xref: /freebsd/sys/powerpc/booke/pmap.c (revision ec65e4f8d0654361df5e97d4de3518edebf76b46)
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 "opt_kstack_pages.h"
55 
56 #include <sys/param.h>
57 #include <sys/conf.h>
58 #include <sys/malloc.h>
59 #include <sys/ktr.h>
60 #include <sys/proc.h>
61 #include <sys/user.h>
62 #include <sys/queue.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/kerneldump.h>
66 #include <sys/linker.h>
67 #include <sys/msgbuf.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/vmmeter.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_pager.h>
84 #include <vm/uma.h>
85 
86 #include <machine/cpu.h>
87 #include <machine/pcb.h>
88 #include <machine/platform.h>
89 
90 #include <machine/tlb.h>
91 #include <machine/spr.h>
92 #include <machine/md_var.h>
93 #include <machine/mmuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/pte.h>
96 
97 #include "mmu_if.h"
98 
99 #define	SPARSE_MAPDEV
100 #ifdef  DEBUG
101 #define debugf(fmt, args...) printf(fmt, ##args)
102 #else
103 #define debugf(fmt, args...)
104 #endif
105 
106 #define TODO			panic("%s: not implemented", __func__);
107 
108 extern unsigned char _etext[];
109 extern unsigned char _end[];
110 
111 extern uint32_t *bootinfo;
112 
113 vm_paddr_t kernload;
114 vm_offset_t kernstart;
115 vm_size_t kernsize;
116 
117 /* Message buffer and tables. */
118 static vm_offset_t data_start;
119 static vm_size_t data_end;
120 
121 /* Phys/avail memory regions. */
122 static struct mem_region *availmem_regions;
123 static int availmem_regions_sz;
124 static struct mem_region *physmem_regions;
125 static int physmem_regions_sz;
126 
127 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
128 static vm_offset_t zero_page_va;
129 static struct mtx zero_page_mutex;
130 
131 static struct mtx tlbivax_mutex;
132 
133 /*
134  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
135  * by idle thred only, no lock required.
136  */
137 static vm_offset_t zero_page_idle_va;
138 
139 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
140 static vm_offset_t copy_page_src_va;
141 static vm_offset_t copy_page_dst_va;
142 static struct mtx copy_page_mutex;
143 
144 /**************************************************************************/
145 /* PMAP */
146 /**************************************************************************/
147 
148 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
149     vm_prot_t, u_int flags, int8_t psind);
150 
151 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
152 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
153 
154 /*
155  * If user pmap is processed with mmu_booke_remove and the resident count
156  * drops to 0, there are no more pages to remove, so we need not continue.
157  */
158 #define PMAP_REMOVE_DONE(pmap) \
159 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
160 
161 extern int elf32_nxstack;
162 
163 /**************************************************************************/
164 /* TLB and TID handling */
165 /**************************************************************************/
166 
167 /* Translation ID busy table */
168 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
169 
170 /*
171  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
172  * core revisions and should be read from h/w registers during early config.
173  */
174 uint32_t tlb0_entries;
175 uint32_t tlb0_ways;
176 uint32_t tlb0_entries_per_way;
177 uint32_t tlb1_entries;
178 
179 #define TLB0_ENTRIES		(tlb0_entries)
180 #define TLB0_WAYS		(tlb0_ways)
181 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
182 
183 #define TLB1_ENTRIES (tlb1_entries)
184 #define TLB1_MAXENTRIES	64
185 
186 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
187 
188 static tlbtid_t tid_alloc(struct pmap *);
189 static void tid_flush(tlbtid_t tid);
190 
191 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
192 
193 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
194 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
195 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
196 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
197 
198 static vm_size_t tsize2size(unsigned int);
199 static unsigned int size2tsize(vm_size_t);
200 static unsigned int ilog2(unsigned int);
201 
202 static void set_mas4_defaults(void);
203 
204 static inline void tlb0_flush_entry(vm_offset_t);
205 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
206 
207 /**************************************************************************/
208 /* Page table management */
209 /**************************************************************************/
210 
211 static struct rwlock_padalign pvh_global_lock;
212 
213 /* Data for the pv entry allocation mechanism */
214 static uma_zone_t pvzone;
215 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
216 
217 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
218 
219 #ifndef PMAP_SHPGPERPROC
220 #define PMAP_SHPGPERPROC	200
221 #endif
222 
223 static void ptbl_init(void);
224 static struct ptbl_buf *ptbl_buf_alloc(void);
225 static void ptbl_buf_free(struct ptbl_buf *);
226 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
227 
228 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
229 static void ptbl_free(mmu_t, pmap_t, unsigned int);
230 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
231 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
232 
233 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
234 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
235 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
236 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
237 static void kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr,
238 			     vm_offset_t pdir);
239 
240 static pv_entry_t pv_alloc(void);
241 static void pv_free(pv_entry_t);
242 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
243 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
244 
245 static void booke_pmap_init_qpages(void);
246 
247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
248 #define PTBL_BUFS		(128 * 16)
249 
250 struct ptbl_buf {
251 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
252 	vm_offset_t kva;		/* va of mapping */
253 };
254 
255 /* ptbl free list and a lock used for access synchronization. */
256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
257 static struct mtx ptbl_buf_freelist_lock;
258 
259 /* Base address of kva space allocated fot ptbl bufs. */
260 static vm_offset_t ptbl_buf_pool_vabase;
261 
262 /* Pointer to ptbl_buf structures. */
263 static struct ptbl_buf *ptbl_bufs;
264 
265 #ifdef SMP
266 extern tlb_entry_t __boot_tlb1[];
267 void pmap_bootstrap_ap(volatile uint32_t *);
268 #endif
269 
270 /*
271  * Kernel MMU interface
272  */
273 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
274 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
275     vm_size_t, vm_offset_t);
276 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
277 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
278     vm_offset_t, vm_page_t *, vm_offset_t, int);
279 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
280     vm_prot_t, u_int flags, int8_t psind);
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 int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
293 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_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_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
313 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
314 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
315 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
316 static void		mmu_booke_activate(mmu_t, struct thread *);
317 static void		mmu_booke_deactivate(mmu_t, struct thread *);
318 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
319 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
320 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
321 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
322 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
323 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
324 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
325 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
326 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
327 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
328     vm_size_t);
329 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
330     void **);
331 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
332     void *);
333 static void		mmu_booke_scan_init(mmu_t);
334 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
335 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
336 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
337     vm_size_t sz, vm_memattr_t mode);
338 
339 static mmu_method_t mmu_booke_methods[] = {
340 	/* pmap dispatcher interface */
341 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
342 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
343 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
344 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
345 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
346 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
347 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
348 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
349 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
350 	MMUMETHOD(mmu_init,		mmu_booke_init),
351 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
352 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
353 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
354 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
355 	MMUMETHOD(mmu_map,		mmu_booke_map),
356 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
357 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
358 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
359 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
360 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
361 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
362 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
363 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
364 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
365 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
366 	MMUMETHOD(mmu_release,		mmu_booke_release),
367 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
368 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
369 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
370 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
371 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
372 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
373 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
374 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
375 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
376 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
377 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
378 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
379 
380 	/* Internal interfaces */
381 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
382 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
383 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
384 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
385 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
386 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
387 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
388 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
389 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
390 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
391 
392 	/* dumpsys() support */
393 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
394 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
395 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
396 
397 	{ 0, 0 }
398 };
399 
400 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
401 
402 static __inline uint32_t
403 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
404 {
405 	uint32_t attrib;
406 	int i;
407 
408 	if (ma != VM_MEMATTR_DEFAULT) {
409 		switch (ma) {
410 		case VM_MEMATTR_UNCACHEABLE:
411 			return (MAS2_I | MAS2_G);
412 		case VM_MEMATTR_WRITE_COMBINING:
413 		case VM_MEMATTR_WRITE_BACK:
414 		case VM_MEMATTR_PREFETCHABLE:
415 			return (MAS2_I);
416 		case VM_MEMATTR_WRITE_THROUGH:
417 			return (MAS2_W | MAS2_M);
418 		case VM_MEMATTR_CACHEABLE:
419 			return (MAS2_M);
420 		}
421 	}
422 
423 	/*
424 	 * Assume the page is cache inhibited and access is guarded unless
425 	 * it's in our available memory array.
426 	 */
427 	attrib = _TLB_ENTRY_IO;
428 	for (i = 0; i < physmem_regions_sz; i++) {
429 		if ((pa >= physmem_regions[i].mr_start) &&
430 		    (pa < (physmem_regions[i].mr_start +
431 		     physmem_regions[i].mr_size))) {
432 			attrib = _TLB_ENTRY_MEM;
433 			break;
434 		}
435 	}
436 
437 	return (attrib);
438 }
439 
440 static inline void
441 tlb_miss_lock(void)
442 {
443 #ifdef SMP
444 	struct pcpu *pc;
445 
446 	if (!smp_started)
447 		return;
448 
449 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
450 		if (pc != pcpup) {
451 
452 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
453 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
454 
455 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
456 			    ("tlb_miss_lock: tried to lock self"));
457 
458 			tlb_lock(pc->pc_booke_tlb_lock);
459 
460 			CTR1(KTR_PMAP, "%s: locked", __func__);
461 		}
462 	}
463 #endif
464 }
465 
466 static inline void
467 tlb_miss_unlock(void)
468 {
469 #ifdef SMP
470 	struct pcpu *pc;
471 
472 	if (!smp_started)
473 		return;
474 
475 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
476 		if (pc != pcpup) {
477 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
478 			    __func__, pc->pc_cpuid);
479 
480 			tlb_unlock(pc->pc_booke_tlb_lock);
481 
482 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
483 		}
484 	}
485 #endif
486 }
487 
488 /* Return number of entries in TLB0. */
489 static __inline void
490 tlb0_get_tlbconf(void)
491 {
492 	uint32_t tlb0_cfg;
493 
494 	tlb0_cfg = mfspr(SPR_TLB0CFG);
495 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
496 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
497 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
498 }
499 
500 /* Return number of entries in TLB1. */
501 static __inline void
502 tlb1_get_tlbconf(void)
503 {
504 	uint32_t tlb1_cfg;
505 
506 	tlb1_cfg = mfspr(SPR_TLB1CFG);
507 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
508 }
509 
510 /**************************************************************************/
511 /* Page table related */
512 /**************************************************************************/
513 
514 /* Initialize pool of kva ptbl buffers. */
515 static void
516 ptbl_init(void)
517 {
518 	int i;
519 
520 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
521 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
522 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
523 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
524 
525 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
526 	TAILQ_INIT(&ptbl_buf_freelist);
527 
528 	for (i = 0; i < PTBL_BUFS; i++) {
529 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
530 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
531 	}
532 }
533 
534 /* Get a ptbl_buf from the freelist. */
535 static struct ptbl_buf *
536 ptbl_buf_alloc(void)
537 {
538 	struct ptbl_buf *buf;
539 
540 	mtx_lock(&ptbl_buf_freelist_lock);
541 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
542 	if (buf != NULL)
543 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
544 	mtx_unlock(&ptbl_buf_freelist_lock);
545 
546 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
547 
548 	return (buf);
549 }
550 
551 /* Return ptbl buff to free pool. */
552 static void
553 ptbl_buf_free(struct ptbl_buf *buf)
554 {
555 
556 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
557 
558 	mtx_lock(&ptbl_buf_freelist_lock);
559 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
560 	mtx_unlock(&ptbl_buf_freelist_lock);
561 }
562 
563 /*
564  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
565  */
566 static void
567 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
568 {
569 	struct ptbl_buf *pbuf;
570 
571 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
572 
573 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
574 
575 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
576 		if (pbuf->kva == (vm_offset_t)ptbl) {
577 			/* Remove from pmap ptbl buf list. */
578 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
579 
580 			/* Free corresponding ptbl buf. */
581 			ptbl_buf_free(pbuf);
582 			break;
583 		}
584 }
585 
586 /* Allocate page table. */
587 static pte_t *
588 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
589 {
590 	vm_page_t mtbl[PTBL_PAGES];
591 	vm_page_t m;
592 	struct ptbl_buf *pbuf;
593 	unsigned int pidx;
594 	pte_t *ptbl;
595 	int i, j;
596 
597 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
598 	    (pmap == kernel_pmap), pdir_idx);
599 
600 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
601 	    ("ptbl_alloc: invalid pdir_idx"));
602 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
603 	    ("pte_alloc: valid ptbl entry exists!"));
604 
605 	pbuf = ptbl_buf_alloc();
606 	if (pbuf == NULL)
607 		panic("pte_alloc: couldn't alloc kernel virtual memory");
608 
609 	ptbl = (pte_t *)pbuf->kva;
610 
611 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
612 
613 	/* Allocate ptbl pages, this will sleep! */
614 	for (i = 0; i < PTBL_PAGES; i++) {
615 		pidx = (PTBL_PAGES * pdir_idx) + i;
616 		while ((m = vm_page_alloc(NULL, pidx,
617 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
618 			PMAP_UNLOCK(pmap);
619 			rw_wunlock(&pvh_global_lock);
620 			if (nosleep) {
621 				ptbl_free_pmap_ptbl(pmap, ptbl);
622 				for (j = 0; j < i; j++)
623 					vm_page_free(mtbl[j]);
624 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
625 				return (NULL);
626 			}
627 			VM_WAIT;
628 			rw_wlock(&pvh_global_lock);
629 			PMAP_LOCK(pmap);
630 		}
631 		mtbl[i] = m;
632 	}
633 
634 	/* Map allocated pages into kernel_pmap. */
635 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
636 
637 	/* Zero whole ptbl. */
638 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
639 
640 	/* Add pbuf to the pmap ptbl bufs list. */
641 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
642 
643 	return (ptbl);
644 }
645 
646 /* Free ptbl pages and invalidate pdir entry. */
647 static void
648 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
649 {
650 	pte_t *ptbl;
651 	vm_paddr_t pa;
652 	vm_offset_t va;
653 	vm_page_t m;
654 	int i;
655 
656 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
657 	    (pmap == kernel_pmap), pdir_idx);
658 
659 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
660 	    ("ptbl_free: invalid pdir_idx"));
661 
662 	ptbl = pmap->pm_pdir[pdir_idx];
663 
664 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
665 
666 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
667 
668 	/*
669 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
670 	 * don't attempt to look up the page tables we are releasing.
