419 
   1420 	asm volatile(
   1421 		"mtspr %0,%1;"
   1422 		"sync;"
   1423 		::  "K"(SPR_ZPR), "r" (0x1b000000));
   1424 }
   1425 
   1426 
   1427 /*
   1428  * We should pass the ctx in from trap code.
   1429  */
   1430 int
   1431 pmap_tlbmiss(vaddr_t va, int ctx)
   1432 {
   1433 	volatile u_int *pte;
   1434 	u_long tte;
   1435 
   1436 	tlbmiss_ev.ev_count++;
   1437 
   1438 	/*
   1439 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
   1440 	 */
   1441 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
   1442 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
   1443 		if (pte == NULL) {
   1444 			/* Map unmanaged addresses directly for kernel access */
   1445 			return 1;
   1446 		}
   1447 		tte = *pte;
   1448 		if (tte == 0) {
   1449 			return 1;
   1450 		}
   1451 	} else {
   1452 		/* Create a 16MB writeable mapping. */
   1453 #ifdef NOCACHE
   1454 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
   1455 #else
   1456 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
   1457 #endif
   1458 	}
   1459 	tlbhit_ev.ev_count++;
   1460 	ppc4xx_tlb_enter(ctx, va, tte);
   1461 
   1462 	return 0;
   1463 }
   1464 
   1465 /*
   1466  * Flush all the entries matching a context from the TLB.
   1467  */
   1468 static int
   1469 ctx_flush(int cnum)
   1470 {
   1471 	int i;
   1472 
   1473 	/* We gotta steal this context */
   1474 	for (i = TLB_NRESERVED; i < NTLB; i++) {
   1475 		if (tlb_info[i].ti_ctx == cnum) {
   1476 			/* Can't steal ctx if it has a locked entry. */
   1477 			if (TLB_LOCKED(i)) {
   1478 #ifdef DIAGNOSTIC
   1479 				printf("ctx_flush: can't invalidate "
   1480 					"locked mapping %d "
   1481 					"for context %d\n", i, cnum);
   1482 				Debugger();
   1483 #endif
   1484 				return (1);
   1485 			}
   1486 #ifdef DIAGNOSTIC
   1487 			if (i < TLB_NRESERVED)
   1488 				panic("TLB entry %d not locked\n", i);
   1489 #endif
   1490 			/* Invalidate particular TLB entry regardless of locked status */
   1491 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
   1492 			tlb_info[i].ti_flags = 0;
   1493 		}
   1494 	}
   1495 	return (0);
   1496 }
   1497 
   1498 /*
   1499  * Allocate a context.  If necessary, steal one from someone else.
   1500  *
   1501  * The new context is flushed from the TLB before returning.
   1502  */
   1503 int
   1504 ctx_alloc(struct pmap *pm)
   1505 {
   1506 	int s, cnum;
   1507 	static int next = MINCTX;
   1508 
   1509 	if (pm == pmap_kernel()) {
   1510 #ifdef DIAGNOSTIC
   1511 		printf("ctx_alloc: kernel pmap!\n");
   1512 #endif
   1513 		return (0);
   1514 	}
   1515 	s = splvm();
   1516 
   1517 	/* Find a likely context. */
   1518 	cnum = next;
   1519 	do {
   1520 		if ((++cnum) > NUMCTX)
   1521 			cnum = MINCTX;
   1522 	} while (ctxbusy[cnum] != NULL && cnum != next);
   1523 
   1524 	/* Now clean it out */
   1525 oops:
   1526 	if (cnum < MINCTX)
   1527 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
   1528 	if (ctx_flush(cnum)) {
   1529 		/* oops -- something's wired. */
   1530 		if ((++cnum) > NUMCTX)
   1531 			cnum = MINCTX;
   1532 		goto oops;
   1533 	}
   1534 
   1535 	if (ctxbusy[cnum]) {
   1536 #ifdef DEBUG
   1537 		/* We should identify this pmap and clear it */
   1538 		printf("Warning: stealing context %d\n", cnum);
   1539 #endif
   1540 		ctxbusy[cnum]->pm_ctx = 0;
   1541 	}
   1542 	ctxbusy[cnum] = pm;
   1543 	next = cnum;
   1544 	splx(s);
   1545 	pm->pm_ctx = cnum;
   1546 
   1547 	return cnum;
   1548 }
   1549 
   1550 /*
   1551  * Give away a context.
