xref: /freebsd/sys/powerpc/booke/pmap.c (revision f7c4bd95ba735bd6a5454b4953945a99cefbb80c)
1 /*-
2  * Copyright (C) 2007 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote products
15  *    derived from this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * Virtual address space layout:
40   * -----------------------------
41   * 0x0000_0000 - 0xbfff_efff	: user process
42   * 0xc000_0000 - 0xc1ff_ffff	: kernel reserved
43   *   0xc000_0000 - kernelend	: kernel code &data
44   *   0xc1ff_c000 - 0xc200_0000	: kstack0
45   * 0xc200_0000 - 0xffef_ffff	: KVA
46   *   0xc200_0000 - 0xc200_3fff : reserved for page zero/copy
47   *   0xc200_4000 - ptbl buf end: reserved for ptbl bufs
48   *   ptbl buf end- 0xffef_ffff	: actual free KVA space
49   * 0xfff0_0000 - 0xffff_ffff	: I/O devices region
50   */
51 
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
54 
55 #include <sys/types.h>
56 #include <sys/param.h>
57 #include <sys/malloc.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/vmmeter.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_pageout.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_param.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_pager.h>
77 #include <vm/uma.h>
78 
79 #include <machine/cpu.h>
80 #include <machine/pcb.h>
81 #include <machine/powerpc.h>
82 
83 #include <machine/tlb.h>
84 #include <machine/spr.h>
85 #include <machine/vmparam.h>
86 #include <machine/md_var.h>
87 #include <machine/mmuvar.h>
88 #include <machine/pmap.h>
89 #include <machine/pte.h>
90 
91 #include "mmu_if.h"
92 
93 #define DEBUG
94 #undef DEBUG
95 
96 #ifdef  DEBUG
97 #define debugf(fmt, args...) printf(fmt, ##args)
98 #else
99 #define debugf(fmt, args...)
100 #endif
101 
102 #define TODO			panic("%s: not implemented", __func__);
103 #define memmove(d, s, l)	bcopy(s, d, l)
104 
105 #include "opt_sched.h"
106 #ifndef SCHED_4BSD
107 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
108 #endif
109 extern struct mtx sched_lock;
110 
111 /* Kernel physical load address. */
112 extern uint32_t kernload;
113 
114 struct mem_region availmem_regions[MEM_REGIONS];
115 int availmem_regions_sz;
116 
117 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
118 static vm_offset_t zero_page_va;
119 static struct mtx zero_page_mutex;
120 
121 /*
122  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
123  * by idle thred only, no lock required.
124  */
125 static vm_offset_t zero_page_idle_va;
126 
127 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
128 static vm_offset_t copy_page_src_va;
129 static vm_offset_t copy_page_dst_va;
130 static struct mtx copy_page_mutex;
131 
132 /**************************************************************************/
133 /* PMAP */
134 /**************************************************************************/
135 
136 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
137     vm_prot_t, boolean_t);
138 
139 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
140 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
141 
142 static int pagedaemon_waken;
143 
144 /*
145  * If user pmap is processed with mmu_booke_remove and the resident count
146  * drops to 0, there are no more pages to remove, so we need not continue.
147  */
148 #define PMAP_REMOVE_DONE(pmap) \
149 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
150 
151 extern void load_pid0(tlbtid_t);
152 
153 /**************************************************************************/
154 /* TLB and TID handling */
155 /**************************************************************************/
156 
157 /* Translation ID busy table */
158 static volatile pmap_t tidbusy[TID_MAX + 1];
159 
160 /*
161  * Actual maximum number of TLB0 entries.
162  * This number differs between e500 core revisions.
163  */
164 u_int32_t tlb0_size;
165 u_int32_t tlb0_nways;
166 u_int32_t tlb0_nentries_per_way;
167 
168 #define TLB0_SIZE		(tlb0_size)
169 #define TLB0_NWAYS		(tlb0_nways)
170 #define TLB0_ENTRIES_PER_WAY	(tlb0_nentries_per_way)
171 
172 /* Pointer to kernel tlb0 table, allocated in mmu_booke_bootstrap() */
173 tlb_entry_t *tlb0;
174 
175 /*
176  * Spinlock to assure proper locking between threads and
177  * between tlb miss handler and kernel.
178  */
179 static struct mtx tlb0_mutex;
180 
181 #define TLB1_SIZE 16
182 
183 /* In-ram copy of the TLB1 */
184 static tlb_entry_t tlb1[TLB1_SIZE];
185 
186 /* Next free entry in the TLB1 */
187 static unsigned int tlb1_idx;
188 
189 static tlbtid_t tid_alloc(struct pmap *);
190 static void tid_flush(tlbtid_t);
191 
192 extern void tlb1_inval_va(vm_offset_t);
193 extern void tlb0_inval_va(vm_offset_t);
194 
195 static void tlb_print_entry(int, u_int32_t, u_int32_t, u_int32_t, u_int32_t);
196 
197 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, u_int32_t);
198 static void __tlb1_set_entry(unsigned int, vm_offset_t, vm_offset_t,
199     vm_size_t, u_int32_t, unsigned int, unsigned int);
200 static void tlb1_write_entry(unsigned int);
201 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
202 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_offset_t, vm_size_t);
203 
204 static vm_size_t tsize2size(unsigned int);
205 static unsigned int size2tsize(vm_size_t);
206 static unsigned int ilog2(unsigned int);
207 
208 static void set_mas4_defaults(void);
209 
210 static void tlb0_inval_entry(vm_offset_t, unsigned int);
211 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
212 static void tlb0_write_entry(unsigned int, unsigned int);
213 static void tlb0_flush_entry(pmap_t, vm_offset_t);
214 static void tlb0_init(void);
215 
216 /**************************************************************************/
217 /* Page table management */
218 /**************************************************************************/
219 
220 /* Data for the pv entry allocation mechanism */
221 static uma_zone_t pvzone;
222 static struct vm_object pvzone_obj;
223 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
224 
225 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
226 
227 #ifndef PMAP_SHPGPERPROC
228 #define PMAP_SHPGPERPROC	200
229 #endif
230 
231 static void ptbl_init(void);
232 static struct ptbl_buf *ptbl_buf_alloc(void);
233 static void ptbl_buf_free(struct ptbl_buf *);
234 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
235 
236 static void ptbl_alloc(mmu_t, pmap_t, unsigned int);
237 static void ptbl_free(mmu_t, pmap_t, unsigned int);
238 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
239 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
240 
241 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
242 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
243 void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, u_int32_t);
244 static int pte_remove(mmu_t, pmap_t, vm_offset_t, u_int8_t);
245 
246 pv_entry_t pv_alloc(void);
247 static void pv_free(pv_entry_t);
248 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
249 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
250 
251 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
252 #define PTBL_BUFS		(128 * 16)
253 
254 struct ptbl_buf {
255 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
256 	vm_offset_t kva;		/* va of mapping */
257 };
258 
259 /* ptbl free list and a lock used for access synchronization. */
260 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
261 static struct mtx ptbl_buf_freelist_lock;
262 
263 /* Base address of kva space allocated fot ptbl bufs. */
264 static vm_offset_t ptbl_buf_pool_vabase;
265 
266 /* Pointer to ptbl_buf structures. */
267 static struct ptbl_buf *ptbl_bufs;
268 
269 /*
270  * Kernel MMU interface
271  */
272 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
273 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
274 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
275 static void		mmu_booke_copy(pmap_t, pmap_t, vm_offset_t, vm_size_t,
276     vm_offset_t);
277 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
278 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
279     vm_prot_t, boolean_t);
280 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
281     vm_page_t, vm_prot_t);
282 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
283     vm_prot_t);
284 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
285 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
286     vm_prot_t);
287 static void		mmu_booke_init(mmu_t);
288 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
289 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
290 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
291 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
292     int);
293 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t);
294 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
295     vm_object_t, vm_pindex_t, vm_size_t);
296 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
297 static void		mmu_booke_page_init(mmu_t, vm_page_t);
298 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
299 static void		mmu_booke_pinit(mmu_t, pmap_t);
300 static void		mmu_booke_pinit0(mmu_t, pmap_t);
301 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
302     vm_prot_t);
303 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
304 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
305 static void		mmu_booke_release(mmu_t, pmap_t);
306 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
308 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
309 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
310 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
311 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
312 static void		mmu_booke_activate(mmu_t, struct thread *);
313 static void		mmu_booke_deactivate(mmu_t, struct thread *);
314 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
315 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
316 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
317 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
318 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
319 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
320 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
321 static boolean_t	mmu_booke_page_executable(mmu_t, vm_page_t);
322 
323 static mmu_method_t mmu_booke_methods[] = {
324 	/* pmap dispatcher interface */
325 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
326 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
327 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
328 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
329 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
330 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
331 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
332 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
333 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
334 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
335 	MMUMETHOD(mmu_init,		mmu_booke_init),
336 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
337 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
338 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
339 	MMUMETHOD(mmu_map,		mmu_booke_map),
340 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
341 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
342 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
343 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
344 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
345 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
346 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
347 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
348 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
349 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
350 	MMUMETHOD(mmu_release,		mmu_booke_release),
351 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
352 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
353 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
354 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
355 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
356 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
357 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
358 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
359 
360 	/* Internal interfaces */
361 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
362 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
363 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
364 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
