xref: /freebsd/sys/powerpc/booke/pmap.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
1 /*-
2  * Copyright (C) 2007-2009 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  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/types.h>
55 #include <sys/param.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.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/smp.h>
67 #include <sys/vmmeter.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_pager.h>
78 #include <vm/uma.h>
79 
80 #include <machine/bootinfo.h>
81 #include <machine/cpu.h>
82 #include <machine/pcb.h>
83 #include <machine/platform.h>
84 
85 #include <machine/tlb.h>
86 #include <machine/spr.h>
87 #include <machine/vmparam.h>
88 #include <machine/md_var.h>
89 #include <machine/mmuvar.h>
90 #include <machine/pmap.h>
91 #include <machine/pte.h>
92 
93 #include "mmu_if.h"
94 
95 #define DEBUG
96 #undef DEBUG
97 
98 #ifdef  DEBUG
99 #define debugf(fmt, args...) printf(fmt, ##args)
100 #else
101 #define debugf(fmt, args...)
102 #endif
103 
104 #define TODO			panic("%s: not implemented", __func__);
105 
106 #include "opt_sched.h"
107 #ifndef SCHED_4BSD
108 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
109 #endif
110 extern struct mtx sched_lock;
111 
112 extern int dumpsys_minidump;
113 
114 extern unsigned char _etext[];
115 extern unsigned char _end[];
116 
117 /* Kernel physical load address. */
118 extern uint32_t kernload;
119 vm_offset_t kernstart;
120 vm_size_t kernsize;
121 
122 /* Message buffer and tables. */
123 static vm_offset_t data_start;
124 static vm_size_t data_end;
125 
126 /* Phys/avail memory regions. */
127 static struct mem_region *availmem_regions;
128 static int availmem_regions_sz;
129 static struct mem_region *physmem_regions;
130 static int physmem_regions_sz;
131 
132 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
133 static vm_offset_t zero_page_va;
134 static struct mtx zero_page_mutex;
135 
136 static struct mtx tlbivax_mutex;
137 
138 /*
139  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
140  * by idle thred only, no lock required.
141  */
142 static vm_offset_t zero_page_idle_va;
143 
144 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
145 static vm_offset_t copy_page_src_va;
146 static vm_offset_t copy_page_dst_va;
147 static struct mtx copy_page_mutex;
148 
149 /**************************************************************************/
150 /* PMAP */
151 /**************************************************************************/
152 
153 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
154     vm_prot_t, boolean_t);
155 
156 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
157 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
158 
159 /*
160  * If user pmap is processed with mmu_booke_remove and the resident count
161  * drops to 0, there are no more pages to remove, so we need not continue.
162  */
163 #define PMAP_REMOVE_DONE(pmap) \
164 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
165 
166 extern void tlb_lock(uint32_t *);
167 extern void tlb_unlock(uint32_t *);
168 extern void tid_flush(tlbtid_t);
169 
170 /**************************************************************************/
171 /* TLB and TID handling */
172 /**************************************************************************/
173 
174 /* Translation ID busy table */
175 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
176 
177 /*
178  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
179  * core revisions and should be read from h/w registers during early config.
180  */
181 uint32_t tlb0_entries;
182 uint32_t tlb0_ways;
183 uint32_t tlb0_entries_per_way;
184 
185 #define TLB0_ENTRIES		(tlb0_entries)
186 #define TLB0_WAYS		(tlb0_ways)
187 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
188 
189 #define TLB1_ENTRIES 16
190 
191 /* In-ram copy of the TLB1 */
192 static tlb_entry_t tlb1[TLB1_ENTRIES];
193 
194 /* Next free entry in the TLB1 */
195 static unsigned int tlb1_idx;
196 
197 static tlbtid_t tid_alloc(struct pmap *);
198 
199 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
200 
201 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
202 static void tlb1_write_entry(unsigned int);
203 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
204 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_offset_t, vm_size_t);
205 
206 static vm_size_t tsize2size(unsigned int);
207 static unsigned int size2tsize(vm_size_t);
208 static unsigned int ilog2(unsigned int);
209 
210 static void set_mas4_defaults(void);
211 
212 static inline void tlb0_flush_entry(vm_offset_t);
213 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
214 
215 /**************************************************************************/
216 /* Page table management */
217 /**************************************************************************/
218 
219 /* Data for the pv entry allocation mechanism */
220 static uma_zone_t pvzone;
221 static struct vm_object pvzone_obj;
222 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
223 
224 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
225 
226 #ifndef PMAP_SHPGPERPROC
227 #define PMAP_SHPGPERPROC	200
228 #endif
229 
230 static void ptbl_init(void);
231 static struct ptbl_buf *ptbl_buf_alloc(void);
232 static void ptbl_buf_free(struct ptbl_buf *);
233 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
234 
235 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
236 static void ptbl_free(mmu_t, pmap_t, unsigned int);
237 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
238 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
239 
240 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
241 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
242 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
243 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
244 
245 static pv_entry_t pv_alloc(void);
246 static void pv_free(pv_entry_t);
247 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
248 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
249 
250 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
251 #define PTBL_BUFS		(128 * 16)
252 
253 struct ptbl_buf {
254 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
255 	vm_offset_t kva;		/* va of mapping */
256 };
257 
258 /* ptbl free list and a lock used for access synchronization. */
259 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
260 static struct mtx ptbl_buf_freelist_lock;
261 
262 /* Base address of kva space allocated fot ptbl bufs. */
263 static vm_offset_t ptbl_buf_pool_vabase;
264 
265 /* Pointer to ptbl_buf structures. */
266 static struct ptbl_buf *ptbl_bufs;
267 
268 void pmap_bootstrap_ap(volatile uint32_t *);
269 
270 /*
271  * Kernel MMU interface
272  */
273 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
274 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
275 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
276 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
277     vm_size_t, vm_offset_t);
278 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
279 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
280     vm_prot_t, boolean_t);
281 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
282     vm_page_t, vm_prot_t);
283 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
284     vm_prot_t);
285 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
286 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
287     vm_prot_t);
288 static void		mmu_booke_init(mmu_t);
289 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
290 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
291 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
292 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
293 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
294     int);
295 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
296     vm_paddr_t *);
297 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
298     vm_object_t, vm_pindex_t, vm_size_t);
299 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
300 static void		mmu_booke_page_init(mmu_t, vm_page_t);
301 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
302 static void		mmu_booke_pinit(mmu_t, pmap_t);
303 static void		mmu_booke_pinit0(mmu_t, pmap_t);
304 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
305     vm_prot_t);
306 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
307 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
308 static void		mmu_booke_release(mmu_t, pmap_t);
309 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
310 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
311 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
312 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
313 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
314 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
315 static void		mmu_booke_activate(mmu_t, struct thread *);
316 static void		mmu_booke_deactivate(mmu_t, struct thread *);
317 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
318 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
319 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
320 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
321 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
322 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
323 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
324 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
325     vm_size_t);
326 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
327     vm_size_t, vm_size_t *);
328 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
329     vm_size_t, vm_offset_t);
330 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
331 
332 static mmu_method_t mmu_booke_methods[] = {
333 	/* pmap dispatcher interface */
334 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
335 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
336 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
337 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
338 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
339 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
340 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
341 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
342 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
343 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
344 	MMUMETHOD(mmu_init,		mmu_booke_init),
345 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
346 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
347 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
348 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
349 	MMUMETHOD(mmu_map,		mmu_booke_map),
350 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
351 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
352 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
353 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
354 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
355 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
356 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
357 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
358 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
359 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
360 	MMUMETHOD(mmu_release,		mmu_booke_release),
361 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
362 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
363 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
364 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
365 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
366 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
367 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
368 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
369 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
370 
371 	/* Internal interfaces */
372 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
373 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
374 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
375 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
376 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
377 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
378 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
379 
380 	/* dumpsys() support */
381 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
382 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
383 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
384 
385 	{ 0, 0 }
386 };
387 
388 static mmu_def_t booke_mmu = {
389 	MMU_TYPE_BOOKE,
390 	mmu_booke_methods,
391 	0
392 };
393 MMU_DEF(booke_mmu);
394 
395 static inline void
396 tlb_miss_lock(void)
397 {
398 #ifdef SMP
399 	struct pcpu *pc;
400 
401 	if (!smp_started)
402 		return;
403 
404 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
405 		if (pc != pcpup) {
406 
407 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
408 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
409 
410 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
411 			    ("tlb_miss_lock: tried to lock self"));
412 
413 			tlb_lock(pc->pc_booke_tlb_lock);
414 
415 			CTR1(KTR_PMAP, "%s: locked", __func__);
416 		}
417 	}
418 #endif
419 }
420 
421 static inline void
422 tlb_miss_unlock(void)
423 {
424 #ifdef SMP
425 	struct pcpu *pc;
426 
427 	if (!smp_started)
428 		return;
429 
430 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
431 		if (pc != pcpup) {
432 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
433 			    __func__, pc->pc_cpuid);
434 
435 			tlb_unlock(pc->pc_booke_tlb_lock);
436 
437 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
438 		}
439 	}
440 #endif
441 }
442 
443 /* Return number of entries in TLB0. */
444 static __inline void
445 tlb0_get_tlbconf(void)
446 {
447 	uint32_t tlb0_cfg;
448 
449 	tlb0_cfg = mfspr(SPR_TLB0CFG);
450 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
451 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
452 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
453 }
454 
455 /* Initialize pool of kva ptbl buffers. */
456 static void
457 ptbl_init(void)
458 {
459 	int i;
460 
461 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
462 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
463 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
464 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
465 
466 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
467 	TAILQ_INIT(&ptbl_buf_freelist);
468 
469 	for (i = 0; i < PTBL_BUFS; i++) {
470 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
471 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
472 	}
473 }
474 
475 /* Get a ptbl_buf from the freelist. */
476 static struct ptbl_buf *
477 ptbl_buf_alloc(void)
478 {
479 	struct ptbl_buf *buf;
480 
481 	mtx_lock(&ptbl_buf_freelist_lock);
482 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
483 	if (buf != NULL)
484 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
485 	mtx_unlock(&ptbl_buf_freelist_lock);
486 
487 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
488 
489 	return (buf);
490 }
491 
492 /* Return ptbl buff to free pool. */
493 static void
494 ptbl_buf_free(struct ptbl_buf *buf)
495 {
496 
497 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
498 
499 	mtx_lock(&ptbl_buf_freelist_lock);
500 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
501 	mtx_unlock(&ptbl_buf_freelist_lock);
502 }
503 
504 /*
505  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
506  */
507 static void
508 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
509 {
510 	struct ptbl_buf *pbuf;
511 
512 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
513 
514 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
515 
516 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
517 		if (pbuf->kva == (vm_offset_t)ptbl) {
518 			/* Remove from pmap ptbl buf list. */
519 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
520 
521 			/* Free corresponding ptbl buf. */
522 			ptbl_buf_free(pbuf);
523 			break;
524 		}
525 }
526 
527 /* Allocate page table. */
528 static pte_t *
529 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
530 {
531 	vm_page_t mtbl[PTBL_PAGES];
532 	vm_page_t m;
533 	struct ptbl_buf *pbuf;
534 	unsigned int pidx;
535 	pte_t *ptbl;
536 	int i;
537 
538 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
539 	    (pmap == kernel_pmap), pdir_idx);
540 
541 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
542 	    ("ptbl_alloc: invalid pdir_idx"));
543 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
544 	    ("pte_alloc: valid ptbl entry exists!"));
545 
546 	pbuf = ptbl_buf_alloc();
547 	if (pbuf == NULL)
548 		panic("pte_alloc: couldn't alloc kernel virtual memory");
549 
550 	ptbl = (pte_t *)pbuf->kva;
551 
552 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
553 
554 	/* Allocate ptbl pages, this will sleep! */
555 	for (i = 0; i < PTBL_PAGES; i++) {
556 		pidx = (PTBL_PAGES * pdir_idx) + i;
557 		while ((m = vm_page_alloc(NULL, pidx,
558 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
559 
560 			PMAP_UNLOCK(pmap);
561 			vm_page_unlock_queues();
562 			VM_WAIT;
563 			vm_page_lock_queues();
564 			PMAP_LOCK(pmap);
565 		}
566 		mtbl[i] = m;
567 	}
568 
569 	/* Map allocated pages into kernel_pmap. */
570 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
571 
572 	/* Zero whole ptbl. */
573 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
574 
575 	/* Add pbuf to the pmap ptbl bufs list. */
576 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
577 
578 	return (ptbl);
579 }
580 
581 /* Free ptbl pages and invalidate pdir entry. */
582 static void
583 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
584 {
585 	pte_t *ptbl;
586 	vm_paddr_t pa;
587 	vm_offset_t va;
588 	vm_page_t m;
589 	int i;
590 
591 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
592 	    (pmap == kernel_pmap), pdir_idx);
593 
594 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
595 	    ("ptbl_free: invalid pdir_idx"));
596 
597 	ptbl = pmap->pm_pdir[pdir_idx];
598 
599 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
600 
601 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
602 
603 	/*
604 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
605 	 * don't attempt to look up the page tables we are releasing.
