xref: /freebsd/sys/powerpc/booke/pmap.c (revision 6c6c03be2ddb04c54e455122799923deaefa4114)
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 a ptbl_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, kstack0_sz;
901 	vm_offset_t kernel_pdir, kstack0;
902 	vm_paddr_t kstack0_phys;
903 
904 	debugf("mmu_booke_bootstrap: entered\n");
905 
906 	/* Align kernel start and end address (kernel image). */
907 	kernelstart = trunc_page(kernelstart);
908 	kernelend = round_page(kernelend);
909 
910 	/* Allocate space for the message buffer. */
911 	msgbufp = (struct msgbuf *)kernelend;
912 	kernelend += MSGBUF_SIZE;
913 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (u_int32_t)msgbufp,
914 	    kernelend);
915 
916 	kernelend = round_page(kernelend);
917 
918 	/* Allocate space for tlb0 table. */
919 	tlb0_get_tlbconf(); /* Read TLB0 size and associativity. */
920 	tlb0 = (tlb_entry_t *)kernelend;
921 	kernelend += 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 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (u_int32_t)ptbl_bufs,
930 	    kernelend);
931 
932 	kernelend = round_page(kernelend);
933 
934 	/* Allocate PTE tables for kernel KVA. */
935 	kernel_pdir = kernelend;
936 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
937 	    PDIR_SIZE - 1) / PDIR_SIZE;
938 	kernelend += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
939 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
940 	debugf(" kernel pdir at 0x%08x\n", kernel_pdir);
941 
942 	if (kernelend - kernelstart > 0x1000000) {
943 		kernelend = (kernelend + 0x3fffff) & ~0x3fffff;
944 		tlb1_mapin_region(kernelstart + 0x1000000,
945 		    kernload + 0x1000000, kernelend - kernelstart - 0x1000000);
946 	} else
947 		kernelend = (kernelend + 0xffffff) & ~0xffffff;
948 
949 	/*
950 	 * Clear the structures - note we can only do it safely after the
951 	 * possible additional TLB1 translations are in place so that
952 	 * all range up to the currently calculated 'kernelend' is covered.
953 	 */
954 	memset((void *)tlb0, 0, sizeof(tlb_entry_t) * tlb0_size);
955 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
956 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
957 
958 	/*******************************************************/
959 	/* Set the start and end of kva. */
960 	/*******************************************************/
961 	virtual_avail = kernelend;
962 	virtual_end = VM_MAX_KERNEL_ADDRESS;
963 
964 	/* Allocate KVA space for page zero/copy operations. */
965 	zero_page_va = virtual_avail;
966 	virtual_avail += PAGE_SIZE;
967 	zero_page_idle_va = virtual_avail;
968 	virtual_avail += PAGE_SIZE;
969 	copy_page_src_va = virtual_avail;
970 	virtual_avail += PAGE_SIZE;
971 	copy_page_dst_va = virtual_avail;
972 	virtual_avail += PAGE_SIZE;
973 
974 	/* Initialize page zero/copy mutexes. */
975 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
976 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
977 
978 	/* Initialize tlb0 table mutex. */
979 	mtx_init(&tlb0_mutex, "tlb0", NULL, MTX_SPIN | MTX_RECURSE);
980 
981 	/* Allocate KVA space for ptbl bufs. */
982 	ptbl_buf_pool_vabase = virtual_avail;
983 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
984 
985 	debugf("ptbl_buf_pool_vabase = 0x%08x\n", ptbl_buf_pool_vabase);
986 	debugf("virtual_avail = %08x\n", virtual_avail);
987 	debugf("virtual_end   = %08x\n", virtual_end);
988 
989 	/* Calculate corresponding physical addresses for the kernel region. */
990 	phys_kernelend = kernload + (kernelend - kernelstart);
991 
992 	debugf("kernel image and allocated data:\n");
993 	debugf(" kernload    = 0x%08x\n", kernload);
994 	debugf(" kernelstart = 0x%08x\n", kernelstart);
995 	debugf(" kernelend   = 0x%08x\n", kernelend);
996 	debugf(" kernel size = 0x%08x\n", kernelend - kernelstart);
997 
998 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
999 		panic("mmu_booke_bootstrap: phys_avail too small");
1000 
1001 	/*
1002 	 * Removed kernel physical address range from avail
1003 	 * regions list. Page align all regions.
1004 	 * Non-page aligned memory isn't very interesting to us.
1005 	 * Also, sort the entries for ascending addresses.
1006 	 */
1007 	sz = 0;
1008 	cnt = availmem_regions_sz;
1009 	debugf("processing avail regions:\n");
1010 	for (mp = availmem_regions; mp->mr_size; mp++) {
1011 		s = mp->mr_start;
1012 		e = mp->mr_start + mp->mr_size;
1013 		debugf(" %08x-%08x -> ", s, e);
1014 		/* Check whether this region holds all of the kernel. */
1015 		if (s < kernload && e > phys_kernelend) {
1016 			availmem_regions[cnt].mr_start = phys_kernelend;
1017 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1018 			e = kernload;
1019 		}
1020 		/* Look whether this regions starts within the kernel. */
1021 		if (s >= kernload && s < phys_kernelend) {
1022 			if (e <= phys_kernelend)
1023 				goto empty;
1024 			s = phys_kernelend;
1025 		}
1026 		/* Now look whether this region ends within the kernel. */
1027 		if (e > kernload && e <= phys_kernelend) {
1028 			if (s >= kernload)
1029 				goto empty;
1030 			e = kernload;
1031 		}
1032 		/* Now page align the start and size of the region. */
1033 		s = round_page(s);
1034 		e = trunc_page(e);
1035 		if (e < s)
1036 			e = s;
1037 		sz = e - s;
1038 		debugf("%08x-%08x = %x\n", s, e, sz);
1039 
1040 		/* Check whether some memory is left here. */
1041 		if (sz == 0) {
1042 		empty:
1043 			memmove(mp, mp + 1,
1044 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1045 			cnt--;
1046 			mp--;
1047 			continue;
1048 		}
1049 
1050 		/* Do an insertion sort. */
1051 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1052 			if (s < mp1->mr_start)
1053 				break;
1054 		if (mp1 < mp) {
1055 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1056 			mp1->mr_start = s;
1057 			mp1->mr_size = sz;
1058 		} else {
1059 			mp->mr_start = s;
1060 			mp->mr_size = sz;
1061 		}
1062 	}
1063 	availmem_regions_sz = cnt;
1064 
1065 	/*******************************************************/
1066 	/* Steal physical memory for kernel stack from the end */
1067 	/* of the first avail region                           */
1068 	/*******************************************************/
1069 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1070 	kstack0_phys = availmem_regions[0].mr_start +
1071 	    availmem_regions[0].mr_size;
1072 	kstack0_phys -= kstack0_sz;
1073 	availmem_regions[0].mr_size -= kstack0_sz;
1074 
1075 	/*******************************************************/
1076 	/* Fill in phys_avail table, based on availmem_regions */
1077 	/*******************************************************/
1078 	phys_avail_count = 0;
1079 	physsz = 0;
1080 	hwphyssz = 0;
1081 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1082 
1083 	debugf("fill in phys_avail:\n");
1084 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1085 
1086 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1087 		    availmem_regions[i].mr_start,
1088 		    availmem_regions[i].mr_start + availmem_regions[i].mr_size,
1089 		    availmem_regions[i].mr_size);
1090 
1091 		if (hwphyssz != 0 &&
1092 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1093 			debugf(" hw.physmem adjust\n");
1094 			if (physsz < hwphyssz) {
1095 				phys_avail[j] = availmem_regions[i].mr_start;
1096 				phys_avail[j + 1] =
1097 				    availmem_regions[i].mr_start +
1098 				    hwphyssz - physsz;
1099 				physsz = hwphyssz;
1100 				phys_avail_count++;
1101 			}
1102 			break;
1103 		}
1104 
1105 		phys_avail[j] = availmem_regions[i].mr_start;
1106 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1107 		    availmem_regions[i].mr_size;
1108 		phys_avail_count++;
1109 		physsz += availmem_regions[i].mr_size;
1110 	}
1111 	physmem = btoc(physsz);
1112 
1113 	/* Calculate the last available physical address. */
1114 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1115 		;
1116 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1117 
1118 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1119 	debugf("phys_avail_count = %d\n", phys_avail_count);
1120 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem, physmem);
1121 
1122 	/*******************************************************/
1123 	/* Initialize (statically allocated) kernel pmap. */
1124 	/*******************************************************/
1125 	PMAP_LOCK_INIT(kernel_pmap);
1126 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1127 
1128 	debugf("kernel_pmap = 0x%08x\n", (u_int32_t)kernel_pmap);
1129 	debugf("kptbl_min = %d, kernel_kptbls = %d\n", kptbl_min, kernel_ptbls);
1130 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1131 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1132 
1133 	/* Initialize kernel pdir */
1134 	for (i = 0; i < kernel_ptbls; i++)
1135 		kernel_pmap->pm_pdir[kptbl_min + i] =
1136 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1137 
1138 	kernel_pmap->pm_tid = KERNEL_TID;
1139 	kernel_pmap->pm_active = ~0;
1140 
1141 	/* Initialize tidbusy with kenel_pmap entry. */
1142 	tidbusy[0] = kernel_pmap;
1143 
1144 	/*******************************************************/
1145 	/* Final setup */
1146 	/*******************************************************/
1147 	/* Enter kstack0 into kernel map, provide guard page */
1148 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1149 	thread0.td_kstack = kstack0;
1150 	thread0.td_kstack_pages = KSTACK_PAGES;
1151 
1152 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1153 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1154 	    kstack0_phys, kstack0_phys + kstack0_sz);
1155 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1156 
1157 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1158 	for (i = 0; i < KSTACK_PAGES; i++) {
1159 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1160 		kstack0 += PAGE_SIZE;
1161 		kstack0_phys += PAGE_SIZE;
1162 	}
1163 
1164 	/* Initialize TLB0 handling. */
1165 	tlb0_init();
1166 
1167 	debugf("mmu_booke_bootstrap: exit\n");
1168 }
1169 
1170 /*
1171  * Get the physical page address for the given pmap/virtual address.