365 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
366 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
367 	MMUMETHOD(mmu_page_executable,	mmu_booke_page_executable),
368 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
369 
370 	{ 0, 0 }
371 };
372 
373 static mmu_def_t booke_mmu = {
374 	MMU_TYPE_BOOKE,
375 	mmu_booke_methods,
376 	0
377 };
378 MMU_DEF(booke_mmu);
379 
380 /* Return number of entries in TLB0. */
381 static __inline void
382 tlb0_get_tlbconf(void)
383 {
384 	uint32_t tlb0_cfg;
385 
386 	tlb0_cfg = mfspr(SPR_TLB0CFG);
387 	tlb0_size = tlb0_cfg & TLBCFG_NENTRY_MASK;
388 	tlb0_nways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
389 	tlb0_nentries_per_way = tlb0_size/tlb0_nways;
390 }
391 
392 /* Initialize pool of kva ptbl buffers. */
393 static void
394 ptbl_init(void)
395 {
396 	int i;
397 
398 	//debugf("ptbl_init: s (ptbl_bufs = 0x%08x size 0x%08x)\n",
399 	//		(u_int32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
400 	//debugf("ptbl_init: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)\n",
401 	//		ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
402 
403 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
404 	TAILQ_INIT(&ptbl_buf_freelist);
405 
406 	for (i = 0; i < PTBL_BUFS; i++) {
407 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
408 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
409 	}
410 
411 	//debugf("ptbl_init: e\n");
412 }
413 
414 /* Get an sf_buf from the freelist. */
415 static struct ptbl_buf *
416 ptbl_buf_alloc(void)
417 {
418 	struct ptbl_buf *buf;
419 
420 	//debugf("ptbl_buf_alloc: s\n");
421 
422 	mtx_lock(&ptbl_buf_freelist_lock);
423 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
424 	if (buf != NULL)
425 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
426 	mtx_unlock(&ptbl_buf_freelist_lock);
427 
428 	//debugf("ptbl_buf_alloc: e (buf = 0x%08x)\n", (u_int32_t)buf);
429 	return (buf);
430 }
431 
432 /* Return ptbl buff to free pool. */
433 static void
434 ptbl_buf_free(struct ptbl_buf *buf)
435 {
436 
437 	//debugf("ptbl_buf_free: s (buf = 0x%08x)\n", (u_int32_t)buf);
438 
439 	mtx_lock(&ptbl_buf_freelist_lock);
440 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
441 	mtx_unlock(&ptbl_buf_freelist_lock);
442 
443 	//debugf("ptbl_buf_free: e\n");
444 }
445 
446 /*
447  * Search the list of allocated ptbl bufs and find
448  * on list of allocated ptbls
449  */
450 static void
451 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
452 {
453 	struct ptbl_buf *pbuf;
454 
455 	//debugf("ptbl_free_pmap_ptbl: s (pmap = 0x%08x ptbl = 0x%08x)\n",
456 	//		(u_int32_t)pmap, (u_int32_t)ptbl);
457 
458 	TAILQ_FOREACH(pbuf, &pmap->ptbl_list, link) {
459 		if (pbuf->kva == (vm_offset_t)ptbl) {
460 			/* Remove from pmap ptbl buf list. */
461 			TAILQ_REMOVE(&pmap->ptbl_list, pbuf, link);
462 
463 			/* Free correspondig ptbl buf. */
464 			ptbl_buf_free(pbuf);
465 
466 			break;
467 		}
468 	}
469 
470 	//debugf("ptbl_free_pmap_ptbl: e\n");
471 }
472 
473 /* Allocate page table. */
474 static void
475 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
476 {
477 	vm_page_t mtbl[PTBL_PAGES];
478 	vm_page_t m;
479 	struct ptbl_buf *pbuf;
480 	unsigned int pidx;
481 	int i;
482 
483 	//int su = (pmap == kernel_pmap);
484 	//debugf("ptbl_alloc: s (pmap = 0x%08x su = %d pdir_idx = %d)\n", (u_int32_t)pmap, su, pdir_idx);
485 
486 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
487 	    ("ptbl_alloc: invalid pdir_idx"));
488 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
489 	    ("pte_alloc: valid ptbl entry exists!"));
490 
491 	pbuf = ptbl_buf_alloc();
492 	if (pbuf == NULL)
493 		panic("pte_alloc: couldn't alloc kernel virtual memory");
494 	pmap->pm_pdir[pdir_idx] = (pte_t *)pbuf->kva;
495 	//debugf("ptbl_alloc: kva = 0x%08x\n", (u_int32_t)pmap->pm_pdir[pdir_idx]);
496 
497 	/* Allocate ptbl pages, this will sleep! */
498 	for (i = 0; i < PTBL_PAGES; i++) {
499 		pidx = (PTBL_PAGES * pdir_idx) + i;
500 		while ((m = vm_page_alloc(NULL, pidx, VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
501 			PMAP_UNLOCK(pmap);
502 			vm_page_unlock_queues();
503 			VM_WAIT;
504 			vm_page_lock_queues();
505 			PMAP_LOCK(pmap);
506 		}
507 		mtbl[i] = m;
508 	}
509 
510 	/* Map in allocated pages into kernel_pmap. */
511 	mmu_booke_qenter(mmu, (vm_offset_t)pmap->pm_pdir[pdir_idx], mtbl, PTBL_PAGES);
512 
513 	/* Zero whole ptbl. */
514 	bzero((caddr_t)pmap->pm_pdir[pdir_idx], PTBL_PAGES * PAGE_SIZE);
515 
516 	/* Add pbuf to the pmap ptbl bufs list. */
517 	TAILQ_INSERT_TAIL(&pmap->ptbl_list, pbuf, link);
518 
519 	//debugf("ptbl_alloc: e\n");
520 }
521 
522 /* Free ptbl pages and invalidate pdir entry. */
523 static void
524 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
525 {
526 	pte_t *ptbl;
527 	vm_paddr_t pa;
528 	vm_offset_t va;
529 	vm_page_t m;
530 	int i;
531 
532 	//int su = (pmap == kernel_pmap);
533 	//debugf("ptbl_free: s (pmap = 0x%08x su = %d pdir_idx = %d)\n", (u_int32_t)pmap, su, pdir_idx);
534 
535 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
536 	    ("ptbl_free: invalid pdir_idx"));
537 
538 	ptbl = pmap->pm_pdir[pdir_idx];
539 
540 	//debugf("ptbl_free: ptbl = 0x%08x\n", (u_int32_t)ptbl);
541 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
542 
543 	for (i = 0; i < PTBL_PAGES; i++) {
544 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
545 		pa = pte_vatopa(mmu, kernel_pmap, va);
546 		m = PHYS_TO_VM_PAGE(pa);
547 		vm_page_free_zero(m);
548 		atomic_subtract_int(&cnt.v_wire_count, 1);
549 		mmu_booke_kremove(mmu, va);
550 	}
551 
552 	ptbl_free_pmap_ptbl(pmap, ptbl);
553 	pmap->pm_pdir[pdir_idx] = NULL;
554 
555 	//debugf("ptbl_free: e\n");
556 }
557 
558 /*
559  * Decrement ptbl pages hold count and attempt to free ptbl pages.
560  * Called when removing pte entry from ptbl.
561  *
562  * Return 1 if ptbl pages were freed.
563  */
564 static int
565 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
566 {
567 	pte_t *ptbl;
568 	vm_paddr_t pa;
569 	vm_page_t m;
570 	int i;
571 
572 	//int su = (pmap == kernel_pmap);
573 	//debugf("ptbl_unhold: s (pmap = %08x su = %d pdir_idx = %d)\n",
574 	//		(u_int32_t)pmap, su, pdir_idx);
575 
576 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
577 	    ("ptbl_unhold: invalid pdir_idx"));
578 	KASSERT((pmap != kernel_pmap),
579 	    ("ptbl_unhold: unholding kernel ptbl!"));
580 
581 	ptbl = pmap->pm_pdir[pdir_idx];
582 
583 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
584 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
585 	    ("ptbl_unhold: non kva ptbl"));
586 
587 	/* decrement hold count */
588 	for (i = 0; i < PTBL_PAGES; i++) {
589 		pa = pte_vatopa(mmu, kernel_pmap, (vm_offset_t)ptbl + (i * PAGE_SIZE));
590 		m = PHYS_TO_VM_PAGE(pa);
591 		m->wire_count--;
592 	}
593 
594 	/*
595 	 * Free ptbl pages if there are no pte etries in this ptbl.
596 	 * wire_count has the same value for all ptbl pages, so check
597 	 * the last page.
598 	 */
599 	if (m->wire_count == 0) {
600 		ptbl_free(mmu, pmap, pdir_idx);
601 
602 		//debugf("ptbl_unhold: e (freed ptbl)\n");
603 		return (1);
604 	}
605 
606 	//debugf("ptbl_unhold: e\n");
607 	return (0);
608 }
609 
610 /*
611  * Increment hold count for ptbl pages. This routine is used when
612  * new pte entry is being inserted into ptbl.
613  */
614 static void
615 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
616 {
617 	vm_paddr_t pa;
618 	pte_t *ptbl;
619 	vm_page_t m;
620 	int i;
621 
622 	//debugf("ptbl_hold: s (pmap = 0x%08x pdir_idx = %d)\n", (u_int32_t)pmap, pdir_idx);
623 
624 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
625 	    ("ptbl_hold: invalid pdir_idx"));
626 	KASSERT((pmap != kernel_pmap),
627 	    ("ptbl_hold: holding kernel ptbl!"));
628 
629 	ptbl = pmap->pm_pdir[pdir_idx];
630 
631 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
632 
633 	for (i = 0; i < PTBL_PAGES; i++) {
634 		pa = pte_vatopa(mmu, kernel_pmap, (vm_offset_t)ptbl + (i * PAGE_SIZE));
635 		m = PHYS_TO_VM_PAGE(pa);
636 		m->wire_count++;
637 	}
638 
639 	//debugf("ptbl_hold: e\n");
640 }
641 
642 /* Allocate pv_entry structure. */
643 pv_entry_t
644 pv_alloc(void)
645 {
646 	pv_entry_t pv;
647 
648 	debugf("pv_alloc: s\n");
649 
650 	pv_entry_count++;
651 	if ((pv_entry_count > pv_entry_high_water) && (pagedaemon_waken == 0)) {
652 		pagedaemon_waken = 1;
653 		wakeup (&vm_pages_needed);
654 	}
655 	pv = uma_zalloc(pvzone, M_NOWAIT);
656 
657 	debugf("pv_alloc: e\n");
658 	return (pv);
659 }
660 
661 /* Free pv_entry structure. */
662 static __inline void
663 pv_free(pv_entry_t pve)
664 {
665 	//debugf("pv_free: s\n");
666 
667 	pv_entry_count--;
668 	uma_zfree(pvzone, pve);
669 
670 	//debugf("pv_free: e\n");
671 }
672 
673 
674 /* Allocate and initialize pv_entry structure. */
675 static void
676 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
677 {
678 	pv_entry_t pve;
679 
680 	//int su = (pmap == kernel_pmap);
681 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
682 	//	(u_int32_t)pmap, va, (u_int32_t)m);
683 
684 	pve = pv_alloc();
685 	if (pve == NULL)
686 		panic("pv_insert: no pv entries!");
687 
688 	pve->pv_pmap = pmap;
689 	pve->pv_va = va;
690 
691 	/* add to pv_list */
692 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
693 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
694 
695 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
696 
697 	//debugf("pv_insert: e\n");
698 }
699 
700 /* Destroy pv entry. */
701 static void
702 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
703 {
704 	pv_entry_t pve;
705 
706 	//int su = (pmap == kernel_pmap);
707 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
708 
709 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
710 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
711 
712 	/* find pv entry */
713 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
714 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
715 			/* remove from pv_list */
716 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
717 			if (TAILQ_EMPTY(&m->md.pv_list))
718 				vm_page_flag_clear(m, PG_WRITEABLE);
719 
720 			/* free pv entry struct */
721 			pv_free(pve);
722 
723 			break;
724 		}
725 	}
726 
727 	//debugf("pv_remove: e\n");
728 }
729 
730 /*
731  * Clean pte entry, try to free page table page if requested.
732  *
733  * Return 1 if ptbl pages were freed, otherwise return 0.
734  */
735 static int
736 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
737 {
738 	unsigned int pdir_idx = PDIR_IDX(va);
739 	unsigned int ptbl_idx = PTBL_IDX(va);
740 	vm_page_t m;
741 	pte_t *ptbl;
742 	pte_t *pte;
743 
744 	//int su = (pmap == kernel_pmap);
745 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
746 	//		su, (u_int32_t)pmap, va, flags);
747 
748 	ptbl = pmap->pm_pdir[pdir_idx];
749 	KASSERT(ptbl, ("pte_remove: null ptbl"));
750 
751 	pte = &ptbl[ptbl_idx];
752 
753 	if (pte == NULL || !PTE_ISVALID(pte))
754 		return (0);
755 
756 	/* Get vm_page_t for mapped pte. */
757 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
758 
759 	if (PTE_ISWIRED(pte))
760 		pmap->pm_stats.wired_count--;
761 
762 	if (!PTE_ISFAKE(pte)) {
763 		/* Handle managed entry. */
764 		if (PTE_ISMANAGED(pte)) {
765 
766 			/* Handle modified pages. */
767 			if (PTE_ISMODIFIED(pte))
768 				vm_page_dirty(m);
769 
770 			/* Referenced pages. */
771 			if (PTE_ISREFERENCED(pte))
772 				vm_page_flag_set(m, PG_REFERENCED);
773 
774 			/* Remove pv_entry from pv_list. */
775 			pv_remove(pmap, va, m);
776 		}
777 	}
778 
779 	pte->flags = 0;
780 	pte->rpn = 0;
781 	pmap->pm_stats.resident_count--;
782 
783 	if (flags & PTBL_UNHOLD) {
784 		//debugf("pte_remove: e (unhold)\n");
785 		return (ptbl_unhold(mmu, pmap, pdir_idx));
786 	}
787 
788 	//debugf("pte_remove: e\n");
789 	return (0);
790 }
791 
792 /*
793  * Insert PTE for a given page and virtual address.
794  */
795 void
796 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, u_int32_t flags)
797 {
798 	unsigned int pdir_idx = PDIR_IDX(va);
799 	unsigned int ptbl_idx = PTBL_IDX(va);
800 	pte_t *ptbl;
801 	pte_t *pte;
802 
803 	//int su = (pmap == kernel_pmap);
804 	//debugf("pte_enter: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
805 
806 	/* Get the page table pointer. */
807 	ptbl = pmap->pm_pdir[pdir_idx];
808 
809 	if (ptbl) {
810 		/*
811 		 * Check if there is valid mapping for requested
812 		 * va, if there is, remove it.