606 	 */
607 	mtx_lock_spin(&tlbivax_mutex);
608 	tlb_miss_lock();
609 
610 	pmap->pm_pdir[pdir_idx] = NULL;
611 
612 	tlb_miss_unlock();
613 	mtx_unlock_spin(&tlbivax_mutex);
614 
615 	for (i = 0; i < PTBL_PAGES; i++) {
616 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
617 		pa = pte_vatopa(mmu, kernel_pmap, va);
618 		m = PHYS_TO_VM_PAGE(pa);
619 		vm_page_free_zero(m);
620 		atomic_subtract_int(&cnt.v_wire_count, 1);
621 		mmu_booke_kremove(mmu, va);
622 	}
623 
624 	ptbl_free_pmap_ptbl(pmap, ptbl);
625 }
626 
627 /*
628  * Decrement ptbl pages hold count and attempt to free ptbl pages.
629  * Called when removing pte entry from ptbl.
630  *
631  * Return 1 if ptbl pages were freed.
632  */
633 static int
634 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
635 {
636 	pte_t *ptbl;
637 	vm_paddr_t pa;
638 	vm_page_t m;
639 	int i;
640 
641 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
642 	    (pmap == kernel_pmap), pdir_idx);
643 
644 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
645 	    ("ptbl_unhold: invalid pdir_idx"));
646 	KASSERT((pmap != kernel_pmap),
647 	    ("ptbl_unhold: unholding kernel ptbl!"));
648 
649 	ptbl = pmap->pm_pdir[pdir_idx];
650 
651 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
652 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
653 	    ("ptbl_unhold: non kva ptbl"));
654 
655 	/* decrement hold count */
656 	for (i = 0; i < PTBL_PAGES; i++) {
657 		pa = pte_vatopa(mmu, kernel_pmap,
658 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
659 		m = PHYS_TO_VM_PAGE(pa);
660 		m->wire_count--;
661 	}
662 
663 	/*
664 	 * Free ptbl pages if there are no pte etries in this ptbl.
665 	 * wire_count has the same value for all ptbl pages, so check the last
666 	 * page.
667 	 */
668 	if (m->wire_count == 0) {
669 		ptbl_free(mmu, pmap, pdir_idx);
670 
671 		//debugf("ptbl_unhold: e (freed ptbl)\n");
672 		return (1);
673 	}
674 
675 	return (0);
676 }
677 
678 /*
679  * Increment hold count for ptbl pages. This routine is used when a new pte
680  * entry is being inserted into the ptbl.
681  */
682 static void
683 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
684 {
685 	vm_paddr_t pa;
686 	pte_t *ptbl;
687 	vm_page_t m;
688 	int i;
689 
690 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
691 	    pdir_idx);
692 
693 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
694 	    ("ptbl_hold: invalid pdir_idx"));
695 	KASSERT((pmap != kernel_pmap),
696 	    ("ptbl_hold: holding kernel ptbl!"));
697 
698 	ptbl = pmap->pm_pdir[pdir_idx];
699 
700 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
701 
702 	for (i = 0; i < PTBL_PAGES; i++) {
703 		pa = pte_vatopa(mmu, kernel_pmap,
704 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
705 		m = PHYS_TO_VM_PAGE(pa);
706 		m->wire_count++;
707 	}
708 }
709 
710 /* Allocate pv_entry structure. */
711 pv_entry_t
712 pv_alloc(void)
713 {
714 	pv_entry_t pv;
715 
716 	pv_entry_count++;
717 	if (pv_entry_count > pv_entry_high_water)
718 		pagedaemon_wakeup();
719 	pv = uma_zalloc(pvzone, M_NOWAIT);
720 
721 	return (pv);
722 }
723 
724 /* Free pv_entry structure. */
725 static __inline void
726 pv_free(pv_entry_t pve)
727 {
728 
729 	pv_entry_count--;
730 	uma_zfree(pvzone, pve);
731 }
732 
733 
734 /* Allocate and initialize pv_entry structure. */
735 static void
736 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
737 {
738 	pv_entry_t pve;
739 
740 	//int su = (pmap == kernel_pmap);
741 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
742 	//	(u_int32_t)pmap, va, (u_int32_t)m);
743 
744 	pve = pv_alloc();
745 	if (pve == NULL)
746 		panic("pv_insert: no pv entries!");
747 
748 	pve->pv_pmap = pmap;
749 	pve->pv_va = va;
750 
751 	/* add to pv_list */
752 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
753 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
754 
755 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
756 
757 	//debugf("pv_insert: e\n");
758 }
759 
760 /* Destroy pv entry. */
761 static void
762 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
763 {
764 	pv_entry_t pve;
765 
766 	//int su = (pmap == kernel_pmap);
767 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
768 
769 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
770 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
771 
772 	/* find pv entry */
773 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
774 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
775 			/* remove from pv_list */
776 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
777 			if (TAILQ_EMPTY(&m->md.pv_list))
778 				vm_page_flag_clear(m, PG_WRITEABLE);
779 
780 			/* free pv entry struct */
781 			pv_free(pve);
782 			break;
783 		}
784 	}
785 
786 	//debugf("pv_remove: e\n");
787 }
788 
789 /*
790  * Clean pte entry, try to free page table page if requested.
791  *
792  * Return 1 if ptbl pages were freed, otherwise return 0.
793  */
794 static int
795 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
796 {
797 	unsigned int pdir_idx = PDIR_IDX(va);
798 	unsigned int ptbl_idx = PTBL_IDX(va);
799 	vm_page_t m;
800 	pte_t *ptbl;
801 	pte_t *pte;
802 
803 	//int su = (pmap == kernel_pmap);
804 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
805 	//		su, (u_int32_t)pmap, va, flags);
806 
807 	ptbl = pmap->pm_pdir[pdir_idx];
808 	KASSERT(ptbl, ("pte_remove: null ptbl"));
809 
810 	pte = &ptbl[ptbl_idx];
811 
812 	if (pte == NULL || !PTE_ISVALID(pte))
813 		return (0);
814 
815 	if (PTE_ISWIRED(pte))
816 		pmap->pm_stats.wired_count--;
817 
818 	/* Handle managed entry. */
819 	if (PTE_ISMANAGED(pte)) {
820 		/* Get vm_page_t for mapped pte. */
821 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
822 
823 		if (PTE_ISMODIFIED(pte))
824 			vm_page_dirty(m);
825 
826 		if (PTE_ISREFERENCED(pte))
827 			vm_page_flag_set(m, PG_REFERENCED);
828 
829 		pv_remove(pmap, va, m);
830 	}
831 
832 	mtx_lock_spin(&tlbivax_mutex);
833 	tlb_miss_lock();
834 
835 	tlb0_flush_entry(va);
836 	pte->flags = 0;
837 	pte->rpn = 0;
838 
839 	tlb_miss_unlock();
840 	mtx_unlock_spin(&tlbivax_mutex);
841 
842 	pmap->pm_stats.resident_count--;
843 
844 	if (flags & PTBL_UNHOLD) {
845 		//debugf("pte_remove: e (unhold)\n");
846 		return (ptbl_unhold(mmu, pmap, pdir_idx));
847 	}
848 
849 	//debugf("pte_remove: e\n");
850 	return (0);
851 }
852 
853 /*
854  * Insert PTE for a given page and virtual address.
855  */
856 static void
857 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
858 {
859 	unsigned int pdir_idx = PDIR_IDX(va);
860 	unsigned int ptbl_idx = PTBL_IDX(va);
861 	pte_t *ptbl, *pte;
862 
863 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
864 	    pmap == kernel_pmap, pmap, va);
865 
866 	/* Get the page table pointer. */
867 	ptbl = pmap->pm_pdir[pdir_idx];
868 
869 	if (ptbl == NULL) {
870 		/* Allocate page table pages. */
871 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
872 	} else {
873 		/*
874 		 * Check if there is valid mapping for requested
875 		 * va, if there is, remove it.
876 		 */
877 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
878 		if (PTE_ISVALID(pte)) {
879 			pte_remove(mmu, pmap, va, PTBL_HOLD);
880 		} else {
881 			/*
882 			 * pte is not used, increment hold count
883 			 * for ptbl pages.
884 			 */
885 			if (pmap != kernel_pmap)
886 				ptbl_hold(mmu, pmap, pdir_idx);
887 		}
888 	}
889 
890 	/*
891 	 * Insert pv_entry into pv_list for mapped page if part of managed
892 	 * memory.
893 	 */
894         if ((m->flags & PG_FICTITIOUS) == 0) {
895 		if ((m->flags & PG_UNMANAGED) == 0) {
896 			flags |= PTE_MANAGED;
897 
898 			/* Create and insert pv entry. */
899 			pv_insert(pmap, va, m);
900 		}
901 	}
902 
903 	pmap->pm_stats.resident_count++;
904 
905 	mtx_lock_spin(&tlbivax_mutex);
906 	tlb_miss_lock();
907 
908 	tlb0_flush_entry(va);
909 	if (pmap->pm_pdir[pdir_idx] == NULL) {
910 		/*
911 		 * If we just allocated a new page table, hook it in
912 		 * the pdir.