1172  */
1173 static vm_paddr_t
1174 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1175 {
1176 	vm_paddr_t pa;
1177 
1178 	PMAP_LOCK(pmap);
1179 	pa = pte_vatopa(mmu, pmap, va);
1180 	PMAP_UNLOCK(pmap);
1181 
1182 	return (pa);
1183 }
1184 
1185 /*
1186  * Extract the physical page address associated with the given
1187  * kernel virtual address.
1188  */
1189 static vm_paddr_t
1190 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1191 {
1192 
1193 	return (pte_vatopa(mmu, kernel_pmap, va));
1194 }
1195 
1196 /*
1197  * Initialize the pmap module.
1198  * Called by vm_init, to initialize any structures that the pmap
1199  * system needs to map virtual memory.
1200  */
1201 static void
1202 mmu_booke_init(mmu_t mmu)
1203 {
1204 	int shpgperproc = PMAP_SHPGPERPROC;
1205 
1206 	//debugf("mmu_booke_init: s\n");
1207 
1208 	/*
1209 	 * Initialize the address space (zone) for the pv entries.  Set a
1210 	 * high water mark so that the system can recover from excessive
1211 	 * numbers of pv entries.
1212 	 */
1213 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1214 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1215 
1216 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1217 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1218 
1219 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1220 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1221 
1222 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1223 
1224 	/* Pre-fill pvzone with initial number of pv entries. */
1225 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1226 
1227 	/* Initialize ptbl allocation. */
1228 	ptbl_init();
1229 
1230 	//debugf("mmu_booke_init: e\n");
1231 }
1232 
1233 /*
1234  * Map a list of wired pages into kernel virtual address space.  This is
1235  * intended for temporary mappings which do not need page modification or
1236  * references recorded.  Existing mappings in the region are overwritten.
1237  */
1238 static void
1239 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1240 {
1241 	vm_offset_t va;
1242 
1243 	//debugf("mmu_booke_qenter: s (sva = 0x%08x count = %d)\n", sva, count);
1244 
1245 	va = sva;
1246 	while (count-- > 0) {
1247 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1248 		va += PAGE_SIZE;
1249 		m++;
1250 	}
1251 
1252 	//debugf("mmu_booke_qenter: e\n");
1253 }
1254 
1255 /*
1256  * Remove page mappings from kernel virtual address space.  Intended for
1257  * temporary mappings entered by mmu_booke_qenter.
1258  */
1259 static void
1260 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1261 {
1262 	vm_offset_t va;
1263 
1264 	//debugf("mmu_booke_qremove: s (sva = 0x%08x count = %d)\n", sva, count);
1265 
1266 	va = sva;
1267 	while (count-- > 0) {
1268 		mmu_booke_kremove(mmu, va);
1269 		va += PAGE_SIZE;
1270 	}
1271 
1272 	//debugf("mmu_booke_qremove: e\n");
1273 }
1274 
1275 /*
1276  * Map a wired page into kernel virtual address space.
1277  */
1278 static void
1279 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1280 {
1281 	unsigned int pdir_idx = PDIR_IDX(va);
1282 	unsigned int ptbl_idx = PTBL_IDX(va);
1283 	u_int32_t flags;
1284 	pte_t *pte;
1285 
1286 	//debugf("mmu_booke_kenter: s (pdir_idx = %d ptbl_idx = %d va=0x%08x pa=0x%08x)\n",
1287 	//		pdir_idx, ptbl_idx, va, pa);
1288 
1289 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)),
1290 			("mmu_booke_kenter: invalid va"));
1291 
1292 #if 0
1293 	/* assume IO mapping, set I, G bits */
1294 	flags = (PTE_G | PTE_I | PTE_FAKE);
1295 
1296 	/* if mapping is within system memory, do not set I, G bits */
1297 	for (i = 0; i < totalmem_regions_sz; i++) {
1298 		if ((pa >= totalmem_regions[i].mr_start) &&
1299 				(pa < (totalmem_regions[i].mr_start +
1300 				       totalmem_regions[i].mr_size))) {
1301 			flags &= ~(PTE_I | PTE_G | PTE_FAKE);
1302 			break;
1303 		}
1304 	}
1305 #else
1306 	flags = 0;
1307 #endif
1308 
1309 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1310 
1311 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1312 
1313 	if (PTE_ISVALID(pte)) {
1314 		//debugf("mmu_booke_kenter: replacing entry!\n");
1315 
1316 		/* Flush entry from TLB0 */
1317 		tlb0_flush_entry(kernel_pmap, va);
1318 	}
1319 
1320 	pte->rpn = pa & ~PTE_PA_MASK;
1321 	pte->flags = flags;
1322 
1323 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1324 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1325 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1326 
1327 	/* Flush the real memory from the instruction cache. */
1328 	if ((flags & (PTE_I | PTE_G)) == 0) {
1329 		__syncicache((void *)va, PAGE_SIZE);
1330 	}
1331 
1332 	//debugf("mmu_booke_kenter: e\n");
1333 }
1334 
1335 /*
1336  * Remove a page from kernel page table.
1337  */
1338 static void
1339 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1340 {
1341 	unsigned int pdir_idx = PDIR_IDX(va);
1342 	unsigned int ptbl_idx = PTBL_IDX(va);
1343 	pte_t *pte;
1344 
1345 	//debugf("mmu_booke_kremove: s (va = 0x%08x)\n", va);
1346 
1347 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)),
1348 	    ("mmu_booke_kremove: invalid va"));
1349 
1350 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1351 
1352 	if (!PTE_ISVALID(pte)) {
1353 		//debugf("mmu_booke_kremove: e (invalid pte)\n");
1354 		return;
1355 	}
1356 
1357 	/* Invalidate entry in TLB0. */
1358 	tlb0_flush_entry(kernel_pmap, va);
1359 
1360 	pte->flags = 0;
1361 	pte->rpn = 0;
1362 
1363 	//debugf("mmu_booke_kremove: e\n");
1364 }
1365 
1366 /*
1367  * Initialize pmap associated with process 0.
1368  */
1369 static void
1370 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1371 {
1372 	//debugf("mmu_booke_pinit0: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
1373 	mmu_booke_pinit(mmu, pmap);
1374 	PCPU_SET(curpmap, pmap);
1375 	//debugf("mmu_booke_pinit0: e\n");
1376 }
1377 
1378 /*
1379  * Initialize a preallocated and zeroed pmap structure,
1380  * such as one in a vmspace structure.
1381  */
1382 static void
1383 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1384 {
1385 
1386 	//struct thread *td;
1387 	//struct proc *p;
1388 
1389 	//td = PCPU_GET(curthread);
1390 	//p = td->td_proc;
1391 	//debugf("mmu_booke_pinit: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
1392 	//printf("mmu_booke_pinit: proc %d '%s'\n", p->p_pid, p->p_comm);
1393 
1394 	KASSERT((pmap != kernel_pmap), ("mmu_booke_pinit: initializing kernel_pmap"));
1395 
1396 	PMAP_LOCK_INIT(pmap);
1397 	pmap->pm_tid = 0;
1398 	pmap->pm_active = 0;
1399 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1400 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1401 
1402 	TAILQ_INIT(&pmap->ptbl_list);
1403 
1404 	//debugf("mmu_booke_pinit: e\n");
1405 }
1406 
1407 /*
1408  * Release any resources held by the given physical map.
1409  * Called when a pmap initialized by mmu_booke_pinit is being released.
1410  * Should only be called if the map contains no valid mappings.