813 		 */
814 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
815 		if (PTE_ISVALID(pte)) {
816 			pte_remove(mmu, pmap, va, PTBL_HOLD);
817 		} else {
818 			/*
819 			 * pte is not used, increment hold count
820 			 * for ptbl pages.
821 			 */
822 			if (pmap != kernel_pmap)
823 				ptbl_hold(mmu, pmap, pdir_idx);
824 		}
825 	} else {
826 		/* Allocate page table pages. */
827 		ptbl_alloc(mmu, pmap, pdir_idx);
828 	}
829 
830 	/* Flush entry from TLB. */
831 	tlb0_flush_entry(pmap, va);
832 
833 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
834 
835 	/*
836 	 * Insert pv_entry into pv_list for mapped page
837 	 * if part of managed memory.
838 	 */
839         if ((m->flags & PG_FICTITIOUS) == 0) {
840 		if ((m->flags & PG_UNMANAGED) == 0) {
841 			pte->flags |= PTE_MANAGED;
842 
843 			/* Create and insert pv entry. */
844 			pv_insert(pmap, va, m);
845 		}
846         } else {
847 		pte->flags |= PTE_FAKE;
848 	}
849 
850 	pmap->pm_stats.resident_count++;
851 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
852 	pte->flags |= (PTE_VALID | flags);
853 
854 	//debugf("pte_enter: e\n");
855 }
856 
857 /* Return the pa for the given pmap/va. */
858 static vm_paddr_t
859 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
860 {
861 	vm_paddr_t pa = 0;
862 	pte_t *pte;
863 
864 	pte = pte_find(mmu, pmap, va);
865 	if ((pte != NULL) && PTE_ISVALID(pte))
866 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
867 	return (pa);
868 }
869 
870 /* Get a pointer to a PTE in a page table. */
871 static pte_t *
872 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
873 {
874 	unsigned int pdir_idx = PDIR_IDX(va);
875 	unsigned int ptbl_idx = PTBL_IDX(va);
876 
877 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
878 
879 	if (pmap->pm_pdir[pdir_idx])
880 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
881 
882 	return (NULL);
883 }
884 
885 /**************************************************************************/
886 /* PMAP related */
887 /**************************************************************************/
888 
889 /*
890  * This is called during e500_init, before the system is really initialized.
891  */
892 static void
893 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t kernelstart, vm_offset_t kernelend)
894 {
895 	vm_offset_t phys_kernelend;
896 	struct mem_region *mp, *mp1;
897 	int cnt, i, j;
898 	u_int s, e, sz;
899 	u_int phys_avail_count;
900 	vm_size_t physsz, hwphyssz;
901 	vm_offset_t kernel_pdir;
902 
903 	debugf("mmu_booke_bootstrap: entered\n");
904 
905 	/* Align kernel start and end address (kernel image). */
906 	kernelstart = trunc_page(kernelstart);
907 	kernelend = round_page(kernelend);
908 
909 	/* Allocate space for the message buffer. */
910 	msgbufp = (struct msgbuf *)kernelend;
911 	kernelend += MSGBUF_SIZE;
912 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (u_int32_t)msgbufp,
913 	    kernelend);
914 
915 	kernelend = round_page(kernelend);
916 
917 	/* Allocate space for tlb0 table. */
918 	tlb0_get_tlbconf(); /* Read TLB0 size and associativity. */
919 	tlb0 = (tlb_entry_t *)kernelend;
920 	kernelend += sizeof(tlb_entry_t) * tlb0_size;
921 	memset((void *)tlb0, 0, sizeof(tlb_entry_t) * tlb0_size);
922 	debugf(" tlb0 at 0x%08x end = 0x%08x\n", (u_int32_t)tlb0, kernelend);
923 
924 	kernelend = round_page(kernelend);
925 
926 	/* Allocate space for ptbl_bufs. */
927 	ptbl_bufs = (struct ptbl_buf *)kernelend;
928 	kernelend += sizeof(struct ptbl_buf) * PTBL_BUFS;
929 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
930 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (u_int32_t)ptbl_bufs,
931 	    kernelend);
932 
933 	kernelend = round_page(kernelend);
934 
935 	/* Allocate PTE tables for kernel KVA. */
936 	kernel_pdir = kernelend;
937 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
938 	    PDIR_SIZE - 1) / PDIR_SIZE;
939 	kernelend += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
940 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
941 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
942 	debugf(" kernel pdir at 0x%08x\n", kernel_pdir);
943 
944 	if (kernelend - kernelstart > 0x1000000) {
945 		kernelend = (kernelend + 0x3fffff) & ~0x3fffff;
946 		tlb1_mapin_region(kernelstart + 0x1000000,
947 		    kernload + 0x1000000, kernelend - kernelstart - 0x1000000);
948 	} else
949 		kernelend = (kernelend + 0xffffff) & ~0xffffff;
950 
951 	/*******************************************************/
952 	/* Set the start and end of kva. */
953 	/*******************************************************/
954 	virtual_avail = kernelend;
955 	virtual_end = VM_MAX_KERNEL_ADDRESS;
956 
957 	/* Allocate KVA space for page zero/copy operations. */
958 	zero_page_va = virtual_avail;
959 	virtual_avail += PAGE_SIZE;
960 	zero_page_idle_va = virtual_avail;
961 	virtual_avail += PAGE_SIZE;
962 	copy_page_src_va = virtual_avail;
963 	virtual_avail += PAGE_SIZE;
964 	copy_page_dst_va = virtual_avail;
965 	virtual_avail += PAGE_SIZE;
966 
967 	/* Initialize page zero/copy mutexes. */
968 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
969 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
970 
971 	/* Initialize tlb0 table mutex. */
972 	mtx_init(&tlb0_mutex, "tlb0", NULL, MTX_SPIN | MTX_RECURSE);
973 
974 	/* Allocate KVA space for ptbl bufs. */
975 	ptbl_buf_pool_vabase = virtual_avail;
976 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
977 
978 	debugf("ptbl_buf_pool_vabase = 0x%08x\n", ptbl_buf_pool_vabase);
979 	debugf("virtual_avail = %08x\n", virtual_avail);
980 	debugf("virtual_end   = %08x\n", virtual_end);
981 
982 	/* Calculate corresponding physical addresses for the kernel region. */
983 	phys_kernelend = kernload + (kernelend - kernelstart);
984 
985 	debugf("kernel image and allocated data:\n");
986 	debugf(" kernload    = 0x%08x\n", kernload);
987 	debugf(" kernelstart = 0x%08x\n", kernelstart);
988 	debugf(" kernelend   = 0x%08x\n", kernelend);
989 	debugf(" kernel size = 0x%08x\n", kernelend - kernelstart);
990 
991 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
992 		panic("mmu_booke_bootstrap: phys_avail too small");
993 
994 	/*
995 	 * Removed kernel physical address range from avail
996 	 * regions list. Page align all regions.
997 	 * Non-page aligned memory isn't very interesting to us.
998 	 * Also, sort the entries for ascending addresses.
999 	 */
1000 	sz = 0;
1001 	cnt = availmem_regions_sz;
1002 	debugf("processing avail regions:\n");
1003 	for (mp = availmem_regions; mp->mr_size; mp++) {
1004 		s = mp->mr_start;
1005 		e = mp->mr_start + mp->mr_size;
1006 		debugf(" %08x-%08x -> ", s, e);
1007 		/* Check whether this region holds all of the kernel. */
1008 		if (s < kernload && e > phys_kernelend) {
1009 			availmem_regions[cnt].mr_start = phys_kernelend;
1010 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1011 			e = kernload;
1012 		}
1013 		/* Look whether this regions starts within the kernel. */
1014 		if (s >= kernload && s < phys_kernelend) {
1015 			if (e <= phys_kernelend)
1016 				goto empty;
1017 			s = phys_kernelend;
1018 		}
1019 		/* Now look whether this region ends within the kernel. */
1020 		if (e > kernload && e <= phys_kernelend) {
1021 			if (s >= kernload)
1022 				goto empty;
1023 			e = kernload;
1024 		}
1025 		/* Now page align the start and size of the region. */
1026 		s = round_page(s);
1027 		e = trunc_page(e);
1028 		if (e < s)
1029 			e = s;
1030 		sz = e - s;
1031 		debugf("%08x-%08x = %x\n", s, e, sz);
1032 
1033 		/* Check whether some memory is left here. */
1034 		if (sz == 0) {
1035 		empty:
1036 			memmove(mp, mp + 1,
1037 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1038 			cnt--;
1039 			mp--;
1040 			continue;
1041 		}
1042 
1043 		/* Do an insertion sort. */
1044 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1045 			if (s < mp1->mr_start)
1046 				break;
1047 		if (mp1 < mp) {
1048 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1049 			mp1->mr_start = s;
1050 			mp1->mr_size = sz;
1051 		} else {
1052 			mp->mr_start = s;
1053 			mp->mr_size = sz;
1054 		}
1055 	}
1056 	availmem_regions_sz = cnt;
1057 
1058 	/*******************************************************/
1059 	/* Fill in phys_avail table, based on availmem_regions */
1060 	/*******************************************************/
1061 	phys_avail_count = 0;
1062 	physsz = 0;
1063 	hwphyssz = 0;
1064 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1065 
1066 	debugf("fill in phys_avail:\n");
1067 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1068 
1069 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1070 		    availmem_regions[i].mr_start,
1071 		    availmem_regions[i].mr_start + availmem_regions[i].mr_size,
1072 		    availmem_regions[i].mr_size);
1073 
1074 		if (hwphyssz != 0 && (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1075 			debugf(" hw.physmem adjust\n");
1076 			if (physsz < hwphyssz) {
1077 				phys_avail[j] = availmem_regions[i].mr_start;
1078 				phys_avail[j + 1] = availmem_regions[i].mr_start +
1079 				    hwphyssz - physsz;
1080 				physsz = hwphyssz;
1081 				phys_avail_count++;
1082 			}
1083 			break;
1084 		}
1085 
1086 		phys_avail[j] = availmem_regions[i].mr_start;
1087 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1088 		    availmem_regions[i].mr_size;
1089 		phys_avail_count++;
1090 		physsz += availmem_regions[i].mr_size;
1091 	}
1092 	physmem = btoc(physsz);
1093 
1094 	/* Calculate the last available physical address. */
1095 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1096 		;
1097 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1098 
1099 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1100 	debugf("phys_avail_count = %d\n", phys_avail_count);
1101 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem, physmem);
1102 
1103 	/*******************************************************/
1104 	/* Initialize (statically allocated) kernel pmap. */
1105 	/*******************************************************/
1106 	PMAP_LOCK_INIT(kernel_pmap);
1107 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1108 
1109 	debugf("kernel_pmap = 0x%08x\n", (u_int32_t)kernel_pmap);
1110 	debugf("kptbl_min = %d, kernel_kptbls = %d\n", kptbl_min, kernel_ptbls);
1111 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1112 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1113 
1114 	/* Initialize kernel pdir */
1115 	for (i = 0; i < kernel_ptbls; i++)
1116 		kernel_pmap->pm_pdir[kptbl_min + i] =
1117 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1118 
1119 	kernel_pmap->pm_tid = KERNEL_TID;
1120 	kernel_pmap->pm_active = ~0;
1121 
1122 	/* Initialize tidbusy with kenel_pmap entry. */
1123 	tidbusy[0] = kernel_pmap;
1124 
1125 	/*******************************************************/
1126 	/* Final setup */
1127 	/*******************************************************/
1128 	/* Initialize TLB0 handling. */
1129 	tlb0_init();
1130 
1131 	debugf("mmu_booke_bootstrap: exit\n");
1132 }
1133 
1134 /*
1135  * Get the physical page address for the given pmap/virtual address.
1136  */
1137 static vm_paddr_t
1138 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1139 {
1140 	vm_paddr_t pa;
1141 
1142 	PMAP_LOCK(pmap);
1143 	pa = pte_vatopa(mmu, pmap, va);
1144 	PMAP_UNLOCK(pmap);
1145 
1146 	return (pa);
1147 }
1148 
1149 /*
1150  * Extract the physical page address associated with the given
1151  * kernel virtual address.