913 		 */
914 		pmap->pm_pdir[pdir_idx] = ptbl;
915 	}
916 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
917 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
918 	pte->flags |= (PTE_VALID | flags);
919 
920 	tlb_miss_unlock();
921 	mtx_unlock_spin(&tlbivax_mutex);
922 }
923 
924 /* Return the pa for the given pmap/va. */
925 static vm_paddr_t
926 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
927 {
928 	vm_paddr_t pa = 0;
929 	pte_t *pte;
930 
931 	pte = pte_find(mmu, pmap, va);
932 	if ((pte != NULL) && PTE_ISVALID(pte))
933 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
934 	return (pa);
935 }
936 
937 /* Get a pointer to a PTE in a page table. */
938 static pte_t *
939 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
940 {
941 	unsigned int pdir_idx = PDIR_IDX(va);
942 	unsigned int ptbl_idx = PTBL_IDX(va);
943 
944 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
945 
946 	if (pmap->pm_pdir[pdir_idx])
947 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
948 
949 	return (NULL);
950 }
951 
952 /**************************************************************************/
953 /* PMAP related */
954 /**************************************************************************/
955 
956 /*
957  * This is called during e500_init, before the system is really initialized.
958  */
959 static void
960 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
961 {
962 	vm_offset_t phys_kernelend;
963 	struct mem_region *mp, *mp1;
964 	int cnt, i, j;
965 	u_int s, e, sz;
966 	u_int phys_avail_count;
967 	vm_size_t physsz, hwphyssz, kstack0_sz;
968 	vm_offset_t kernel_pdir, kstack0, va;
969 	vm_paddr_t kstack0_phys;
970 	void *dpcpu;
971 	pte_t *pte;
972 
973 	debugf("mmu_booke_bootstrap: entered\n");
974 
975 	/* Initialize invalidation mutex */
976 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
977 
978 	/* Read TLB0 size and associativity. */
979 	tlb0_get_tlbconf();
980 
981 	/* Align kernel start and end address (kernel image). */
982 	kernstart = trunc_page(start);
983 	data_start = round_page(kernelend);
984 	kernsize = data_start - kernstart;
985 
986 	data_end = data_start;
987 
988 	/* Allocate space for the message buffer. */
989 	msgbufp = (struct msgbuf *)data_end;
990 	data_end += MSGBUF_SIZE;
991 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
992 	    data_end);
993 
994 	data_end = round_page(data_end);
995 
996 	/* Allocate the dynamic per-cpu area. */
997 	dpcpu = (void *)data_end;
998 	data_end += DPCPU_SIZE;
999 	dpcpu_init(dpcpu, 0);
1000 
1001 	/* Allocate space for ptbl_bufs. */
1002 	ptbl_bufs = (struct ptbl_buf *)data_end;
1003 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1004 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1005 	    data_end);
1006 
1007 	data_end = round_page(data_end);
1008 
1009 	/* Allocate PTE tables for kernel KVA. */
1010 	kernel_pdir = data_end;
1011 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1012 	    PDIR_SIZE - 1) / PDIR_SIZE;
1013 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1014 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1015 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1016 
1017 	debugf(" data_end: 0x%08x\n", data_end);
1018 	if (data_end - kernstart > 0x1000000) {
1019 		data_end = (data_end + 0x3fffff) & ~0x3fffff;
1020 		tlb1_mapin_region(kernstart + 0x1000000,
1021 		    kernload + 0x1000000, data_end - kernstart - 0x1000000);
1022 	} else
1023 		data_end = (data_end + 0xffffff) & ~0xffffff;
1024 
1025 	debugf(" updated data_end: 0x%08x\n", data_end);
1026 
1027 	kernsize += data_end - data_start;
1028 
1029 	/*
1030 	 * Clear the structures - note we can only do it safely after the
1031 	 * possible additional TLB1 translations are in place (above) so that
1032 	 * all range up to the currently calculated 'data_end' is covered.
1033 	 */
1034 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1035 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1036 
1037 	/*******************************************************/
1038 	/* Set the start and end of kva. */
1039 	/*******************************************************/
1040 	virtual_avail = round_page(data_end);
1041 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1042 
1043 	/* Allocate KVA space for page zero/copy operations. */
1044 	zero_page_va = virtual_avail;
1045 	virtual_avail += PAGE_SIZE;
1046 	zero_page_idle_va = virtual_avail;
1047 	virtual_avail += PAGE_SIZE;
1048 	copy_page_src_va = virtual_avail;
1049 	virtual_avail += PAGE_SIZE;
1050 	copy_page_dst_va = virtual_avail;
1051 	virtual_avail += PAGE_SIZE;
1052 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1053 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1054 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1055 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1056 
1057 	/* Initialize page zero/copy mutexes. */
1058 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1059 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1060 
1061 	/* Allocate KVA space for ptbl bufs. */
1062 	ptbl_buf_pool_vabase = virtual_avail;
1063 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1064 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1065 	    ptbl_buf_pool_vabase, virtual_avail);
1066 
1067 	/* Calculate corresponding physical addresses for the kernel region. */
1068 	phys_kernelend = kernload + kernsize;
1069 	debugf("kernel image and allocated data:\n");
1070 	debugf(" kernload    = 0x%08x\n", kernload);
1071 	debugf(" kernstart   = 0x%08x\n", kernstart);
1072 	debugf(" kernsize    = 0x%08x\n", kernsize);
1073 
1074 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1075 		panic("mmu_booke_bootstrap: phys_avail too small");
1076 
1077 	/*
1078 	 * Remove kernel physical address range from avail regions list. Page
1079 	 * align all regions.  Non-page aligned memory isn't very interesting
1080 	 * to us.  Also, sort the entries for ascending addresses.
1081 	 */
1082 
1083 	/* Retrieve phys/avail mem regions */
1084 	mem_regions(&physmem_regions, &physmem_regions_sz,
1085 	    &availmem_regions, &availmem_regions_sz);
1086 	sz = 0;
1087 	cnt = availmem_regions_sz;
1088 	debugf("processing avail regions:\n");
1089 	for (mp = availmem_regions; mp->mr_size; mp++) {
1090 		s = mp->mr_start;
1091 		e = mp->mr_start + mp->mr_size;
1092 		debugf(" %08x-%08x -> ", s, e);
1093 		/* Check whether this region holds all of the kernel. */
1094 		if (s < kernload && e > phys_kernelend) {
1095 			availmem_regions[cnt].mr_start = phys_kernelend;
1096 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1097 			e = kernload;
1098 		}
1099 		/* Look whether this regions starts within the kernel. */
1100 		if (s >= kernload && s < phys_kernelend) {
1101 			if (e <= phys_kernelend)
1102 				goto empty;
1103 			s = phys_kernelend;
1104 		}
1105 		/* Now look whether this region ends within the kernel. */
1106 		if (e > kernload && e <= phys_kernelend) {
1107 			if (s >= kernload)
1108 				goto empty;
1109 			e = kernload;
1110 		}
1111 		/* Now page align the start and size of the region. */
1112 		s = round_page(s);
1113 		e = trunc_page(e);
1114 		if (e < s)
1115 			e = s;
1116 		sz = e - s;
1117 		debugf("%08x-%08x = %x\n", s, e, sz);
1118 
1119 		/* Check whether some memory is left here. */
1120 		if (sz == 0) {
1121 		empty:
1122 			memmove(mp, mp + 1,
1123 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1124 			cnt--;
1125 			mp--;
1126 			continue;
1127 		}
1128 
1129 		/* Do an insertion sort. */
1130 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1131 			if (s < mp1->mr_start)
1132 				break;
1133 		if (mp1 < mp) {
1134 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1135 			mp1->mr_start = s;
1136 			mp1->mr_size = sz;
1137 		} else {
1138 			mp->mr_start = s;
1139 			mp->mr_size = sz;
1140 		}
1141 	}
1142 	availmem_regions_sz = cnt;
1143 
1144 	/*******************************************************/
1145 	/* Steal physical memory for kernel stack from the end */
1146 	/* of the first avail region                           */
1147 	/*******************************************************/
1148 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1149 	kstack0_phys = availmem_regions[0].mr_start +
1150 	    availmem_regions[0].mr_size;
1151 	kstack0_phys -= kstack0_sz;
1152 	availmem_regions[0].mr_size -= kstack0_sz;
1153 
1154 	/*******************************************************/
1155 	/* Fill in phys_avail table, based on availmem_regions */
1156 	/*******************************************************/
1157 	phys_avail_count = 0;
1158 	physsz = 0;
1159 	hwphyssz = 0;
1160 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1161 
1162 	debugf("fill in phys_avail:\n");
1163 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1164 
1165 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1166 		    availmem_regions[i].mr_start,
1167 		    availmem_regions[i].mr_start +
1168 		        availmem_regions[i].mr_size,
1169 		    availmem_regions[i].mr_size);
1170 
1171 		if (hwphyssz != 0 &&
1172 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1173 			debugf(" hw.physmem adjust\n");
1174 			if (physsz < hwphyssz) {
1175 				phys_avail[j] = availmem_regions[i].mr_start;
1176 				phys_avail[j + 1] =
1177 				    availmem_regions[i].mr_start +
1178 				    hwphyssz - physsz;
1179 				physsz = hwphyssz;
1180 				phys_avail_count++;
1181 			}
1182 			break;
1183 		}
1184 
1185 		phys_avail[j] = availmem_regions[i].mr_start;
1186 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1187 		    availmem_regions[i].mr_size;
1188 		phys_avail_count++;
1189 		physsz += availmem_regions[i].mr_size;
1190 	}
1191 	physmem = btoc(physsz);
1192 
1193 	/* Calculate the last available physical address. */
1194 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1195 		;
1196 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1197 
1198 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1199 	debugf("phys_avail_count = %d\n", phys_avail_count);
1200 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1201 	    physmem);
1202 
1203 	/*******************************************************/
1204 	/* Initialize (statically allocated) kernel pmap. */
1205 	/*******************************************************/
1206 	PMAP_LOCK_INIT(kernel_pmap);
1207 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1208 
1209 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1210 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1211 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1212 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1213 
1214 	/* Initialize kernel pdir */
1215 	for (i = 0; i < kernel_ptbls; i++)
1216 		kernel_pmap->pm_pdir[kptbl_min + i] =
1217 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1218 
1219 	for (i = 0; i < MAXCPU; i++) {
1220 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1221 
1222 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1223 		tidbusy[i][0] = kernel_pmap;
1224 	}
1225 
1226 	/*
1227 	 * Fill in PTEs covering kernel code and data. They are not required
1228 	 * for address translation, as this area is covered by static TLB1
1229 	 * entries, but for pte_vatopa() to work correctly with kernel area
1230 	 * addresses.