1411  */
1412 static void
1413 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1414 {
1415 
1416 	//debugf("mmu_booke_release: s\n");
1417 
1418 	PMAP_LOCK_DESTROY(pmap);
1419 
1420 	//debugf("mmu_booke_release: e\n");
1421 }
1422 
1423 #if 0
1424 /* Not needed, kernel page tables are statically allocated. */
1425 void
1426 mmu_booke_growkernel(vm_offset_t maxkvaddr)
1427 {
1428 }
1429 #endif
1430 
1431 /*
1432  * Insert the given physical page at the specified virtual address in the
1433  * target physical map with the protection requested. If specified the page
1434  * will be wired down.
1435  */
1436 static void
1437 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1438     vm_prot_t prot, boolean_t wired)
1439 {
1440 	vm_page_lock_queues();
1441 	PMAP_LOCK(pmap);
1442 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1443 	vm_page_unlock_queues();
1444 	PMAP_UNLOCK(pmap);
1445 }
1446 
1447 static void
1448 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1449     vm_prot_t prot, boolean_t wired)
1450 {
1451 	pte_t *pte;
1452 	vm_paddr_t pa;
1453 	u_int32_t flags;
1454 	int su, sync;
1455 
1456 	pa = VM_PAGE_TO_PHYS(m);
1457 	su = (pmap == kernel_pmap);
1458 	sync = 0;
1459 
1460 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1461 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1462 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1463 	//		(u_int32_t)m, va, pa, prot, wired);
1464 
1465 	if (su) {
1466 		KASSERT(((va >= virtual_avail) && (va <= VM_MAX_KERNEL_ADDRESS)),
1467 				("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1468 	} else {
1469 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1470 				("mmu_booke_enter_locked: user pmap, non user va"));
1471 	}
1472 
1473 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1474 
1475 	/*
1476 	 * If there is an existing mapping, and the physical address has not
1477 	 * changed, must be protection or wiring change.
1478 	 */
1479 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1480 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1481 
1482 		//debugf("mmu_booke_enter_locked: update\n");
1483 
1484 		/* Wiring change, just update stats. */
1485 		if (wired) {
1486 			if (!PTE_ISWIRED(pte)) {
1487 				pte->flags |= PTE_WIRED;
1488 				pmap->pm_stats.wired_count++;
1489 			}
1490 		} else {
1491 			if (PTE_ISWIRED(pte)) {
1492 				pte->flags &= ~PTE_WIRED;
1493 				pmap->pm_stats.wired_count--;
1494 			}
1495 		}
1496 
1497 		/* Save the old bits and clear the ones we're interested in. */
1498 		flags = pte->flags;
1499 		pte->flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1500 
1501 		if (prot & VM_PROT_WRITE) {
1502 			/* Add write permissions. */
1503 			pte->flags |= PTE_SW;
1504 			if (!su)
1505 				pte->flags |= PTE_UW;
1506 		} else {
1507 			/* Handle modified pages, sense modify status. */
1508 			if (PTE_ISMODIFIED(pte))
1509 				vm_page_dirty(m);
1510 		}
1511 
1512 		/* If we're turning on execute permissions, flush the icache. */
1513 		if (prot & VM_PROT_EXECUTE) {
1514 			pte->flags |= PTE_SX;
1515 			if (!su)
1516 				pte->flags |= PTE_UX;
1517 
1518 			if ((flags & (PTE_UX | PTE_SX)) == 0)
1519 				sync++;
1520 		}
1521 
1522 		/* Flush the old mapping from TLB0. */
1523 		pte->flags &= ~PTE_REFERENCED;
1524 		tlb0_flush_entry(pmap, va);
1525 	} else {
1526 		/*
1527 		 * If there is an existing mapping, but its for a different
1528 		 * physical address, pte_enter() will delete the old mapping.
1529 		 */
1530 		//if ((pte != NULL) && PTE_ISVALID(pte))
1531 		//	debugf("mmu_booke_enter_locked: replace\n");
1532 		//else
1533 		//	debugf("mmu_booke_enter_locked: new\n");
1534 
1535 		/* Now set up the flags and install the new mapping. */
1536 		flags = (PTE_SR | PTE_VALID);
1537 
1538 		if (!su)
1539 			flags |= PTE_UR;
1540 
1541 		if (prot & VM_PROT_WRITE) {
1542 			flags |= PTE_SW;
1543 			if (!su)
1544 				flags |= PTE_UW;
1545 		}
1546 
1547 		if (prot & VM_PROT_EXECUTE) {
1548 			flags |= PTE_SX;
1549 			if (!su)
1550 				flags |= PTE_UX;
1551 		}
1552 
1553 		/* If its wired update stats. */
1554 		if (wired) {
1555 			pmap->pm_stats.wired_count++;
1556 			flags |= PTE_WIRED;
1557 		}
1558 
1559 		pte_enter(mmu, pmap, m, va, flags);
1560 
1561 		/* Flush the real memory from the instruction cache. */
1562 		if (prot & VM_PROT_EXECUTE)
1563 			sync++;
1564 	}
1565 
1566 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1567 		__syncicache((void *)va, PAGE_SIZE);
1568 		sync = 0;
1569 	}
1570 
1571 	if (sync) {
1572 		/* Create a temporary mapping. */
1573 		pmap = PCPU_GET(curpmap);
1574 
1575 		va = 0;
1576 		pte = pte_find(mmu, pmap, va);
1577 		KASSERT(pte == NULL, ("%s:%d", __func__, __LINE__));
1578 
1579 		flags = PTE_SR | PTE_VALID | PTE_UR;
1580 		pte_enter(mmu, pmap, m, va, flags);
1581 		__syncicache((void *)va, PAGE_SIZE);
1582 		pte_remove(mmu, pmap, va, PTBL_UNHOLD);
1583 	}
1584 
1585 	//debugf("mmu_booke_enter_locked: e\n");
1586 }
1587 
1588 /*
1589  * Maps a sequence of resident pages belonging to the same object.
1590  * The sequence begins with the given page m_start.  This page is
1591  * mapped at the given virtual address start.  Each subsequent page is
1592  * mapped at a virtual address that is offset from start by the same
1593  * amount as the page is offset from m_start within the object.  The
1594  * last page in the sequence is the page with the largest offset from
1595  * m_start that can be mapped at a virtual address less than the given
1596  * virtual address end.  Not every virtual page between start and end
1597  * is mapped; only those for which a resident page exists with the
1598  * corresponding offset from m_start are mapped.
1599  */
1600 static void
1601 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1602     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1603 {
1604 	vm_page_t m;
1605 	vm_pindex_t diff, psize;
1606 
1607 	psize = atop(end - start);
1608 	m = m_start;
1609 	PMAP_LOCK(pmap);
1610 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1611 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m, prot &
1612 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1613 		m = TAILQ_NEXT(m, listq);
1614 	}
1615 	PMAP_UNLOCK(pmap);
1616 }
1617 
1618 static void
1619 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1620     vm_prot_t prot)
1621 {
1622 
1623 	//debugf("mmu_booke_enter_quick: s\n");
1624 
1625 	PMAP_LOCK(pmap);
1626 	mmu_booke_enter_locked(mmu, pmap, va, m,
1627 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1628 	PMAP_UNLOCK(pmap);
1629 
1630 	//debugf("mmu_booke_enter_quick e\n");
1631 }
1632 
1633 /*
1634  * Remove the given range of addresses from the specified map.
1635  *
1636  * It is assumed that the start and end are properly rounded to the page size.
1637  */
1638 static void
1639 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1640 {
1641 	pte_t *pte;
1642 	u_int8_t hold_flag;
1643 
1644 	int su = (pmap == kernel_pmap);
1645 
1646 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1647 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1648 
1649 	if (su) {
1650 		KASSERT(((va >= virtual_avail) && (va <= VM_MAX_KERNEL_ADDRESS)),
1651 		    ("mmu_booke_enter: kernel pmap, non kernel va"));
1652 	} else {
1653 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1654 		    ("mmu_booke_enter: user pmap, non user va"));
1655 	}
1656 
1657 	if (PMAP_REMOVE_DONE(pmap)) {
1658 		//debugf("mmu_booke_remove: e (empty)\n");
1659 		return;
1660 	}
1661 
1662 	hold_flag = PTBL_HOLD_FLAG(pmap);
1663 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1664 
1665 	vm_page_lock_queues();
1666 	PMAP_LOCK(pmap);
1667 	for (; va < endva; va += PAGE_SIZE) {
1668 		pte = pte_find(mmu, pmap, va);
1669 		if ((pte != NULL) && PTE_ISVALID(pte)) {
1670 			pte_remove(mmu, pmap, va, hold_flag);
1671 
1672 			/* Flush mapping from TLB0. */
1673 			tlb0_flush_entry(pmap, va);
1674 		}
1675 	}
1676 	PMAP_UNLOCK(pmap);
1677 	vm_page_unlock_queues();
1678 
1679 	//debugf("mmu_booke_remove: e\n");
1680 }
1681 
1682 /*
1683  * Remove physical page from all pmaps in which it resides.