1152  */
1153 static vm_paddr_t
1154 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1155 {
1156 
1157 	return (pte_vatopa(mmu, kernel_pmap, va));
1158 }
1159 
1160 /*
1161  * Initialize the pmap module.
1162  * Called by vm_init, to initialize any structures that the pmap
1163  * system needs to map virtual memory.
1164  */
1165 static void
1166 mmu_booke_init(mmu_t mmu)
1167 {
1168 	int shpgperproc = PMAP_SHPGPERPROC;
1169 
1170 	//debugf("mmu_booke_init: s\n");
1171 
1172 	/*
1173 	 * Initialize the address space (zone) for the pv entries.  Set a
1174 	 * high water mark so that the system can recover from excessive
1175 	 * numbers of pv entries.
1176 	 */
1177 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1178 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1179 
1180 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1181 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1182 
1183 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1184 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1185 
1186 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1187 
1188 	/* Pre-fill pvzone with initial number of pv entries. */
1189 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1190 
1191 	/* Initialize ptbl allocation. */
1192 	ptbl_init();
1193 
1194 	//debugf("mmu_booke_init: e\n");
1195 }
1196 
1197 /*
1198  * Map a list of wired pages into kernel virtual address space.  This is
1199  * intended for temporary mappings which do not need page modification or
1200  * references recorded.  Existing mappings in the region are overwritten.
1201  */
1202 static void
1203 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1204 {
1205 	vm_offset_t va;
1206 
1207 	//debugf("mmu_booke_qenter: s (sva = 0x%08x count = %d)\n", sva, count);
1208 
1209 	va = sva;
1210 	while (count-- > 0) {
1211 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1212 		va += PAGE_SIZE;
1213 		m++;
1214 	}
1215 
1216 	//debugf("mmu_booke_qenter: e\n");
1217 }
1218 
1219 /*
1220  * Remove page mappings from kernel virtual address space.  Intended for
1221  * temporary mappings entered by mmu_booke_qenter.
1222  */
1223 static void
1224 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1225 {
1226 	vm_offset_t va;
1227 
1228 	//debugf("mmu_booke_qremove: s (sva = 0x%08x count = %d)\n", sva, count);
1229 
1230 	va = sva;
1231 	while (count-- > 0) {
1232 		mmu_booke_kremove(mmu, va);
1233 		va += PAGE_SIZE;
1234 	}
1235 
1236 	//debugf("mmu_booke_qremove: e\n");
1237 }
1238 
1239 /*
1240  * Map a wired page into kernel virtual address space.
1241  */
1242 static void
1243 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1244 {
1245 	unsigned int pdir_idx = PDIR_IDX(va);
1246 	unsigned int ptbl_idx = PTBL_IDX(va);
1247 	u_int32_t flags;
1248 	pte_t *pte;
1249 
1250 	//debugf("mmu_booke_kenter: s (pdir_idx = %d ptbl_idx = %d va=0x%08x pa=0x%08x)\n",
1251 	//		pdir_idx, ptbl_idx, va, pa);
1252 
1253 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)),
1254 			("mmu_booke_kenter: invalid va"));
1255 
1256 #if 0
1257 	/* assume IO mapping, set I, G bits */
1258 	flags = (PTE_G | PTE_I | PTE_FAKE);
1259 
1260 	/* if mapping is within system memory, do not set I, G bits */
1261 	for (i = 0; i < totalmem_regions_sz; i++) {
1262 		if ((pa >= totalmem_regions[i].mr_start) &&
1263 				(pa < (totalmem_regions[i].mr_start +
1264 				       totalmem_regions[i].mr_size))) {
1265 			flags &= ~(PTE_I | PTE_G | PTE_FAKE);
1266 			break;
1267 		}
1268 	}
1269 #else
1270 	flags = 0;
1271 #endif
1272 
1273 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1274 
1275 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1276 
1277 	if (PTE_ISVALID(pte)) {
1278 		//debugf("mmu_booke_kenter: replacing entry!\n");
1279 
1280 		/* Flush entry from TLB0 */
1281 		tlb0_flush_entry(kernel_pmap, va);
1282 	}
1283 
1284 	pte->rpn = pa & ~PTE_PA_MASK;
1285 	pte->flags = flags;
1286 
1287 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1288 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1289 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1290 
1291 	/* Flush the real memory from the instruction cache. */
1292 	if ((flags & (PTE_I | PTE_G)) == 0) {
1293 		__syncicache((void *)va, PAGE_SIZE);
1294 	}
1295 
1296 	//debugf("mmu_booke_kenter: e\n");
1297 }
1298 
1299 /*
1300  * Remove a page from kernel page table.
1301  */
1302 static void
1303 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1304 {
1305 	unsigned int pdir_idx = PDIR_IDX(va);
1306 	unsigned int ptbl_idx = PTBL_IDX(va);
1307 	pte_t *pte;
1308 
1309 	//debugf("mmu_booke_kremove: s (va = 0x%08x)\n", va);
1310 
1311 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)),
1312 	    ("mmu_booke_kremove: invalid va"));
1313 
1314 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1315 
1316 	if (!PTE_ISVALID(pte)) {
1317 		//debugf("mmu_booke_kremove: e (invalid pte)\n");
1318 		return;
1319 	}
1320 
1321 	/* Invalidate entry in TLB0. */
1322 	tlb0_flush_entry(kernel_pmap, va);
1323 
1324 	pte->flags = 0;
1325 	pte->rpn = 0;
1326 
1327 	//debugf("mmu_booke_kremove: e\n");
1328 }
1329 
1330 /*
1331  * Initialize pmap associated with process 0.
1332  */
1333 static void
1334 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1335 {
1336 	//debugf("mmu_booke_pinit0: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
1337 	mmu_booke_pinit(mmu, pmap);
1338 	PCPU_SET(curpmap, pmap);
1339 	//debugf("mmu_booke_pinit0: e\n");
1340 }
1341 
1342 /*
1343  * Initialize a preallocated and zeroed pmap structure,
1344  * such as one in a vmspace structure.
1345  */
1346 static void
1347 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1348 {
1349 
1350 	//struct thread *td;
1351 	//struct proc *p;
1352 
1353 	//td = PCPU_GET(curthread);
1354 	//p = td->td_proc;
1355 	//debugf("mmu_booke_pinit: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
1356 	//printf("mmu_booke_pinit: proc %d '%s'\n", p->p_pid, p->p_comm);
1357 
1358 	KASSERT((pmap != kernel_pmap), ("mmu_booke_pinit: initializing kernel_pmap"));
1359 
1360 	PMAP_LOCK_INIT(pmap);
1361 	pmap->pm_tid = 0;
1362 	pmap->pm_active = 0;
1363 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1364 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1365 
1366 	TAILQ_INIT(&pmap->ptbl_list);
1367 
1368 	//debugf("mmu_booke_pinit: e\n");
1369 }
1370 
1371 /*
1372  * Release any resources held by the given physical map.
1373  * Called when a pmap initialized by mmu_booke_pinit is being released.
1374  * Should only be called if the map contains no valid mappings.
1375  */
1376 static void
1377 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1378 {
1379 
1380 	//debugf("mmu_booke_release: s\n");
1381 
1382 	PMAP_LOCK_DESTROY(pmap);
1383 
1384 	//debugf("mmu_booke_release: e\n");
1385 }
1386 
1387 #if 0
1388 /* Not needed, kernel page tables are statically allocated. */
1389 void
1390 mmu_booke_growkernel(vm_offset_t maxkvaddr)
1391 {
1392 }
1393 #endif
1394 
1395 /*
1396  * Insert the given physical page at the specified virtual address in the
1397  * target physical map with the protection requested. If specified the page
1398  * will be wired down.
1399  */
1400 static void
1401 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1402     vm_prot_t prot, boolean_t wired)
1403 {
1404 	vm_page_lock_queues();
1405 	PMAP_LOCK(pmap);
1406 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1407 	vm_page_unlock_queues();
1408 	PMAP_UNLOCK(pmap);
1409 }
1410 
1411 static void
1412 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1413     vm_prot_t prot, boolean_t wired)
1414 {
1415 	pte_t *pte;
1416 	vm_paddr_t pa;
1417 	u_int32_t flags;
1418 	int su, sync;
1419 
1420 	pa = VM_PAGE_TO_PHYS(m);
1421 	su = (pmap == kernel_pmap);
1422 	sync = 0;
1423 
1424 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1425 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1426 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1427 	//		(u_int32_t)m, va, pa, prot, wired);
1428 
1429 	if (su) {
1430 		KASSERT(((va >= virtual_avail) && (va <= VM_MAX_KERNEL_ADDRESS)),
1431 				("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1432 	} else {
1433 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1434 				("mmu_booke_enter_locked: user pmap, non user va"));
1435 	}
1436 
1437 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1438 
1439 	/*
1440 	 * If there is an existing mapping, and the physical address has not
1441 	 * changed, must be protection or wiring change.
1442 	 */
1443 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1444 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1445 
1446 		//debugf("mmu_booke_enter_locked: update\n");
1447 
1448 		/* Wiring change, just update stats. */
1449 		if (wired) {
1450 			if (!PTE_ISWIRED(pte)) {
1451 				pte->flags |= PTE_WIRED;
1452 				pmap->pm_stats.wired_count++;
1453 			}
1454 		} else {
1455 			if (PTE_ISWIRED(pte)) {
1456 				pte->flags &= ~PTE_WIRED;
1457 				pmap->pm_stats.wired_count--;
1458 			}
1459 		}
1460 
1461 		/* Save the old bits and clear the ones we're interested in. */
1462 		flags = pte->flags;
1463 		pte->flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1464 
1465 		if (prot & VM_PROT_WRITE) {
1466 			/* Add write permissions. */
1467 			pte->flags |= PTE_SW;
1468 			if (!su)
1469 				pte->flags |= PTE_UW;
1470 		} else {
1471 			/* Handle modified pages, sense modify status. */
1472 			if (PTE_ISMODIFIED(pte))
1473 				vm_page_dirty(m);
1474 		}
1475 
1476 		/* If we're turning on execute permissions, flush the icache. */
1477 		if (prot & VM_PROT_EXECUTE) {
1478 			pte->flags |= PTE_SX;
1479 			if (!su)
1480 				pte->flags |= PTE_UX;
1481 
1482 			if ((flags & (PTE_UX | PTE_SX)) == 0)
1483 				sync++;
1484 		}
1485 
1486 		/* Flush the old mapping from TLB0. */
1487 		pte->flags &= ~PTE_REFERENCED;
1488 		tlb0_flush_entry(pmap, va);
1489 	} else {
1490 		/*
1491 		 * If there is an existing mapping, but its for a different
1492 		 * physical address, pte_enter() will delete the old mapping.
1493 		 */
1494 		//if ((pte != NULL) && PTE_ISVALID(pte))
1495 		//	debugf("mmu_booke_enter_locked: replace\n");
1496 		//else
1497 		//	debugf("mmu_booke_enter_locked: new\n");
1498 
1499 		/* Now set up the flags and install the new mapping. */
1500 		flags = (PTE_SR | PTE_VALID);
1501 
1502 		if (!su)
1503 			flags |= PTE_UR;
1504 
1505 		if (prot & VM_PROT_WRITE) {
1506 			flags |= PTE_SW;
1507 			if (!su)
1508 				flags |= PTE_UW;
1509 		}
1510 
1511 		if (prot & VM_PROT_EXECUTE) {
1512 			flags |= PTE_SX;
1513 			if (!su)
1514 				flags |= PTE_UX;
1515 		}
1516 
1517 		/* If its wired update stats. */
1518 		if (wired) {
1519 			pmap->pm_stats.wired_count++;
1520 			flags |= PTE_WIRED;
1521 		}
1522 
1523 		pte_enter(mmu, pmap, m, va, flags);
1524 
1525 		/* Flush the real memory from the instruction cache. */
1526 		if (prot & VM_PROT_EXECUTE)
1527 			sync++;
1528 	}
1529 
1530 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1531 		__syncicache((void *)va, PAGE_SIZE);
1532 		sync = 0;
1533 	}
1534 
1535 	if (sync) {
1536 		/* Create a temporary mapping. */
1537 		pmap = PCPU_GET(curpmap);
1538 
1539 		va = 0;
1540 		pte = pte_find(mmu, pmap, va);
1541 		KASSERT(pte == NULL, ("%s:%d", __func__, __LINE__));
1542 
1543 		flags = PTE_SR | PTE_VALID | PTE_UR;
1544 		pte_enter(mmu, pmap, m, va, flags);
1545 		__syncicache((void *)va, PAGE_SIZE);
1546 		pte_remove(mmu, pmap, va, PTBL_UNHOLD);
1547 	}
1548 
1549 	//debugf("mmu_booke_enter_locked: e\n");
1550 }
1551 
1552 /*
1553  * Maps a sequence of resident pages belonging to the same object.