1231 	 */
1232 	for (va = KERNBASE; va < data_end; va += PAGE_SIZE) {
1233 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1234 		pte->rpn = kernload + (va - KERNBASE);
1235 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1236 		    PTE_VALID;
1237 	}
1238 	/* Mark kernel_pmap active on all CPUs */
1239 	kernel_pmap->pm_active = ~0;
1240 
1241 	/*******************************************************/
1242 	/* Final setup */
1243 	/*******************************************************/
1244 
1245 	/* Enter kstack0 into kernel map, provide guard page */
1246 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1247 	thread0.td_kstack = kstack0;
1248 	thread0.td_kstack_pages = KSTACK_PAGES;
1249 
1250 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1251 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1252 	    kstack0_phys, kstack0_phys + kstack0_sz);
1253 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1254 
1255 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1256 	for (i = 0; i < KSTACK_PAGES; i++) {
1257 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1258 		kstack0 += PAGE_SIZE;
1259 		kstack0_phys += PAGE_SIZE;
1260 	}
1261 
1262 	debugf("virtual_avail = %08x\n", virtual_avail);
1263 	debugf("virtual_end   = %08x\n", virtual_end);
1264 
1265 	debugf("mmu_booke_bootstrap: exit\n");
1266 }
1267 
1268 void
1269 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1270 {
1271 	int i;
1272 
1273 	/*
1274 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1275 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1276 	 * these values directly to (re)program AP's TLB1 hardware.
1277 	 */
1278 	for (i = 0; i < tlb1_idx; i ++) {
1279 		/* Skip invalid entries */
1280 		if (!(tlb1[i].mas1 & MAS1_VALID))
1281 			continue;
1282 
1283 		tlb1_write_entry(i);
1284 	}
1285 
1286 	set_mas4_defaults();
1287 }
1288 
1289 /*
1290  * Get the physical page address for the given pmap/virtual address.
1291  */
1292 static vm_paddr_t
1293 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1294 {
1295 	vm_paddr_t pa;
1296 
1297 	PMAP_LOCK(pmap);
1298 	pa = pte_vatopa(mmu, pmap, va);
1299 	PMAP_UNLOCK(pmap);
1300 
1301 	return (pa);
1302 }
1303 
1304 /*
1305  * Extract the physical page address associated with the given
1306  * kernel virtual address.
1307  */
1308 static vm_paddr_t
1309 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1310 {
1311 
1312 	return (pte_vatopa(mmu, kernel_pmap, va));
1313 }
1314 
1315 /*
1316  * Initialize the pmap module.
1317  * Called by vm_init, to initialize any structures that the pmap
1318  * system needs to map virtual memory.
1319  */
1320 static void
1321 mmu_booke_init(mmu_t mmu)
1322 {
1323 	int shpgperproc = PMAP_SHPGPERPROC;
1324 
1325 	/*
1326 	 * Initialize the address space (zone) for the pv entries.  Set a
1327 	 * high water mark so that the system can recover from excessive
1328 	 * numbers of pv entries.
1329 	 */
1330 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1331 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1332 
1333 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1334 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1335 
1336 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1337 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1338 
1339 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1340 
1341 	/* Pre-fill pvzone with initial number of pv entries. */
1342 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1343 
1344 	/* Initialize ptbl allocation. */
1345 	ptbl_init();
1346 }
1347 
1348 /*
1349  * Map a list of wired pages into kernel virtual address space.  This is
1350  * intended for temporary mappings which do not need page modification or
1351  * references recorded.  Existing mappings in the region are overwritten.
1352  */
1353 static void
1354 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1355 {
1356 	vm_offset_t va;
1357 
1358 	va = sva;
1359 	while (count-- > 0) {
1360 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1361 		va += PAGE_SIZE;
1362 		m++;
1363 	}
1364 }
1365 
1366 /*
1367  * Remove page mappings from kernel virtual address space.  Intended for
1368  * temporary mappings entered by mmu_booke_qenter.
1369  */
1370 static void
1371 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1372 {
1373 	vm_offset_t va;
1374 
1375 	va = sva;
1376 	while (count-- > 0) {
1377 		mmu_booke_kremove(mmu, va);
1378 		va += PAGE_SIZE;
1379 	}
1380 }
1381 
1382 /*
1383  * Map a wired page into kernel virtual address space.
1384  */
1385 static void
1386 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1387 {
1388 	unsigned int pdir_idx = PDIR_IDX(va);
1389 	unsigned int ptbl_idx = PTBL_IDX(va);
1390 	uint32_t flags;
1391 	pte_t *pte;
1392 
1393 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1394 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1395 
1396 	flags = 0;
1397 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1398 	flags |= PTE_M;
1399 
1400 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1401 
1402 	mtx_lock_spin(&tlbivax_mutex);
1403 	tlb_miss_lock();
1404 
1405 	if (PTE_ISVALID(pte)) {
1406 
1407 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1408 
1409 		/* Flush entry from TLB0 */
1410 		tlb0_flush_entry(va);
1411 	}
1412 
1413 	pte->rpn = pa & ~PTE_PA_MASK;
1414 	pte->flags = flags;
1415 
1416 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1417 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1418 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1419 
1420 	/* Flush the real memory from the instruction cache. */
1421 	if ((flags & (PTE_I | PTE_G)) == 0) {
1422 		__syncicache((void *)va, PAGE_SIZE);
1423 	}
1424 
1425 	tlb_miss_unlock();
1426 	mtx_unlock_spin(&tlbivax_mutex);
1427 }
1428 
1429 /*
1430  * Remove a page from kernel page table.
1431  */
1432 static void
1433 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1434 {
1435 	unsigned int pdir_idx = PDIR_IDX(va);
1436 	unsigned int ptbl_idx = PTBL_IDX(va);
1437 	pte_t *pte;
1438 
1439 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1440 
1441 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1442 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1443 	    ("mmu_booke_kremove: invalid va"));
1444 
1445 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1446 
1447 	if (!PTE_ISVALID(pte)) {
1448 
1449 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1450 
1451 		return;
1452 	}
1453 
1454 	mtx_lock_spin(&tlbivax_mutex);
1455 	tlb_miss_lock();
1456 
1457 	/* Invalidate entry in TLB0, update PTE. */
1458 	tlb0_flush_entry(va);
1459 	pte->flags = 0;
1460 	pte->rpn = 0;
1461 
1462 	tlb_miss_unlock();
1463 	mtx_unlock_spin(&tlbivax_mutex);
1464 }
1465 
1466 /*
1467  * Initialize pmap associated with process 0.
1468  */
1469 static void
1470 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1471 {
1472 
1473 	mmu_booke_pinit(mmu, pmap);
1474 	PCPU_SET(curpmap, pmap);
1475 }
1476 
1477 /*
1478  * Initialize a preallocated and zeroed pmap structure,
1479  * such as one in a vmspace structure.
1480  */
1481 static void
1482 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1483 {
1484 	int i;
1485 
1486 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1487 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1488 
1489 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1490 
1491 	PMAP_LOCK_INIT(pmap);
1492 	for (i = 0; i < MAXCPU; i++)
1493 		pmap->pm_tid[i] = TID_NONE;
1494 	pmap->pm_active = 0;
1495 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1496 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1497 	TAILQ_INIT(&pmap->pm_ptbl_list);
1498 }
1499 
1500 /*
1501  * Release any resources held by the given physical map.
1502  * Called when a pmap initialized by mmu_booke_pinit is being released.
1503  * Should only be called if the map contains no valid mappings.
1504  */
1505 static void
1506 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1507 {
1508 
1509 	printf("mmu_booke_release: s\n");
1510 
1511 	KASSERT(pmap->pm_stats.resident_count == 0,
1512 	    ("pmap_release: pmap resident count %ld != 0",
1513 	    pmap->pm_stats.resident_count));
1514 
1515 	PMAP_LOCK_DESTROY(pmap);
1516 }
1517 
1518 /*
1519  * Insert the given physical page at the specified virtual address in the
1520  * target physical map with the protection requested. If specified the page
1521  * will be wired down.
1522  */
1523 static void
1524 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1525     vm_prot_t prot, boolean_t wired)
1526 {
1527 
1528 	vm_page_lock_queues();
1529 	PMAP_LOCK(pmap);
1530 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1531 	vm_page_unlock_queues();
1532 	PMAP_UNLOCK(pmap);
1533 }
1534 
1535 static void
1536 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1537     vm_prot_t prot, boolean_t wired)
1538 {
1539 	pte_t *pte;
1540 	vm_paddr_t pa;
1541 	uint32_t flags;
1542 	int su, sync;
1543 
1544 	pa = VM_PAGE_TO_PHYS(m);
1545 	su = (pmap == kernel_pmap);
1546 	sync = 0;
1547 
1548 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1549 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1550 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1551 	//		(u_int32_t)m, va, pa, prot, wired);
1552 
1553 	if (su) {
1554 		KASSERT(((va >= virtual_avail) &&
1555 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1556 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1557 	} else {
1558 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1559 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1560 	}
1561 	KASSERT((m->oflags & VPO_BUSY) != 0 || VM_OBJECT_LOCKED(m->object),
1562 	    ("mmu_booke_enter_locked: page %p is not busy", m));
1563 
1564 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1565 
1566 	/*
1567 	 * If there is an existing mapping, and the physical address has not
1568 	 * changed, must be protection or wiring change.
1569 	 */
1570 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1571 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1572 
1573 		/*
1574 		 * Before actually updating pte->flags we calculate and
1575 		 * prepare its new value in a helper var.
1576 		 */
1577 		flags = pte->flags;
1578 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1579 
1580 		/* Wiring change, just update stats. */
1581 		if (wired) {
1582 			if (!PTE_ISWIRED(pte)) {
1583 				flags |= PTE_WIRED;
1584 				pmap->pm_stats.wired_count++;
1585 			}
1586 		} else {
1587 			if (PTE_ISWIRED(pte)) {
1588 				flags &= ~PTE_WIRED;
1589 				pmap->pm_stats.wired_count--;
1590 			}
1591 		}
1592 
1593 		if (prot & VM_PROT_WRITE) {
1594 			/* Add write permissions. */
1595 			flags |= PTE_SW;
1596 			if (!su)
1597 				flags |= PTE_UW;
1598 
1599 			if ((flags & PTE_MANAGED) != 0)
1600 				vm_page_flag_set(m, PG_WRITEABLE);
1601 		} else {
1602 			/* Handle modified pages, sense modify status. */
1603 
1604 			/*
1605 			 * The PTE_MODIFIED flag could be set by underlying
1606 			 * TLB misses since we last read it (above), possibly
1607 			 * other CPUs could update it so we check in the PTE
1608 			 * directly rather than rely on that saved local flags
1609 			 * copy.
1610 			 */
1611 			if (PTE_ISMODIFIED(pte))
1612 				vm_page_dirty(m);
1613 		}
1614 
1615 		if (prot & VM_PROT_EXECUTE) {
1616 			flags |= PTE_SX;
1617 			if (!su)
1618 				flags |= PTE_UX;
1619 
1620 			/*
1621 			 * Check existing flags for execute permissions: if we
1622 			 * are turning execute permissions on, icache should
1623 			 * be flushed.
1624 			 */
1625 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1626 				sync++;
1627 		}
1628 
1629 		flags &= ~PTE_REFERENCED;
1630 
1631 		/*
1632 		 * The new flags value is all calculated -- only now actually
1633 		 * update the PTE.