671 	 */
672 	mtx_lock_spin(&tlbivax_mutex);
673 	tlb_miss_lock();
674 
675 	pmap->pm_pdir[pdir_idx] = NULL;
676 
677 	tlb_miss_unlock();
678 	mtx_unlock_spin(&tlbivax_mutex);
679 
680 	for (i = 0; i < PTBL_PAGES; i++) {
681 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
682 		pa = pte_vatopa(mmu, kernel_pmap, va);
683 		m = PHYS_TO_VM_PAGE(pa);
684 		vm_page_free_zero(m);
685 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
686 		mmu_booke_kremove(mmu, va);
687 	}
688 
689 	ptbl_free_pmap_ptbl(pmap, ptbl);
690 }
691 
692 /*
693  * Decrement ptbl pages hold count and attempt to free ptbl pages.
694  * Called when removing pte entry from ptbl.
695  *
696  * Return 1 if ptbl pages were freed.
697  */
698 static int
699 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
700 {
701 	pte_t *ptbl;
702 	vm_paddr_t pa;
703 	vm_page_t m;
704 	int i;
705 
706 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
707 	    (pmap == kernel_pmap), pdir_idx);
708 
709 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
710 	    ("ptbl_unhold: invalid pdir_idx"));
711 	KASSERT((pmap != kernel_pmap),
712 	    ("ptbl_unhold: unholding kernel ptbl!"));
713 
714 	ptbl = pmap->pm_pdir[pdir_idx];
715 
716 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
717 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
718 	    ("ptbl_unhold: non kva ptbl"));
719 
720 	/* decrement hold count */
721 	for (i = 0; i < PTBL_PAGES; i++) {
722 		pa = pte_vatopa(mmu, kernel_pmap,
723 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
724 		m = PHYS_TO_VM_PAGE(pa);
725 		m->wire_count--;
726 	}
727 
728 	/*
729 	 * Free ptbl pages if there are no pte etries in this ptbl.
730 	 * wire_count has the same value for all ptbl pages, so check the last
731 	 * page.
732 	 */
733 	if (m->wire_count == 0) {
734 		ptbl_free(mmu, pmap, pdir_idx);
735 
736 		//debugf("ptbl_unhold: e (freed ptbl)\n");
737 		return (1);
738 	}
739 
740 	return (0);
741 }
742 
743 /*
744  * Increment hold count for ptbl pages. This routine is used when a new pte
745  * entry is being inserted into the ptbl.
746  */
747 static void
748 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
749 {
750 	vm_paddr_t pa;
751 	pte_t *ptbl;
752 	vm_page_t m;
753 	int i;
754 
755 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
756 	    pdir_idx);
757 
758 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
759 	    ("ptbl_hold: invalid pdir_idx"));
760 	KASSERT((pmap != kernel_pmap),
761 	    ("ptbl_hold: holding kernel ptbl!"));
762 
763 	ptbl = pmap->pm_pdir[pdir_idx];
764 
765 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
766 
767 	for (i = 0; i < PTBL_PAGES; i++) {
768 		pa = pte_vatopa(mmu, kernel_pmap,
769 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
770 		m = PHYS_TO_VM_PAGE(pa);
771 		m->wire_count++;
772 	}
773 }
774 
775 /* Allocate pv_entry structure. */
776 pv_entry_t
777 pv_alloc(void)
778 {
779 	pv_entry_t pv;
780 
781 	pv_entry_count++;
782 	if (pv_entry_count > pv_entry_high_water)
783 		pagedaemon_wakeup();
784 	pv = uma_zalloc(pvzone, M_NOWAIT);
785 
786 	return (pv);
787 }
788 
789 /* Free pv_entry structure. */
790 static __inline void
791 pv_free(pv_entry_t pve)
792 {
793 
794 	pv_entry_count--;
795 	uma_zfree(pvzone, pve);
796 }
797 
798 
799 /* Allocate and initialize pv_entry structure. */
800 static void
801 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
802 {
803 	pv_entry_t pve;
804 
805 	//int su = (pmap == kernel_pmap);
806 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
807 	//	(u_int32_t)pmap, va, (u_int32_t)m);
808 
809 	pve = pv_alloc();
810 	if (pve == NULL)
811 		panic("pv_insert: no pv entries!");
812 
813 	pve->pv_pmap = pmap;
814 	pve->pv_va = va;
815 
816 	/* add to pv_list */
817 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
818 	rw_assert(&pvh_global_lock, RA_WLOCKED);
819 
820 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
821 
822 	//debugf("pv_insert: e\n");
823 }
824 
825 /* Destroy pv entry. */
826 static void
827 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
828 {
829 	pv_entry_t pve;
830 
831 	//int su = (pmap == kernel_pmap);
832 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
833 
834 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
835 	rw_assert(&pvh_global_lock, RA_WLOCKED);
836 
837 	/* find pv entry */
838 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
839 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
840 			/* remove from pv_list */
841 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
842 			if (TAILQ_EMPTY(&m->md.pv_list))
843 				vm_page_aflag_clear(m, PGA_WRITEABLE);
844 
845 			/* free pv entry struct */
846 			pv_free(pve);
847 			break;
848 		}
849 	}
850 
851 	//debugf("pv_remove: e\n");
852 }
853 
854 /*
855  * Clean pte entry, try to free page table page if requested.
856  *
857  * Return 1 if ptbl pages were freed, otherwise return 0.
858  */
859 static int
860 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
861 {
862 	unsigned int pdir_idx = PDIR_IDX(va);
863 	unsigned int ptbl_idx = PTBL_IDX(va);
864 	vm_page_t m;
865 	pte_t *ptbl;
866 	pte_t *pte;
867 
868 	//int su = (pmap == kernel_pmap);
869 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
870 	//		su, (u_int32_t)pmap, va, flags);
871 
872 	ptbl = pmap->pm_pdir[pdir_idx];
873 	KASSERT(ptbl, ("pte_remove: null ptbl"));
874 
875 	pte = &ptbl[ptbl_idx];
876 
877 	if (pte == NULL || !PTE_ISVALID(pte))
878 		return (0);
879 
880 	if (PTE_ISWIRED(pte))
881 		pmap->pm_stats.wired_count--;
882 
883 	/* Handle managed entry. */
884 	if (PTE_ISMANAGED(pte)) {
885 		/* Get vm_page_t for mapped pte. */
886 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
887 
888 		if (PTE_ISMODIFIED(pte))
889 			vm_page_dirty(m);
890 
891 		if (PTE_ISREFERENCED(pte))
892 			vm_page_aflag_set(m, PGA_REFERENCED);
893 
894 		pv_remove(pmap, va, m);
895 	}
896 
897 	mtx_lock_spin(&tlbivax_mutex);
898 	tlb_miss_lock();
899 
900 	tlb0_flush_entry(va);
901 	*pte = 0;
902 
903 	tlb_miss_unlock();
904 	mtx_unlock_spin(&tlbivax_mutex);
905 
906 	pmap->pm_stats.resident_count--;
907 
908 	if (flags & PTBL_UNHOLD) {
909 		//debugf("pte_remove: e (unhold)\n");
910 		return (ptbl_unhold(mmu, pmap, pdir_idx));
911 	}
912 
913 	//debugf("pte_remove: e\n");
914 	return (0);
915 }
916 
917 /*
918  * Insert PTE for a given page and virtual address.
919  */
920 static int
921 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
922     boolean_t nosleep)
923 {
924 	unsigned int pdir_idx = PDIR_IDX(va);
925 	unsigned int ptbl_idx = PTBL_IDX(va);
926 	pte_t *ptbl, *pte;
927 
928 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
929 	    pmap == kernel_pmap, pmap, va);
930 
931 	/* Get the page table pointer. */
932 	ptbl = pmap->pm_pdir[pdir_idx];
933 
934 	if (ptbl == NULL) {
935 		/* Allocate page table pages. */
936 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
937 		if (ptbl == NULL) {
938 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
939 			return (ENOMEM);
940 		}
941 	} else {
942 		/*
943 		 * Check if there is valid mapping for requested
944 		 * va, if there is, remove it.
945 		 */
946 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
947 		if (PTE_ISVALID(pte)) {
948 			pte_remove(mmu, pmap, va, PTBL_HOLD);
949 		} else {
950 			/*
951 			 * pte is not used, increment hold count
952 			 * for ptbl pages.
953 			 */
954 			if (pmap != kernel_pmap)
955 				ptbl_hold(mmu, pmap, pdir_idx);
956 		}
957 	}
958 
959 	/*
960 	 * Insert pv_entry into pv_list for mapped page if part of managed
961 	 * memory.
962 	 */
963 	if ((m->oflags & VPO_UNMANAGED) == 0) {
964 		flags |= PTE_MANAGED;
965 
966 		/* Create and insert pv entry. */
967 		pv_insert(pmap, va, m);
968 	}
969 
970 	pmap->pm_stats.resident_count++;
971 
972 	mtx_lock_spin(&tlbivax_mutex);
973 	tlb_miss_lock();
974 
975 	tlb0_flush_entry(va);
976 	if (pmap->pm_pdir[pdir_idx] == NULL) {
977 		/*
978 		 * If we just allocated a new page table, hook it in
979 		 * the pdir.
980 		 */
981 		pmap->pm_pdir[pdir_idx] = ptbl;
982 	}
983 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
984 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
985 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
986 
987 	tlb_miss_unlock();
988 	mtx_unlock_spin(&tlbivax_mutex);
989 	return (0);
990 }
991 
992 /* Return the pa for the given pmap/va. */
993 static vm_paddr_t
994 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
995 {
996 	vm_paddr_t pa = 0;
997 	pte_t *pte;
998 
999 	pte = pte_find(mmu, pmap, va);
1000 	if ((pte != NULL) && PTE_ISVALID(pte))
1001 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1002 	return (pa);
1003 }
1004 
1005 /* Get a pointer to a PTE in a page table. */
1006 static pte_t *
1007 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1008 {
1009 	unsigned int pdir_idx = PDIR_IDX(va);
1010 	unsigned int ptbl_idx = PTBL_IDX(va);
1011 
1012 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1013 
1014 	if (pmap->pm_pdir[pdir_idx])
1015 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1016 
1017 	return (NULL);
1018 }
1019 
1020 /* Set up kernel page tables. */
1021 static void
1022 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1023 {
1024 	int		i;
1025 	vm_offset_t	va;
1026 	pte_t		*pte;
1027 
1028 	/* Initialize kernel pdir */
1029 	for (i = 0; i < kernel_ptbls; i++)
1030 		kernel_pmap->pm_pdir[kptbl_min + i] =
1031 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1032 
1033 	/*
1034 	 * Fill in PTEs covering kernel code and data. They are not required
1035 	 * for address translation, as this area is covered by static TLB1
1036 	 * entries, but for pte_vatopa() to work correctly with kernel area
1037 	 * addresses.