1684  */
1685 static void
1686 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1687 {
1688 	pv_entry_t pv, pvn;
1689 	u_int8_t hold_flag;
1690 
1691 	//debugf("mmu_booke_remove_all: s\n");
1692 
1693 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1694 
1695 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1696 		pvn = TAILQ_NEXT(pv, pv_link);
1697 
1698 		PMAP_LOCK(pv->pv_pmap);
1699 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1700 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1701 
1702 		/* Flush mapping from TLB0. */
1703 		tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
1704 		PMAP_UNLOCK(pv->pv_pmap);
1705 	}
1706 	vm_page_flag_clear(m, PG_WRITEABLE);
1707 
1708 	//debugf("mmu_booke_remove_all: e\n");
1709 }
1710 
1711 /*
1712  * Map a range of physical addresses into kernel virtual address space.
1713  *
1714  * The value passed in *virt is a suggested virtual address for the mapping.
1715  * Architectures which can support a direct-mapped physical to virtual region
1716  * can return the appropriate address within that region, leaving '*virt'
1717  * unchanged.  We cannot and therefore do not; *virt is updated with the
1718  * first usable address after the mapped region.
1719  */
1720 static vm_offset_t
1721 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1722     vm_offset_t pa_end, int prot)
1723 {
1724 	vm_offset_t sva = *virt;
1725 	vm_offset_t va = sva;
1726 
1727 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1728 	//		sva, pa_start, pa_end);
1729 
1730 	while (pa_start < pa_end) {
1731 		mmu_booke_kenter(mmu, va, pa_start);
1732 		va += PAGE_SIZE;
1733 		pa_start += PAGE_SIZE;
1734 	}
1735 	*virt = va;
1736 
1737 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1738 	return (sva);
1739 }
1740 
1741 /*
1742  * The pmap must be activated before it's address space can be accessed in any
1743  * way.
1744  */
1745 static void
1746 mmu_booke_activate(mmu_t mmu, struct thread *td)
1747 {
1748 	pmap_t pmap;
1749 
1750 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1751 
1752 	//debugf("mmu_booke_activate: s (proc = '%s', id = %d, pmap = 0x%08x)\n",
1753 	//		td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1754 
1755 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1756 
1757 	mtx_lock_spin(&sched_lock);
1758 
1759 	pmap->pm_active |= PCPU_GET(cpumask);
1760 	PCPU_SET(curpmap, pmap);
1761 
1762 	if (!pmap->pm_tid)
1763 		tid_alloc(pmap);
1764 
1765 	/* Load PID0 register with pmap tid value. */
1766 	load_pid0(pmap->pm_tid);
1767 
1768 	mtx_unlock_spin(&sched_lock);
1769 
1770 	//debugf("mmu_booke_activate: e (tid = %d for '%s')\n", pmap->pm_tid,
1771 	//		td->td_proc->p_comm);
1772 }
1773 
1774 /*
1775  * Deactivate the specified process's address space.
1776  */
1777 static void
1778 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1779 {
1780 	pmap_t pmap;
1781 
1782 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1783 	pmap->pm_active &= ~(PCPU_GET(cpumask));
1784 	PCPU_SET(curpmap, NULL);
1785 }
1786 
1787 /*
1788  * Copy the range specified by src_addr/len
1789  * from the source map to the range dst_addr/len
1790  * in the destination map.
1791  *
1792  * This routine is only advisory and need not do anything.
1793  */
1794 static void
1795 mmu_booke_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
1796     vm_size_t len, vm_offset_t src_addr)
1797 {
1798 
1799 }
1800 
1801 /*
1802  * Set the physical protection on the specified range of this map as requested.
1803  */
1804 static void
1805 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1806     vm_prot_t prot)
1807 {
1808 	vm_offset_t va;
1809 	vm_page_t m;
1810 	pte_t *pte;
1811 
1812 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1813 		mmu_booke_remove(mmu, pmap, sva, eva);
1814 		return;
1815 	}
1816 
1817 	if (prot & VM_PROT_WRITE)
1818 		return;
1819 
1820 	vm_page_lock_queues();
1821 	PMAP_LOCK(pmap);
1822 	for (va = sva; va < eva; va += PAGE_SIZE) {
1823 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1824 			if (PTE_ISVALID(pte)) {
1825 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1826 
1827 				/* Handle modified pages. */
1828 				if (PTE_ISMODIFIED(pte))
1829 					vm_page_dirty(m);
1830 
1831 				/* Referenced pages. */
1832 				if (PTE_ISREFERENCED(pte))
1833 					vm_page_flag_set(m, PG_REFERENCED);
1834 
1835 				/* Flush mapping from TLB0. */
1836 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1837 				    PTE_REFERENCED);
1838 				tlb0_flush_entry(pmap, va);
1839 			}
1840 		}
1841 	}
1842 	PMAP_UNLOCK(pmap);
1843 	vm_page_unlock_queues();
1844 }
1845 
1846 /*
1847  * Clear the write and modified bits in each of the given page's mappings.
1848  */
1849 static void
1850 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1851 {
1852 	pv_entry_t pv;
1853 	pte_t *pte;
1854 
1855 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1856 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1857 	    (m->flags & PG_WRITEABLE) == 0)
1858 		return;
1859 
1860 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1861 		PMAP_LOCK(pv->pv_pmap);
1862 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1863 			if (PTE_ISVALID(pte)) {
1864 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1865 
1866 				/* Handle modified pages. */
1867 				if (PTE_ISMODIFIED(pte))
1868 					vm_page_dirty(m);
1869 
1870 				/* Referenced pages. */
1871 				if (PTE_ISREFERENCED(pte))
1872 					vm_page_flag_set(m, PG_REFERENCED);
1873 
1874 				/* Flush mapping from TLB0. */
1875 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1876 				    PTE_REFERENCED);
1877 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
1878 			}
1879 		}
1880 		PMAP_UNLOCK(pv->pv_pmap);
1881 	}
1882 	vm_page_flag_clear(m, PG_WRITEABLE);
1883 }
1884 
1885 static boolean_t
1886 mmu_booke_page_executable(mmu_t mmu, vm_page_t m)
1887 {
1888 	pv_entry_t pv;
1889 	pte_t *pte;
1890 	boolean_t executable;
1891 
1892 	executable = FALSE;
1893 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1894 		PMAP_LOCK(pv->pv_pmap);
1895 		pte = pte_find(mmu, pv->pv_pmap, pv->pv_va);
1896 		if (pte != NULL && PTE_ISVALID(pte) && (pte->flags & PTE_UX))
1897 			executable = TRUE;
1898 		PMAP_UNLOCK(pv->pv_pmap);
1899 		if (executable)
1900 			break;
1901 	}
1902 
1903 	return (executable);
1904 }
1905 
1906 /*
1907  * Atomically extract and hold the physical page with the given
1908  * pmap and virtual address pair if that mapping permits the given
1909  * protection.
1910  */
1911 static vm_page_t
1912 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
1913     vm_prot_t prot)
1914 {
1915 	pte_t *pte;
1916 	vm_page_t m;
1917 	u_int32_t pte_wbit;
1918 
1919 	m = NULL;
1920 	vm_page_lock_queues();
1921 	PMAP_LOCK(pmap);
1922 	pte = pte_find(mmu, pmap, va);
1923 
1924 	if ((pte != NULL) && PTE_ISVALID(pte)) {
1925 		if (pmap == kernel_pmap)
1926 			pte_wbit = PTE_SW;
1927 		else
1928 			pte_wbit = PTE_UW;
1929 
1930 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
1931 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1932 			vm_page_hold(m);
1933 		}
1934 	}
1935 
1936 	vm_page_unlock_queues();
1937 	PMAP_UNLOCK(pmap);
1938 	return (m);
1939 }
1940 
1941 /*
1942  * Initialize a vm_page's machine-dependent fields.
1943  */
1944 static void
1945 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
1946 {
1947 
1948 	TAILQ_INIT(&m->md.pv_list);
1949 }
1950 
1951 /*
1952  * mmu_booke_zero_page_area zeros the specified hardware page by
1953  * mapping it into virtual memory and using bzero to clear
1954  * its contents.
1955  *
1956  * off and size must reside within a single page.
1957  */
1958 static void
1959 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1960 {
1961 	vm_offset_t va;
1962 
1963 	//debugf("mmu_booke_zero_page_area: s\n");
1964 
1965 	mtx_lock(&zero_page_mutex);
1966 	va = zero_page_va;
1967 
1968 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
1969 	bzero((caddr_t)va + off, size);
1970 	mmu_booke_kremove(mmu, va);
1971 
1972 	mtx_unlock(&zero_page_mutex);
1973 
1974 	//debugf("mmu_booke_zero_page_area: e\n");
1975 }
1976 
1977 /*
1978  * mmu_booke_zero_page zeros the specified hardware page.
1979  */
1980 static void
1981 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
1982 {
1983 
1984 	//debugf("mmu_booke_zero_page: s\n");
1985 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
1986 	//debugf("mmu_booke_zero_page: e\n");
1987 }
1988 
1989 /*
1990  * mmu_booke_copy_page copies the specified (machine independent) page by
1991  * mapping the page into virtual memory and using memcopy to copy the page,
1992  * one machine dependent page at a time.