1634 		 */
1635 		mtx_lock_spin(&tlbivax_mutex);
1636 		tlb_miss_lock();
1637 
1638 		tlb0_flush_entry(va);
1639 		pte->flags = flags;
1640 
1641 		tlb_miss_unlock();
1642 		mtx_unlock_spin(&tlbivax_mutex);
1643 
1644 	} else {
1645 		/*
1646 		 * If there is an existing mapping, but it's for a different
1647 		 * physical address, pte_enter() will delete the old mapping.
1648 		 */
1649 		//if ((pte != NULL) && PTE_ISVALID(pte))
1650 		//	debugf("mmu_booke_enter_locked: replace\n");
1651 		//else
1652 		//	debugf("mmu_booke_enter_locked: new\n");
1653 
1654 		/* Now set up the flags and install the new mapping. */
1655 		flags = (PTE_SR | PTE_VALID);
1656 		flags |= PTE_M;
1657 
1658 		if (!su)
1659 			flags |= PTE_UR;
1660 
1661 		if (prot & VM_PROT_WRITE) {
1662 			flags |= PTE_SW;
1663 			if (!su)
1664 				flags |= PTE_UW;
1665 
1666 			if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0)
1667 				vm_page_flag_set(m, PG_WRITEABLE);
1668 		}
1669 
1670 		if (prot & VM_PROT_EXECUTE) {
1671 			flags |= PTE_SX;
1672 			if (!su)
1673 				flags |= PTE_UX;
1674 		}
1675 
1676 		/* If its wired update stats. */
1677 		if (wired) {
1678 			pmap->pm_stats.wired_count++;
1679 			flags |= PTE_WIRED;
1680 		}
1681 
1682 		pte_enter(mmu, pmap, m, va, flags);
1683 
1684 		/* Flush the real memory from the instruction cache. */
1685 		if (prot & VM_PROT_EXECUTE)
1686 			sync++;
1687 	}
1688 
1689 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1690 		__syncicache((void *)va, PAGE_SIZE);
1691 		sync = 0;
1692 	}
1693 }
1694 
1695 /*
1696  * Maps a sequence of resident pages belonging to the same object.
1697  * The sequence begins with the given page m_start.  This page is
1698  * mapped at the given virtual address start.  Each subsequent page is
1699  * mapped at a virtual address that is offset from start by the same
1700  * amount as the page is offset from m_start within the object.  The
1701  * last page in the sequence is the page with the largest offset from
1702  * m_start that can be mapped at a virtual address less than the given
1703  * virtual address end.  Not every virtual page between start and end
1704  * is mapped; only those for which a resident page exists with the
1705  * corresponding offset from m_start are mapped.
1706  */
1707 static void
1708 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1709     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1710 {
1711 	vm_page_t m;
1712 	vm_pindex_t diff, psize;
1713 
1714 	psize = atop(end - start);
1715 	m = m_start;
1716 	vm_page_lock_queues();
1717 	PMAP_LOCK(pmap);
1718 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1719 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1720 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1721 		m = TAILQ_NEXT(m, listq);
1722 	}
1723 	vm_page_unlock_queues();
1724 	PMAP_UNLOCK(pmap);
1725 }
1726 
1727 static void
1728 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1729     vm_prot_t prot)
1730 {
1731 
1732 	vm_page_lock_queues();
1733 	PMAP_LOCK(pmap);
1734 	mmu_booke_enter_locked(mmu, pmap, va, m,
1735 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1736 	vm_page_unlock_queues();
1737 	PMAP_UNLOCK(pmap);
1738 }
1739 
1740 /*
1741  * Remove the given range of addresses from the specified map.
1742  *
1743  * It is assumed that the start and end are properly rounded to the page size.
1744  */
1745 static void
1746 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1747 {
1748 	pte_t *pte;
1749 	uint8_t hold_flag;
1750 
1751 	int su = (pmap == kernel_pmap);
1752 
1753 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1754 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1755 
1756 	if (su) {
1757 		KASSERT(((va >= virtual_avail) &&
1758 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1759 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1760 	} else {
1761 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1762 		    ("mmu_booke_remove: user pmap, non user va"));
1763 	}
1764 
1765 	if (PMAP_REMOVE_DONE(pmap)) {
1766 		//debugf("mmu_booke_remove: e (empty)\n");
1767 		return;
1768 	}
1769 
1770 	hold_flag = PTBL_HOLD_FLAG(pmap);
1771 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1772 
1773 	vm_page_lock_queues();
1774 	PMAP_LOCK(pmap);
1775 	for (; va < endva; va += PAGE_SIZE) {
1776 		pte = pte_find(mmu, pmap, va);
1777 		if ((pte != NULL) && PTE_ISVALID(pte))
1778 			pte_remove(mmu, pmap, va, hold_flag);
1779 	}
1780 	PMAP_UNLOCK(pmap);
1781 	vm_page_unlock_queues();
1782 
1783 	//debugf("mmu_booke_remove: e\n");
1784 }
1785 
1786 /*
1787  * Remove physical page from all pmaps in which it resides.
1788  */
1789 static void
1790 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1791 {
1792 	pv_entry_t pv, pvn;
1793 	uint8_t hold_flag;
1794 
1795 	vm_page_lock_queues();
1796 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1797 		pvn = TAILQ_NEXT(pv, pv_link);
1798 
1799 		PMAP_LOCK(pv->pv_pmap);
1800 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1801 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1802 		PMAP_UNLOCK(pv->pv_pmap);
1803 	}
1804 	vm_page_flag_clear(m, PG_WRITEABLE);
1805 	vm_page_unlock_queues();
1806 }
1807 
1808 /*
1809  * Map a range of physical addresses into kernel virtual address space.
1810  */
1811 static vm_offset_t
1812 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1813     vm_offset_t pa_end, int prot)
1814 {
1815 	vm_offset_t sva = *virt;
1816 	vm_offset_t va = sva;
1817 
1818 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1819 	//		sva, pa_start, pa_end);
1820 
1821 	while (pa_start < pa_end) {
1822 		mmu_booke_kenter(mmu, va, pa_start);
1823 		va += PAGE_SIZE;
1824 		pa_start += PAGE_SIZE;
1825 	}
1826 	*virt = va;
1827 
1828 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1829 	return (sva);
1830 }
1831 
1832 /*
1833  * The pmap must be activated before it's address space can be accessed in any
1834  * way.
1835  */
1836 static void
1837 mmu_booke_activate(mmu_t mmu, struct thread *td)
1838 {
1839 	pmap_t pmap;
1840 
1841 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1842 
1843 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1844 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1845 
1846 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1847 
1848 	mtx_lock_spin(&sched_lock);
1849 
1850 	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
1851 	PCPU_SET(curpmap, pmap);
1852 
1853 	if (pmap->pm_tid[PCPU_GET(cpuid)] == TID_NONE)
1854 		tid_alloc(pmap);
1855 
1856 	/* Load PID0 register with pmap tid value. */
1857 	mtspr(SPR_PID0, pmap->pm_tid[PCPU_GET(cpuid)]);
1858 	__asm __volatile("isync");
1859 
1860 	mtx_unlock_spin(&sched_lock);
1861 
1862 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1863 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1864 }
1865 
1866 /*
1867  * Deactivate the specified process's address space.
1868  */
1869 static void
1870 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1871 {
1872 	pmap_t pmap;
1873 
1874 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1875 
1876 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1877 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1878 
1879 	atomic_clear_int(&pmap->pm_active, PCPU_GET(cpumask));
1880 	PCPU_SET(curpmap, NULL);
1881 }
1882 
1883 /*
1884  * Copy the range specified by src_addr/len
1885  * from the source map to the range dst_addr/len
1886  * in the destination map.
1887  *
1888  * This routine is only advisory and need not do anything.
1889  */
1890 static void
1891 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1892     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1893 {
1894 
1895 }
1896 
1897 /*
1898  * Set the physical protection on the specified range of this map as requested.
1899  */
1900 static void
1901 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1902     vm_prot_t prot)
1903 {
1904 	vm_offset_t va;
1905 	vm_page_t m;
1906 	pte_t *pte;
1907 
1908 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1909 		mmu_booke_remove(mmu, pmap, sva, eva);
1910 		return;
1911 	}
1912 
1913 	if (prot & VM_PROT_WRITE)
1914 		return;
1915 
1916 	vm_page_lock_queues();
1917 	PMAP_LOCK(pmap);
1918 	for (va = sva; va < eva; va += PAGE_SIZE) {
1919 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1920 			if (PTE_ISVALID(pte)) {
1921 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1922 
1923 				mtx_lock_spin(&tlbivax_mutex);
1924 				tlb_miss_lock();
1925 
1926 				/* Handle modified pages. */
1927 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1928 					vm_page_dirty(m);
1929 
1930 				tlb0_flush_entry(va);
1931 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1932 
1933 				tlb_miss_unlock();
1934 				mtx_unlock_spin(&tlbivax_mutex);
1935 			}
1936 		}
1937 	}
1938 	PMAP_UNLOCK(pmap);
1939 	vm_page_unlock_queues();
1940 }
1941 
1942 /*
1943  * Clear the write and modified bits in each of the given page's mappings.
1944  */
1945 static void
1946 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1947 {
1948 	pv_entry_t pv;
1949 	pte_t *pte;
1950 
1951 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
1952 	    ("mmu_booke_remove_write: page %p is not managed", m));
1953 
1954 	/*
1955 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be set by
1956 	 * another thread while the object is locked.  Thus, if PG_WRITEABLE
1957 	 * is clear, no page table entries need updating.
1958 	 */
1959 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1960 	if ((m->oflags & VPO_BUSY) == 0 &&
1961 	    (m->flags & PG_WRITEABLE) == 0)
1962 		return;
1963 	vm_page_lock_queues();
1964 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1965 		PMAP_LOCK(pv->pv_pmap);
1966 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1967 			if (PTE_ISVALID(pte)) {
1968 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1969 
1970 				mtx_lock_spin(&tlbivax_mutex);
1971 				tlb_miss_lock();
1972 
1973 				/* Handle modified pages. */
1974 				if (PTE_ISMODIFIED(pte))
1975 					vm_page_dirty(m);
1976 
1977 				/* Flush mapping from TLB0. */
1978 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1979 
1980 				tlb_miss_unlock();
1981 				mtx_unlock_spin(&tlbivax_mutex);
1982 			}
1983 		}
1984 		PMAP_UNLOCK(pv->pv_pmap);
1985 	}
1986 	vm_page_flag_clear(m, PG_WRITEABLE);
1987 	vm_page_unlock_queues();
1988 }
1989 
1990 static void
1991 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1992 {
1993 	pte_t *pte;
1994 	pmap_t pmap;
1995 	vm_page_t m;
1996 	vm_offset_t addr;
1997 	vm_paddr_t pa;
1998 	int active, valid;
1999 
2000 	va = trunc_page(va);
2001 	sz = round_page(sz);
2002 
2003 	vm_page_lock_queues();
2004 	pmap = PCPU_GET(curpmap);
2005 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2006 	while (sz > 0) {
2007 		PMAP_LOCK(pm);
2008 		pte = pte_find(mmu, pm, va);
2009 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2010 		if (valid)
2011 			pa = PTE_PA(pte);
2012 		PMAP_UNLOCK(pm);
2013 		if (valid) {
2014 			if (!active) {
2015 				/* Create a mapping in the active pmap. */
2016 				addr = 0;
2017 				m = PHYS_TO_VM_PAGE(pa);
2018 				PMAP_LOCK(pmap);
2019 				pte_enter(mmu, pmap, m, addr,
2020 				    PTE_SR | PTE_VALID | PTE_UR);
2021 				__syncicache((void *)addr, PAGE_SIZE);
2022 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2023 				PMAP_UNLOCK(pmap);
2024 			} else
2025 				__syncicache((void *)va, PAGE_SIZE);
2026 		}
2027 		va += PAGE_SIZE;
2028 		sz -= PAGE_SIZE;
2029 	}
2030 	vm_page_unlock_queues();
2031 }
2032 
2033 /*
2034  * Atomically extract and hold the physical page with the given
2035  * pmap and virtual address pair if that mapping permits the given
2036  * protection.