1554  * The sequence begins with the given page m_start.  This page is
1555  * mapped at the given virtual address start.  Each subsequent page is
1556  * mapped at a virtual address that is offset from start by the same
1557  * amount as the page is offset from m_start within the object.  The
1558  * last page in the sequence is the page with the largest offset from
1559  * m_start that can be mapped at a virtual address less than the given
1560  * virtual address end.  Not every virtual page between start and end
1561  * is mapped; only those for which a resident page exists with the
1562  * corresponding offset from m_start are mapped.
1563  */
1564 static void
1565 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1566     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1567 {
1568 	vm_page_t m;
1569 	vm_pindex_t diff, psize;
1570 
1571 	psize = atop(end - start);
1572 	m = m_start;
1573 	PMAP_LOCK(pmap);
1574 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1575 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m, prot &
1576 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1577 		m = TAILQ_NEXT(m, listq);
1578 	}
1579 	PMAP_UNLOCK(pmap);
1580 }
1581 
1582 static void
1583 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1584     vm_prot_t prot)
1585 {
1586 
1587 	//debugf("mmu_booke_enter_quick: s\n");
1588 
1589 	PMAP_LOCK(pmap);
1590 	mmu_booke_enter_locked(mmu, pmap, va, m,
1591 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1592 	PMAP_UNLOCK(pmap);
1593 
1594 	//debugf("mmu_booke_enter_quick e\n");
1595 }
1596 
1597 /*
1598  * Remove the given range of addresses from the specified map.
1599  *
1600  * It is assumed that the start and end are properly rounded to the page size.
1601  */
1602 static void
1603 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1604 {
1605 	pte_t *pte;
1606 	u_int8_t hold_flag;
1607 
1608 	int su = (pmap == kernel_pmap);
1609 
1610 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1611 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1612 
1613 	if (su) {
1614 		KASSERT(((va >= virtual_avail) && (va <= VM_MAX_KERNEL_ADDRESS)),
1615 		    ("mmu_booke_enter: kernel pmap, non kernel va"));
1616 	} else {
1617 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1618 		    ("mmu_booke_enter: user pmap, non user va"));
1619 	}
1620 
1621 	if (PMAP_REMOVE_DONE(pmap)) {
1622 		//debugf("mmu_booke_remove: e (empty)\n");
1623 		return;
1624 	}
1625 
1626 	hold_flag = PTBL_HOLD_FLAG(pmap);
1627 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1628 
1629 	vm_page_lock_queues();
1630 	PMAP_LOCK(pmap);
1631 	for (; va < endva; va += PAGE_SIZE) {
1632 		pte = pte_find(mmu, pmap, va);
1633 		if ((pte != NULL) && PTE_ISVALID(pte)) {
1634 			pte_remove(mmu, pmap, va, hold_flag);
1635 
1636 			/* Flush mapping from TLB0. */
1637 			tlb0_flush_entry(pmap, va);
1638 		}
1639 	}
1640 	PMAP_UNLOCK(pmap);
1641 	vm_page_unlock_queues();
1642 
1643 	//debugf("mmu_booke_remove: e\n");
1644 }
1645 
1646 /*
1647  * Remove physical page from all pmaps in which it resides.
1648  */
1649 static void
1650 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1651 {
1652 	pv_entry_t pv, pvn;
1653 	u_int8_t hold_flag;
1654 
1655 	//debugf("mmu_booke_remove_all: s\n");
1656 
1657 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1658 
1659 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1660 		pvn = TAILQ_NEXT(pv, pv_link);
1661 
1662 		PMAP_LOCK(pv->pv_pmap);
1663 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1664 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1665 
1666 		/* Flush mapping from TLB0. */
1667 		tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
1668 		PMAP_UNLOCK(pv->pv_pmap);
1669 	}
1670 	vm_page_flag_clear(m, PG_WRITEABLE);
1671 
1672 	//debugf("mmu_booke_remove_all: e\n");
1673 }
1674 
1675 /*
1676  * Map a range of physical addresses into kernel virtual address space.
1677  *
1678  * The value passed in *virt is a suggested virtual address for the mapping.
1679  * Architectures which can support a direct-mapped physical to virtual region
1680  * can return the appropriate address within that region, leaving '*virt'
1681  * unchanged.  We cannot and therefore do not; *virt is updated with the
1682  * first usable address after the mapped region.
1683  */
1684 static vm_offset_t
1685 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1686     vm_offset_t pa_end, int prot)
1687 {
1688 	vm_offset_t sva = *virt;
1689 	vm_offset_t va = sva;
1690 
1691 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1692 	//		sva, pa_start, pa_end);
1693 
1694 	while (pa_start < pa_end) {
1695 		mmu_booke_kenter(mmu, va, pa_start);
1696 		va += PAGE_SIZE;
1697 		pa_start += PAGE_SIZE;
1698 	}
1699 	*virt = va;
1700 
1701 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1702 	return (sva);
1703 }
1704 
1705 /*
1706  * The pmap must be activated before it's address space can be accessed in any
1707  * way.
1708  */
1709 static void
1710 mmu_booke_activate(mmu_t mmu, struct thread *td)
1711 {
1712 	pmap_t pmap;
1713 
1714 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1715 
1716 	//debugf("mmu_booke_activate: s (proc = '%s', id = %d, pmap = 0x%08x)\n",
1717 	//		td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1718 
1719 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1720 
1721 	mtx_lock_spin(&sched_lock);
1722 
1723 	pmap->pm_active |= PCPU_GET(cpumask);
1724 	PCPU_SET(curpmap, pmap);
1725 
1726 	if (!pmap->pm_tid)
1727 		tid_alloc(pmap);
1728 
1729 	/* Load PID0 register with pmap tid value. */
1730 	load_pid0(pmap->pm_tid);
1731 
1732 	mtx_unlock_spin(&sched_lock);
1733 
1734 	//debugf("mmu_booke_activate: e (tid = %d for '%s')\n", pmap->pm_tid,
1735 	//		td->td_proc->p_comm);
1736 }
1737 
1738 /*
1739  * Deactivate the specified process's address space.
1740  */
1741 static void
1742 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1743 {
1744 	pmap_t pmap;
1745 
1746 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1747 	pmap->pm_active &= ~(PCPU_GET(cpumask));
1748 	PCPU_SET(curpmap, NULL);
1749 }
1750 
1751 /*
1752  * Copy the range specified by src_addr/len
1753  * from the source map to the range dst_addr/len
1754  * in the destination map.
1755  *
1756  * This routine is only advisory and need not do anything.
1757  */
1758 static void
1759 mmu_booke_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
1760     vm_size_t len, vm_offset_t src_addr)
1761 {
1762 
1763 }
1764 
1765 /*
1766  * Set the physical protection on the specified range of this map as requested.
1767  */
1768 static void
1769 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1770     vm_prot_t prot)
1771 {
1772 	vm_offset_t va;
1773 	vm_page_t m;
1774 	pte_t *pte;
1775 
1776 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1777 		mmu_booke_remove(mmu, pmap, sva, eva);
1778 		return;
1779 	}
1780 
1781 	if (prot & VM_PROT_WRITE)
1782 		return;
1783 
1784 	vm_page_lock_queues();
1785 	PMAP_LOCK(pmap);
1786 	for (va = sva; va < eva; va += PAGE_SIZE) {
1787 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1788 			if (PTE_ISVALID(pte)) {
1789 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1790 
1791 				/* Handle modified pages. */
1792 				if (PTE_ISMODIFIED(pte))
1793 					vm_page_dirty(m);
1794 
1795 				/* Referenced pages. */
1796 				if (PTE_ISREFERENCED(pte))
1797 					vm_page_flag_set(m, PG_REFERENCED);
1798 
1799 				/* Flush mapping from TLB0. */
1800 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1801 				    PTE_REFERENCED);
1802 				tlb0_flush_entry(pmap, va);
1803 			}
1804 		}
1805 	}
1806 	PMAP_UNLOCK(pmap);
1807 	vm_page_unlock_queues();
1808 }
1809 
1810 /*
1811  * Clear the write and modified bits in each of the given page's mappings.
1812  */
1813 static void
1814 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1815 {
1816 	pv_entry_t pv;
1817 	pte_t *pte;
1818 
1819 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1820 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1821 	    (m->flags & PG_WRITEABLE) == 0)
1822 		return;
1823 
1824 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1825 		PMAP_LOCK(pv->pv_pmap);
1826 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1827 			if (PTE_ISVALID(pte)) {
1828 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1829 
1830 				/* Handle modified pages. */
1831 				if (PTE_ISMODIFIED(pte))
1832 					vm_page_dirty(m);
1833 
1834 				/* Referenced pages. */
1835 				if (PTE_ISREFERENCED(pte))
1836 					vm_page_flag_set(m, PG_REFERENCED);
1837 
1838 				/* Flush mapping from TLB0. */
1839 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1840 				    PTE_REFERENCED);
1841 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
1842 			}
1843 		}
1844 		PMAP_UNLOCK(pv->pv_pmap);
1845 	}
1846 	vm_page_flag_clear(m, PG_WRITEABLE);
1847 }
1848 
1849 static boolean_t
1850 mmu_booke_page_executable(mmu_t mmu, vm_page_t m)
1851 {
1852 	pv_entry_t pv;
1853 	pte_t *pte;
1854 	boolean_t executable;
1855 
1856 	executable = FALSE;
1857 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1858 		PMAP_LOCK(pv->pv_pmap);
1859 		pte = pte_find(mmu, pv->pv_pmap, pv->pv_va);
1860 		if (pte != NULL && PTE_ISVALID(pte) && (pte->flags & PTE_UX))
1861 			executable = TRUE;
1862 		PMAP_UNLOCK(pv->pv_pmap);
1863 		if (executable)
1864 			break;
1865 	}
1866 
1867 	return (executable);
1868 }
1869 
1870 /*
1871  * Atomically extract and hold the physical page with the given
1872  * pmap and virtual address pair if that mapping permits the given
1873  * protection.
1874  */
1875 static vm_page_t
1876 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
1877     vm_prot_t prot)
1878 {
1879 	pte_t *pte;
1880 	vm_page_t m;
1881 	u_int32_t pte_wbit;
1882 
1883 	m = NULL;
1884 	vm_page_lock_queues();
1885 	PMAP_LOCK(pmap);
1886 	pte = pte_find(mmu, pmap, va);
1887 
1888 	if ((pte != NULL) && PTE_ISVALID(pte)) {
1889 		if (pmap == kernel_pmap)
1890 			pte_wbit = PTE_SW;
1891 		else
1892 			pte_wbit = PTE_UW;
1893 
1894 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
1895 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1896 			vm_page_hold(m);
1897 		}
1898 	}
1899 
1900 	vm_page_unlock_queues();
1901 	PMAP_UNLOCK(pmap);
1902 	return (m);
1903 }
1904 
1905 /*
1906  * Initialize a vm_page's machine-dependent fields.
1907  */
1908 static void
1909 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
1910 {
1911 
1912 	TAILQ_INIT(&m->md.pv_list);
1913 }
1914 
1915 /*
1916  * mmu_booke_zero_page_area zeros the specified hardware page by
1917  * mapping it into virtual memory and using bzero to clear
1918  * its contents.
1919  *
1920  * off and size must reside within a single page.