1038 	 */
1039 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1040 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1041 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1042 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1043 		    PTE_VALID | PTE_PS_4KB;
1044 	}
1045 }
1046 
1047 /**************************************************************************/
1048 /* PMAP related */
1049 /**************************************************************************/
1050 
1051 /*
1052  * This is called during booke_init, before the system is really initialized.
1053  */
1054 static void
1055 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1056 {
1057 	vm_paddr_t phys_kernelend;
1058 	struct mem_region *mp, *mp1;
1059 	int cnt, i, j;
1060 	vm_paddr_t s, e, sz;
1061 	vm_paddr_t physsz, hwphyssz;
1062 	u_int phys_avail_count;
1063 	vm_size_t kstack0_sz;
1064 	vm_offset_t kernel_pdir, kstack0;
1065 	vm_paddr_t kstack0_phys;
1066 	void *dpcpu;
1067 
1068 	debugf("mmu_booke_bootstrap: entered\n");
1069 
1070 	/* Set interesting system properties */
1071 	hw_direct_map = 0;
1072 	elf32_nxstack = 1;
1073 
1074 	/* Initialize invalidation mutex */
1075 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1076 
1077 	/* Read TLB0 size and associativity. */
1078 	tlb0_get_tlbconf();
1079 
1080 	/*
1081 	 * Align kernel start and end address (kernel image).
1082 	 * Note that kernel end does not necessarily relate to kernsize.
1083 	 * kernsize is the size of the kernel that is actually mapped.
1084 	 */
1085 	kernstart = trunc_page(start);
1086 	data_start = round_page(kernelend);
1087 	data_end = data_start;
1088 
1089 	/*
1090 	 * Addresses of preloaded modules (like file systems) use
1091 	 * physical addresses. Make sure we relocate those into
1092 	 * virtual addresses.
1093 	 */
1094 	preload_addr_relocate = kernstart - kernload;
1095 
1096 	/* Allocate the dynamic per-cpu area. */
1097 	dpcpu = (void *)data_end;
1098 	data_end += DPCPU_SIZE;
1099 
1100 	/* Allocate space for the message buffer. */
1101 	msgbufp = (struct msgbuf *)data_end;
1102 	data_end += msgbufsize;
1103 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1104 	    data_end);
1105 
1106 	data_end = round_page(data_end);
1107 
1108 	/* Allocate space for ptbl_bufs. */
1109 	ptbl_bufs = (struct ptbl_buf *)data_end;
1110 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1111 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1112 	    data_end);
1113 
1114 	data_end = round_page(data_end);
1115 
1116 	/* Allocate PTE tables for kernel KVA. */
1117 	kernel_pdir = data_end;
1118 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1119 	    PDIR_SIZE);
1120 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1121 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1122 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1123 
1124 	debugf(" data_end: 0x%08x\n", data_end);
1125 	if (data_end - kernstart > kernsize) {
1126 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1127 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1128 	}
1129 	data_end = kernstart + kernsize;
1130 	debugf(" updated data_end: 0x%08x\n", data_end);
1131 
1132 	/*
1133 	 * Clear the structures - note we can only do it safely after the
1134 	 * possible additional TLB1 translations are in place (above) so that
1135 	 * all range up to the currently calculated 'data_end' is covered.
1136 	 */
1137 	dpcpu_init(dpcpu, 0);
1138 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1139 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1140 
1141 	/*******************************************************/
1142 	/* Set the start and end of kva. */
1143 	/*******************************************************/
1144 	virtual_avail = round_page(data_end);
1145 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1146 
1147 	/* Allocate KVA space for page zero/copy operations. */
1148 	zero_page_va = virtual_avail;
1149 	virtual_avail += PAGE_SIZE;
1150 	zero_page_idle_va = virtual_avail;
1151 	virtual_avail += PAGE_SIZE;
1152 	copy_page_src_va = virtual_avail;
1153 	virtual_avail += PAGE_SIZE;
1154 	copy_page_dst_va = virtual_avail;
1155 	virtual_avail += PAGE_SIZE;
1156 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1157 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1158 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1159 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1160 
1161 	/* Initialize page zero/copy mutexes. */
1162 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1163 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1164 
1165 	/* Allocate KVA space for ptbl bufs. */
1166 	ptbl_buf_pool_vabase = virtual_avail;
1167 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1168 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1169 	    ptbl_buf_pool_vabase, virtual_avail);
1170 
1171 	/* Calculate corresponding physical addresses for the kernel region. */
1172 	phys_kernelend = kernload + kernsize;
1173 	debugf("kernel image and allocated data:\n");
1174 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1175 	debugf(" kernstart   = 0x%08x\n", kernstart);
1176 	debugf(" kernsize    = 0x%08x\n", kernsize);
1177 
1178 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1179 		panic("mmu_booke_bootstrap: phys_avail too small");
1180 
1181 	/*
1182 	 * Remove kernel physical address range from avail regions list. Page
1183 	 * align all regions.  Non-page aligned memory isn't very interesting
1184 	 * to us.  Also, sort the entries for ascending addresses.
1185 	 */
1186 
1187 	/* Retrieve phys/avail mem regions */
1188 	mem_regions(&physmem_regions, &physmem_regions_sz,
1189 	    &availmem_regions, &availmem_regions_sz);
1190 	sz = 0;
1191 	cnt = availmem_regions_sz;
1192 	debugf("processing avail regions:\n");
1193 	for (mp = availmem_regions; mp->mr_size; mp++) {
1194 		s = mp->mr_start;
1195 		e = mp->mr_start + mp->mr_size;
1196 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1197 		/* Check whether this region holds all of the kernel. */
1198 		if (s < kernload && e > phys_kernelend) {
1199 			availmem_regions[cnt].mr_start = phys_kernelend;
1200 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1201 			e = kernload;
1202 		}
1203 		/* Look whether this regions starts within the kernel. */
1204 		if (s >= kernload && s < phys_kernelend) {
1205 			if (e <= phys_kernelend)
1206 				goto empty;
1207 			s = phys_kernelend;
1208 		}
1209 		/* Now look whether this region ends within the kernel. */
1210 		if (e > kernload && e <= phys_kernelend) {
1211 			if (s >= kernload)
1212 				goto empty;
1213 			e = kernload;
1214 		}
1215 		/* Now page align the start and size of the region. */
1216 		s = round_page(s);
1217 		e = trunc_page(e);
1218 		if (e < s)
1219 			e = s;
1220 		sz = e - s;
1221 		debugf("%09jx-%09jx = %jx\n",
1222 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1223 
1224 		/* Check whether some memory is left here. */
1225 		if (sz == 0) {
1226 		empty:
1227 			memmove(mp, mp + 1,
1228 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1229 			cnt--;
1230 			mp--;
1231 			continue;
1232 		}
1233 
1234 		/* Do an insertion sort. */
1235 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1236 			if (s < mp1->mr_start)
1237 				break;
1238 		if (mp1 < mp) {
1239 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1240 			mp1->mr_start = s;
1241 			mp1->mr_size = sz;
1242 		} else {
1243 			mp->mr_start = s;
1244 			mp->mr_size = sz;
1245 		}
1246 	}
1247 	availmem_regions_sz = cnt;
1248 
1249 	/*******************************************************/
1250 	/* Steal physical memory for kernel stack from the end */
1251 	/* of the first avail region                           */
1252 	/*******************************************************/
1253 	kstack0_sz = kstack_pages * PAGE_SIZE;
1254 	kstack0_phys = availmem_regions[0].mr_start +
1255 	    availmem_regions[0].mr_size;
1256 	kstack0_phys -= kstack0_sz;
1257 	availmem_regions[0].mr_size -= kstack0_sz;
1258 
1259 	/*******************************************************/
1260 	/* Fill in phys_avail table, based on availmem_regions */
1261 	/*******************************************************/
1262 	phys_avail_count = 0;
1263 	physsz = 0;
1264 	hwphyssz = 0;
1265 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1266 
1267 	debugf("fill in phys_avail:\n");
1268 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1269 
1270 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1271 		    (uintmax_t)availmem_regions[i].mr_start,
1272 		    (uintmax_t)availmem_regions[i].mr_start +
1273 		        availmem_regions[i].mr_size,
1274 		    (uintmax_t)availmem_regions[i].mr_size);
1275 
1276 		if (hwphyssz != 0 &&
1277 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1278 			debugf(" hw.physmem adjust\n");
1279 			if (physsz < hwphyssz) {
1280 				phys_avail[j] = availmem_regions[i].mr_start;
1281 				phys_avail[j + 1] =
1282 				    availmem_regions[i].mr_start +
1283 				    hwphyssz - physsz;
1284 				physsz = hwphyssz;
1285 				phys_avail_count++;
1286 			}
1287 			break;
1288 		}
1289 
1290 		phys_avail[j] = availmem_regions[i].mr_start;
1291 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1292 		    availmem_regions[i].mr_size;
1293 		phys_avail_count++;
1294 		physsz += availmem_regions[i].mr_size;
1295 	}
1296 	physmem = btoc(physsz);
1297 
1298 	/* Calculate the last available physical address. */
1299 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1300 		;
1301 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1302 
1303 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1304 	debugf("phys_avail_count = %d\n", phys_avail_count);
1305 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1306 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1307 
1308 	/*******************************************************/
1309 	/* Initialize (statically allocated) kernel pmap. */
1310 	/*******************************************************/
1311 	PMAP_LOCK_INIT(kernel_pmap);
1312 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1313 
1314 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1315 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1316 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1317 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1318 
1319 	kernel_pte_alloc(data_end, kernstart, kernel_pdir);
1320 	for (i = 0; i < MAXCPU; i++) {
1321 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1322 
1323 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1324 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1325 	}
1326 
1327 	/* Mark kernel_pmap active on all CPUs */
1328 	CPU_FILL(&kernel_pmap->pm_active);
1329 
1330  	/*
1331 	 * Initialize the global pv list lock.
1332 	 */
1333 	rw_init(&pvh_global_lock, "pmap pv global");
1334 
1335 	/*******************************************************/
1336 	/* Final setup */
1337 	/*******************************************************/
1338 
1339 	/* Enter kstack0 into kernel map, provide guard page */
1340 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1341 	thread0.td_kstack = kstack0;
1342 	thread0.td_kstack_pages = kstack_pages;
1343 
1344 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1345 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1346 	    kstack0_phys, kstack0_phys + kstack0_sz);
1347 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1348 
1349 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1350 	for (i = 0; i < kstack_pages; i++) {
1351 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1352 		kstack0 += PAGE_SIZE;
1353 		kstack0_phys += PAGE_SIZE;
1354 	}
1355 
1356 	pmap_bootstrapped = 1;
1357 
1358 	debugf("virtual_avail = %08x\n", virtual_avail);
1359 	debugf("virtual_end   = %08x\n", virtual_end);
1360 
1361 	debugf("mmu_booke_bootstrap: exit\n");
1362 }
1363 
1364 #ifdef SMP
1365  void
1366 tlb1_ap_prep(void)
1367 {
1368 	tlb_entry_t *e, tmp;
1369 	unsigned int i;
1370 
1371 	/* Prepare TLB1 image for AP processors */
1372 	e = __boot_tlb1;
1373 	for (i = 0; i < TLB1_ENTRIES; i++) {
1374 		tlb1_read_entry(&tmp, i);
1375 
1376 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1377 			memcpy(e++, &tmp, sizeof(tmp));
1378 	}
1379 }
1380 
1381 void
1382 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1383 {
1384 	int i;
1385 
1386 	/*
1387 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1388 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1389 	 * created by tlb1_ap_prep(), so use these values directly to
1390 	 * (re)program AP's TLB1 hardware.
1391 	 *
1392 	 * Start at index 1 because index 0 has the kernel map.
1393 	 */
1394 	for (i = 1; i < TLB1_ENTRIES; i++) {
1395 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1396 			tlb1_write_entry(&__boot_tlb1[i], i);
1397 	}
1398 
1399 	set_mas4_defaults();
1400 }
1401 #endif
1402 
1403 static void
1404 booke_pmap_init_qpages(void)
1405 {
1406 	struct pcpu *pc;
1407 	int i;
1408 
1409 	CPU_FOREACH(i) {
1410 		pc = pcpu_find(i);
1411 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1412 		if (pc->pc_qmap_addr == 0)
1413 			panic("pmap_init_qpages: unable to allocate KVA");
1414 	}
1415 }
1416 
1417 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1418 
1419 /*
1420  * Get the physical page address for the given pmap/virtual address.
1421  */
1422 static vm_paddr_t
1423 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1424 {
1425 	vm_paddr_t pa;
1426 
1427 	PMAP_LOCK(pmap);
1428 	pa = pte_vatopa(mmu, pmap, va);
1429 	PMAP_UNLOCK(pmap);
1430 
1431 	return (pa);
1432 }
1433 
1434 /*
1435  * Extract the physical page address associated with the given
1436  * kernel virtual address.