   1552  */
   1553 void
   1554 ctx_free(struct pmap *pm)
   1555 {
   1556 	int oldctx;
   1557 
   1558 	oldctx = pm->pm_ctx;
   1559 
   1560 	if (oldctx == 0)
   1561 		panic("ctx_free: freeing kernel context");
   1562 #ifdef DIAGNOSTIC
   1563 	if (ctxbusy[oldctx] == 0)
   1564 		printf("ctx_free: freeing free context %d\n", oldctx);
   1565 	if (ctxbusy[oldctx] != pm) {
   1566 		printf("ctx_free: freeing someone esle's context\n "
   1567 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
   1568 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
   1569 		Debugger();
   1570 	}
   1571 #endif
   1572 	/* We should verify it has not been stolen and reallocated... */
   1573 	ctxbusy[oldctx] = NULL;
   1574 	ctx_flush(oldctx);
   1575 }
   1576 
   1577 
   1578 #ifdef DEBUG
   1579 /*
   1580  * Test ref/modify handling.
   1581  */
   1582 void pmap_testout __P((void));
   1583 void
   1584 pmap_testout()
   1585 {
   1586 	vaddr_t va;
   1587 	volatile int *loc;
   1588 	int val = 0;
   1589 	paddr_t pa;
   1590 	struct vm_page *pg;
   1591 	int ref, mod;
   1592 
   1593 	/* Allocate a page */
   1594 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
   1595 	loc = (int*)va;
   1596 
   1597 	pmap_extract(pmap_kernel(), va, &pa);
   1598 	pg = PHYS_TO_VM_PAGE(pa);
   1599 	pmap_unwire(pmap_kernel(), va);
   1600 
   1601 	pmap_remove(pmap_kernel(), va, va+1);
   1602 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1603 	pmap_update(pmap_kernel());
   1604 
   1605 	/* Now clear reference and modify */
   1606 	ref = pmap_clear_reference(pg);
   1607 	mod = pmap_clear_modify(pg);
   1608 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1609 	       (void *)(u_long)va, (long)pa,
   1610 	       ref, mod);
   1611 
   1612 	/* Check it's properly cleared */
   1613 	ref = pmap_is_referenced(pg);
   1614 	mod = pmap_is_modified(pg);
   1615 	printf("Checking cleared page: ref %d, mod %d\n",
   1616 	       ref, mod);
   1617 
   1618 	/* Reference page */
   1619 	val = *loc;
   1620 
   1621 	ref = pmap_is_referenced(pg);
   1622 	mod = pmap_is_modified(pg);
   1623 	printf("Referenced page: ref %d, mod %d val %x\n",
   1624 	       ref, mod, val);
   1625 
   1626 	/* Now clear reference and modify */
   1627 	ref = pmap_clear_reference(pg);
   1628 	mod = pmap_clear_modify(pg);
   1629 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1630 	       (void *)(u_long)va, (long)pa,
   1631 	       ref, mod);
   1632 
   1633 	/* Modify page */
   1634 	*loc = 1;
   1635 
   1636 	ref = pmap_is_referenced(pg);
   1637 	mod = pmap_is_modified(pg);
   1638 	printf("Modified page: ref %d, mod %d\n",
   1639 	       ref, mod);
   1640 
   1641 	/* Now clear reference and modify */
   1642 	ref = pmap_clear_reference(pg);
   1643 	mod = pmap_clear_modify(pg);
   1644 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1645 	       (void *)(u_long)va, (long)pa,
   1646 	       ref, mod);
   1647 
   1648 	/* Check it's properly cleared */
   1649 	ref = pmap_is_referenced(pg);
   1650 	mod = pmap_is_modified(pg);
   1651 	printf("Checking cleared page: ref %d, mod %d\n",
   1652 	       ref, mod);
   1653 
   1654 	/* Modify page */
   1655 	*loc = 1;
   1656 
   1657 	ref = pmap_is_referenced(pg);