1921  */
1922 static void
1923 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1924 {
1925 	vm_offset_t va;
1926 
1927 	//debugf("mmu_booke_zero_page_area: s\n");
1928 
1929 	mtx_lock(&zero_page_mutex);
1930 	va = zero_page_va;
1931 
1932 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
1933 	bzero((caddr_t)va + off, size);
1934 	mmu_booke_kremove(mmu, va);
1935 
1936 	mtx_unlock(&zero_page_mutex);
1937 
1938 	//debugf("mmu_booke_zero_page_area: e\n");
1939 }
1940 
1941 /*
1942  * mmu_booke_zero_page zeros the specified hardware page.
1943  */
1944 static void
1945 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
1946 {
1947 
1948 	//debugf("mmu_booke_zero_page: s\n");
1949 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
1950 	//debugf("mmu_booke_zero_page: e\n");
1951 }
1952 
1953 /*
1954  * mmu_booke_copy_page copies the specified (machine independent) page by
1955  * mapping the page into virtual memory and using memcopy to copy the page,
1956  * one machine dependent page at a time.
1957  */
1958 static void
1959 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
1960 {
1961 	vm_offset_t sva, dva;
1962 
1963 	//debugf("mmu_booke_copy_page: s\n");
1964 
1965 	mtx_lock(&copy_page_mutex);
1966 	sva = copy_page_src_va;
1967 	dva = copy_page_dst_va;
1968 
1969 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
1970 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
1971 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
1972 	mmu_booke_kremove(mmu, dva);
1973 	mmu_booke_kremove(mmu, sva);
1974 
1975 	mtx_unlock(&copy_page_mutex);
1976 
1977 	//debugf("mmu_booke_copy_page: e\n");
1978 }
1979 
1980 #if 0
1981 /*
1982  * Remove all pages from specified address space, this aids process exit
1983  * speeds. This is much faster than mmu_booke_remove in the case of running
1984  * down an entire address space. Only works for the current pmap.
1985  */
1986 void
1987 mmu_booke_remove_pages(pmap_t pmap)
1988 {
1989 }
1990 #endif
1991 
1992 /*
1993  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
1994  * into virtual memory and using bzero to clear its contents. This is intended
1995  * to be called from the vm_pagezero process only and outside of Giant. No
1996  * lock is required.
1997  */
1998 static void
1999 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2000 {
2001 	vm_offset_t va;
2002 
2003 	//debugf("mmu_booke_zero_page_idle: s\n");
2004 
2005 	va = zero_page_idle_va;
2006 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2007 	bzero((caddr_t)va, PAGE_SIZE);
2008 	mmu_booke_kremove(mmu, va);
2009 
2010 	//debugf("mmu_booke_zero_page_idle: e\n");
2011 }
2012 
2013 /*
2014  * Return whether or not the specified physical page was modified
2015  * in any of physical maps.
2016  */
2017 static boolean_t
2018 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2019 {
2020 	pte_t *pte;
2021 	pv_entry_t pv;
2022 
2023 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2024 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2025 		return (FALSE);
2026 
2027 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2028 		PMAP_LOCK(pv->pv_pmap);
2029 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2030 			if (!PTE_ISVALID(pte))
2031 				goto make_sure_to_unlock;
2032 
2033 			if (PTE_ISMODIFIED(pte)) {
2034 				PMAP_UNLOCK(pv->pv_pmap);
2035 				return (TRUE);
2036 			}
2037 		}
2038 make_sure_to_unlock:
2039 		PMAP_UNLOCK(pv->pv_pmap);
2040 	}
2041 	return (FALSE);
2042 }
2043 
2044 /*
2045  * Return whether or not the specified virtual address is elgible
2046  * for prefault.
2047  */
2048 static boolean_t
2049 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2050 {
2051 
2052 	return (FALSE);
2053 }
2054 
2055 /*
2056  * Clear the modify bits on the specified physical page.
2057  */
2058 static void
2059 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2060 {
2061 	pte_t *pte;
2062 	pv_entry_t pv;
2063 
2064 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2065 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2066 		return;
2067 
2068 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2069 		PMAP_LOCK(pv->pv_pmap);
2070 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2071 			if (!PTE_ISVALID(pte))
2072 				goto make_sure_to_unlock;
2073 
2074 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2075 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2076 				    PTE_REFERENCED);
2077 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2078 			}
2079 		}
2080 make_sure_to_unlock:
2081 		PMAP_UNLOCK(pv->pv_pmap);
2082 	}
2083 }
2084 
2085 /*
2086  * Return a count of reference bits for a page, clearing those bits.
2087  * It is not necessary for every reference bit to be cleared, but it
2088  * is necessary that 0 only be returned when there are truly no
2089  * reference bits set.
2090  *
2091  * XXX: The exact number of bits to check and clear is a matter that
2092  * should be tested and standardized at some point in the future for
2093  * optimal aging of shared pages.
2094  */
2095 static int
2096 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2097 {
2098 	pte_t *pte;
2099 	pv_entry_t pv;
2100 	int count;
2101 
2102 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2103 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2104 		return (0);
2105 
2106 	count = 0;
2107 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2108 		PMAP_LOCK(pv->pv_pmap);
2109 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2110 			if (!PTE_ISVALID(pte))
2111 				goto make_sure_to_unlock;
2112 
2113 			if (PTE_ISREFERENCED(pte)) {
2114 				pte->flags &= ~PTE_REFERENCED;
2115 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2116 
2117 				if (++count > 4) {
2118 					PMAP_UNLOCK(pv->pv_pmap);
2119 					break;
2120 				}
2121 			}
2122 		}
2123 make_sure_to_unlock:
2124 		PMAP_UNLOCK(pv->pv_pmap);
2125 	}
2126 	return (count);
2127 }
2128 
2129 /*
2130  * Clear the reference bit on the specified physical page.
2131  */
2132 static void
2133 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2134 {
2135 	pte_t *pte;
2136 	pv_entry_t pv;
2137 
2138 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2139 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2140 		return;
2141 
2142 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2143 		PMAP_LOCK(pv->pv_pmap);
2144 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2145 			if (!PTE_ISVALID(pte))
2146 				goto make_sure_to_unlock;
2147 
2148 			if (PTE_ISREFERENCED(pte)) {
2149 				pte->flags &= ~PTE_REFERENCED;
2150 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2151 			}
2152 		}
2153 make_sure_to_unlock:
2154 		PMAP_UNLOCK(pv->pv_pmap);
2155 	}
2156 }
2157 
2158 /*
2159  * Change wiring attribute for a map/virtual-address pair.
2160  */
2161 static void
2162 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2163 {
2164 	pte_t *pte;;
2165 
2166 	PMAP_LOCK(pmap);
2167 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2168 		if (wired) {
2169 			if (!PTE_ISWIRED(pte)) {
2170 				pte->flags |= PTE_WIRED;
2171 				pmap->pm_stats.wired_count++;
2172 			}
2173 		} else {
2174 			if (PTE_ISWIRED(pte)) {
2175 				pte->flags &= ~PTE_WIRED;
2176 				pmap->pm_stats.wired_count--;
2177 			}
2178 		}
2179 	}
2180 	PMAP_UNLOCK(pmap);
2181 }
2182 
2183 /*
2184  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2185  * page.  This count may be changed upwards or downwards in the future; it is
2186  * only necessary that true be returned for a small subset of pmaps for proper
2187  * page aging.
2188  */
2189 static boolean_t
2190 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2191 {
2192 	pv_entry_t pv;
2193 	int loops;
2194 
2195 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2196 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2197 		return (FALSE);
2198 
2199 	loops = 0;
2200 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2201 
2202 		if (pv->pv_pmap == pmap)
2203 			return (TRUE);
2204 
2205 		if (++loops >= 16)
2206 			break;
2207 	}
2208 	return (FALSE);
2209 }
2210 
2211 /*
2212  * Return the number of managed mappings to the given physical page that are
2213  * wired.
2214  */
2215 static int
2216 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2217 {
2218 	pv_entry_t pv;
2219 	pte_t *pte;
2220 	int count = 0;
2221 
2222 	if ((m->flags & PG_FICTITIOUS) != 0)
2223 		return (count);
2224 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2225 
2226 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2227 		PMAP_LOCK(pv->pv_pmap);
2228 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2229 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2230 				count++;
2231 		PMAP_UNLOCK(pv->pv_pmap);
2232 	}
2233 
2234 	return (count);
2235 }
2236 
2237 static int
2238 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2239 {
2240 	int i;
2241 	vm_offset_t va;
2242 
2243 	/*
2244 	 * This currently does not work for entries that
2245 	 * overlap TLB1 entries.
2246 	 */
2247 	for (i = 0; i < tlb1_idx; i ++) {
2248 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2249 			return (0);
2250 	}
2251 
2252 	return (EFAULT);
2253 }
2254 
2255 /*
2256  * Map a set of physical memory pages into the kernel virtual address space.
2257  * Return a pointer to where it is mapped. This routine is intended to be used
2258  * for mapping device memory, NOT real memory.
2259  */
2260 static void *
2261 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2262 {
2263 	uintptr_t va;
2264 
2265 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2266 	if (bootverbose)
2267 		printf("Wiring VA=%x to PA=%x (size=%x), using TLB1[%d]\n",
2268 		    va, pa, size, tlb1_idx);
2269 	tlb1_set_entry(va, pa, size, _TLB_ENTRY_IO);
2270 	return ((void *)va);
2271 }
2272 
2273 /*
2274  * 'Unmap' a range mapped by mmu_booke_mapdev().
2275  */
2276 static void
2277 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2278 {
2279 	vm_offset_t base, offset;
2280 
2281 	//debugf("mmu_booke_unmapdev: s (va = 0x%08x)\n", va);
2282 
2283 	/*
2284 	 * Unmap only if this is inside kernel virtual space.
2285 	 */
2286 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2287 		base = trunc_page(va);
2288 		offset = va & PAGE_MASK;
2289 		size = roundup(offset + size, PAGE_SIZE);
2290 		kmem_free(kernel_map, base, size);
2291 	}
2292 
2293 	//debugf("mmu_booke_unmapdev: e\n");
2294 }
2295 
2296 /*
2297  * mmu_booke_object_init_pt preloads the ptes for a given object
2298  * into the specified pmap. This eliminates the blast of soft
2299  * faults on process startup and immediately after an mmap.
2300  */
2301 static void
2302 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2303     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2304 {
2305 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2306 	KASSERT(object->type == OBJT_DEVICE,
2307 	    ("mmu_booke_object_init_pt: non-device object"));
2308 }
2309 
2310 /*
2311  * Perform the pmap work for mincore.
2312  */
2313 static int
2314 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2315 {
2316 
2317 	TODO;
2318 	return (0);
2319 }
2320 
2321 /**************************************************************************/
2322 /* TID handling */
2323 /**************************************************************************/
2324 /*
2325  * Flush all entries from TLB0 matching given tid.
2326  */
2327 static void
2328 tid_flush(tlbtid_t tid)
2329 {
2330 	int i, entryidx, way;
2331 
2332 	//debugf("tid_flush: s (tid = %d)\n", tid);
2333 
2334 	mtx_lock_spin(&tlb0_mutex);
2335 
2336 	for (i = 0; i < TLB0_SIZE; i++) {
2337 		if (MAS1_GETTID(tlb0[i].mas1) == tid) {
2338 			way = i / TLB0_ENTRIES_PER_WAY;
2339 			entryidx = i - (way * TLB0_ENTRIES_PER_WAY);
2340 
2341 			//debugf("tid_flush: inval tlb0 entry %d\n", i);
2342 			tlb0_inval_entry(entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT, way);
2343 		}
2344 	}
2345 
2346 	mtx_unlock_spin(&tlb0_mutex);
2347 
2348 	//debugf("tid_flush: e\n");
2349 }
2350 
2351 /*
2352  * Allocate a TID. If necessary, steal one from someone else.
2353  * The new TID is flushed from the TLB before returning.