1437  */
1438 static vm_paddr_t
1439 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1440 {
1441 	tlb_entry_t e;
1442 	int i;
1443 
1444 	/* Check TLB1 mappings */
1445 	for (i = 0; i < TLB1_ENTRIES; i++) {
1446 		tlb1_read_entry(&e, i);
1447 		if (!(e.mas1 & MAS1_VALID))
1448 			continue;
1449 		if (va >= e.virt && va < e.virt + e.size)
1450 			return (e.phys + (va - e.virt));
1451 	}
1452 
1453 	return (pte_vatopa(mmu, kernel_pmap, va));
1454 }
1455 
1456 /*
1457  * Initialize the pmap module.
1458  * Called by vm_init, to initialize any structures that the pmap
1459  * system needs to map virtual memory.
1460  */
1461 static void
1462 mmu_booke_init(mmu_t mmu)
1463 {
1464 	int shpgperproc = PMAP_SHPGPERPROC;
1465 
1466 	/*
1467 	 * Initialize the address space (zone) for the pv entries.  Set a
1468 	 * high water mark so that the system can recover from excessive
1469 	 * numbers of pv entries.
1470 	 */
1471 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1472 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1473 
1474 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1475 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1476 
1477 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1478 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1479 
1480 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1481 
1482 	/* Pre-fill pvzone with initial number of pv entries. */
1483 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1484 
1485 	/* Initialize ptbl allocation. */
1486 	ptbl_init();
1487 }
1488 
1489 /*
1490  * Map a list of wired pages into kernel virtual address space.  This is
1491  * intended for temporary mappings which do not need page modification or
1492  * references recorded.  Existing mappings in the region are overwritten.
1493  */
1494 static void
1495 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1496 {
1497 	vm_offset_t va;
1498 
1499 	va = sva;
1500 	while (count-- > 0) {
1501 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1502 		va += PAGE_SIZE;
1503 		m++;
1504 	}
1505 }
1506 
1507 /*
1508  * Remove page mappings from kernel virtual address space.  Intended for
1509  * temporary mappings entered by mmu_booke_qenter.
1510  */
1511 static void
1512 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1513 {
1514 	vm_offset_t va;
1515 
1516 	va = sva;
1517 	while (count-- > 0) {
1518 		mmu_booke_kremove(mmu, va);
1519 		va += PAGE_SIZE;
1520 	}
1521 }
1522 
1523 /*
1524  * Map a wired page into kernel virtual address space.
1525  */
1526 static void
1527 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1528 {
1529 
1530 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1531 }
1532 
1533 static void
1534 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1535 {
1536 	uint32_t flags;
1537 	pte_t *pte;
1538 
1539 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1540 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1541 
1542 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1543 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1544 	flags |= PTE_PS_4KB;
1545 
1546 	pte = pte_find(mmu, kernel_pmap, va);
1547 
1548 	mtx_lock_spin(&tlbivax_mutex);
1549 	tlb_miss_lock();
1550 
1551 	if (PTE_ISVALID(pte)) {
1552 
1553 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1554 
1555 		/* Flush entry from TLB0 */
1556 		tlb0_flush_entry(va);
1557 	}
1558 
1559 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1560 
1561 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1562 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1563 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1564 
1565 	/* Flush the real memory from the instruction cache. */
1566 	if ((flags & (PTE_I | PTE_G)) == 0)
1567 		__syncicache((void *)va, PAGE_SIZE);
1568 
1569 	tlb_miss_unlock();
1570 	mtx_unlock_spin(&tlbivax_mutex);
1571 }
1572 
1573 /*
1574  * Remove a page from kernel page table.
1575  */
1576 static void
1577 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1578 {
1579 	pte_t *pte;
1580 
1581 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
1582 
1583 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1584 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1585 	    ("mmu_booke_kremove: invalid va"));
1586 
1587 	pte = pte_find(mmu, kernel_pmap, va);
1588 
1589 	if (!PTE_ISVALID(pte)) {
1590 
1591 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1592 
1593 		return;
1594 	}
1595 
1596 	mtx_lock_spin(&tlbivax_mutex);
1597 	tlb_miss_lock();
1598 
1599 	/* Invalidate entry in TLB0, update PTE. */
1600 	tlb0_flush_entry(va);
1601 	*pte = 0;
1602 
1603 	tlb_miss_unlock();
1604 	mtx_unlock_spin(&tlbivax_mutex);
1605 }
1606 
1607 /*
1608  * Initialize pmap associated with process 0.
1609  */
1610 static void
1611 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1612 {
1613 
1614 	PMAP_LOCK_INIT(pmap);
1615 	mmu_booke_pinit(mmu, pmap);
1616 	PCPU_SET(curpmap, pmap);
1617 }
1618 
1619 /*
1620  * Initialize a preallocated and zeroed pmap structure,
1621  * such as one in a vmspace structure.
1622  */
1623 static void
1624 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1625 {
1626 	int i;
1627 
1628 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1629 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1630 
1631 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1632 
1633 	for (i = 0; i < MAXCPU; i++)
1634 		pmap->pm_tid[i] = TID_NONE;
1635 	CPU_ZERO(&kernel_pmap->pm_active);
1636 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1637 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1638 	TAILQ_INIT(&pmap->pm_ptbl_list);
1639 }
1640 
1641 /*
1642  * Release any resources held by the given physical map.
1643  * Called when a pmap initialized by mmu_booke_pinit is being released.
1644  * Should only be called if the map contains no valid mappings.
1645  */
1646 static void
1647 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1648 {
1649 
1650 	KASSERT(pmap->pm_stats.resident_count == 0,
1651 	    ("pmap_release: pmap resident count %ld != 0",
1652 	    pmap->pm_stats.resident_count));
1653 }
1654 
1655 /*
1656  * Insert the given physical page at the specified virtual address in the
1657  * target physical map with the protection requested. If specified the page
1658  * will be wired down.
1659  */
1660 static int
1661 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1662     vm_prot_t prot, u_int flags, int8_t psind)
1663 {
1664 	int error;
1665 
1666 	rw_wlock(&pvh_global_lock);
1667 	PMAP_LOCK(pmap);
1668 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1669 	rw_wunlock(&pvh_global_lock);
1670 	PMAP_UNLOCK(pmap);
1671 	return (error);
1672 }
1673 
1674 static int
1675 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1676     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1677 {
1678 	pte_t *pte;
1679 	vm_paddr_t pa;
1680 	uint32_t flags;
1681 	int error, su, sync;
1682 
1683 	pa = VM_PAGE_TO_PHYS(m);
1684 	su = (pmap == kernel_pmap);
1685 	sync = 0;
1686 
1687 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1688 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1689 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1690 	//		(u_int32_t)m, va, pa, prot, flags);
1691 
1692 	if (su) {
1693 		KASSERT(((va >= virtual_avail) &&
1694 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1695 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1696 	} else {
1697 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1698 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1699 	}
1700 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1701 		VM_OBJECT_ASSERT_LOCKED(m->object);
1702 
1703 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1704 
1705 	/*
1706 	 * If there is an existing mapping, and the physical address has not
1707 	 * changed, must be protection or wiring change.
1708 	 */
1709 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1710 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1711 
1712 		/*
1713 		 * Before actually updating pte->flags we calculate and
1714 		 * prepare its new value in a helper var.
1715 		 */
1716 		flags = *pte;
1717 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1718 
1719 		/* Wiring change, just update stats. */
1720 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1721 			if (!PTE_ISWIRED(pte)) {
1722 				flags |= PTE_WIRED;
1723 				pmap->pm_stats.wired_count++;
1724 			}
1725 		} else {
1726 			if (PTE_ISWIRED(pte)) {
1727 				flags &= ~PTE_WIRED;
1728 				pmap->pm_stats.wired_count--;
1729 			}
1730 		}
1731 
1732 		if (prot & VM_PROT_WRITE) {
1733 			/* Add write permissions. */
1734 			flags |= PTE_SW;
1735 			if (!su)
1736 				flags |= PTE_UW;
1737 
1738 			if ((flags & PTE_MANAGED) != 0)
1739 				vm_page_aflag_set(m, PGA_WRITEABLE);
1740 		} else {
1741 			/* Handle modified pages, sense modify status. */
1742 
1743 			/*
1744 			 * The PTE_MODIFIED flag could be set by underlying
1745 			 * TLB misses since we last read it (above), possibly
1746 			 * other CPUs could update it so we check in the PTE
1747 			 * directly rather than rely on that saved local flags
1748 			 * copy.
1749 			 */
1750 			if (PTE_ISMODIFIED(pte))
1751 				vm_page_dirty(m);
1752 		}
1753 
1754 		if (prot & VM_PROT_EXECUTE) {
1755 			flags |= PTE_SX;
1756 			if (!su)
1757 				flags |= PTE_UX;
1758 
1759 			/*
1760 			 * Check existing flags for execute permissions: if we
1761 			 * are turning execute permissions on, icache should
1762 			 * be flushed.
1763 			 */
1764 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1765 				sync++;
1766 		}
1767 
1768 		flags &= ~PTE_REFERENCED;
1769 
1770 		/*
1771 		 * The new flags value is all calculated -- only now actually
1772 		 * update the PTE.
1773 		 */
1774 		mtx_lock_spin(&tlbivax_mutex);
1775 		tlb_miss_lock();
1776 
1777 		tlb0_flush_entry(va);
1778 		*pte &= ~PTE_FLAGS_MASK;
1779 		*pte |= flags;
1780 
1781 		tlb_miss_unlock();
1782 		mtx_unlock_spin(&tlbivax_mutex);
1783 
1784 	} else {
1785 		/*
1786 		 * If there is an existing mapping, but it's for a different
1787 		 * physical address, pte_enter() will delete the old mapping.
1788 		 */
1789 		//if ((pte != NULL) && PTE_ISVALID(pte))
1790 		//	debugf("mmu_booke_enter_locked: replace\n");
1791 		//else
1792 		//	debugf("mmu_booke_enter_locked: new\n");
1793 
1794 		/* Now set up the flags and install the new mapping. */
1795 		flags = (PTE_SR | PTE_VALID);
1796 		flags |= PTE_M;
1797 
1798 		if (!su)
1799 			flags |= PTE_UR;
1800 
1801 		if (prot & VM_PROT_WRITE) {
1802 			flags |= PTE_SW;
1803 			if (!su)
1804 				flags |= PTE_UW;
1805 
1806 			if ((m->oflags & VPO_UNMANAGED) == 0)
1807 				vm_page_aflag_set(m, PGA_WRITEABLE);
1808 		}
1809 
1810 		if (prot & VM_PROT_EXECUTE) {
1811 			flags |= PTE_SX;
1812 			if (!su)
1813 				flags |= PTE_UX;
1814 		}
1815 
1816 		/* If its wired update stats. */
1817 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1818 			flags |= PTE_WIRED;
1819 
1820 		error = pte_enter(mmu, pmap, m, va, flags,
1821 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1822 		if (error != 0)
1823 			return (KERN_RESOURCE_SHORTAGE);
1824 
1825 		if ((flags & PMAP_ENTER_WIRED) != 0)
1826 			pmap->pm_stats.wired_count++;
1827 
1828 		/* Flush the real memory from the instruction cache. */
1829 		if (prot & VM_PROT_EXECUTE)
1830 			sync++;
1831 	}
1832 
1833 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1834 		__syncicache((void *)va, PAGE_SIZE);
1835 		sync = 0;
1836 	}
1837 
1838 	return (KERN_SUCCESS);
1839 }
1840 
1841 /*
1842  * Maps a sequence of resident pages belonging to the same object.
1843  * The sequence begins with the given page m_start.  This page is
1844  * mapped at the given virtual address start.  Each subsequent page is
1845  * mapped at a virtual address that is offset from start by the same
1846  * amount as the page is offset from m_start within the object.  The
1847  * last page in the sequence is the page with the largest offset from
1848  * m_start that can be mapped at a virtual address less than the given
1849  * virtual address end.  Not every virtual page between start and end
1850  * is mapped; only those for which a resident page exists with the
1851  * corresponding offset from m_start are mapped.