   1658 	mod = pmap_is_modified(pg);
   1659 	printf("Modified page: ref %d, mod %d\n",
   1660 	       ref, mod);
   1661 
   1662 	/* Check pmap_protect() */
   1663 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
   1664 	pmap_update(pmap_kernel());
   1665 	ref = pmap_is_referenced(pg);
   1666 	mod = pmap_is_modified(pg);
   1667 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
   1668 	       ref, mod);
   1669 
   1670 	/* Now clear reference and modify */
   1671 	ref = pmap_clear_reference(pg);
   1672 	mod = pmap_clear_modify(pg);
   1673 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1674 	       (void *)(u_long)va, (long)pa,
   1675 	       ref, mod);
   1676 
   1677 	/* Reference page */
   1678 	val = *loc;
   1679 
   1680 	ref = pmap_is_referenced(pg);
   1681 	mod = pmap_is_modified(pg);
   1682 	printf("Referenced page: ref %d, mod %d val %x\n",
   1683 	       ref, mod, val);
   1684 
   1685 	/* Now clear reference and modify */
   1686 	ref = pmap_clear_reference(pg);
   1687 	mod = pmap_clear_modify(pg);
   1688 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1689 	       (void *)(u_long)va, (long)pa,
   1690 	       ref, mod);
   1691 
   1692 	/* Modify page */
   1693 #if 0
   1694 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1695 	pmap_update(pmap_kernel());
   1696 #endif
   1697 	*loc = 1;
   1698 
   1699 	ref = pmap_is_referenced(pg);
   1700 	mod = pmap_is_modified(pg);
   1701 	printf("Modified page: ref %d, mod %d\n",
   1702 	       ref, mod);
   1703 
   1704 	/* Check pmap_protect() */
   1705 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
   1706 	pmap_update(pmap_kernel());
   1707 	ref = pmap_is_referenced(pg);
   1708 	mod = pmap_is_modified(pg);
   1709 	printf("pmap_protect(): ref %d, mod %d\n",
   1710 	       ref, mod);
   1711 
   1712 	/* Now clear reference and modify */
   1713 	ref = pmap_clear_reference(pg);
   1714 	mod = pmap_clear_modify(pg);
   1715 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1716 	       (void *)(u_long)va, (long)pa,
   1717 	       ref, mod);
   1718 
   1719 	/* Reference page */
   1720 	val = *loc;
   1721 
   1722 	ref = pmap_is_referenced(pg);
   1723 	mod = pmap_is_modified(pg);
   1724 	printf("Referenced page: ref %d, mod %d val %x\n",
   1725 	       ref, mod, val);
   1726 
   1727 	/* Now clear reference and modify */
   1728 	ref = pmap_clear_reference(pg);
   1729 	mod = pmap_clear_modify(pg);
   1730 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1731 	       (void *)(u_long)va, (long)pa,
   1732 	       ref, mod);
   1733 
   1734 	/* Modify page */
   1735 #if 0
   1736 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1737 	pmap_update(pmap_kernel());
   1738 #endif
   1739 	*loc = 1;
   1740 
   1741 	ref = pmap_is_referenced(pg);
   1742 	mod = pmap_is_modified(pg);
   1743 	printf("Modified page: ref %d, mod %d\n",
   1744 	       ref, mod);
   1745 
   1746 	/* Check pmap_pag_protect() */
   1747 	pmap_page_protect(pg, VM_PROT_READ);
   1748 	ref = pmap_is_referenced(pg);
   1749 	mod = pmap_is_modified(pg);
   1750 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
   1751 	       ref, mod);
   1752 
   1753 	/* Now clear reference and modify */
   1754 	ref = pmap_clear_reference(pg);