1993  */
1994 static void
1995 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
1996 {
1997 	vm_offset_t sva, dva;
1998 
1999 	//debugf("mmu_booke_copy_page: s\n");
2000 
2001 	mtx_lock(&copy_page_mutex);
2002 	sva = copy_page_src_va;
2003 	dva = copy_page_dst_va;
2004 
2005 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2006 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2007 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2008 	mmu_booke_kremove(mmu, dva);
2009 	mmu_booke_kremove(mmu, sva);
2010 
2011 	mtx_unlock(&copy_page_mutex);
2012 
2013 	//debugf("mmu_booke_copy_page: e\n");
2014 }
2015 
2016 #if 0
2017 /*
2018  * Remove all pages from specified address space, this aids process exit
2019  * speeds. This is much faster than mmu_booke_remove in the case of running
2020  * down an entire address space. Only works for the current pmap.
2021  */
2022 void
2023 mmu_booke_remove_pages(pmap_t pmap)
2024 {
2025 }
2026 #endif
2027 
2028 /*
2029  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2030  * into virtual memory and using bzero to clear its contents. This is intended
2031  * to be called from the vm_pagezero process only and outside of Giant. No
2032  * lock is required.
2033  */
2034 static void
2035 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2036 {
2037 	vm_offset_t va;
2038 
2039 	//debugf("mmu_booke_zero_page_idle: s\n");
2040 
2041 	va = zero_page_idle_va;
2042 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2043 	bzero((caddr_t)va, PAGE_SIZE);
2044 	mmu_booke_kremove(mmu, va);
2045 
2046 	//debugf("mmu_booke_zero_page_idle: e\n");
2047 }
2048 
2049 /*
2050  * Return whether or not the specified physical page was modified
2051  * in any of physical maps.
2052  */
2053 static boolean_t
2054 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2055 {
2056 	pte_t *pte;
2057 	pv_entry_t pv;
2058 
2059 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2060 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2061 		return (FALSE);
2062 
2063 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2064 		PMAP_LOCK(pv->pv_pmap);
2065 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2066 			if (!PTE_ISVALID(pte))
2067 				goto make_sure_to_unlock;
2068 
2069 			if (PTE_ISMODIFIED(pte)) {
2070 				PMAP_UNLOCK(pv->pv_pmap);
2071 				return (TRUE);
2072 			}
2073 		}
2074 make_sure_to_unlock:
2075 		PMAP_UNLOCK(pv->pv_pmap);
2076 	}
2077 	return (FALSE);
2078 }
2079 
2080 /*
2081  * Return whether or not the specified virtual address is elgible
2082  * for prefault.
2083  */
2084 static boolean_t
2085 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2086 {
2087 
2088 	return (FALSE);
2089 }
2090 
2091 /*
2092  * Clear the modify bits on the specified physical page.
2093  */
2094 static void
2095 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2096 {
2097 	pte_t *pte;
2098 	pv_entry_t pv;
2099 
2100 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2101 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2102 		return;
2103 
2104 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2105 		PMAP_LOCK(pv->pv_pmap);
2106 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2107 			if (!PTE_ISVALID(pte))
2108 				goto make_sure_to_unlock;
2109 
2110 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2111 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2112 				    PTE_REFERENCED);
2113 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2114 			}
2115 		}
2116 make_sure_to_unlock:
2117 		PMAP_UNLOCK(pv->pv_pmap);
2118 	}
2119 }
2120 
2121 /*
2122  * Return a count of reference bits for a page, clearing those bits.
2123  * It is not necessary for every reference bit to be cleared, but it
2124  * is necessary that 0 only be returned when there are truly no
2125  * reference bits set.
2126  *
2127  * XXX: The exact number of bits to check and clear is a matter that
2128  * should be tested and standardized at some point in the future for
2129  * optimal aging of shared pages.
2130  */
2131 static int
2132 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2133 {
2134 	pte_t *pte;
2135 	pv_entry_t pv;
2136 	int count;
2137 
2138 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2139 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2140 		return (0);
2141 
2142 	count = 0;
2143 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2144 		PMAP_LOCK(pv->pv_pmap);
2145 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2146 			if (!PTE_ISVALID(pte))
2147 				goto make_sure_to_unlock;
2148 
2149 			if (PTE_ISREFERENCED(pte)) {
2150 				pte->flags &= ~PTE_REFERENCED;
2151 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2152 
2153 				if (++count > 4) {
2154 					PMAP_UNLOCK(pv->pv_pmap);
2155 					break;
2156 				}
2157 			}
2158 		}
2159 make_sure_to_unlock:
2160 		PMAP_UNLOCK(pv->pv_pmap);
2161 	}
2162 	return (count);
2163 }
2164 
2165 /*
2166  * Clear the reference bit on the specified physical page.
2167  */
2168 static void
2169 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2170 {
2171 	pte_t *pte;
2172 	pv_entry_t pv;
2173 
2174 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2175 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2176 		return;
2177 
2178 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2179 		PMAP_LOCK(pv->pv_pmap);
2180 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2181 			if (!PTE_ISVALID(pte))
2182 				goto make_sure_to_unlock;
2183 
2184 			if (PTE_ISREFERENCED(pte)) {
2185 				pte->flags &= ~PTE_REFERENCED;
2186 				tlb0_flush_entry(pv->pv_pmap, pv->pv_va);
2187 			}
2188 		}
2189 make_sure_to_unlock:
2190 		PMAP_UNLOCK(pv->pv_pmap);
2191 	}
2192 }
2193 
2194 /*
2195  * Change wiring attribute for a map/virtual-address pair.
2196  */
2197 static void
2198 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2199 {
2200 	pte_t *pte;;
2201 
2202 	PMAP_LOCK(pmap);
2203 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2204 		if (wired) {
2205 			if (!PTE_ISWIRED(pte)) {
2206 				pte->flags |= PTE_WIRED;
2207 				pmap->pm_stats.wired_count++;
2208 			}
2209 		} else {
2210 			if (PTE_ISWIRED(pte)) {
2211 				pte->flags &= ~PTE_WIRED;
2212 				pmap->pm_stats.wired_count--;
2213 			}
2214 		}
2215 	}
2216 	PMAP_UNLOCK(pmap);
2217 }
2218 
2219 /*
2220  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2221  * page.  This count may be changed upwards or downwards in the future; it is
2222  * only necessary that true be returned for a small subset of pmaps for proper
2223  * page aging.
2224  */
2225 static boolean_t
2226 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2227 {
2228 	pv_entry_t pv;
2229 	int loops;
2230 
2231 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2232 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2233 		return (FALSE);
2234 
2235 	loops = 0;
2236 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2237 
2238 		if (pv->pv_pmap == pmap)
2239 			return (TRUE);
2240 
2241 		if (++loops >= 16)
2242 			break;
2243 	}
2244 	return (FALSE);
2245 }
2246 
2247 /*
2248  * Return the number of managed mappings to the given physical page that are
2249  * wired.
2250  */
2251 static int
2252 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2253 {
2254 	pv_entry_t pv;
2255 	pte_t *pte;
2256 	int count = 0;
2257 
2258 	if ((m->flags & PG_FICTITIOUS) != 0)
2259 		return (count);
2260 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2261 
2262 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2263 		PMAP_LOCK(pv->pv_pmap);
2264 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2265 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2266 				count++;
2267 		PMAP_UNLOCK(pv->pv_pmap);
2268 	}
2269 
2270 	return (count);
2271 }
2272 
2273 static int
2274 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2275 {
2276 	int i;
2277 	vm_offset_t va;
2278 
2279 	/*
2280 	 * This currently does not work for entries that
2281 	 * overlap TLB1 entries.
2282 	 */
2283 	for (i = 0; i < tlb1_idx; i ++) {
2284 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2285 			return (0);
2286 	}
2287 
2288 	return (EFAULT);
2289 }
2290 
2291 /*
2292  * Map a set of physical memory pages into the kernel virtual address space.
2293  * Return a pointer to where it is mapped. This routine is intended to be used
2294  * for mapping device memory, NOT real memory.
2295  */
2296 static void *
2297 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2298 {
2299 	void *res;
2300 	uintptr_t va;
2301 	vm_size_t sz;
2302 
2303 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2304 	res = (void *)va;
2305 
2306 	do {
2307 		sz = 1 << (ilog2(size) & ~1);
2308 		if (bootverbose)
2309 			printf("Wiring VA=%x to PA=%x (size=%x), "
2310 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2311 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2312 		size -= sz;
2313 		pa += sz;
2314 		va += sz;
2315 	} while (size > 0);
2316 
2317 	return (res);
2318 }
2319 
2320 /*
2321  * 'Unmap' a range mapped by mmu_booke_mapdev().
2322  */
2323 static void
2324 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2325 {
2326 	vm_offset_t base, offset;
2327 
2328 	//debugf("mmu_booke_unmapdev: s (va = 0x%08x)\n", va);
2329 
2330 	/*
2331 	 * Unmap only if this is inside kernel virtual space.