1852  */
1853 static void
1854 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1855     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1856 {
1857 	vm_page_t m;
1858 	vm_pindex_t diff, psize;
1859 
1860 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1861 
1862 	psize = atop(end - start);
1863 	m = m_start;
1864 	rw_wlock(&pvh_global_lock);
1865 	PMAP_LOCK(pmap);
1866 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1867 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1868 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1869 		    PMAP_ENTER_NOSLEEP, 0);
1870 		m = TAILQ_NEXT(m, listq);
1871 	}
1872 	rw_wunlock(&pvh_global_lock);
1873 	PMAP_UNLOCK(pmap);
1874 }
1875 
1876 static void
1877 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1878     vm_prot_t prot)
1879 {
1880 
1881 	rw_wlock(&pvh_global_lock);
1882 	PMAP_LOCK(pmap);
1883 	mmu_booke_enter_locked(mmu, pmap, va, m,
1884 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1885 	    0);
1886 	rw_wunlock(&pvh_global_lock);
1887 	PMAP_UNLOCK(pmap);
1888 }
1889 
1890 /*
1891  * Remove the given range of addresses from the specified map.
1892  *
1893  * It is assumed that the start and end are properly rounded to the page size.
1894  */
1895 static void
1896 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1897 {
1898 	pte_t *pte;
1899 	uint8_t hold_flag;
1900 
1901 	int su = (pmap == kernel_pmap);
1902 
1903 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1904 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1905 
1906 	if (su) {
1907 		KASSERT(((va >= virtual_avail) &&
1908 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1909 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1910 	} else {
1911 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1912 		    ("mmu_booke_remove: user pmap, non user va"));
1913 	}
1914 
1915 	if (PMAP_REMOVE_DONE(pmap)) {
1916 		//debugf("mmu_booke_remove: e (empty)\n");
1917 		return;
1918 	}
1919 
1920 	hold_flag = PTBL_HOLD_FLAG(pmap);
1921 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1922 
1923 	rw_wlock(&pvh_global_lock);
1924 	PMAP_LOCK(pmap);
1925 	for (; va < endva; va += PAGE_SIZE) {
1926 		pte = pte_find(mmu, pmap, va);
1927 		if ((pte != NULL) && PTE_ISVALID(pte))
1928 			pte_remove(mmu, pmap, va, hold_flag);
1929 	}
1930 	PMAP_UNLOCK(pmap);
1931 	rw_wunlock(&pvh_global_lock);
1932 
1933 	//debugf("mmu_booke_remove: e\n");
1934 }
1935 
1936 /*
1937  * Remove physical page from all pmaps in which it resides.
1938  */
1939 static void
1940 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1941 {
1942 	pv_entry_t pv, pvn;
1943 	uint8_t hold_flag;
1944 
1945 	rw_wlock(&pvh_global_lock);
1946 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1947 		pvn = TAILQ_NEXT(pv, pv_link);
1948 
1949 		PMAP_LOCK(pv->pv_pmap);
1950 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1951 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1952 		PMAP_UNLOCK(pv->pv_pmap);
1953 	}
1954 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1955 	rw_wunlock(&pvh_global_lock);
1956 }
1957 
1958 /*
1959  * Map a range of physical addresses into kernel virtual address space.
1960  */
1961 static vm_offset_t
1962 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1963     vm_paddr_t pa_end, int prot)
1964 {
1965 	vm_offset_t sva = *virt;
1966 	vm_offset_t va = sva;
1967 
1968 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1969 	//		sva, pa_start, pa_end);
1970 
1971 	while (pa_start < pa_end) {
1972 		mmu_booke_kenter(mmu, va, pa_start);
1973 		va += PAGE_SIZE;
1974 		pa_start += PAGE_SIZE;
1975 	}
1976 	*virt = va;
1977 
1978 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1979 	return (sva);
1980 }
1981 
1982 /*
1983  * The pmap must be activated before it's address space can be accessed in any
1984  * way.
1985  */
1986 static void
1987 mmu_booke_activate(mmu_t mmu, struct thread *td)
1988 {
1989 	pmap_t pmap;
1990 	u_int cpuid;
1991 
1992 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1993 
1994 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1995 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1996 
1997 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1998 
1999 	sched_pin();
2000 
2001 	cpuid = PCPU_GET(cpuid);
2002 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2003 	PCPU_SET(curpmap, pmap);
2004 
2005 	if (pmap->pm_tid[cpuid] == TID_NONE)
2006 		tid_alloc(pmap);
2007 
2008 	/* Load PID0 register with pmap tid value. */
2009 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2010 	__asm __volatile("isync");
2011 
2012 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2013 
2014 	sched_unpin();
2015 
2016 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2017 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2018 }
2019 
2020 /*
2021  * Deactivate the specified process's address space.
2022  */
2023 static void
2024 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2025 {
2026 	pmap_t pmap;
2027 
2028 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2029 
2030 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2031 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2032 
2033 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2034 
2035 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2036 	PCPU_SET(curpmap, NULL);
2037 }
2038 
2039 /*
2040  * Copy the range specified by src_addr/len
2041  * from the source map to the range dst_addr/len
2042  * in the destination map.
2043  *
2044  * This routine is only advisory and need not do anything.
2045  */
2046 static void
2047 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2048     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2049 {
2050 
2051 }
2052 
2053 /*
2054  * Set the physical protection on the specified range of this map as requested.
2055  */
2056 static void
2057 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2058     vm_prot_t prot)
2059 {
2060 	vm_offset_t va;
2061 	vm_page_t m;
2062 	pte_t *pte;
2063 
2064 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2065 		mmu_booke_remove(mmu, pmap, sva, eva);
2066 		return;
2067 	}
2068 
2069 	if (prot & VM_PROT_WRITE)
2070 		return;
2071 
2072 	PMAP_LOCK(pmap);
2073 	for (va = sva; va < eva; va += PAGE_SIZE) {
2074 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2075 			if (PTE_ISVALID(pte)) {
2076 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2077 
2078 				mtx_lock_spin(&tlbivax_mutex);
2079 				tlb_miss_lock();
2080 
2081 				/* Handle modified pages. */
2082 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2083 					vm_page_dirty(m);
2084 
2085 				tlb0_flush_entry(va);
2086 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2087 
2088 				tlb_miss_unlock();
2089 				mtx_unlock_spin(&tlbivax_mutex);
2090 			}
2091 		}
2092 	}
2093 	PMAP_UNLOCK(pmap);
2094 }
2095 
2096 /*
2097  * Clear the write and modified bits in each of the given page's mappings.
2098  */
2099 static void
2100 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2101 {
2102 	pv_entry_t pv;
2103 	pte_t *pte;
2104 
2105 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2106 	    ("mmu_booke_remove_write: page %p is not managed", m));
2107 
2108 	/*
2109 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2110 	 * set by another thread while the object is locked.  Thus,
2111 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2112 	 */
2113 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2114 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2115 		return;
2116 	rw_wlock(&pvh_global_lock);
2117 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2118 		PMAP_LOCK(pv->pv_pmap);
2119 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2120 			if (PTE_ISVALID(pte)) {
2121 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2122 
2123 				mtx_lock_spin(&tlbivax_mutex);
2124 				tlb_miss_lock();
2125 
2126 				/* Handle modified pages. */
2127 				if (PTE_ISMODIFIED(pte))
2128 					vm_page_dirty(m);
2129 
2130 				/* Flush mapping from TLB0. */
2131 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2132 
2133 				tlb_miss_unlock();
2134 				mtx_unlock_spin(&tlbivax_mutex);
2135 			}
2136 		}
2137 		PMAP_UNLOCK(pv->pv_pmap);
2138 	}
2139 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2140 	rw_wunlock(&pvh_global_lock);
2141 }
2142 
2143 static void
2144 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2145 {
2146 	pte_t *pte;
2147 	pmap_t pmap;
2148 	vm_page_t m;
2149 	vm_offset_t addr;
2150 	vm_paddr_t pa = 0;
2151 	int active, valid;
2152 
2153 	va = trunc_page(va);
2154 	sz = round_page(sz);
2155 
2156 	rw_wlock(&pvh_global_lock);
2157 	pmap = PCPU_GET(curpmap);
2158 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2159 	while (sz > 0) {
2160 		PMAP_LOCK(pm);
2161 		pte = pte_find(mmu, pm, va);
2162 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2163 		if (valid)
2164 			pa = PTE_PA(pte);
2165 		PMAP_UNLOCK(pm);
2166 		if (valid) {
2167 			if (!active) {
2168 				/* Create a mapping in the active pmap. */
2169 				addr = 0;
2170 				m = PHYS_TO_VM_PAGE(pa);
2171 				PMAP_LOCK(pmap);
2172 				pte_enter(mmu, pmap, m, addr,
2173 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2174 				__syncicache((void *)addr, PAGE_SIZE);
2175 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2176 				PMAP_UNLOCK(pmap);
2177 			} else
2178 				__syncicache((void *)va, PAGE_SIZE);
2179 		}
2180 		va += PAGE_SIZE;
2181 		sz -= PAGE_SIZE;
2182 	}
2183 	rw_wunlock(&pvh_global_lock);
2184 }
2185 
2186 /*
2187  * Atomically extract and hold the physical page with the given
2188  * pmap and virtual address pair if that mapping permits the given
2189  * protection.
2190  */
2191 static vm_page_t
2192 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2193     vm_prot_t prot)
2194 {
2195 	pte_t *pte;
2196 	vm_page_t m;
2197 	uint32_t pte_wbit;
2198 	vm_paddr_t pa;
2199 
2200 	m = NULL;
2201 	pa = 0;
2202 	PMAP_LOCK(pmap);
2203 retry:
2204 	pte = pte_find(mmu, pmap, va);
2205 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2206 		if (pmap == kernel_pmap)
2207 			pte_wbit = PTE_SW;
2208 		else
2209 			pte_wbit = PTE_UW;
2210 
2211 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2212 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2213 				goto retry;
2214 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2215 			vm_page_hold(m);
2216 		}
2217 	}
2218 
2219 	PA_UNLOCK_COND(pa);
2220 	PMAP_UNLOCK(pmap);
2221 	return (m);
2222 }
2223 
2224 /*
2225  * Initialize a vm_page's machine-dependent fields.
2226  */
2227 static void
2228 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2229 {
2230 
2231 	TAILQ_INIT(&m->md.pv_list);
2232 }
2233 
2234 /*
2235  * mmu_booke_zero_page_area zeros the specified hardware page by
2236  * mapping it into virtual memory and using bzero to clear
2237  * its contents.
2238  *
2239  * off and size must reside within a single page.
2240  */
2241 static void
2242 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2243 {
2244 	vm_offset_t va;
2245 
2246 	/* XXX KASSERT off and size are within a single page? */
2247 
2248 	mtx_lock(&zero_page_mutex);
2249 	va = zero_page_va;
2250 
2251 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2252 	bzero((caddr_t)va + off, size);
2253 	mmu_booke_kremove(mmu, va);
2254 
2255 	mtx_unlock(&zero_page_mutex);
2256 }
2257 
2258 /*
2259  * mmu_booke_zero_page zeros the specified hardware page.
2260  */
2261 static void
2262 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2263 {
2264 	vm_offset_t off, va;
2265 
2266 	mtx_lock(&zero_page_mutex);
2267 	va = zero_page_va;
2268 
2269 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2270 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2271 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2272 	mmu_booke_kremove(mmu, va);
2273 
2274 	mtx_unlock(&zero_page_mutex);
2275 }
2276 
2277 /*
2278  * mmu_booke_copy_page copies the specified (machine independent) page by
2279  * mapping the page into virtual memory and using memcopy to copy the page,
2280  * one machine dependent page at a time.
2281  */
2282 static void
2283 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2284 {
2285 	vm_offset_t sva, dva;
2286 
2287 	sva = copy_page_src_va;
2288 	dva = copy_page_dst_va;
2289 
2290 	mtx_lock(&copy_page_mutex);
2291 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2292 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2293 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2294 	mmu_booke_kremove(mmu, dva);
2295 	mmu_booke_kremove(mmu, sva);
2296 	mtx_unlock(&copy_page_mutex);
2297 }
2298 
2299 static inline void
2300 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2301     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2302 {
2303 	void *a_cp, *b_cp;
2304 	vm_offset_t a_pg_offset, b_pg_offset;
2305 	int cnt;
2306 
2307 	mtx_lock(&copy_page_mutex);
2308 	while (xfersize > 0) {
2309 		a_pg_offset = a_offset & PAGE_MASK;
2310 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2311 		mmu_booke_kenter(mmu, copy_page_src_va,
2312 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2313 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2314 		b_pg_offset = b_offset & PAGE_MASK;
2315 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2316 		mmu_booke_kenter(mmu, copy_page_dst_va,
2317 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2318 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2319 		bcopy(a_cp, b_cp, cnt);
2320 		mmu_booke_kremove(mmu, copy_page_dst_va);
2321 		mmu_booke_kremove(mmu, copy_page_src_va);
2322 		a_offset += cnt;
2323 		b_offset += cnt;
2324 		xfersize -= cnt;
2325 	}
2326 	mtx_unlock(&copy_page_mutex);
2327 }
2328 
2329 /*
2330  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2331  * into virtual memory and using bzero to clear its contents. This is intended
2332  * to be called from the vm_pagezero process only and outside of Giant. No
2333  * lock is required.
