xref: /freebsd/sys/powerpc/booke/pmap.c (revision 7d536dc855c85c15bf45f033d108a61b1f3cecc3)
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 #ifdef  DEBUG
100 #define debugf(fmt, args...) printf(fmt, ##args)
101 #else
102 #define debugf(fmt, args...)
103 #endif
104 
105 #define TODO			panic("%s: not implemented", __func__);
106 
107 extern unsigned char _etext[];
108 extern unsigned char _end[];
109 
110 extern uint32_t *bootinfo;
111 
112 #ifdef SMP
113 extern uint32_t bp_ntlb1s;
114 #endif
115 
116 vm_paddr_t kernload;
117 vm_offset_t kernstart;
118 vm_size_t kernsize;
119 
120 /* Message buffer and tables. */
121 static vm_offset_t data_start;
122 static vm_size_t data_end;
123 
124 /* Phys/avail memory regions. */
125 static struct mem_region *availmem_regions;
126 static int availmem_regions_sz;
127 static struct mem_region *physmem_regions;
128 static int physmem_regions_sz;
129 
130 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
131 static vm_offset_t zero_page_va;
132 static struct mtx zero_page_mutex;
133 
134 static struct mtx tlbivax_mutex;
135 
136 /*
137  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
138  * by idle thred only, no lock required.
139  */
140 static vm_offset_t zero_page_idle_va;
141 
142 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
143 static vm_offset_t copy_page_src_va;
144 static vm_offset_t copy_page_dst_va;
145 static struct mtx copy_page_mutex;
146 
147 /**************************************************************************/
148 /* PMAP */
149 /**************************************************************************/
150 
151 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
152     vm_prot_t, u_int flags, int8_t psind);
153 
154 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
155 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
156 
157 /*
158  * If user pmap is processed with mmu_booke_remove and the resident count
159  * drops to 0, there are no more pages to remove, so we need not continue.
160  */
161 #define PMAP_REMOVE_DONE(pmap) \
162 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
163 
164 extern int elf32_nxstack;
165 
166 /**************************************************************************/
167 /* TLB and TID handling */
168 /**************************************************************************/
169 
170 /* Translation ID busy table */
171 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
172 
173 /*
174  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
175  * core revisions and should be read from h/w registers during early config.
176  */
177 uint32_t tlb0_entries;
178 uint32_t tlb0_ways;
179 uint32_t tlb0_entries_per_way;
180 uint32_t tlb1_entries;
181 
182 #define TLB0_ENTRIES		(tlb0_entries)
183 #define TLB0_WAYS		(tlb0_ways)
184 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
185 
186 #define TLB1_ENTRIES (tlb1_entries)
187 #define TLB1_MAXENTRIES	64
188 
189 /* In-ram copy of the TLB1 */
190 static tlb_entry_t tlb1[TLB1_MAXENTRIES];
191 
192 /* Next free entry in the TLB1 */
193 static unsigned int tlb1_idx;
194 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
195 
196 static tlbtid_t tid_alloc(struct pmap *);
197 static void tid_flush(tlbtid_t tid);
198 
199 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
200 
201 static void tlb1_write_entry(unsigned int);
202 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
203 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
204 
205 static vm_size_t tsize2size(unsigned int);
206 static unsigned int size2tsize(vm_size_t);
207 static unsigned int ilog2(unsigned int);
208 
209 static void set_mas4_defaults(void);
210 
211 static inline void tlb0_flush_entry(vm_offset_t);
212 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
213 
214 /**************************************************************************/
215 /* Page table management */
216 /**************************************************************************/
217 
218 static struct rwlock_padalign pvh_global_lock;
219 
220 /* Data for the pv entry allocation mechanism */
221 static uma_zone_t pvzone;
222 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
223 
224 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
225 
226 #ifndef PMAP_SHPGPERPROC
227 #define PMAP_SHPGPERPROC	200
228 #endif
229 
230 static void ptbl_init(void);
231 static struct ptbl_buf *ptbl_buf_alloc(void);
232 static void ptbl_buf_free(struct ptbl_buf *);
233 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
234 
235 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
236 static void ptbl_free(mmu_t, pmap_t, unsigned int);
237 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
238 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
239 
240 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
241 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
242 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
243 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
244 static void kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr,
245 			     vm_offset_t pdir);
246 
247 static pv_entry_t pv_alloc(void);
248 static void pv_free(pv_entry_t);
249 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
250 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
251 
252 static void booke_pmap_init_qpages(void);
253 
254 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
255 #define PTBL_BUFS		(128 * 16)
256 
257 struct ptbl_buf {
258 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
259 	vm_offset_t kva;		/* va of mapping */
260 };
261 
262 /* ptbl free list and a lock used for access synchronization. */
263 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
264 static struct mtx ptbl_buf_freelist_lock;
265 
266 /* Base address of kva space allocated fot ptbl bufs. */
267 static vm_offset_t ptbl_buf_pool_vabase;
268 
269 /* Pointer to ptbl_buf structures. */
270 static struct ptbl_buf *ptbl_bufs;
271 
272 #ifdef SMP
273 void pmap_bootstrap_ap(volatile uint32_t *);
274 #endif
275 
276 /*
277  * Kernel MMU interface
278  */
279 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
280 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
281     vm_size_t, vm_offset_t);
282 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
283 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
284     vm_offset_t, vm_page_t *, vm_offset_t, int);
285 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
286     vm_prot_t, u_int flags, int8_t psind);
287 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
288     vm_page_t, vm_prot_t);