2332 	 */
2333 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2334 		base = trunc_page(va);
2335 		offset = va & PAGE_MASK;
2336 		size = roundup(offset + size, PAGE_SIZE);
2337 		kmem_free(kernel_map, base, size);
2338 	}
2339 
2340 	//debugf("mmu_booke_unmapdev: e\n");
2341 }
2342 
2343 /*
2344  * mmu_booke_object_init_pt preloads the ptes for a given object
2345  * into the specified pmap. This eliminates the blast of soft
2346  * faults on process startup and immediately after an mmap.
2347  */
2348 static void
2349 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2350     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2351 {
2352 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2353 	KASSERT(object->type == OBJT_DEVICE,
2354 	    ("mmu_booke_object_init_pt: non-device object"));
2355 }
2356 
2357 /*
2358  * Perform the pmap work for mincore.
2359  */
2360 static int
2361 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2362 {
2363 
2364 	TODO;
2365 	return (0);
2366 }
2367 
2368 /**************************************************************************/
2369 /* TID handling */
2370 /**************************************************************************/
2371 /*
2372  * Flush all entries from TLB0 matching given tid.
2373  */
2374 static void
2375 tid_flush(tlbtid_t tid)
2376 {
2377 	int i, entryidx, way;
2378 
2379 	//debugf("tid_flush: s (tid = %d)\n", tid);
2380 
2381 	mtx_lock_spin(&tlb0_mutex);
2382 
2383 	for (i = 0; i < TLB0_SIZE; i++) {
2384 		if (MAS1_GETTID(tlb0[i].mas1) == tid) {
2385 			way = i / TLB0_ENTRIES_PER_WAY;
2386 			entryidx = i - (way * TLB0_ENTRIES_PER_WAY);
2387 
2388 			//debugf("tid_flush: inval tlb0 entry %d\n", i);
2389 			tlb0_inval_entry(entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT, way);
2390 		}
2391 	}
2392 
2393 	mtx_unlock_spin(&tlb0_mutex);
2394 
2395 	//debugf("tid_flush: e\n");
2396 }
2397 
2398 /*
2399  * Allocate a TID. If necessary, steal one from someone else.
2400  * The new TID is flushed from the TLB before returning.
2401  */
2402 static tlbtid_t
2403 tid_alloc(pmap_t pmap)
2404 {
2405 	tlbtid_t tid;
2406 	static tlbtid_t next_tid = TID_MIN;
2407 
2408 	//struct thread *td;
2409 	//struct proc *p;
2410 
2411 	//td = PCPU_GET(curthread);
2412 	//p = td->td_proc;
2413 	//debugf("tid_alloc: s (pmap = 0x%08x)\n", (u_int32_t)pmap);
2414 	//printf("tid_alloc: proc %d '%s'\n", p->p_pid, p->p_comm);
2415 
2416 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2417 
2418 	/*
2419 	 * Find a likely TID, allocate unused if possible,
2420 	 * skip reserved entries.
2421 	 */
2422 	tid = next_tid;
2423 	while (tidbusy[tid] != NULL) {
2424 		if (tid == next_tid)
2425 			break;
2426 
2427 		if (tid == TID_MAX)
2428 			tid = TID_MIN;
2429 		else
2430 			tid++;
2431 
2432 	}
2433 
2434 	/* Now clean it out */
2435 	tid_flush(tid);
2436 
2437 	/* If we are stealing pmap then clear its tid */
2438 	if (tidbusy[tid]) {
2439 		//debugf("warning: stealing tid %d\n", tid);
2440 		tidbusy[tid]->pm_tid = 0;
2441 	}
2442 
2443 	/* Calculate next tid */
2444 	if (tid == TID_MAX)
2445 		next_tid = TID_MIN;
2446 	else
2447 		next_tid = tid + 1;
2448 
2449 	tidbusy[tid] = pmap;
2450 	pmap->pm_tid = tid;
2451 
2452 	//debugf("tid_alloc: e (%02d next = %02d)\n", tid, next_tid);
2453 	return (tid);
2454 }
2455 
2456 #if 0
2457 /*
2458  * Free this pmap's TID.
2459  */
2460 static void
2461 tid_free(pmap_t pmap)
2462 {
2463 	tlbtid_t oldtid;
2464 
2465 	oldtid = pmap->pm_tid;
2466 
2467 	if (oldtid == 0) {
2468 		panic("tid_free: freeing kernel tid");
2469 	}
2470 
2471 #ifdef DEBUG
2472 	if (tidbusy[oldtid] == 0)
2473 		debugf("tid_free: freeing free tid %d\n", oldtid);
2474 	if (tidbusy[oldtid] != pmap) {
2475 		debugf("tid_free: freeing someone esle's tid\n "
2476 		       "tidbusy[%d] = 0x%08x pmap = 0x%08x\n",
2477 		       oldtid, (u_int32_t)tidbusy[oldtid], (u_int32_t)pmap);
2478 	}
2479 #endif
2480 
2481 	tidbusy[oldtid] = NULL;
2482 	tid_flush(oldtid);
2483 }
2484 #endif
2485 
2486 #if 0
2487 #if DEBUG
2488 static void
2489 tid_print_busy(void)
2490 {
2491 	int i;
2492 
2493 	for (i = 0; i < TID_MAX; i++) {
2494 		debugf("tid %d = pmap 0x%08x", i, (u_int32_t)tidbusy[i]);
2495 		if (tidbusy[i])
2496 			debugf(" pmap->tid = %d", tidbusy[i]->pm_tid);
2497 		debugf("\n");
2498 	}
2499 
2500 }
2501 #endif /* DEBUG */
2502 #endif
2503 
2504 /**************************************************************************/
2505 /* TLB0 handling */
2506 /**************************************************************************/
2507 
2508 static void
2509 tlb_print_entry(int i, u_int32_t mas1, u_int32_t mas2, u_int32_t mas3, u_int32_t mas7)
2510 {
2511 	int as;
2512 	char desc[3];
2513 	tlbtid_t tid;
2514 	vm_size_t size;
2515 	unsigned int tsize;
2516 
2517 	desc[2] = '\0';
2518 	if (mas1 & MAS1_VALID)
2519 		desc[0] = 'V';
2520 	else
2521 		desc[0] = ' ';
2522 
2523 	if (mas1 & MAS1_IPROT)
2524 		desc[1] = 'P';
2525 	else
2526 		desc[1] = ' ';
2527 
2528 	as = (mas1 & MAS1_TS) ? 1 : 0;
2529 	tid = MAS1_GETTID(mas1);
2530 
2531 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2532 	size = 0;
2533 	if (tsize)
2534 		size = tsize2size(tsize);
2535 
2536 	debugf("%3d: (%s) [AS=%d] "
2537 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2538 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2539 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2540 }
2541 
2542 /* Convert TLB0 va and way number to tlb0[] table index. */
2543 static inline unsigned int
2544 tlb0_tableidx(vm_offset_t va, unsigned int way)
2545 {
2546 	unsigned int idx;
2547 
2548 	idx = (way * TLB0_ENTRIES_PER_WAY);
2549 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2550 	return (idx);
2551 }
2552 
2553 /*
2554  * Write given entry to TLB0 hardware.
2555  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2556  */
2557 static void
2558 tlb0_write_entry(unsigned int idx, unsigned int way)
2559 {
2560 	u_int32_t mas0, mas7, nv;
2561 
2562 	/* Clear high order RPN bits. */
2563 	mas7 = 0;
2564 
2565 	/* Preserve NV. */
2566 	mas0 = mfspr(SPR_MAS0);
2567 	nv = mas0 & (TLB0_NWAYS - 1);
2568 
2569 	/* Select entry. */
2570 	mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way) | nv;
2571 
2572 	//debugf("tlb0_write_entry: s (idx=%d way=%d mas0=0x%08x "
2573 	//		"mas1=0x%08x mas2=0x%08x mas3=0x%08x)\n",
2574 	//		idx, way, mas0, tlb0[idx].mas1,
2575 	//		tlb0[idx].mas2, tlb0[idx].mas3);
2576 
2577 	mtspr(SPR_MAS0, mas0);
2578 	__asm volatile("isync");
2579 	mtspr(SPR_MAS1, tlb0[idx].mas1);
2580 	__asm volatile("isync");
2581 	mtspr(SPR_MAS2, tlb0[idx].mas2);
2582 	__asm volatile("isync");
2583 	mtspr(SPR_MAS3, tlb0[idx].mas3);
2584 	__asm volatile("isync");
2585 	mtspr(SPR_MAS7, mas7);
2586 	__asm volatile("isync; tlbwe; isync; msync");
2587 
2588 	//debugf("tlb0_write_entry: e\n");
2589 }
2590 
2591 /*
2592  * Invalidate TLB0 entry, clear correspondig tlb0 table element.