2334  */
2335 static void
2336 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2337 {
2338 	vm_offset_t va;
2339 
2340 	va = zero_page_idle_va;
2341 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2342 	bzero((caddr_t)va, PAGE_SIZE);
2343 	mmu_booke_kremove(mmu, va);
2344 }
2345 
2346 static vm_offset_t
2347 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2348 {
2349 	vm_paddr_t paddr;
2350 	vm_offset_t qaddr;
2351 	uint32_t flags;
2352 	pte_t *pte;
2353 
2354 	paddr = VM_PAGE_TO_PHYS(m);
2355 
2356 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2357 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2358 	flags |= PTE_PS_4KB;
2359 
2360 	critical_enter();
2361 	qaddr = PCPU_GET(qmap_addr);
2362 
2363 	pte = pte_find(mmu, kernel_pmap, qaddr);
2364 
2365 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2366 
2367 	/*
2368 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2369  	 * not be present in other TLBs.  Is there a better instruction
2370 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2371 	 */
2372 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2373 	__asm __volatile("isync; msync");
2374 
2375 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
2376 
2377 	/* Flush the real memory from the instruction cache. */
2378 	if ((flags & (PTE_I | PTE_G)) == 0)
2379 		__syncicache((void *)qaddr, PAGE_SIZE);
2380 
2381 	return (qaddr);
2382 }
2383 
2384 static void
2385 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2386 {
2387 	pte_t *pte;
2388 
2389 	pte = pte_find(mmu, kernel_pmap, addr);
2390 
2391 	KASSERT(PCPU_GET(qmap_addr) == addr,
2392 	    ("mmu_booke_quick_remove_page: invalid address"));
2393 	KASSERT(*pte != 0,
2394 	    ("mmu_booke_quick_remove_page: PTE not in use"));
2395 
2396 	*pte = 0;
2397 	critical_exit();
2398 }
2399 
2400 /*
2401  * Return whether or not the specified physical page was modified
2402  * in any of physical maps.
2403  */
2404 static boolean_t
2405 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2406 {
2407 	pte_t *pte;
2408 	pv_entry_t pv;
2409 	boolean_t rv;
2410 
2411 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2412 	    ("mmu_booke_is_modified: page %p is not managed", m));
2413 	rv = FALSE;
2414 
2415 	/*
2416 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2417 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2418 	 * is clear, no PTEs can be modified.
2419 	 */
2420 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2421 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2422 		return (rv);
2423 	rw_wlock(&pvh_global_lock);
2424 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2425 		PMAP_LOCK(pv->pv_pmap);
2426 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2427 		    PTE_ISVALID(pte)) {
2428 			if (PTE_ISMODIFIED(pte))
2429 				rv = TRUE;
2430 		}
2431 		PMAP_UNLOCK(pv->pv_pmap);
2432 		if (rv)
2433 			break;
2434 	}
2435 	rw_wunlock(&pvh_global_lock);
2436 	return (rv);
2437 }
2438 
2439 /*
2440  * Return whether or not the specified virtual address is eligible
2441  * for prefault.
2442  */
2443 static boolean_t
2444 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2445 {
2446 
2447 	return (FALSE);
2448 }
2449 
2450 /*
2451  * Return whether or not the specified physical page was referenced
2452  * in any physical maps.
2453  */
2454 static boolean_t
2455 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2456 {
2457 	pte_t *pte;
2458 	pv_entry_t pv;
2459 	boolean_t rv;
2460 
2461 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2462 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2463 	rv = FALSE;
2464 	rw_wlock(&pvh_global_lock);
2465 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2466 		PMAP_LOCK(pv->pv_pmap);
2467 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2468 		    PTE_ISVALID(pte)) {
2469 			if (PTE_ISREFERENCED(pte))
2470 				rv = TRUE;
2471 		}
2472 		PMAP_UNLOCK(pv->pv_pmap);
2473 		if (rv)
2474 			break;
2475 	}
2476 	rw_wunlock(&pvh_global_lock);
2477 	return (rv);
2478 }
2479 
2480 /*
2481  * Clear the modify bits on the specified physical page.
2482  */
2483 static void
2484 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2485 {
2486 	pte_t *pte;
2487 	pv_entry_t pv;
2488 
2489 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2490 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2491 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2492 	KASSERT(!vm_page_xbusied(m),
2493 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2494 
2495 	/*
2496 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2497 	 * If the object containing the page is locked and the page is not
2498 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2499 	 */
2500 	if ((m->aflags & PGA_WRITEABLE) == 0)
2501 		return;
2502 	rw_wlock(&pvh_global_lock);
2503 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2504 		PMAP_LOCK(pv->pv_pmap);
2505 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2506 		    PTE_ISVALID(pte)) {
2507 			mtx_lock_spin(&tlbivax_mutex);
2508 			tlb_miss_lock();
2509 
2510 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2511 				tlb0_flush_entry(pv->pv_va);
2512 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2513 				    PTE_REFERENCED);
2514 			}
2515 
2516 			tlb_miss_unlock();
2517 			mtx_unlock_spin(&tlbivax_mutex);
2518 		}
2519 		PMAP_UNLOCK(pv->pv_pmap);
2520 	}
2521 	rw_wunlock(&pvh_global_lock);
2522 }
2523 
2524 /*
2525  * Return a count of reference bits for a page, clearing those bits.
2526  * It is not necessary for every reference bit to be cleared, but it
2527  * is necessary that 0 only be returned when there are truly no
2528  * reference bits set.
2529  *
2530  * XXX: The exact number of bits to check and clear is a matter that
2531  * should be tested and standardized at some point in the future for
2532  * optimal aging of shared pages.
2533  */
2534 static int
2535 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2536 {
2537 	pte_t *pte;
2538 	pv_entry_t pv;
2539 	int count;
2540 
2541 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2542 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2543 	count = 0;
2544 	rw_wlock(&pvh_global_lock);
2545 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2546 		PMAP_LOCK(pv->pv_pmap);
2547 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2548 		    PTE_ISVALID(pte)) {
2549 			if (PTE_ISREFERENCED(pte)) {
2550 				mtx_lock_spin(&tlbivax_mutex);
2551 				tlb_miss_lock();
2552 
2553 				tlb0_flush_entry(pv->pv_va);
2554 				*pte &= ~PTE_REFERENCED;
2555 
2556 				tlb_miss_unlock();
2557 				mtx_unlock_spin(&tlbivax_mutex);
2558 
2559 				if (++count > 4) {
2560 					PMAP_UNLOCK(pv->pv_pmap);
2561 					break;
2562 				}
2563 			}
2564 		}
2565 		PMAP_UNLOCK(pv->pv_pmap);
2566 	}
2567 	rw_wunlock(&pvh_global_lock);
2568 	return (count);
2569 }
2570 
2571 /*
2572  * Clear the wired attribute from the mappings for the specified range of
2573  * addresses in the given pmap.  Every valid mapping within that range must
2574  * have the wired attribute set.  In contrast, invalid mappings cannot have
2575  * the wired attribute set, so they are ignored.
2576  *
2577  * The wired attribute of the page table entry is not a hardware feature, so
2578  * there is no need to invalidate any TLB entries.
2579  */
2580 static void
2581 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2582 {
2583 	vm_offset_t va;
2584 	pte_t *pte;
2585 
2586 	PMAP_LOCK(pmap);
2587 	for (va = sva; va < eva; va += PAGE_SIZE) {
2588 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2589 		    PTE_ISVALID(pte)) {
2590 			if (!PTE_ISWIRED(pte))
2591 				panic("mmu_booke_unwire: pte %p isn't wired",
2592 				    pte);
2593 			*pte &= ~PTE_WIRED;
2594 			pmap->pm_stats.wired_count--;
2595 		}
2596 	}
2597 	PMAP_UNLOCK(pmap);
2598 
2599 }
2600 
2601 /*
2602  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2603  * page.  This count may be changed upwards or downwards in the future; it is
2604  * only necessary that true be returned for a small subset of pmaps for proper
2605  * page aging.
2606  */
2607 static boolean_t
2608 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2609 {
2610 	pv_entry_t pv;
2611 	int loops;
2612 	boolean_t rv;
2613 
2614 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2615 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2616 	loops = 0;
2617 	rv = FALSE;
2618 	rw_wlock(&pvh_global_lock);
2619 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2620 		if (pv->pv_pmap == pmap) {
2621 			rv = TRUE;
2622 			break;
2623 		}
2624 		if (++loops >= 16)
2625 			break;
2626 	}
2627 	rw_wunlock(&pvh_global_lock);
2628 	return (rv);
2629 }
2630 
2631 /*
2632  * Return the number of managed mappings to the given physical page that are
2633  * wired.
2634  */
2635 static int
2636 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2637 {
2638 	pv_entry_t pv;
2639 	pte_t *pte;
2640 	int count = 0;
2641 
2642 	if ((m->oflags & VPO_UNMANAGED) != 0)
2643 		return (count);
2644 	rw_wlock(&pvh_global_lock);
2645 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2646 		PMAP_LOCK(pv->pv_pmap);
2647 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2648 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2649 				count++;
2650 		PMAP_UNLOCK(pv->pv_pmap);
2651 	}
2652 	rw_wunlock(&pvh_global_lock);
2653 	return (count);
2654 }
2655 
2656 static int
2657 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2658 {
2659 	int i;
2660 	vm_offset_t va;
2661 
2662 	/*
2663 	 * This currently does not work for entries that
2664 	 * overlap TLB1 entries.
2665 	 */
2666 	for (i = 0; i < TLB1_ENTRIES; i ++) {
2667 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2668 			return (0);
2669 	}
2670 
2671 	return (EFAULT);
2672 }
2673 
2674 void
2675 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2676 {
2677 	vm_paddr_t ppa;
2678 	vm_offset_t ofs;
2679 	vm_size_t gran;
2680 
2681 	/* Minidumps are based on virtual memory addresses. */
2682 	if (do_minidump) {
2683 		*va = (void *)(vm_offset_t)pa;
2684 		return;
2685 	}
2686 
2687 	/* Raw physical memory dumps don't have a virtual address. */
2688 	/* We always map a 256MB page at 256M. */
2689 	gran = 256 * 1024 * 1024;
2690 	ppa = rounddown2(pa, gran);
2691 	ofs = pa - ppa;
2692 	*va = (void *)gran;
2693 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2694 
2695 	if (sz > (gran - ofs))
2696 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2697 		    _TLB_ENTRY_IO);
2698 }
2699 
2700 void
2701 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2702 {
2703 	vm_paddr_t ppa;
2704 	vm_offset_t ofs;
2705 	vm_size_t gran;
2706 	tlb_entry_t e;
2707 	int i;
2708 
2709 	/* Minidumps are based on virtual memory addresses. */
2710 	/* Nothing to do... */
2711 	if (do_minidump)
2712 		return;
2713 
2714 	for (i = 0; i < TLB1_ENTRIES; i++) {
2715 		tlb1_read_entry(&e, i);
2716 		if (!(e.mas1 & MAS1_VALID))
2717 			break;
2718 	}
2719 
2720 	/* Raw physical memory dumps don't have a virtual address. */
2721 	i--;
2722 	e.mas1 = 0;
2723 	e.mas2 = 0;
2724 	e.mas3 = 0;
2725 	tlb1_write_entry(&e, i);
2726 
2727 	gran = 256 * 1024 * 1024;
2728 	ppa = rounddown2(pa, gran);
2729 	ofs = pa - ppa;
2730 	if (sz > (gran - ofs)) {
2731 		i--;
2732 		e.mas1 = 0;
2733 		e.mas2 = 0;
2734 		e.mas3 = 0;
2735 		tlb1_write_entry(&e, i);
2736 	}
2737 }
2738 
2739 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2740 
2741 void
2742 mmu_booke_scan_init(mmu_t mmu)
2743 {
2744 	vm_offset_t va;
2745 	pte_t *pte;
2746 	int i;
2747 
2748 	if (!do_minidump) {
2749 		/* Initialize phys. segments for dumpsys(). */
2750 		memset(&dump_map, 0, sizeof(dump_map));
2751 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2752 		    &availmem_regions_sz);
2753 		for (i = 0; i < physmem_regions_sz; i++) {
2754 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2755 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2756 		}
2757 		return;
2758 	}
2759 
2760 	/* Virtual segments for minidumps: */
2761 	memset(&dump_map, 0, sizeof(dump_map));
2762 
2763 	/* 1st: kernel .data and .bss. */
2764 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2765 	dump_map[0].pa_size =
2766 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2767 
2768 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2769 	dump_map[1].pa_start = data_start;
2770 	dump_map[1].pa_size = data_end - data_start;
2771 
2772 	/* 3rd: kernel VM. */
2773 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2774 	/* Find start of next chunk (from va). */
2775 	while (va < virtual_end) {
2776 		/* Don't dump the buffer cache. */
2777 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2778 			va = kmi.buffer_eva;
2779 			continue;
2780 		}
2781 		pte = pte_find(mmu, kernel_pmap, va);
2782 		if (pte != NULL && PTE_ISVALID(pte))
2783 			break;
2784 		va += PAGE_SIZE;
2785 	}
2786 	if (va < virtual_end) {
2787 		dump_map[2].pa_start = va;
2788 		va += PAGE_SIZE;
2789 		/* Find last page in chunk. */
2790 		while (va < virtual_end) {
2791 			/* Don't run into the buffer cache. */
2792 			if (va == kmi.buffer_sva)
2793 				break;
2794 			pte = pte_find(mmu, kernel_pmap, va);
2795 			if (pte == NULL || !PTE_ISVALID(pte))
2796 				break;
2797 			va += PAGE_SIZE;
2798 		}
2799 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2800 	}
2801 }
2802 
2803 /*
2804  * Map a set of physical memory pages into the kernel virtual address space.