289 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
290     vm_prot_t);
291 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
292 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
293     vm_prot_t);
294 static void		mmu_booke_init(mmu_t);
295 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
296 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
297 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
298 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
299 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
300     int);
301 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
302     vm_paddr_t *);
303 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
304     vm_object_t, vm_pindex_t, vm_size_t);
305 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
306 static void		mmu_booke_page_init(mmu_t, vm_page_t);
307 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
308 static void		mmu_booke_pinit(mmu_t, pmap_t);
309 static void		mmu_booke_pinit0(mmu_t, pmap_t);
310 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
311     vm_prot_t);
312 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
313 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
314 static void		mmu_booke_release(mmu_t, pmap_t);
315 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
316 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
317 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
318 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
319 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
320 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
321 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
322 static void		mmu_booke_activate(mmu_t, struct thread *);
323 static void		mmu_booke_deactivate(mmu_t, struct thread *);
324 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
325 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
326 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
327 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
328 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
329 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
330 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
331 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
332 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
333 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
334     vm_size_t);
335 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
336     void **);
337 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
338     void *);
339 static void		mmu_booke_scan_init(mmu_t);
340 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
341 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
342 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
343     vm_size_t sz, vm_memattr_t mode);
344 
345 static mmu_method_t mmu_booke_methods[] = {
346 	/* pmap dispatcher interface */
347 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
348 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
349 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
350 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
351 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
352 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
353 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
354 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
355 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
356 	MMUMETHOD(mmu_init,		mmu_booke_init),
357 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
358 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
359 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
360 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
361 	MMUMETHOD(mmu_map,		mmu_booke_map),
362 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
363 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
364 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
365 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
366 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
367 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
368 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
369 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
370 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
371 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
372 	MMUMETHOD(mmu_release,		mmu_booke_release),
373 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
374 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
375 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
376 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
377 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
378 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
379 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
380 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
381 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
382 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
383 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
384 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
385 
386 	/* Internal interfaces */
387 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
388 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
389 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
390 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
391 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
392 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
393 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
394 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
395 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
396 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
397 
398 	/* dumpsys() support */
399 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
400 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
401 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
402 
403 	{ 0, 0 }
404 };
405 
406 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
407 
408 static __inline uint32_t
409 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
410 {
411 	uint32_t attrib;
412 	int i;
413 
414 	if (ma != VM_MEMATTR_DEFAULT) {
415 		switch (ma) {
416 		case VM_MEMATTR_UNCACHEABLE:
417 			return (MAS2_I | MAS2_G);
418 		case VM_MEMATTR_WRITE_COMBINING:
419 		case VM_MEMATTR_WRITE_BACK:
420 		case VM_MEMATTR_PREFETCHABLE:
421 			return (MAS2_I);
422 		case VM_MEMATTR_WRITE_THROUGH:
423 			return (MAS2_W | MAS2_M);
424 		case VM_MEMATTR_CACHEABLE:
425 			return (MAS2_M);
426 		}
427 	}
428 
429 	/*
430 	 * Assume the page is cache inhibited and access is guarded unless
431 	 * it's in our available memory array.