2593  */
2594 static void
2595 tlb0_inval_entry(vm_offset_t va, unsigned int way)
2596 {
2597 	int idx = tlb0_tableidx(va, way);
2598 
2599 	//debugf("tlb0_inval_entry: s (va=0x%08x way=%d idx=%d)\n",
2600 	//		va, way, idx);
2601 
2602 	tlb0[idx].mas1 = 1 << MAS1_TSIZE_SHIFT;	/* !MAS1_VALID */
2603 	tlb0[idx].mas2 = va & MAS2_EPN;
2604 	tlb0[idx].mas3 = 0;
2605 
2606 	tlb0_write_entry(idx, way);
2607 
2608 	//debugf("tlb0_inval_entry: e\n");
2609 }
2610 
2611 /*
2612  * Invalidate TLB0 entry that corresponds to pmap/va.
2613  */
2614 static void
2615 tlb0_flush_entry(pmap_t pmap, vm_offset_t va)
2616 {
2617 	int idx, way;
2618 
2619 	//debugf("tlb0_flush_entry: s (pmap=0x%08x va=0x%08x)\n",
2620 	//		(u_int32_t)pmap, va);
2621 
2622 	mtx_lock_spin(&tlb0_mutex);
2623 
2624 	/* Check all TLB0 ways. */
2625 	for (way = 0; way < TLB0_NWAYS; way ++) {
2626 		idx = tlb0_tableidx(va, way);
2627 
2628 		/* Invalidate only if entry matches va and pmap tid. */
2629 		if (((MAS1_GETTID(tlb0[idx].mas1) == pmap->pm_tid) &&
2630 				((tlb0[idx].mas2 & MAS2_EPN) == va))) {
2631 			tlb0_inval_entry(va, way);
2632 		}
2633 	}
2634 
2635 	mtx_unlock_spin(&tlb0_mutex);
2636 
2637 	//debugf("tlb0_flush_entry: e\n");
2638 }
2639 
2640 /* Clean TLB0 hardware and tlb0[] table. */
2641 static void
2642 tlb0_init(void)
2643 {
2644 	int entryidx, way;
2645 
2646 	debugf("tlb0_init: TLB0_SIZE = %d TLB0_NWAYS = %d\n",
2647 	    TLB0_SIZE, TLB0_NWAYS);
2648 
2649 	mtx_lock_spin(&tlb0_mutex);
2650 
2651 	for (way = 0; way < TLB0_NWAYS; way ++) {
2652 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2653 			tlb0_inval_entry(entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT, way);
2654 		}
2655 	}
2656 
2657 	mtx_unlock_spin(&tlb0_mutex);
2658 }
2659 
2660 #if 0
2661 #if DEBUG
2662 /* Print out tlb0 entries for given va. */
2663 static void
2664 tlb0_print_tlbentries_va(vm_offset_t va)
2665 {
2666 	u_int32_t mas0, mas1, mas2, mas3, mas7;
2667 	int way, idx;
2668 
2669 	debugf("TLB0 entries for va = 0x%08x:\n", va);
2670 	for (way = 0; way < TLB0_NWAYS; way ++) {
2671 		mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2672 		mtspr(SPR_MAS0, mas0);
2673 		__asm volatile("isync");
2674 
2675 		mas2 = va & MAS2_EPN;
2676 		mtspr(SPR_MAS2, mas2);
2677 		__asm volatile("isync; tlbre");
2678 
2679 		mas1 = mfspr(SPR_MAS1);
2680 		mas2 = mfspr(SPR_MAS2);
2681 		mas3 = mfspr(SPR_MAS3);
2682 		mas7 = mfspr(SPR_MAS7);
2683 
2684 		idx = tlb0_tableidx(va, way);
2685 		tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2686 	}
2687 }
2688 
2689 /* Print out contents of the MAS registers for each TLB0 entry */
2690 static void
2691 tlb0_print_tlbentries(void)
2692 {
2693 	u_int32_t mas0, mas1, mas2, mas3, mas7;
2694 	int entryidx, way, idx;
2695 
2696 	debugf("TLB0 entries:\n");
2697 	for (way = 0; way < TLB0_NWAYS; way ++) {
2698 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2699 
2700 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2701 			mtspr(SPR_MAS0, mas0);
2702 			__asm volatile("isync");
2703 
2704 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2705 			mtspr(SPR_MAS2, mas2);
2706 
2707 			__asm volatile("isync; tlbre");
2708 
2709 			mas1 = mfspr(SPR_MAS1);
2710 			mas2 = mfspr(SPR_MAS2);
2711 			mas3 = mfspr(SPR_MAS3);
2712 			mas7 = mfspr(SPR_MAS7);
2713 
2714 			idx = tlb0_tableidx(mas2, way);
2715 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2716 		}
2717 	}
2718 }
2719 
2720 /* Print out kernel tlb0[] table. */
2721 static void
2722 tlb0_print_entries(void)
2723 {
2724 	int i;
2725 
2726 	debugf("tlb0[] table entries:\n");
2727 	for (i = 0; i < TLB0_SIZE; i++) {
2728 		tlb_print_entry(i, tlb0[i].mas1,
2729 				tlb0[i].mas2, tlb0[i].mas3, 0);
2730 	}
2731 }
2732 #endif /* DEBUG */
2733 #endif
2734 
2735 /**************************************************************************/
2736 /* TLB1 handling */
2737 /**************************************************************************/
2738 /*
2739  * Write given entry to TLB1 hardware.
2740  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2741  */
2742 static void
2743 tlb1_write_entry(unsigned int idx)
2744 {
2745 	u_int32_t mas0, mas7;
2746 
2747 	//debugf("tlb1_write_entry: s\n");
2748 
2749 	/* Clear high order RPN bits */
2750 	mas7 = 0;
2751 
2752 	/* Select entry */
2753 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2754 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2755 
2756 	mtspr(SPR_MAS0, mas0);
2757 	__asm volatile("isync");
2758 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2759 	__asm volatile("isync");
2760 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2761 	__asm volatile("isync");
2762 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2763 	__asm volatile("isync");
2764 	mtspr(SPR_MAS7, mas7);
2765 	__asm volatile("isync; tlbwe; isync; msync");
2766 
2767 	//debugf("tlb1_write_entry: e\n");;
2768 }
2769 
2770 /*
2771  * Return the largest uint value log such that 2^log <= num.
2772  */
2773 static unsigned int
2774 ilog2(unsigned int num)
2775 {
2776 	int lz;
2777 
2778 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2779 	return (31 - lz);
2780 }
2781 
2782 /*
2783  * Convert TLB TSIZE value to mapped region size.
2784  */
2785 static vm_size_t
2786 tsize2size(unsigned int tsize)
2787 {
2788 
2789 	/*
2790 	 * size = 4^tsize KB
2791 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2792 	 */
2793 
2794 	return ((1 << (2 * tsize)) * 1024);
2795 }
2796 
2797 /*
2798  * Convert region size (must be power of 4) to TLB TSIZE value.
2799  */
2800 static unsigned int
2801 size2tsize(vm_size_t size)
2802 {
2803 
2804 	/*
2805 	 * tsize = log2(size) / 2 - 5
2806 	 */
2807 
2808 	return (ilog2(size) / 2 - 5);
2809 }
2810 
2811 /*
2812  * Setup entry in a sw tlb1 table, write entry to TLB1 hardware.
2813  * This routine is used for low level operations on the TLB1,
2814  * for creating temporaray as well as permanent mappings (tlb_set_entry).
2815  *
2816  * We assume kernel mappings only, thus all entries created have supervisor
2817  * permission bits set nad user permission bits cleared.
2818  *
2819  * Provided mapping size must be a power of 4.
2820  * Mapping flags must be a combination of MAS2_[WIMG].
2821  * Entry TID is set to _tid which must not exceed 8 bit value.
2822  * Entry TS is set to either 0 or MAS1_TS based on provided _ts.
2823  */
2824 static void
2825 __tlb1_set_entry(unsigned int idx, vm_offset_t va, vm_offset_t pa,
2826     vm_size_t size, u_int32_t flags, unsigned int _tid, unsigned int _ts)
2827 {
2828 	int tsize;
2829 	u_int32_t ts, tid;
2830 
2831 	//debugf("__tlb1_set_entry: s (idx = %d va = 0x%08x pa = 0x%08x "
2832 	//		"size = 0x%08x flags = 0x%08x _tid = %d _ts = %d\n",
2833 	//		idx, va, pa, size, flags, _tid, _ts);
2834 
2835 	/* Convert size to TSIZE */
2836 	tsize = size2tsize(size);
2837 	//debugf("__tlb1_set_entry: tsize = %d\n", tsize);
2838 
2839 	tid = (_tid <<  MAS1_TID_SHIFT) & MAS1_TID_MASK;
2840 	ts = (_ts) ? MAS1_TS : 0;
2841 	tlb1[idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2842 	tlb1[idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2843 
2844 	tlb1[idx].mas2 = (va & MAS2_EPN) | flags;
2845 
2846 	/* Set supervisor rwx permission bits */
2847 	tlb1[idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2848 
2849 	//debugf("__tlb1_set_entry: mas1 = %08x mas2 = %08x mas3 = 0x%08x\n",
2850 	//		tlb1[idx].mas1, tlb1[idx].mas2, tlb1[idx].mas3);
2851 
2852 	tlb1_write_entry(idx);
2853 	//debugf("__tlb1_set_entry: e\n");
2854 }
2855 
2856 /*
2857  * Register permanent kernel mapping in TLB1.