2805  * Return a pointer to where it is mapped. This routine is intended to be used
2806  * for mapping device memory, NOT real memory.
2807  */
2808 static void *
2809 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2810 {
2811 
2812 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2813 }
2814 
2815 static void *
2816 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2817 {
2818 	tlb_entry_t e;
2819 	void *res;
2820 	uintptr_t va, tmpva;
2821 	vm_size_t sz;
2822 	int i;
2823 
2824 	/*
2825 	 * Check if this is premapped in TLB1. Note: this should probably also
2826 	 * check whether a sequence of TLB1 entries exist that match the
2827 	 * requirement, but now only checks the easy case.
2828 	 */
2829 	if (ma == VM_MEMATTR_DEFAULT) {
2830 		for (i = 0; i < TLB1_ENTRIES; i++) {
2831 			tlb1_read_entry(&e, i);
2832 			if (!(e.mas1 & MAS1_VALID))
2833 				continue;
2834 			if (pa >= e.phys &&
2835 			    (pa + size) <= (e.phys + e.size))
2836 				return (void *)(e.virt +
2837 				    (vm_offset_t)(pa - e.phys));
2838 		}
2839 	}
2840 
2841 	size = roundup(size, PAGE_SIZE);
2842 
2843 	/*
2844 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
2845 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
2846 	 */
2847 #ifdef SPARSE_MAPDEV
2848 	/*
2849 	 * With a sparse mapdev, align to the largest starting region.  This
2850 	 * could feasibly be optimized for a 'best-fit' alignment, but that
2851 	 * calculation could be very costly.
2852 	 */
2853 	do {
2854 	    tmpva = tlb1_map_base;
2855 	    va = roundup(tlb1_map_base, 1 << flsl(size));
2856 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
2857 #else
2858 	va = atomic_fetchadd_int(&tlb1_map_base, size);
2859 #endif
2860 	res = (void *)va;
2861 
2862 	do {
2863 		sz = 1 << (ilog2(size) & ~1);
2864 		if (va % sz != 0) {
2865 			do {
2866 				sz >>= 2;
2867 			} while (va % sz != 0);
2868 		}
2869 		if (bootverbose)
2870 			printf("Wiring VA=%x to PA=%jx (size=%x)\n",
2871 			    va, (uintmax_t)pa, sz);
2872 		tlb1_set_entry(va, pa, sz,
2873 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma));
2874 		size -= sz;
2875 		pa += sz;
2876 		va += sz;
2877 	} while (size > 0);
2878 
2879 	return (res);
2880 }
2881 
2882 /*
2883  * 'Unmap' a range mapped by mmu_booke_mapdev().
2884  */
2885 static void
2886 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2887 {
2888 #ifdef SUPPORTS_SHRINKING_TLB1
2889 	vm_offset_t base, offset;
2890 
2891 	/*
2892 	 * Unmap only if this is inside kernel virtual space.
2893 	 */
2894 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2895 		base = trunc_page(va);
2896 		offset = va & PAGE_MASK;
2897 		size = roundup(offset + size, PAGE_SIZE);
2898 		kva_free(base, size);
2899 	}
2900 #endif
2901 }
2902 
2903 /*
2904  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2905  * specified pmap. This eliminates the blast of soft faults on process startup
2906  * and immediately after an mmap.
2907  */
2908 static void
2909 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2910     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2911 {
2912 
2913 	VM_OBJECT_ASSERT_WLOCKED(object);
2914 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2915 	    ("mmu_booke_object_init_pt: non-device object"));
2916 }
2917 
2918 /*
2919  * Perform the pmap work for mincore.
2920  */
2921 static int
2922 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2923     vm_paddr_t *locked_pa)
2924 {
2925 
2926 	/* XXX: this should be implemented at some point */
2927 	return (0);
2928 }
2929 
2930 static int
2931 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
2932     vm_memattr_t mode)
2933 {
2934 	vm_offset_t va;
2935 	pte_t *pte;
2936 	int i, j;
2937 	tlb_entry_t e;
2938 
2939 	/* Check TLB1 mappings */
2940 	for (i = 0; i < TLB1_ENTRIES; i++) {
2941 		tlb1_read_entry(&e, i);
2942 		if (!(e.mas1 & MAS1_VALID))
2943 			continue;
2944 		if (addr >= e.virt && addr < e.virt + e.size)
2945 			break;
2946 	}
2947 	if (i < TLB1_ENTRIES) {
2948 		/* Only allow full mappings to be modified for now. */
2949 		/* Validate the range. */
2950 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
2951 			tlb1_read_entry(&e, j);
2952 			if (va != e.virt || (sz - (va - addr) < e.size))
2953 				return (EINVAL);
2954 		}
2955 		for (va = addr; va < addr + sz; va += e.size, i++) {
2956 			tlb1_read_entry(&e, i);
2957 			e.mas2 &= ~MAS2_WIMGE_MASK;
2958 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
2959 
2960 			/*
2961 			 * Write it out to the TLB.  Should really re-sync with other
2962 			 * cores.
2963 			 */
2964 			tlb1_write_entry(&e, i);
2965 		}
2966 		return (0);
2967 	}
2968 
2969 	/* Not in TLB1, try through pmap */
2970 	/* First validate the range. */
2971 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2972 		pte = pte_find(mmu, kernel_pmap, va);
2973 		if (pte == NULL || !PTE_ISVALID(pte))
2974 			return (EINVAL);
2975 	}
2976 
2977 	mtx_lock_spin(&tlbivax_mutex);
2978 	tlb_miss_lock();
2979 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2980 		pte = pte_find(mmu, kernel_pmap, va);
2981 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2982 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode << PTE_MAS2_SHIFT);
2983 		tlb0_flush_entry(va);
2984 	}
2985 	tlb_miss_unlock();
2986 	mtx_unlock_spin(&tlbivax_mutex);
2987 
2988 	return (pte_vatopa(mmu, kernel_pmap, va));
2989 }
2990 
2991 /**************************************************************************/
2992 /* TID handling */
2993 /**************************************************************************/
2994 
2995 /*
2996  * Allocate a TID. If necessary, steal one from someone else.
2997  * The new TID is flushed from the TLB before returning.
2998  */
2999 static tlbtid_t
3000 tid_alloc(pmap_t pmap)
3001 {
3002 	tlbtid_t tid;
3003 	int thiscpu;
3004 
3005 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3006 
3007 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3008 
3009 	thiscpu = PCPU_GET(cpuid);
3010 
3011 	tid = PCPU_GET(tid_next);
3012 	if (tid > TID_MAX)
3013 		tid = TID_MIN;
3014 	PCPU_SET(tid_next, tid + 1);
3015 
3016 	/* If we are stealing TID then clear the relevant pmap's field */
3017 	if (tidbusy[thiscpu][tid] != NULL) {
3018 
3019 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3020 
3021 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3022 
3023 		/* Flush all entries from TLB0 matching this TID. */
3024 		tid_flush(tid);
3025 	}
3026 
3027 	tidbusy[thiscpu][tid] = pmap;
3028 	pmap->pm_tid[thiscpu] = tid;
3029 	__asm __volatile("msync; isync");
3030 
3031 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3032 	    PCPU_GET(tid_next));
3033 
3034 	return (tid);
3035 }
3036 
3037 /**************************************************************************/
3038 /* TLB0 handling */
3039 /**************************************************************************/
3040 
3041 static void
3042 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3043     uint32_t mas7)
3044 {
3045 	int as;
3046 	char desc[3];
3047 	tlbtid_t tid;
3048 	vm_size_t size;
3049 	unsigned int tsize;
3050 
3051 	desc[2] = '\0';
3052 	if (mas1 & MAS1_VALID)
3053 		desc[0] = 'V';
3054 	else
3055 		desc[0] = ' ';
3056 
3057 	if (mas1 & MAS1_IPROT)
3058 		desc[1] = 'P';
3059 	else
3060 		desc[1] = ' ';
3061 
3062 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3063 	tid = MAS1_GETTID(mas1);
3064 
3065 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3066 	size = 0;
3067 	if (tsize)
3068 		size = tsize2size(tsize);
3069 
3070 	debugf("%3d: (%s) [AS=%d] "
3071 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3072 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3073 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3074 }
3075 
3076 /* Convert TLB0 va and way number to tlb0[] table index. */
3077 static inline unsigned int
3078 tlb0_tableidx(vm_offset_t va, unsigned int way)
3079 {
3080 	unsigned int idx;
3081 
3082 	idx = (way * TLB0_ENTRIES_PER_WAY);
3083 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3084 	return (idx);
3085 }
3086 
3087 /*
3088  * Invalidate TLB0 entry.
3089  */
3090 static inline void
3091 tlb0_flush_entry(vm_offset_t va)
3092 {
3093 
3094 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3095 
3096 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3097 
3098 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3099 	__asm __volatile("isync; msync");
3100 	__asm __volatile("tlbsync; msync");
3101 
3102 	CTR1(KTR_PMAP, "%s: e", __func__);
3103 }
3104 
3105 /* Print out contents of the MAS registers for each TLB0 entry */
3106 void
3107 tlb0_print_tlbentries(void)
3108 {
3109 	uint32_t mas0, mas1, mas2, mas3, mas7;
3110 	int entryidx, way, idx;
3111 
3112 	debugf("TLB0 entries:\n");
3113 	for (way = 0; way < TLB0_WAYS; way ++)
3114 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3115 
3116 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3117 			mtspr(SPR_MAS0, mas0);
3118 			__asm __volatile("isync");
3119 
3120 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3121 			mtspr(SPR_MAS2, mas2);
3122 
3123 			__asm __volatile("isync; tlbre");
3124 
3125 			mas1 = mfspr(SPR_MAS1);
3126 			mas2 = mfspr(SPR_MAS2);
3127 			mas3 = mfspr(SPR_MAS3);
3128 			mas7 = mfspr(SPR_MAS7);
3129 
3130 			idx = tlb0_tableidx(mas2, way);
3131 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3132 		}
3133 }
3134 
3135 /**************************************************************************/
3136 /* TLB1 handling */
3137 /**************************************************************************/
3138 
3139 /*
3140  * TLB1 mapping notes:
3141  *
3142  * TLB1[0]	Kernel text and data.
3143  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3144  *		windows, other devices mappings.
3145  */
3146 
3147  /*
3148  * Read an entry from given TLB1 slot.
3149  */
3150 void
3151 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3152 {
3153 	uint32_t mas0;
3154 
3155 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3156 
3157 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3158 	mtspr(SPR_MAS0, mas0);
3159 	__asm __volatile("isync; tlbre");
3160 
3161 	entry->mas1 = mfspr(SPR_MAS1);
3162 	entry->mas2 = mfspr(SPR_MAS2);
3163 	entry->mas3 = mfspr(SPR_MAS3);
3164 
3165 	switch ((mfpvr() >> 16) & 0xFFFF) {
3166 	case FSL_E500v2:
3167 	case FSL_E500mc:
3168 	case FSL_E5500:
3169 		entry->mas7 = mfspr(SPR_MAS7);
3170 		break;
3171 	default:
3172 		entry->mas7 = 0;
3173 		break;
3174 	}
3175 
3176 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3177 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3178 	    (entry->mas3 & MAS3_RPN);
3179 	entry->size =
3180 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3181 }
3182 
3183 /*
3184  * Write given entry to TLB1 hardware.