   1755 	mod = pmap_clear_modify(pg);
   1756 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1757 	       (void *)(u_long)va, (long)pa,
   1758 	       ref, mod);
   1759 
   1760 	/* Reference page */
   1761 	val = *loc;
   1762 
   1763 	ref = pmap_is_referenced(pg);
   1764 	mod = pmap_is_modified(pg);
   1765 	printf("Referenced page: ref %d, mod %d val %x\n",
   1766 	       ref, mod, val);
   1767 
   1768 	/* Now clear reference and modify */
   1769 	ref = pmap_clear_reference(pg);
   1770 	mod = pmap_clear_modify(pg);
   1771 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1772 	       (void *)(u_long)va, (long)pa,
   1773 	       ref, mod);
   1774 
   1775 	/* Modify page */
   1776 #if 0
   1777 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1778 	pmap_update(pmap_kernel());
   1779 #endif
   1780 	*loc = 1;
   1781 
   1782 	ref = pmap_is_referenced(pg);
   1783 	mod = pmap_is_modified(pg);
   1784 	printf("Modified page: ref %d, mod %d\n",
   1785 	       ref, mod);
   1786 
   1787 	/* Check pmap_pag_protect() */
   1788 	pmap_page_protect(pg, VM_PROT_NONE);
   1789 	ref = pmap_is_referenced(pg);
   1790 	mod = pmap_is_modified(pg);
   1791 	printf("pmap_page_protect(): ref %d, mod %d\n",
   1792 	       ref, mod);
   1793 
   1794 	/* Now clear reference and modify */
   1795 	ref = pmap_clear_reference(pg);
   1796 	mod = pmap_clear_modify(pg);
   1797 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1798 	       (void *)(u_long)va, (long)pa,
   1799 	       ref, mod);
   1800 
   1801 
   1802 	/* Reference page */
   1803 	val = *loc;
   1804 
   1805 	ref = pmap_is_referenced(pg);
   1806 	mod = pmap_is_modified(pg);
   1807 	printf("Referenced page: ref %d, mod %d val %x\n",
   1808 	       ref, mod, val);
   1809 
   1810 	/* Now clear reference and modify */
   1811 	ref = pmap_clear_reference(pg);
   1812 	mod = pmap_clear_modify(pg);
   1813 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1814 	       (void *)(u_long)va, (long)pa,
   1815 	       ref, mod);
   1816 
   1817 	/* Modify page */
   1818 #if 0
   1819 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
   1820 	pmap_update(pmap_kernel());
   1821 #endif
   1822 	*loc = 1;
   1823 
   1824 	ref = pmap_is_referenced(pg);
   1825 	mod = pmap_is_modified(pg);
   1826 	printf("Modified page: ref %d, mod %d\n",
   1827 	       ref, mod);
   1828 
   1829 	/* Unmap page */
   1830 	pmap_remove(pmap_kernel(), va, va+1);
   1831 	pmap_update(pmap_kernel());
   1832 	ref = pmap_is_referenced(pg);
   1833 	mod = pmap_is_modified(pg);
   1834 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
   1835 
   1836 	/* Now clear reference and modify */
   1837 	ref = pmap_clear_reference(pg);
   1838 	mod = pmap_clear_modify(pg);
   1839 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
   1840 	       (void *)(u_long)va, (long)pa, ref, mod);
   1841 
   1842 	/* Check it's properly cleared */
   1843 	ref = pmap_is_referenced(pg);
   1844 	mod = pmap_is_modified(pg);
   1845 	printf("Checking cleared page: ref %d, mod %d\n",
   1846 	       ref, mod);
   1847 
   1848 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
   1849 		VM_PROT_ALL|PMAP_WIRED);
   1850 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
   1851 }
   1852 #endif
   1853