2858  *
2859  * Entries are created starting from index 0 (current free entry is
2860  * kept in tlb1_idx) and are not supposed to be invalidated.
2861  */
2862 static int
2863 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size, u_int32_t flags)
2864 {
2865 	//debugf("tlb1_set_entry: s (tlb1_idx = %d va = 0x%08x pa = 0x%08x "
2866 	//		"size = 0x%08x flags = 0x%08x\n",
2867 	//		tlb1_idx, va, pa, size, flags);
2868 
2869 	if (tlb1_idx >= TLB1_SIZE) {
2870 		//debugf("tlb1_set_entry: e (tlb1 full!)\n");
2871 		return (-1);
2872 	}
2873 
2874 	/* TS = 0, TID = 0 */
2875 	__tlb1_set_entry(tlb1_idx++, va, pa, size, flags, KERNEL_TID, 0);
2876 	//debugf("tlb1_set_entry: e\n");
2877 	return (0);
2878 }
2879 
2880 /*
2881  * Invalidate TLB1 entry, clear correspondig tlb1 table element.
2882  * This routine is used to clear temporary entries created
2883  * early in a locore.S or through the use of __tlb1_set_entry().
2884  */
2885 void
2886 tlb1_inval_entry(unsigned int idx)
2887 {
2888 	vm_offset_t va;
2889 
2890 	va = tlb1[idx].mas2 & MAS2_EPN;
2891 
2892 	tlb1[idx].mas1 = 0; /* !MAS1_VALID */
2893 	tlb1[idx].mas2 = 0;
2894 	tlb1[idx].mas3 = 0;
2895 
2896 	tlb1_write_entry(idx);
2897 }
2898 
2899 static int
2900 tlb1_entry_size_cmp(const void *a, const void *b)
2901 {
2902 	const vm_size_t *sza;
2903 	const vm_size_t *szb;
2904 
2905 	sza = a;
2906 	szb = b;
2907 	if (*sza > *szb)
2908 		return (-1);
2909 	else if (*sza < *szb)
2910 		return (1);
2911 	else
2912 		return (0);
2913 }
2914 
2915 /*
2916  * Mapin contiguous RAM region into the TLB1 using maximum of
2917  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2918  *
2919  * If necessarry round up last entry size and return total size
2920  * used by all allocated entries.
2921  */
2922 vm_size_t
2923 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2924 {
2925 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2926 	vm_size_t mapped_size, sz, esz;
2927 	unsigned int log;
2928 	int i;
2929 
2930 	debugf("tlb1_mapin_region:\n");
2931 	debugf(" region size = 0x%08x va = 0x%08x pa = 0x%08x\n", size, va, pa);
2932 
2933 	mapped_size = 0;
2934 	sz = size;
2935 	memset(entry_size, 0, sizeof(entry_size));
2936 
2937 	/* Calculate entry sizes. */
2938 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2939 
2940 		/* Largest region that is power of 4 and fits within size */
2941 		log = ilog2(sz)/2;
2942 		esz = 1 << (2 * log);
2943 
2944 		/* Minimum region size is 4KB */
2945 		if (esz < (1 << 12))
2946 			esz = 1 << 12;
2947 
2948 		/* If this is last entry cover remaining size. */
2949 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2950 			while (esz < sz)
2951 				esz = esz << 2;
2952 		}
2953 
2954 		entry_size[i] = esz;
2955 		mapped_size += esz;
2956 		if (esz < sz)
2957 			sz -= esz;
2958 		else
2959 			sz = 0;
2960 	}
2961 
2962 	/* Sort entry sizes, required to get proper entry address alignment. */
2963 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2964 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2965 
2966 	/* Load TLB1 entries. */
2967 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
2968 		esz = entry_size[i];
2969 		if (!esz)
2970 			break;
2971 		debugf("  entry %d: sz  = 0x%08x (va = 0x%08x pa = 0x%08x)\n",
2972 		    tlb1_idx, esz, va, pa);
2973 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
2974 
2975 		va += esz;
2976 		pa += esz;
2977 	}
2978 
2979 	debugf(" mapped size 0x%08x (wasted space 0x%08x)\n",
2980 	    mapped_size, mapped_size - size);
2981 
2982 	return (mapped_size);
2983 }
2984 
2985 /*
2986  * TLB1 initialization routine, to be called after the very first
2987  * assembler level setup done in locore.S.
2988  */
2989 void
2990 tlb1_init(vm_offset_t ccsrbar)
2991 {
2992 	uint32_t mas0;
2993 
2994 	/* TBL1[1] is used to map the kernel. Save that entry. */
2995 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(1);
2996 	mtspr(SPR_MAS0, mas0);
2997 	__asm __volatile("isync; tlbre");
2998 
2999 	tlb1[1].mas1 = mfspr(SPR_MAS1);
3000 	tlb1[1].mas2 = mfspr(SPR_MAS2);
3001 	tlb1[1].mas3 = mfspr(SPR_MAS3);
3002 
3003 	/* Mapin CCSRBAR in TLB1[0] */
3004 	__tlb1_set_entry(0, CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE,
3005 	    _TLB_ENTRY_IO, KERNEL_TID, 0);
3006 
3007 	/* Setup TLB miss defaults */
3008 	set_mas4_defaults();
3009 
3010 	/* Reset next available TLB1 entry index. */
3011 	tlb1_idx = 2;
3012 }
3013 
3014 /*
3015  * Setup MAS4 defaults.
3016  * These values are loaded to MAS0-2 on a TLB miss.
3017  */
3018 static void
3019 set_mas4_defaults(void)
3020 {
3021 	u_int32_t mas4;
3022 
3023 	/* Defaults: TLB0, PID0, TSIZED=4K */
3024 	mas4 = MAS4_TLBSELD0;
3025 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3026 
3027 	mtspr(SPR_MAS4, mas4);
3028 	__asm volatile("isync");
3029 }
3030 
3031 /*
3032  * Print out contents of the MAS registers for each TLB1 entry
3033  */
3034 void
3035 tlb1_print_tlbentries(void)
3036 {
3037 	u_int32_t mas0, mas1, mas2, mas3, mas7;
3038 	int i;
3039 
3040 	debugf("TLB1 entries:\n");
3041 	for (i = 0; i < TLB1_SIZE; i++) {
3042 
3043 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3044 		mtspr(SPR_MAS0, mas0);
3045 
3046 		__asm volatile("isync; tlbre");
3047 
3048 		mas1 = mfspr(SPR_MAS1);
3049 		mas2 = mfspr(SPR_MAS2);
3050 		mas3 = mfspr(SPR_MAS3);
3051 		mas7 = mfspr(SPR_MAS7);
3052 
3053 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3054 	}
3055 }
3056 
3057 /*
3058  * Print out contents of the in-ram tlb1 table.
3059  */
3060 void
3061 tlb1_print_entries(void)
3062 {
3063 	int i;
3064 
3065 	debugf("tlb1[] table entries:\n");
3066 	for (i = 0; i < TLB1_SIZE; i++)
3067 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3068 }
3069 
3070 /*
3071  * Return 0 if the physical IO range is encompassed by one of the
3072  * the TLB1 entries, otherwise return related error code.
3073  */
3074 static int
3075 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3076 {
3077 	u_int32_t prot;
3078 	vm_paddr_t pa_start;
3079 	vm_paddr_t pa_end;
3080 	unsigned int entry_tsize;
3081 	vm_size_t entry_size;
3082 
3083 	*va = (vm_offset_t)NULL;
3084 
3085 	/* Skip invalid entries */
3086 	if (!(tlb1[i].mas1 & MAS1_VALID))
3087 		return (EINVAL);
3088 
3089 	/*
3090 	 * The entry must be cache-inhibited, guarded, and r/w
3091 	 * so it can function as an i/o page
3092 	 */
3093 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3094 	if (prot != (MAS2_I | MAS2_G))
3095 		return (EPERM);
3096 
3097 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3098 	if (prot != (MAS3_SR | MAS3_SW))
3099 		return (EPERM);
3100 
3101 	/* The address should be within the entry range. */
3102 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3103 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3104 
3105 	entry_size = tsize2size(entry_tsize);
3106 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3107 	pa_end = pa_start + entry_size - 1;
3108 
3109 	if ((pa < pa_start) || ((pa + size) > pa_end))
3110 		return (ERANGE);
3111 
3112 	/* Return virtual address of this mapping. */
3113 	*va = (tlb1[i].mas2 & MAS2_EPN) + (pa - pa_start);
3114 	return (0);
3115 }
3116