2354  */
2355 static tlbtid_t
2356 tid_alloc(pmap_t pmap)
2357 {
2358 	tlbtid_t tid;
2359 	static tlbtid_t next_tid = TID_MIN;
2360 
2361 	//struct thread *td;
2362 	//struct proc *p;
2363 
2364 	//td = PCPU_GET(curthread);
2365 	//p = td->td_proc;
2366 	//debugf("tid_alloc: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
2367 	//printf("tid_alloc: proc %d '%s'\n", p->p_pid, p->p_comm);
2368 
2369 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2370 
2371 	/*
2372 	 * Find a likely TID, allocate unused if possible,
2373 	 * skip reserved entries.
2374 	 */
2375 	tid = next_tid;
2376 	while (tidbusy[tid] != NULL) {
2377 		if (tid == next_tid)
2378 			break;
2379 
2380 		if (tid == TID_MAX)
2381 			tid = TID_MIN;
2382 		else
2383 			tid++;
2384 
2385 	}
2386 
2387 	/* Now clean it out */
2388 	tid_flush(tid);
2389 
2390 	/* If we are stealing pmap then clear its tid */
2391 	if (tidbusy[tid]) {
2392 		//debugf("warning: stealing tid %d\n", tid);
2393 		tidbusy[tid]->pm_tid = 0;
2394 	}
2395 
2396 	/* Calculate next tid */
2397 	if (tid == TID_MAX)
2398 		next_tid = TID_MIN;
2399 	else
2400 		next_tid = tid + 1;
2401 
2402 	tidbusy[tid] = pmap;
2403 	pmap->pm_tid = tid;
2404 
2405 	//debugf("tid_alloc: e (%02d next = %02d)\n", tid, next_tid);
2406 	return (tid);
2407 }
2408 
2409 #if 0
2410 /*
2411  * Free this pmap's TID.
2412  */
2413 static void
2414 tid_free(pmap_t pmap)
2415 {
2416 	tlbtid_t oldtid;
2417 
2418 	oldtid = pmap->pm_tid;
2419 
2420 	if (oldtid == 0) {
2421 		panic("tid_free: freeing kernel tid");
2422 	}
2423 
2424 #ifdef DEBUG
2425 	if (tidbusy[oldtid] == 0)
2426 		debugf("tid_free: freeing free tid %d\n", oldtid);
2427 	if (tidbusy[oldtid] != pmap) {
2428 		debugf("tid_free: freeing someone esle's tid\n "
2429 		       "tidbusy[%d] = 0x%08x pmap = 0x%08x\n",
2430 		       oldtid, (u_int32_t)tidbusy[oldtid], (u_int32_t)pmap);
2431 	}
2432 #endif
2433 
2434 	tidbusy[oldtid] = NULL;
2435 	tid_flush(oldtid);
2436 }
2437 #endif
2438 
2439 #if 0
2440 #if DEBUG
2441 static void
2442 tid_print_busy(void)
2443 {
2444 	int i;
2445 
2446 	for (i = 0; i < TID_MAX; i++) {
2447 		debugf("tid %d = pmap 0x%08x", i, (u_int32_t)tidbusy[i]);
2448 		if (tidbusy[i])
2449 			debugf(" pmap->tid = %d", tidbusy[i]->pm_tid);
2450 		debugf("\n");
2451 	}
2452 
2453 }
2454 #endif /* DEBUG */
2455 #endif
2456 
2457 /**************************************************************************/
2458 /* TLB0 handling */
2459 /**************************************************************************/
2460 
2461 static void
2462 tlb_print_entry(int i, u_int32_t mas1, u_int32_t mas2, u_int32_t mas3, u_int32_t mas7)
2463 {
2464 	int as;
2465 	char desc[3];
2466 	tlbtid_t tid;
2467 	vm_size_t size;
2468 	unsigned int tsize;
2469 
2470 	desc[2] = '\0';
2471 	if (mas1 & MAS1_VALID)
2472 		desc[0] = 'V';
2473 	else
2474 		desc[0] = ' ';
2475 
2476 	if (mas1 & MAS1_IPROT)
2477 		desc[1] = 'P';
2478 	else
2479 		desc[1] = ' ';
2480 
2481 	as = (mas1 & MAS1_TS) ? 1 : 0;
2482 	tid = MAS1_GETTID(mas1);
2483 
2484 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2485 	size = 0;
2486 	if (tsize)
2487 		size = tsize2size(tsize);
2488 
2489 	debugf("%3d: (%s) [AS=%d] "
2490 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2491 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2492 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2493 }
2494 
2495 /* Convert TLB0 va and way number to tlb0[] table index. */
2496 static inline unsigned int
2497 tlb0_tableidx(vm_offset_t va, unsigned int way)
2498 {
2499 	unsigned int idx;
2500 
2501 	idx = (way * TLB0_ENTRIES_PER_WAY);
2502 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2503 	return (idx);
2504 }
2505 
2506 /*
2507  * Write given entry to TLB0 hardware.
2508  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2509  */
2510 static void
2511 tlb0_write_entry(unsigned int idx, unsigned int way)
2512 {
2513 	u_int32_t mas0, mas7, nv;
2514 
2515 	/* Clear high order RPN bits. */
2516 	mas7 = 0;
2517 
2518 	/* Preserve NV. */
2519 	mas0 = mfspr(SPR_MAS0);
2520 	nv = mas0 & (TLB0_NWAYS - 1);
2521 
2522 	/* Select entry. */
2523 	mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way) | nv;
2524 
2525 	//debugf("tlb0_write_entry: s (idx=%d way=%d mas0=0x%08x "
2526 	//		"mas1=0x%08x mas2=0x%08x mas3=0x%08x)\n",
2527 	//		idx, way, mas0, tlb0[idx].mas1,
2528 	//		tlb0[idx].mas2, tlb0[idx].mas3);
2529 
2530 	mtspr(SPR_MAS0, mas0);
2531 	__asm volatile("isync");
2532 	mtspr(SPR_MAS1, tlb0[idx].mas1);
2533 	__asm volatile("isync");
2534 	mtspr(SPR_MAS2, tlb0[idx].mas2);
2535 	__asm volatile("isync");
2536 	mtspr(SPR_MAS3, tlb0[idx].mas3);
2537 	__asm volatile("isync");
2538 	mtspr(SPR_MAS7, mas7);
2539 	__asm volatile("isync; tlbwe; isync; msync");
2540 
2541 	//debugf("tlb0_write_entry: e\n");
2542 }
2543 
2544 /*
2545  * Invalidate TLB0 entry, clear correspondig tlb0 table element.
2546  */
2547 static void
2548 tlb0_inval_entry(vm_offset_t va, unsigned int way)
2549 {
2550 	int idx = tlb0_tableidx(va, way);
2551 
2552 	//debugf("tlb0_inval_entry: s (va=0x%08x way=%d idx=%d)\n",
2553 	//		va, way, idx);
2554 
2555 	tlb0[idx].mas1 = 1 << MAS1_TSIZE_SHIFT;	/* !MAS1_VALID */
2556 	tlb0[idx].mas2 = va & MAS2_EPN;
2557 	tlb0[idx].mas3 = 0;
2558 
2559 	tlb0_write_entry(idx, way);
2560 
2561 	//debugf("tlb0_inval_entry: e\n");
2562 }
2563 
2564 /*
2565  * Invalidate TLB0 entry that corresponds to pmap/va.
2566  */
2567 static void
2568 tlb0_flush_entry(pmap_t pmap, vm_offset_t va)
2569 {
2570 	int idx, way;
2571 
2572 	//debugf("tlb0_flush_entry: s (pmap=0x%08x va=0x%08x)\n",
2573 	//		(u_int32_t)pmap, va);
2574 
2575 	mtx_lock_spin(&tlb0_mutex);
2576 
2577 	/* Check all TLB0 ways. */
2578 	for (way = 0; way < TLB0_NWAYS; way ++) {
2579 		idx = tlb0_tableidx(va, way);
2580 
2581 		/* Invalidate only if entry matches va and pmap tid. */
2582 		if (((MAS1_GETTID(tlb0[idx].mas1) == pmap->pm_tid) &&
2583 				((tlb0[idx].mas2 & MAS2_EPN) == va))) {
2584 			tlb0_inval_entry(va, way);
2585 		}
2586 	}
2587 
2588 	mtx_unlock_spin(&tlb0_mutex);
2589 
2590 	//debugf("tlb0_flush_entry: e\n");
2591 }
2592 
2593 /* Clean TLB0 hardware and tlb0[] table. */
2594 static void
2595 tlb0_init(void)
2596 {
2597 	int entryidx, way;
2598 
2599 	debugf("tlb0_init: TLB0_SIZE = %d TLB0_NWAYS = %d\n",
2600 	    TLB0_SIZE, TLB0_NWAYS);
2601 
2602 	mtx_lock_spin(&tlb0_mutex);
2603 
2604 	for (way = 0; way < TLB0_NWAYS; way ++) {
2605 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2606 			tlb0_inval_entry(entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT, way);
2607 		}
2608 	}
2609 
2610 	mtx_unlock_spin(&tlb0_mutex);
2611 }
2612 
2613 #if 0
2614 #if DEBUG
2615 /* Print out tlb0 entries for given va. */
2616 static void
2617 tlb0_print_tlbentries_va(vm_offset_t va)
2618 {
2619 	u_int32_t mas0, mas1, mas2, mas3, mas7;
2620 	int way, idx;
2621 
2622 	debugf("TLB0 entries for va = 0x%08x:\n", va);
2623 	for (way = 0; way < TLB0_NWAYS; way ++) {
2624 		mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2625 		mtspr(SPR_MAS0, mas0);
2626 		__asm volatile("isync");
2627 
2628 		mas2 = va & MAS2_EPN;
2629 		mtspr(SPR_MAS2, mas2);
2630 		__asm volatile("isync; tlbre");
2631 
2632 		mas1 = mfspr(SPR_MAS1);
2633 		mas2 = mfspr(SPR_MAS2);
2634 		mas3 = mfspr(SPR_MAS3);
2635 		mas7 = mfspr(SPR_MAS7);
2636 
2637 		idx = tlb0_tableidx(va, way);
2638 		tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2639 	}
2640 }
2641 
2642 /* Print out contents of the MAS registers for each TLB0 entry */
2643 static void
2644 tlb0_print_tlbentries(void)
2645 {
2646 	u_int32_t mas0, mas1, mas2, mas3, mas7;
2647 	int entryidx, way, idx;
2648 
2649 	debugf("TLB0 entries:\n");
2650 	for (way = 0; way < TLB0_NWAYS; way ++) {
2651 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2652 
2653 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2654 			mtspr(SPR_MAS0, mas0);
2655 			__asm volatile("isync");
2656 
2657 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2658 			mtspr(SPR_MAS2, mas2);
2659 
2660 			__asm volatile("isync; tlbre");
2661 
2662 			mas1 = mfspr(SPR_MAS1);
2663 			mas2 = mfspr(SPR_MAS2);
2664 			mas3 = mfspr(SPR_MAS3);
2665 			mas7 = mfspr(SPR_MAS7);
2666 
2667 			idx = tlb0_tableidx(mas2, way);
2668 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2669 		}
2670 	}
2671 }
2672 
2673 /* Print out kernel tlb0[] table. */
2674 static void
2675 tlb0_print_entries(void)
2676 {
2677 	int i;
2678 
2679 	debugf("tlb0[] table entries:\n");
2680 	for (i = 0; i < TLB0_SIZE; i++) {
2681 		tlb_print_entry(i, tlb0[i].mas1,
2682 				tlb0[i].mas2, tlb0[i].mas3, 0);
2683 	}
2684 }
2685 #endif /* DEBUG */
2686 #endif
2687 
2688 /**************************************************************************/
2689 /* TLB1 handling */
2690 /**************************************************************************/
2691 /*
2692  * Write given entry to TLB1 hardware.
2693  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2694  */
2695 static void
2696 tlb1_write_entry(unsigned int idx)
2697 {
2698 	u_int32_t mas0, mas7;
2699 
2700 	//debugf("tlb1_write_entry: s\n");
2701 
2702 	/* Clear high order RPN bits */
2703 	mas7 = 0;
2704 
2705 	/* Select entry */
2706 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2707 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2708 
2709 	mtspr(SPR_MAS0, mas0);
2710 	__asm volatile("isync");
2711 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2712 	__asm volatile("isync");
2713 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2714 	__asm volatile("isync");
2715 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2716 	__asm volatile("isync");
2717 	mtspr(SPR_MAS7, mas7);
2718 	__asm volatile("isync; tlbwe; isync; msync");
2719 
2720 	//debugf("tlb1_write_entry: e\n");;
2721 }
2722 
2723 /*
2724  * Return the largest uint value log such that 2^log <= num.