3185  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
3186  */
3187 static void
3188 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3189 {
3190 	uint32_t mas0;
3191 
3192 	//debugf("tlb1_write_entry: s\n");
3193 
3194 	/* Select entry */
3195 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3196 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
3197 
3198 	mtspr(SPR_MAS0, mas0);
3199 	__asm __volatile("isync");
3200 	mtspr(SPR_MAS1, e->mas1);
3201 	__asm __volatile("isync");
3202 	mtspr(SPR_MAS2, e->mas2);
3203 	__asm __volatile("isync");
3204 	mtspr(SPR_MAS3, e->mas3);
3205 	__asm __volatile("isync");
3206 	switch ((mfpvr() >> 16) & 0xFFFF) {
3207 	case FSL_E500mc:
3208 	case FSL_E5500:
3209 		mtspr(SPR_MAS8, 0);
3210 		__asm __volatile("isync");
3211 		/* FALLTHROUGH */
3212 	case FSL_E500v2:
3213 		mtspr(SPR_MAS7, e->mas7);
3214 		__asm __volatile("isync");
3215 		break;
3216 	default:
3217 		break;
3218 	}
3219 
3220 	__asm __volatile("tlbwe; isync; msync");
3221 
3222 	//debugf("tlb1_write_entry: e\n");
3223 }
3224 
3225 /*
3226  * Return the largest uint value log such that 2^log <= num.
3227  */
3228 static unsigned int
3229 ilog2(unsigned int num)
3230 {
3231 	int lz;
3232 
3233 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3234 	return (31 - lz);
3235 }
3236 
3237 /*
3238  * Convert TLB TSIZE value to mapped region size.
3239  */
3240 static vm_size_t
3241 tsize2size(unsigned int tsize)
3242 {
3243 
3244 	/*
3245 	 * size = 4^tsize KB
3246 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3247 	 */
3248 
3249 	return ((1 << (2 * tsize)) * 1024);
3250 }
3251 
3252 /*
3253  * Convert region size (must be power of 4) to TLB TSIZE value.
3254  */
3255 static unsigned int
3256 size2tsize(vm_size_t size)
3257 {
3258 
3259 	return (ilog2(size) / 2 - 5);
3260 }
3261 
3262 /*
3263  * Register permanent kernel mapping in TLB1.
3264  *
3265  * Entries are created starting from index 0 (current free entry is
3266  * kept in tlb1_idx) and are not supposed to be invalidated.
3267  */
3268 int
3269 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3270     uint32_t flags)
3271 {
3272 	tlb_entry_t e;
3273 	uint32_t ts, tid;
3274 	int tsize, index;
3275 
3276 	for (index = 0; index < TLB1_ENTRIES; index++) {
3277 		tlb1_read_entry(&e, index);
3278 		if ((e.mas1 & MAS1_VALID) == 0)
3279 			break;
3280 		/* Check if we're just updating the flags, and update them. */
3281 		if (e.phys == pa && e.virt == va && e.size == size) {
3282 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
3283 			tlb1_write_entry(&e, index);
3284 			return (0);
3285 		}
3286 	}
3287 	if (index >= TLB1_ENTRIES) {
3288 		printf("tlb1_set_entry: TLB1 full!\n");
3289 		return (-1);
3290 	}
3291 
3292 	/* Convert size to TSIZE */
3293 	tsize = size2tsize(size);
3294 
3295 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3296 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3297 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3298 
3299 	e.phys = pa;
3300 	e.virt = va;
3301 	e.size = size;
3302 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3303 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3304 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
3305 
3306 	/* Set supervisor RWX permission bits */
3307 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3308 	e.mas7 = (pa >> 32) & MAS7_RPN;
3309 
3310 	tlb1_write_entry(&e, index);
3311 
3312 	/*
3313 	 * XXX in general TLB1 updates should be propagated between CPUs,
3314 	 * since current design assumes to have the same TLB1 set-up on all
3315 	 * cores.
3316 	 */
3317 	return (0);
3318 }
3319 
3320 /*
3321  * Map in contiguous RAM region into the TLB1 using maximum of
3322  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3323  *
3324  * If necessary round up last entry size and return total size
3325  * used by all allocated entries.
3326  */
3327 vm_size_t
3328 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3329 {
3330 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3331 	vm_size_t mapped, pgsz, base, mask;
3332 	int idx, nents;
3333 
3334 	/* Round up to the next 1M */
3335 	size = roundup2(size, 1 << 20);
3336 
3337 	mapped = 0;
3338 	idx = 0;
3339 	base = va;
3340 	pgsz = 64*1024*1024;
3341 	while (mapped < size) {
3342 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3343 			while (pgsz > (size - mapped))
3344 				pgsz >>= 2;
3345 			pgs[idx++] = pgsz;
3346 			mapped += pgsz;
3347 		}
3348 
3349 		/* We under-map. Correct for this. */
3350 		if (mapped < size) {
3351 			while (pgs[idx - 1] == pgsz) {
3352 				idx--;
3353 				mapped -= pgsz;
3354 			}
3355 			/* XXX We may increase beyond out starting point. */
3356 			pgsz <<= 2;
3357 			pgs[idx++] = pgsz;
3358 			mapped += pgsz;
3359 		}
3360 	}
3361 
3362 	nents = idx;
3363 	mask = pgs[0] - 1;
3364 	/* Align address to the boundary */
3365 	if (va & mask) {
3366 		va = (va + mask) & ~mask;
3367 		pa = (pa + mask) & ~mask;
3368 	}
3369 
3370 	for (idx = 0; idx < nents; idx++) {
3371 		pgsz = pgs[idx];
3372 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
3373 		tlb1_set_entry(va, pa, pgsz,
3374 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
3375 		pa += pgsz;
3376 		va += pgsz;
3377 	}
3378 
3379 	mapped = (va - base);
3380 #ifdef __powerpc64__
3381 	printf("mapped size 0x%016lx (wasted space 0x%16lx)\n",
3382 #else
3383 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3384 #endif
3385 	    mapped, mapped - size);
3386 	return (mapped);
3387 }
3388 
3389 /*
3390  * TLB1 initialization routine, to be called after the very first
3391  * assembler level setup done in locore.S.
3392  */
3393 void
3394 tlb1_init()
3395 {
3396 	uint32_t mas0, mas1, mas2, mas3, mas7;
3397 	uint32_t tsz;
3398 
3399 	tlb1_get_tlbconf();
3400 
3401 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3402 	mtspr(SPR_MAS0, mas0);
3403 	__asm __volatile("isync; tlbre");
3404 
3405 	mas1 = mfspr(SPR_MAS1);
3406 	mas2 = mfspr(SPR_MAS2);
3407 	mas3 = mfspr(SPR_MAS3);
3408 	mas7 = mfspr(SPR_MAS7);
3409 
3410 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
3411 	    (mas3 & MAS3_RPN);
3412 
3413 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3414 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
3415 
3416 	/* Setup TLB miss defaults */
3417 	set_mas4_defaults();
3418 }
3419 
3420 vm_offset_t
3421 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3422 {
3423 	vm_paddr_t pa_base;
3424 	vm_offset_t va, sz;
3425 	int i;
3426 	tlb_entry_t e;
3427 
3428 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3429 
3430 	for (i = 0; i < TLB1_ENTRIES; i++) {
3431 		tlb1_read_entry(&e, i);
3432 		if (!(e.mas1 & MAS1_VALID))
3433 			continue;
3434 		if (pa >= e.phys && (pa + size) <=
3435 		    (e.phys + e.size))
3436 			return (e.virt + (pa - e.phys));
3437 	}
3438 
3439 	pa_base = rounddown(pa, PAGE_SIZE);
3440 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3441 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3442 	va = tlb1_map_base + (pa - pa_base);
3443 
3444 	do {
3445 		sz = 1 << (ilog2(size) & ~1);
3446 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
3447 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
3448 		size -= sz;
3449 		pa_base += sz;
3450 		tlb1_map_base += sz;
3451 	} while (size > 0);
3452 
3453 	return (va);
3454 }
3455 
3456 /*
3457  * Setup MAS4 defaults.
3458  * These values are loaded to MAS0-2 on a TLB miss.
3459  */
3460 static void
3461 set_mas4_defaults(void)
3462 {
3463 	uint32_t mas4;
3464 
3465 	/* Defaults: TLB0, PID0, TSIZED=4K */
3466 	mas4 = MAS4_TLBSELD0;
3467 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3468 #ifdef SMP
3469 	mas4 |= MAS4_MD;
3470 #endif
3471 	mtspr(SPR_MAS4, mas4);
3472 	__asm __volatile("isync");
3473 }
3474 
3475 /*
3476  * Print out contents of the MAS registers for each TLB1 entry
3477  */
3478 void
3479 tlb1_print_tlbentries(void)
3480 {
3481 	uint32_t mas0, mas1, mas2, mas3, mas7;
3482 	int i;
3483 
3484 	debugf("TLB1 entries:\n");
3485 	for (i = 0; i < TLB1_ENTRIES; i++) {
3486 
3487 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3488 		mtspr(SPR_MAS0, mas0);
3489 
3490 		__asm __volatile("isync; tlbre");
3491 
3492 		mas1 = mfspr(SPR_MAS1);
3493 		mas2 = mfspr(SPR_MAS2);
3494 		mas3 = mfspr(SPR_MAS3);
3495 		mas7 = mfspr(SPR_MAS7);
3496 
3497 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3498 	}
3499 }
3500 
3501 /*
3502  * Return 0 if the physical IO range is encompassed by one of the
3503  * the TLB1 entries, otherwise return related error code.
3504  */
3505 static int
3506 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3507 {
3508 	uint32_t prot;
3509 	vm_paddr_t pa_start;
3510 	vm_paddr_t pa_end;
3511 	unsigned int entry_tsize;
3512 	vm_size_t entry_size;
3513 	tlb_entry_t e;
3514 
3515 	*va = (vm_offset_t)NULL;
3516 
3517 	tlb1_read_entry(&e, i);
3518 	/* Skip invalid entries */
3519 	if (!(e.mas1 & MAS1_VALID))
3520 		return (EINVAL);
3521 
3522 	/*
3523 	 * The entry must be cache-inhibited, guarded, and r/w
3524 	 * so it can function as an i/o page
3525 	 */
3526 	prot = e.mas2 & (MAS2_I | MAS2_G);
3527 	if (prot != (MAS2_I | MAS2_G))
3528 		return (EPERM);
3529 
3530 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
3531 	if (prot != (MAS3_SR | MAS3_SW))
3532 		return (EPERM);
3533 
3534 	/* The address should be within the entry range. */
3535 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3536 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3537 
3538 	entry_size = tsize2size(entry_tsize);
3539 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
3540 	    (e.mas3 & MAS3_RPN);
3541 	pa_end = pa_start + entry_size;
3542 
3543 	if ((pa < pa_start) || ((pa + size) > pa_end))
3544 		return (ERANGE);
3545 
3546 	/* Return virtual address of this mapping. */
3547 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3548 	return (0);
3549 }
3550 
3551 /*
3552  * Invalidate all TLB0 entries which match the given TID. Note this is
3553  * dedicated for cases when invalidations should NOT be propagated to other
3554  * CPUs.
3555  */
3556 static void
3557 tid_flush(tlbtid_t tid)
3558 {
3559 	register_t msr;
3560 	uint32_t mas0, mas1, mas2;
3561 	int entry, way;
3562 
3563 
3564 	/* Don't evict kernel translations */
3565 	if (tid == TID_KERNEL)
3566 		return;
3567 
3568 	msr = mfmsr();
3569 	__asm __volatile("wrteei 0");
3570 
3571 	for (way = 0; way < TLB0_WAYS; way++)
3572 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
3573 
3574 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3575 			mtspr(SPR_MAS0, mas0);
3576 			__asm __volatile("isync");
3577 
3578 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
3579 			mtspr(SPR_MAS2, mas2);
3580 
3581 			__asm __volatile("isync; tlbre");
3582 
3583 			mas1 = mfspr(SPR_MAS1);
3584 
3585 			if (!(mas1 & MAS1_VALID))
3586 				continue;
3587 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
3588 				continue;
3589 			mas1 &= ~MAS1_VALID;
3590 			mtspr(SPR_MAS1, mas1);
3591 			__asm __volatile("isync; tlbwe; isync; msync");
3592 		}
3593 	mtmsr(msr);
3594 }
3595