2037  */
2038 static vm_page_t
2039 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2040     vm_prot_t prot)
2041 {
2042 	pte_t *pte;
2043 	vm_page_t m;
2044 	uint32_t pte_wbit;
2045 	vm_paddr_t pa;
2046 
2047 	m = NULL;
2048 	pa = 0;
2049 	PMAP_LOCK(pmap);
2050 retry:
2051 	pte = pte_find(mmu, pmap, va);
2052 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2053 		if (pmap == kernel_pmap)
2054 			pte_wbit = PTE_SW;
2055 		else
2056 			pte_wbit = PTE_UW;
2057 
2058 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2059 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2060 				goto retry;
2061 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2062 			vm_page_hold(m);
2063 		}
2064 	}
2065 
2066 	PA_UNLOCK_COND(pa);
2067 	PMAP_UNLOCK(pmap);
2068 	return (m);
2069 }
2070 
2071 /*
2072  * Initialize a vm_page's machine-dependent fields.
2073  */
2074 static void
2075 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2076 {
2077 
2078 	TAILQ_INIT(&m->md.pv_list);
2079 }
2080 
2081 /*
2082  * mmu_booke_zero_page_area zeros the specified hardware page by
2083  * mapping it into virtual memory and using bzero to clear
2084  * its contents.
2085  *
2086  * off and size must reside within a single page.
2087  */
2088 static void
2089 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2090 {
2091 	vm_offset_t va;
2092 
2093 	/* XXX KASSERT off and size are within a single page? */
2094 
2095 	mtx_lock(&zero_page_mutex);
2096 	va = zero_page_va;
2097 
2098 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2099 	bzero((caddr_t)va + off, size);
2100 	mmu_booke_kremove(mmu, va);
2101 
2102 	mtx_unlock(&zero_page_mutex);
2103 }
2104 
2105 /*
2106  * mmu_booke_zero_page zeros the specified hardware page.
2107  */
2108 static void
2109 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2110 {
2111 
2112 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2113 }
2114 
2115 /*
2116  * mmu_booke_copy_page copies the specified (machine independent) page by
2117  * mapping the page into virtual memory and using memcopy to copy the page,
2118  * one machine dependent page at a time.
2119  */
2120 static void
2121 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2122 {
2123 	vm_offset_t sva, dva;
2124 
2125 	sva = copy_page_src_va;
2126 	dva = copy_page_dst_va;
2127 
2128 	mtx_lock(&copy_page_mutex);
2129 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2130 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2131 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2132 	mmu_booke_kremove(mmu, dva);
2133 	mmu_booke_kremove(mmu, sva);
2134 	mtx_unlock(&copy_page_mutex);
2135 }
2136 
2137 /*
2138  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2139  * into virtual memory and using bzero to clear its contents. This is intended
2140  * to be called from the vm_pagezero process only and outside of Giant. No
2141  * lock is required.
2142  */
2143 static void
2144 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2145 {
2146 	vm_offset_t va;
2147 
2148 	va = zero_page_idle_va;
2149 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2150 	bzero((caddr_t)va, PAGE_SIZE);
2151 	mmu_booke_kremove(mmu, va);
2152 }
2153 
2154 /*
2155  * Return whether or not the specified physical page was modified
2156  * in any of physical maps.
2157  */
2158 static boolean_t
2159 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2160 {
2161 	pte_t *pte;
2162 	pv_entry_t pv;
2163 	boolean_t rv;
2164 
2165 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2166 	    ("mmu_booke_is_modified: page %p is not managed", m));
2167 	rv = FALSE;
2168 
2169 	/*
2170 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be
2171 	 * concurrently set while the object is locked.  Thus, if PG_WRITEABLE
2172 	 * is clear, no PTEs can be modified.
2173 	 */
2174 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2175 	if ((m->oflags & VPO_BUSY) == 0 &&
2176 	    (m->flags & PG_WRITEABLE) == 0)
2177 		return (rv);
2178 	vm_page_lock_queues();
2179 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2180 		PMAP_LOCK(pv->pv_pmap);
2181 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2182 		    PTE_ISVALID(pte)) {
2183 			if (PTE_ISMODIFIED(pte))
2184 				rv = TRUE;
2185 		}
2186 		PMAP_UNLOCK(pv->pv_pmap);
2187 		if (rv)
2188 			break;
2189 	}
2190 	vm_page_unlock_queues();
2191 	return (rv);
2192 }
2193 
2194 /*
2195  * Return whether or not the specified virtual address is eligible
2196  * for prefault.
2197  */
2198 static boolean_t
2199 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2200 {
2201 
2202 	return (FALSE);
2203 }
2204 
2205 /*
2206  * Return whether or not the specified physical page was referenced
2207  * in any physical maps.
2208  */
2209 static boolean_t
2210 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2211 {
2212 	pte_t *pte;
2213 	pv_entry_t pv;
2214 	boolean_t rv;
2215 
2216 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2217 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2218 	rv = FALSE;
2219 	vm_page_lock_queues();
2220 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2221 		PMAP_LOCK(pv->pv_pmap);
2222 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2223 		    PTE_ISVALID(pte)) {
2224 			if (PTE_ISREFERENCED(pte))
2225 				rv = TRUE;
2226 		}
2227 		PMAP_UNLOCK(pv->pv_pmap);
2228 		if (rv)
2229 			break;
2230 	}
2231 	vm_page_unlock_queues();
2232 	return (rv);
2233 }
2234 
2235 /*
2236  * Clear the modify bits on the specified physical page.
2237  */
2238 static void
2239 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2240 {
2241 	pte_t *pte;
2242 	pv_entry_t pv;
2243 
2244 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2245 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2246 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2247 	KASSERT((m->oflags & VPO_BUSY) == 0,
2248 	    ("mmu_booke_clear_modify: page %p is busy", m));
2249 
2250 	/*
2251 	 * If the page is not PG_WRITEABLE, then no PTEs can be modified.
2252 	 * If the object containing the page is locked and the page is not
2253 	 * VPO_BUSY, then PG_WRITEABLE cannot be concurrently set.
2254 	 */
2255 	if ((m->flags & PG_WRITEABLE) == 0)
2256 		return;
2257 	vm_page_lock_queues();
2258 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2259 		PMAP_LOCK(pv->pv_pmap);
2260 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2261 		    PTE_ISVALID(pte)) {
2262 			mtx_lock_spin(&tlbivax_mutex);
2263 			tlb_miss_lock();
2264 
2265 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2266 				tlb0_flush_entry(pv->pv_va);
2267 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2268 				    PTE_REFERENCED);
2269 			}
2270 
2271 			tlb_miss_unlock();
2272 			mtx_unlock_spin(&tlbivax_mutex);
2273 		}
2274 		PMAP_UNLOCK(pv->pv_pmap);
2275 	}
2276 	vm_page_unlock_queues();
2277 }
2278 
2279 /*
2280  * Return a count of reference bits for a page, clearing those bits.
2281  * It is not necessary for every reference bit to be cleared, but it
2282  * is necessary that 0 only be returned when there are truly no
2283  * reference bits set.
2284  *
2285  * XXX: The exact number of bits to check and clear is a matter that
2286  * should be tested and standardized at some point in the future for
2287  * optimal aging of shared pages.
2288  */
2289 static int
2290 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2291 {
2292 	pte_t *pte;
2293 	pv_entry_t pv;
2294 	int count;
2295 
2296 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2297 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2298 		return (0);
2299 
2300 	count = 0;
2301 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2302 		PMAP_LOCK(pv->pv_pmap);
2303 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2304 			if (!PTE_ISVALID(pte))
2305 				goto make_sure_to_unlock;
2306 
2307 			if (PTE_ISREFERENCED(pte)) {
2308 				mtx_lock_spin(&tlbivax_mutex);
2309 				tlb_miss_lock();
2310 
2311 				tlb0_flush_entry(pv->pv_va);
2312 				pte->flags &= ~PTE_REFERENCED;
2313 
2314 				tlb_miss_unlock();
2315 				mtx_unlock_spin(&tlbivax_mutex);
2316 
2317 				if (++count > 4) {
2318 					PMAP_UNLOCK(pv->pv_pmap);
2319 					break;
2320 				}
2321 			}
2322 		}
2323 make_sure_to_unlock:
2324 		PMAP_UNLOCK(pv->pv_pmap);
2325 	}
2326 	return (count);
2327 }
2328 
2329 /*
2330  * Clear the reference bit on the specified physical page.
2331  */
2332 static void
2333 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2334 {
2335 	pte_t *pte;
2336 	pv_entry_t pv;
2337 
2338 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2339 	    ("mmu_booke_clear_reference: page %p is not managed", m));
2340 	vm_page_lock_queues();
2341 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2342 		PMAP_LOCK(pv->pv_pmap);
2343 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2344 		    PTE_ISVALID(pte)) {
2345 			if (PTE_ISREFERENCED(pte)) {
2346 				mtx_lock_spin(&tlbivax_mutex);
2347 				tlb_miss_lock();
2348 
2349 				tlb0_flush_entry(pv->pv_va);
2350 				pte->flags &= ~PTE_REFERENCED;
2351 
2352 				tlb_miss_unlock();
2353 				mtx_unlock_spin(&tlbivax_mutex);
2354 			}
2355 		}
2356 		PMAP_UNLOCK(pv->pv_pmap);
2357 	}
2358 	vm_page_unlock_queues();
2359 }
2360 
2361 /*
2362  * Change wiring attribute for a map/virtual-address pair.
2363  */
2364 static void
2365 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2366 {
2367 	pte_t *pte;
2368 
2369 	PMAP_LOCK(pmap);
2370 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2371 		if (wired) {
2372 			if (!PTE_ISWIRED(pte)) {
2373 				pte->flags |= PTE_WIRED;
2374 				pmap->pm_stats.wired_count++;
2375 			}
2376 		} else {
2377 			if (PTE_ISWIRED(pte)) {
2378 				pte->flags &= ~PTE_WIRED;
2379 				pmap->pm_stats.wired_count--;
2380 			}
2381 		}
2382 	}
2383 	PMAP_UNLOCK(pmap);
2384 }
2385 
2386 /*
2387  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2388  * page.  This count may be changed upwards or downwards in the future; it is
2389  * only necessary that true be returned for a small subset of pmaps for proper
2390  * page aging.