2725  */
2726 static unsigned int
2727 ilog2(unsigned int num)
2728 {
2729 	int lz;
2730 
2731 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2732 	return (31 - lz);
2733 }
2734 
2735 /*
2736  * Convert TLB TSIZE value to mapped region size.
2737  */
2738 static vm_size_t
2739 tsize2size(unsigned int tsize)
2740 {
2741 
2742 	/*
2743 	 * size = 4^tsize KB
2744 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2745 	 */
2746 
2747 	return ((1 << (2 * tsize)) * 1024);
2748 }
2749 
2750 /*
2751  * Convert region size (must be power of 4) to TLB TSIZE value.
2752  */
2753 static unsigned int
2754 size2tsize(vm_size_t size)
2755 {
2756 
2757 	/*
2758 	 * tsize = log2(size) / 2 - 5
2759 	 */
2760 
2761 	return (ilog2(size) / 2 - 5);
2762 }
2763 
2764 /*
2765  * Setup entry in a sw tlb1 table, write entry to TLB1 hardware.
2766  * This routine is used for low level operations on the TLB1,
2767  * for creating temporaray as well as permanent mappings (tlb_set_entry).
2768  *
2769  * We assume kernel mappings only, thus all entries created have supervisor
2770  * permission bits set nad user permission bits cleared.
2771  *
2772  * Provided mapping size must be a power of 4.
2773  * Mapping flags must be a combination of MAS2_[WIMG].
2774  * Entry TID is set to _tid which must not exceed 8 bit value.
2775  * Entry TS is set to either 0 or MAS1_TS based on provided _ts.
2776  */
2777 static void
2778 __tlb1_set_entry(unsigned int idx, vm_offset_t va, vm_offset_t pa,
2779     vm_size_t size, u_int32_t flags, unsigned int _tid, unsigned int _ts)
2780 {
2781 	int tsize;
2782 	u_int32_t ts, tid;
2783 
2784 	//debugf("__tlb1_set_entry: s (idx = %d va = 0x%08x pa = 0x%08x "
2785 	//		"size = 0x%08x flags = 0x%08x _tid = %d _ts = %d\n",
2786 	//		idx, va, pa, size, flags, _tid, _ts);
2787 
2788 	/* Convert size to TSIZE */
2789 	tsize = size2tsize(size);
2790 	//debugf("__tlb1_set_entry: tsize = %d\n", tsize);
2791 
2792 	tid = (_tid <<  MAS1_TID_SHIFT) & MAS1_TID_MASK;
2793 	ts = (_ts) ? MAS1_TS : 0;
2794 	tlb1[idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2795 	tlb1[idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2796 
2797 	tlb1[idx].mas2 = (va & MAS2_EPN) | flags;
2798 
2799 	/* Set supervisor rwx permission bits */
2800 	tlb1[idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2801 
2802 	//debugf("__tlb1_set_entry: mas1 = %08x mas2 = %08x mas3 = 0x%08x\n",
2803 	//		tlb1[idx].mas1, tlb1[idx].mas2, tlb1[idx].mas3);
2804 
2805 	tlb1_write_entry(idx);
2806 	//debugf("__tlb1_set_entry: e\n");
2807 }
2808 
2809 /*
2810  * Register permanent kernel mapping in TLB1.
2811  *
2812  * Entries are created starting from index 0 (current free entry is
2813  * kept in tlb1_idx) and are not supposed to be invalidated.
2814  */
2815 static int
2816 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size, u_int32_t flags)
2817 {
2818 	//debugf("tlb1_set_entry: s (tlb1_idx = %d va = 0x%08x pa = 0x%08x "
2819 	//		"size = 0x%08x flags = 0x%08x\n",
2820 	//		tlb1_idx, va, pa, size, flags);
2821 
2822 	if (tlb1_idx >= TLB1_SIZE) {
2823 		//debugf("tlb1_set_entry: e (tlb1 full!)\n");
2824 		return (-1);
2825 	}
2826 
2827 	/* TS = 0, TID = 0 */
2828 	__tlb1_set_entry(tlb1_idx++, va, pa, size, flags, KERNEL_TID, 0);
2829 	//debugf("tlb1_set_entry: e\n");
2830 	return (0);
2831 }
2832 
2833 /*
2834  * Invalidate TLB1 entry, clear correspondig tlb1 table element.
2835  * This routine is used to clear temporary entries created
2836  * early in a locore.S or through the use of __tlb1_set_entry().
2837  */
2838 void
2839 tlb1_inval_entry(unsigned int idx)
2840 {
2841 	vm_offset_t va;
2842 
2843 	va = tlb1[idx].mas2 & MAS2_EPN;
2844 
2845 	tlb1[idx].mas1 = 0; /* !MAS1_VALID */
2846 	tlb1[idx].mas2 = 0;
2847 	tlb1[idx].mas3 = 0;
2848 
2849 	tlb1_write_entry(idx);
2850 }
2851 
2852 static int
2853 tlb1_entry_size_cmp(const void *a, const void *b)
2854 {
2855 	const vm_size_t *sza;
2856 	const vm_size_t *szb;
2857 
2858 	sza = a;
2859 	szb = b;
2860 	if (*sza > *szb)
2861 		return (-1);
2862 	else if (*sza < *szb)
2863 		return (1);
2864 	else
2865 		return (0);
2866 }
2867 
2868 /*
2869  * Mapin contiguous RAM region into the TLB1 using maximum of
2870  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2871  *
2872  * If necessarry round up last entry size and return total size
2873  * used by all allocated entries.
2874  */
2875 vm_size_t
2876 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2877 {
2878 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2879 	vm_size_t mapped_size, sz, esz;
2880 	unsigned int log;
2881 	int i;
2882 
2883 	debugf("tlb1_mapin_region:\n");
2884 	debugf(" region size = 0x%08x va = 0x%08x pa = 0x%08x\n", size, va, pa);
2885 
2886 	mapped_size = 0;
2887 	sz = size;
2888 	memset(entry_size, 0, sizeof(entry_size));
2889 
2890 	/* Calculate entry sizes. */
2891 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2892 
2893 		/* Largest region that is power of 4 and fits within size */
2894 		log = ilog2(sz)/2;
2895 		esz = 1 << (2 * log);
2896 
2897 		/* Minimum region size is 4KB */
2898 		if (esz < (1 << 12))
2899 			esz = 1 << 12;
2900 
2901 		/* If this is last entry cover remaining size. */
2902 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2903 			while (esz < sz)
2904 				esz = esz << 2;
2905 		}
2906 
2907 		entry_size[i] = esz;
2908 		mapped_size += esz;
2909 		if (esz < sz)
2910 			sz -= esz;
2911 		else
2912 			sz = 0;
2913 	}
2914 
2915 	/* Sort entry sizes, required to get proper entry address alignment. */
2916 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2917 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2918 
2919 	/* Load TLB1 entries. */
2920 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
2921 		esz = entry_size[i];
2922 		if (!esz)
2923 			break;
2924 		debugf("  entry %d: sz  = 0x%08x (va = 0x%08x pa = 0x%08x)\n",
2925 		    tlb1_idx, esz, va, pa);
2926 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
2927 
2928 		va += esz;
2929 		pa += esz;
2930 	}
2931 
2932 	debugf(" mapped size 0x%08x (wasted space 0x%08x)\n",
2933 	    mapped_size, mapped_size - size);
2934 
2935 	return (mapped_size);
2936 }
2937 
2938 /*
2939  * TLB1 initialization routine, to be called after the very first
2940  * assembler level setup done in locore.S.
2941  */
2942 void
2943 tlb1_init(vm_offset_t ccsrbar)
2944 {
2945 	uint32_t mas0;
2946 
2947 	/* TBL1[1] is used to map the kernel. Save that entry. */
2948 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(1);
2949 	mtspr(SPR_MAS0, mas0);
2950 	__asm __volatile("isync; tlbre");
2951 
2952 	tlb1[1].mas1 = mfspr(SPR_MAS1);
2953 	tlb1[1].mas2 = mfspr(SPR_MAS2);
2954 	tlb1[1].mas3 = mfspr(SPR_MAS3);
2955 
2956 	/* Mapin CCSRBAR in TLB1[0] */
2957 	__tlb1_set_entry(0, CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE,
2958 	    _TLB_ENTRY_IO, KERNEL_TID, 0);
2959 
2960 	/* Setup TLB miss defaults */
2961 	set_mas4_defaults();
2962 
2963 	/* Reset next available TLB1 entry index. */
2964 	tlb1_idx = 2;
2965 }
2966 
2967 /*
2968  * Setup MAS4 defaults.
2969  * These values are loaded to MAS0-2 on a TLB miss.
2970  */
2971 static void
2972 set_mas4_defaults(void)
2973 {
2974 	u_int32_t mas4;
2975 
2976 	/* Defaults: TLB0, PID0, TSIZED=4K */
2977 	mas4 = MAS4_TLBSELD0;
2978 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
2979 
2980 	mtspr(SPR_MAS4, mas4);
2981 	__asm volatile("isync");
2982 }
2983 
2984 /*
2985  * Print out contents of the MAS registers for each TLB1 entry
2986  */
2987 void
2988 tlb1_print_tlbentries(void)
2989 {
2990 	u_int32_t mas0, mas1, mas2, mas3, mas7;
2991 	int i;
2992 
2993 	debugf("TLB1 entries:\n");
2994 	for (i = 0; i < TLB1_SIZE; i++) {
2995 
2996 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
2997 		mtspr(SPR_MAS0, mas0);
2998 
2999 		__asm volatile("isync; tlbre");
3000 
3001 		mas1 = mfspr(SPR_MAS1);
3002 		mas2 = mfspr(SPR_MAS2);
3003 		mas3 = mfspr(SPR_MAS3);
3004 		mas7 = mfspr(SPR_MAS7);
3005 
3006 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3007 	}
3008 }
3009 
3010 /*
3011  * Print out contents of the in-ram tlb1 table.
3012  */
3013 void
3014 tlb1_print_entries(void)
3015 {
3016 	int i;
3017 
3018 	debugf("tlb1[] table entries:\n");
3019 	for (i = 0; i < TLB1_SIZE; i++)
3020 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3021 }
3022 
3023 /*
3024  * Return 0 if the physical IO range is encompassed by one of the
3025  * the TLB1 entries, otherwise return related error code.
3026  */
3027 static int
3028 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3029 {
3030 	u_int32_t prot;
3031 	vm_paddr_t pa_start;
3032 	vm_paddr_t pa_end;
3033 	unsigned int entry_tsize;
3034 	vm_size_t entry_size;
3035 
3036 	*va = (vm_offset_t)NULL;
3037 
3038 	/* Skip invalid entries */
3039 	if (!(tlb1[i].mas1 & MAS1_VALID))
3040 		return (EINVAL);
3041 
3042 	/*
3043 	 * The entry must be cache-inhibited, guarded, and r/w
3044 	 * so it can function as an i/o page
3045 	 */
3046 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3047 	if (prot != (MAS2_I | MAS2_G))
3048 		return (EPERM);
3049 
3050 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3051 	if (prot != (MAS3_SR | MAS3_SW))
3052 		return (EPERM);
3053 
3054 	/* The address should be within the entry range. */
3055 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3056 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3057 
3058 	entry_size = tsize2size(entry_tsize);
3059 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3060 	pa_end = pa_start + entry_size - 1;
3061 
3062 	if ((pa < pa_start) || ((pa + size) > pa_end))
3063 		return (ERANGE);
3064 
3065 	/* Return virtual address of this mapping. */
3066 	*va = (tlb1[i].mas2 & MAS2_EPN) + (pa - pa_start);
3067 	return (0);
3068 }
3069