2391  */
2392 static boolean_t
2393 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2394 {
2395 	pv_entry_t pv;
2396 	int loops;
2397 
2398 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2399 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2400 		return (FALSE);
2401 
2402 	loops = 0;
2403 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2404 		if (pv->pv_pmap == pmap)
2405 			return (TRUE);
2406 
2407 		if (++loops >= 16)
2408 			break;
2409 	}
2410 	return (FALSE);
2411 }
2412 
2413 /*
2414  * Return the number of managed mappings to the given physical page that are
2415  * wired.
2416  */
2417 static int
2418 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2419 {
2420 	pv_entry_t pv;
2421 	pte_t *pte;
2422 	int count = 0;
2423 
2424 	if ((m->flags & PG_FICTITIOUS) != 0)
2425 		return (count);
2426 	vm_page_lock_queues();
2427 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2428 		PMAP_LOCK(pv->pv_pmap);
2429 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2430 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2431 				count++;
2432 		PMAP_UNLOCK(pv->pv_pmap);
2433 	}
2434 	vm_page_unlock_queues();
2435 	return (count);
2436 }
2437 
2438 static int
2439 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2440 {
2441 	int i;
2442 	vm_offset_t va;
2443 
2444 	/*
2445 	 * This currently does not work for entries that
2446 	 * overlap TLB1 entries.
2447 	 */
2448 	for (i = 0; i < tlb1_idx; i ++) {
2449 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2450 			return (0);
2451 	}
2452 
2453 	return (EFAULT);
2454 }
2455 
2456 vm_offset_t
2457 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2458     vm_size_t *sz)
2459 {
2460 	vm_paddr_t pa, ppa;
2461 	vm_offset_t va;
2462 	vm_size_t gran;
2463 
2464 	/* Raw physical memory dumps don't have a virtual address. */
2465 	if (md->md_vaddr == ~0UL) {
2466 		/* We always map a 256MB page at 256M. */
2467 		gran = 256 * 1024 * 1024;
2468 		pa = md->md_paddr + ofs;
2469 		ppa = pa & ~(gran - 1);
2470 		ofs = pa - ppa;
2471 		va = gran;
2472 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2473 		if (*sz > (gran - ofs))
2474 			*sz = gran - ofs;
2475 		return (va + ofs);
2476 	}
2477 
2478 	/* Minidumps are based on virtual memory addresses. */
2479 	va = md->md_vaddr + ofs;
2480 	if (va >= kernstart + kernsize) {
2481 		gran = PAGE_SIZE - (va & PAGE_MASK);
2482 		if (*sz > gran)
2483 			*sz = gran;
2484 	}
2485 	return (va);
2486 }
2487 
2488 void
2489 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2490     vm_offset_t va)
2491 {
2492 
2493 	/* Raw physical memory dumps don't have a virtual address. */
2494 	if (md->md_vaddr == ~0UL) {
2495 		tlb1_idx--;
2496 		tlb1[tlb1_idx].mas1 = 0;
2497 		tlb1[tlb1_idx].mas2 = 0;
2498 		tlb1[tlb1_idx].mas3 = 0;
2499 		tlb1_write_entry(tlb1_idx);
2500 		return;
2501 	}
2502 
2503 	/* Minidumps are based on virtual memory addresses. */
2504 	/* Nothing to do... */
2505 }
2506 
2507 struct pmap_md *
2508 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2509 {
2510 	static struct pmap_md md;
2511 	struct bi_mem_region *mr;
2512 	pte_t *pte;
2513 	vm_offset_t va;
2514 
2515 	if (dumpsys_minidump) {
2516 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2517 		if (prev == NULL) {
2518 			/* 1st: kernel .data and .bss. */
2519 			md.md_index = 1;
2520 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2521 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2522 			return (&md);
2523 		}
2524 		switch (prev->md_index) {
2525 		case 1:
2526 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2527 			md.md_index = 2;
2528 			md.md_vaddr = data_start;
2529 			md.md_size = data_end - data_start;
2530 			break;
2531 		case 2:
2532 			/* 3rd: kernel VM. */
2533 			va = prev->md_vaddr + prev->md_size;
2534 			/* Find start of next chunk (from va). */
2535 			while (va < virtual_end) {
2536 				/* Don't dump the buffer cache. */
2537 				if (va >= kmi.buffer_sva &&
2538 				    va < kmi.buffer_eva) {
2539 					va = kmi.buffer_eva;
2540 					continue;
2541 				}
2542 				pte = pte_find(mmu, kernel_pmap, va);
2543 				if (pte != NULL && PTE_ISVALID(pte))
2544 					break;
2545 				va += PAGE_SIZE;
2546 			}
2547 			if (va < virtual_end) {
2548 				md.md_vaddr = va;
2549 				va += PAGE_SIZE;
2550 				/* Find last page in chunk. */
2551 				while (va < virtual_end) {
2552 					/* Don't run into the buffer cache. */
2553 					if (va == kmi.buffer_sva)
2554 						break;
2555 					pte = pte_find(mmu, kernel_pmap, va);
2556 					if (pte == NULL || !PTE_ISVALID(pte))
2557 						break;
2558 					va += PAGE_SIZE;
2559 				}
2560 				md.md_size = va - md.md_vaddr;
2561 				break;
2562 			}
2563 			md.md_index = 3;
2564 			/* FALLTHROUGH */
2565 		default:
2566 			return (NULL);
2567 		}
2568 	} else { /* minidumps */
2569 		mr = bootinfo_mr();
2570 		if (prev == NULL) {
2571 			/* first physical chunk. */
2572 			md.md_paddr = mr->mem_base;
2573 			md.md_size = mr->mem_size;
2574 			md.md_vaddr = ~0UL;
2575 			md.md_index = 1;
2576 		} else if (md.md_index < bootinfo->bi_mem_reg_no) {
2577 			md.md_paddr = mr[md.md_index].mem_base;
2578 			md.md_size = mr[md.md_index].mem_size;
2579 			md.md_vaddr = ~0UL;
2580 			md.md_index++;
2581 		} else {
2582 			/* There's no next physical chunk. */
2583 			return (NULL);
2584 		}
2585 	}
2586 
2587 	return (&md);
2588 }
2589 
2590 /*
2591  * Map a set of physical memory pages into the kernel virtual address space.
2592  * Return a pointer to where it is mapped. This routine is intended to be used
2593  * for mapping device memory, NOT real memory.
2594  */
2595 static void *
2596 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2597 {
2598 	void *res;
2599 	uintptr_t va;
2600 	vm_size_t sz;
2601 
2602 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2603 	res = (void *)va;
2604 
2605 	do {
2606 		sz = 1 << (ilog2(size) & ~1);
2607 		if (bootverbose)
2608 			printf("Wiring VA=%x to PA=%x (size=%x), "
2609 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2610 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2611 		size -= sz;
2612 		pa += sz;
2613 		va += sz;
2614 	} while (size > 0);
2615 
2616 	return (res);
2617 }
2618 
2619 /*
2620  * 'Unmap' a range mapped by mmu_booke_mapdev().
2621  */
2622 static void
2623 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2624 {
2625 	vm_offset_t base, offset;
2626 
2627 	/*
2628 	 * Unmap only if this is inside kernel virtual space.
2629 	 */
2630 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2631 		base = trunc_page(va);
2632 		offset = va & PAGE_MASK;
2633 		size = roundup(offset + size, PAGE_SIZE);
2634 		kmem_free(kernel_map, base, size);
2635 	}
2636 }
2637 
2638 /*
2639  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2640  * specified pmap. This eliminates the blast of soft faults on process startup
2641  * and immediately after an mmap.
2642  */
2643 static void
2644 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2645     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2646 {
2647 
2648 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2649 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2650 	    ("mmu_booke_object_init_pt: non-device object"));
2651 }
2652 
2653 /*
2654  * Perform the pmap work for mincore.
2655  */
2656 static int
2657 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2658     vm_paddr_t *locked_pa)
2659 {
2660 
2661 	TODO;
2662 	return (0);
2663 }
2664 
2665 /**************************************************************************/
2666 /* TID handling */
2667 /**************************************************************************/
2668 
2669 /*
2670  * Allocate a TID. If necessary, steal one from someone else.
2671  * The new TID is flushed from the TLB before returning.
2672  */
2673 static tlbtid_t
2674 tid_alloc(pmap_t pmap)
2675 {
2676 	tlbtid_t tid;
2677 	int thiscpu;
2678 
2679 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2680 
2681 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2682 
2683 	thiscpu = PCPU_GET(cpuid);
2684 
2685 	tid = PCPU_GET(tid_next);
2686 	if (tid > TID_MAX)
2687 		tid = TID_MIN;
2688 	PCPU_SET(tid_next, tid + 1);
2689 
2690 	/* If we are stealing TID then clear the relevant pmap's field */
2691 	if (tidbusy[thiscpu][tid] != NULL) {
2692 
2693 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2694 
2695 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2696 
2697 		/* Flush all entries from TLB0 matching this TID. */
2698 		tid_flush(tid);
2699 	}
2700 
2701 	tidbusy[thiscpu][tid] = pmap;
2702 	pmap->pm_tid[thiscpu] = tid;
2703 	__asm __volatile("msync; isync");
2704 
2705 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2706 	    PCPU_GET(tid_next));
2707 
2708 	return (tid);
2709 }
2710 
2711 /**************************************************************************/
2712 /* TLB0 handling */
2713 /**************************************************************************/
2714 
2715 static void
2716 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2717     uint32_t mas7)
2718 {
2719 	int as;
2720 	char desc[3];
2721 	tlbtid_t tid;
2722 	vm_size_t size;
2723 	unsigned int tsize;
2724 
2725 	desc[2] = '\0';
2726 	if (mas1 & MAS1_VALID)
2727 		desc[0] = 'V';
2728 	else
2729 		desc[0] = ' ';
2730 
2731 	if (mas1 & MAS1_IPROT)
2732 		desc[1] = 'P';
2733 	else
2734 		desc[1] = ' ';
2735 
2736 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2737 	tid = MAS1_GETTID(mas1);
2738 
2739 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2740 	size = 0;
2741 	if (tsize)
2742 		size = tsize2size(tsize);
2743 
2744 	debugf("%3d: (%s) [AS=%d] "
2745 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2746 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2747 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2748 }
2749 
2750 /* Convert TLB0 va and way number to tlb0[] table index. */
2751 static inline unsigned int
2752 tlb0_tableidx(vm_offset_t va, unsigned int way)
2753 {
2754 	unsigned int idx;
2755 
2756 	idx = (way * TLB0_ENTRIES_PER_WAY);
2757 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2758 	return (idx);
2759 }
2760 
2761 /*
2762  * Invalidate TLB0 entry.
2763  */
2764 static inline void
2765 tlb0_flush_entry(vm_offset_t va)
2766 {
2767 
2768 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2769 
2770 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2771 
2772 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2773 	__asm __volatile("isync; msync");
2774 	__asm __volatile("tlbsync; msync");
2775 
2776 	CTR1(KTR_PMAP, "%s: e", __func__);
2777 }
2778 
2779 /* Print out contents of the MAS registers for each TLB0 entry */
2780 void
2781 tlb0_print_tlbentries(void)
2782 {
2783 	uint32_t mas0, mas1, mas2, mas3, mas7;
2784 	int entryidx, way, idx;
2785 
2786 	debugf("TLB0 entries:\n");
2787 	for (way = 0; way < TLB0_WAYS; way ++)
2788 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2789 
2790 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2791 			mtspr(SPR_MAS0, mas0);
2792 			__asm __volatile("isync");
2793 
2794 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2795 			mtspr(SPR_MAS2, mas2);
2796 
2797 			__asm __volatile("isync; tlbre");
2798 
2799 			mas1 = mfspr(SPR_MAS1);
2800 			mas2 = mfspr(SPR_MAS2);
2801 			mas3 = mfspr(SPR_MAS3);
2802 			mas7 = mfspr(SPR_MAS7);
2803 
2804 			idx = tlb0_tableidx(mas2, way);
2805 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2806 		}
2807 }
2808 
2809 /**************************************************************************/
2810 /* TLB1 handling */
2811 /**************************************************************************/
2812 
2813 /*
2814  * TLB1 mapping notes:
2815  *
2816  * TLB1[0]	CCSRBAR
2817  * TLB1[1]	Kernel text and data.
2818  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2819  *		windows, other devices mappings.
2820  */
2821 
2822 /*
2823  * Write given entry to TLB1 hardware.
2824  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2825  */
2826 static void
2827 tlb1_write_entry(unsigned int idx)
2828 {
2829 	uint32_t mas0, mas7;
2830 
2831 	//debugf("tlb1_write_entry: s\n");
2832 
2833 	/* Clear high order RPN bits */
2834 	mas7 = 0;
2835 
2836 	/* Select entry */
2837 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2838 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2839 
2840 	mtspr(SPR_MAS0, mas0);
2841 	__asm __volatile("isync");
2842 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2843 	__asm __volatile("isync");
2844 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2845 	__asm __volatile("isync");
2846 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2847 	__asm __volatile("isync");
2848 	mtspr(SPR_MAS7, mas7);
2849 	__asm __volatile("isync; tlbwe; isync; msync");
2850 
2851 	//debugf("tlb1_write_entry: e\n");
2852 }
2853 
2854 /*
2855  * Return the largest uint value log such that 2^log <= num.
2856  */
2857 static unsigned int
2858 ilog2(unsigned int num)
2859 {
2860 	int lz;
2861 
2862 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2863 	return (31 - lz);
2864 }
2865 
2866 /*
2867  * Convert TLB TSIZE value to mapped region size.
2868  */
2869 static vm_size_t
2870 tsize2size(unsigned int tsize)
2871 {
2872 
2873 	/*
2874 	 * size = 4^tsize KB
2875 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2876 	 */
2877 
2878 	return ((1 << (2 * tsize)) * 1024);
2879 }
2880 
2881 /*
2882  * Convert region size (must be power of 4) to TLB TSIZE value.
2883  */
2884 static unsigned int
2885 size2tsize(vm_size_t size)
2886 {
2887 
2888 	return (ilog2(size) / 2 - 5);
2889 }
2890 
2891 /*
2892  * Register permanent kernel mapping in TLB1.
2893  *
2894  * Entries are created starting from index 0 (current free entry is
2895  * kept in tlb1_idx) and are not supposed to be invalidated.
2896  */
2897 static int
2898 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2899     uint32_t flags)
2900 {
2901 	uint32_t ts, tid;
2902 	int tsize;
2903 
2904 	if (tlb1_idx >= TLB1_ENTRIES) {
2905 		printf("tlb1_set_entry: TLB1 full!\n");
2906 		return (-1);
2907 	}
2908 
2909 	/* Convert size to TSIZE */
2910 	tsize = size2tsize(size);
2911 
2912 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2913 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2914 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2915 
2916 	/* XXX LOCK tlb1[] */
2917 
2918 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2919 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2920 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2921 
2922 	/* Set supervisor RWX permission bits */
2923 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2924 
2925 	tlb1_write_entry(tlb1_idx++);
2926 
2927 	/* XXX UNLOCK tlb1[] */
2928 
2929 	/*
2930 	 * XXX in general TLB1 updates should be propagated between CPUs,
2931 	 * since current design assumes to have the same TLB1 set-up on all
2932 	 * cores.
2933 	 */
2934 	return (0);
2935 }
2936 
2937 static int
2938 tlb1_entry_size_cmp(const void *a, const void *b)
2939 {
2940 	const vm_size_t *sza;
2941 	const vm_size_t *szb;
2942 
2943 	sza = a;
2944 	szb = b;
2945 	if (*sza > *szb)
2946 		return (-1);
2947 	else if (*sza < *szb)
2948 		return (1);
2949 	else
2950 		return (0);
2951 }
2952 
2953 /*
2954  * Map in contiguous RAM region into the TLB1 using maximum of
2955  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2956  *
2957  * If necessary round up last entry size and return total size
2958  * used by all allocated entries.
2959  */
2960 vm_size_t
2961 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2962 {
2963 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2964 	vm_size_t mapped_size, sz, esz;
2965 	unsigned int log;
2966 	int i;
2967 
2968 	CTR4(KTR_PMAP, "%s: region size = 0x%08x va = 0x%08x pa = 0x%08x",
2969 	    __func__, size, va, pa);
2970 
2971 	mapped_size = 0;
2972 	sz = size;
2973 	memset(entry_size, 0, sizeof(entry_size));
2974 
2975 	/* Calculate entry sizes. */
2976 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2977 
2978 		/* Largest region that is power of 4 and fits within size */
2979 		log = ilog2(sz) / 2;
2980 		esz = 1 << (2 * log);
2981 
2982 		/* If this is last entry cover remaining size. */
2983 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2984 			while (esz < sz)
2985 				esz = esz << 2;
2986 		}
2987 
2988 		entry_size[i] = esz;
2989 		mapped_size += esz;
2990 		if (esz < sz)
2991 			sz -= esz;
2992 		else
2993 			sz = 0;
2994 	}
2995 
2996 	/* Sort entry sizes, required to get proper entry address alignment. */
2997 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2998 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2999 
3000 	/* Load TLB1 entries. */
3001 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
3002 		esz = entry_size[i];
3003 		if (!esz)
3004 			break;
3005 
3006 		CTR5(KTR_PMAP, "%s: entry %d: sz  = 0x%08x (va = 0x%08x "
3007 		    "pa = 0x%08x)", __func__, tlb1_idx, esz, va, pa);
3008 
3009 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
3010 
3011 		va += esz;
3012 		pa += esz;
3013 	}
3014 
3015 	CTR3(KTR_PMAP, "%s: mapped size 0x%08x (wasted space 0x%08x)",
3016 	    __func__, mapped_size, mapped_size - size);
3017 
3018 	return (mapped_size);
3019 }
3020 
3021 /*
3022  * TLB1 initialization routine, to be called after the very first
3023  * assembler level setup done in locore.S.
3024  */
3025 void
3026 tlb1_init(vm_offset_t ccsrbar)
3027 {
3028 	uint32_t mas0;
3029 
3030 	/* TLB1[1] is used to map the kernel. Save that entry. */
3031 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(1);
3032 	mtspr(SPR_MAS0, mas0);
3033 	__asm __volatile("isync; tlbre");
3034 
3035 	tlb1[1].mas1 = mfspr(SPR_MAS1);
3036 	tlb1[1].mas2 = mfspr(SPR_MAS2);
3037 	tlb1[1].mas3 = mfspr(SPR_MAS3);
3038 
3039 	/* Map in CCSRBAR in TLB1[0] */
3040 	tlb1_idx = 0;
3041 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
3042 	/*
3043 	 * Set the next available TLB1 entry index. Note TLB[1] is reserved
3044 	 * for initial mapping of kernel text+data, which was set early in
3045 	 * locore, we need to skip this [busy] entry.
3046 	 */
3047 	tlb1_idx = 2;
3048 
3049 	/* Setup TLB miss defaults */
3050 	set_mas4_defaults();
3051 }
3052 
3053 /*
3054  * Setup MAS4 defaults.
3055  * These values are loaded to MAS0-2 on a TLB miss.
3056  */
3057 static void
3058 set_mas4_defaults(void)
3059 {
3060 	uint32_t mas4;
3061 
3062 	/* Defaults: TLB0, PID0, TSIZED=4K */
3063 	mas4 = MAS4_TLBSELD0;
3064 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3065 #ifdef SMP
3066 	mas4 |= MAS4_MD;
3067 #endif
3068 	mtspr(SPR_MAS4, mas4);
3069 	__asm __volatile("isync");
3070 }
3071 
3072 /*
3073  * Print out contents of the MAS registers for each TLB1 entry
3074  */
3075 void
3076 tlb1_print_tlbentries(void)
3077 {
3078 	uint32_t mas0, mas1, mas2, mas3, mas7;
3079 	int i;
3080 
3081 	debugf("TLB1 entries:\n");
3082 	for (i = 0; i < TLB1_ENTRIES; i++) {
3083 
3084 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3085 		mtspr(SPR_MAS0, mas0);
3086 
3087 		__asm __volatile("isync; tlbre");
3088 
3089 		mas1 = mfspr(SPR_MAS1);
3090 		mas2 = mfspr(SPR_MAS2);
3091 		mas3 = mfspr(SPR_MAS3);
3092 		mas7 = mfspr(SPR_MAS7);
3093 
3094 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3095 	}
3096 }
3097 
3098 /*
3099  * Print out contents of the in-ram tlb1 table.
3100  */
3101 void
3102 tlb1_print_entries(void)
3103 {
3104 	int i;
3105 
3106 	debugf("tlb1[] table entries:\n");
3107 	for (i = 0; i < TLB1_ENTRIES; i++)
3108 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3109 }
3110 
3111 /*
3112  * Return 0 if the physical IO range is encompassed by one of the
3113  * the TLB1 entries, otherwise return related error code.
3114  */
3115 static int
3116 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3117 {
3118 	uint32_t prot;
3119 	vm_paddr_t pa_start;
3120 	vm_paddr_t pa_end;
3121 	unsigned int entry_tsize;
3122 	vm_size_t entry_size;
3123 
3124 	*va = (vm_offset_t)NULL;
3125 
3126 	/* Skip invalid entries */
3127 	if (!(tlb1[i].mas1 & MAS1_VALID))
3128 		return (EINVAL);
3129 
3130 	/*
3131 	 * The entry must be cache-inhibited, guarded, and r/w
3132 	 * so it can function as an i/o page
3133 	 */
3134 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3135 	if (prot != (MAS2_I | MAS2_G))
3136 		return (EPERM);
3137 
3138 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3139 	if (prot != (MAS3_SR | MAS3_SW))
3140 		return (EPERM);
3141 
3142 	/* The address should be within the entry range. */
3143 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3144 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3145 
3146 	entry_size = tsize2size(entry_tsize);
3147 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3148 	pa_end = pa_start + entry_size - 1;
3149 
3150 	if ((pa < pa_start) || ((pa + size) > pa_end))
3151 		return (ERANGE);
3152 
3153 	/* Return virtual address of this mapping. */
3154 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3155 	return (0);
3156 }
3157