432 	 */
433 	attrib = _TLB_ENTRY_IO;
434 	for (i = 0; i < physmem_regions_sz; i++) {
435 		if ((pa >= physmem_regions[i].mr_start) &&
436 		    (pa < (physmem_regions[i].mr_start +
437 		     physmem_regions[i].mr_size))) {
438 			attrib = _TLB_ENTRY_MEM;
439 			break;
440 		}
441 	}
442 
443 	return (attrib);
444 }
445 
446 static inline void
447 tlb_miss_lock(void)
448 {
449 #ifdef SMP
450 	struct pcpu *pc;
451 
452 	if (!smp_started)
453 		return;
454 
455 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
456 		if (pc != pcpup) {
457 
458 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
459 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
460 
461 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
462 			    ("tlb_miss_lock: tried to lock self"));
463 
464 			tlb_lock(pc->pc_booke_tlb_lock);
465 
466 			CTR1(KTR_PMAP, "%s: locked", __func__);
467 		}
468 	}
469 #endif
470 }
471 
472 static inline void
473 tlb_miss_unlock(void)
474 {
475 #ifdef SMP
476 	struct pcpu *pc;
477 
478 	if (!smp_started)
479 		return;
480 
481 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
482 		if (pc != pcpup) {
483 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
484 			    __func__, pc->pc_cpuid);
485 
486 			tlb_unlock(pc->pc_booke_tlb_lock);
487 
488 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
489 		}
490 	}
491 #endif
492 }
493 
494 /* Return number of entries in TLB0. */
495 static __inline void
496 tlb0_get_tlbconf(void)
497 {
498 	uint32_t tlb0_cfg;
499 
500 	tlb0_cfg = mfspr(SPR_TLB0CFG);
501 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
502 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
503 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
504 }
505 
506 /* Return number of entries in TLB1. */
507 static __inline void
508 tlb1_get_tlbconf(void)
509 {
510 	uint32_t tlb1_cfg;
511 
512 	tlb1_cfg = mfspr(SPR_TLB1CFG);
513 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
514 }
515 
516 /**************************************************************************/
517 /* Page table related */
518 /**************************************************************************/
519 
520 /* Initialize pool of kva ptbl buffers. */
521 static void
522 ptbl_init(void)
523 {
524 	int i;
525 
526 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
527 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
528 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
529 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
530 
531 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
532 	TAILQ_INIT(&ptbl_buf_freelist);
533 
534 	for (i = 0; i < PTBL_BUFS; i++) {
535 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
536 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
537 	}
538 }
539 
540 /* Get a ptbl_buf from the freelist. */
541 static struct ptbl_buf *
542 ptbl_buf_alloc(void)
543 {
544 	struct ptbl_buf *buf;
545 
546 	mtx_lock(&ptbl_buf_freelist_lock);
547 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
548 	if (buf != NULL)
549 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
550 	mtx_unlock(&ptbl_buf_freelist_lock);
551 
552 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
553 
554 	return (buf);
555 }
556 
557 /* Return ptbl buff to free pool. */
558 static void
559 ptbl_buf_free(struct ptbl_buf *buf)
560 {
561 
562 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
563 
564 	mtx_lock(&ptbl_buf_freelist_lock);
565 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
566 	mtx_unlock(&ptbl_buf_freelist_lock);
567 }
568 
569 /*
570  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
571  */
572 static void
573 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
574 {
575 	struct ptbl_buf *pbuf;
576 
577 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
578 
579 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
580 
581 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
582 		if (pbuf->kva == (vm_offset_t)ptbl) {
583 			/* Remove from pmap ptbl buf list. */
584 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
585 
586 			/* Free corresponding ptbl buf. */
587 			ptbl_buf_free(pbuf);
588 			break;
589 		}
590 }
591 
592 /* Allocate page table. */
593 static pte_t *
594 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
595 {
596 	vm_page_t mtbl[PTBL_PAGES];
597 	vm_page_t m;
598 	struct ptbl_buf *pbuf;
599 	unsigned int pidx;
600 	pte_t *ptbl;
601 	int i, j;
602 
603 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
604 	    (pmap == kernel_pmap), pdir_idx);
605 
606 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
607 	    ("ptbl_alloc: invalid pdir_idx"));
608 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
609 	    ("pte_alloc: valid ptbl entry exists!"));
610 
611 	pbuf = ptbl_buf_alloc();
612 	if (pbuf == NULL)
613 		panic("pte_alloc: couldn't alloc kernel virtual memory");
614 
615 	ptbl = (pte_t *)pbuf->kva;
616 
617 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
618 
619 	/* Allocate ptbl pages, this will sleep! */
620 	for (i = 0; i < PTBL_PAGES; i++) {
621 		pidx = (PTBL_PAGES * pdir_idx) + i;
622 		while ((m = vm_page_alloc(NULL, pidx,
623 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
624 			PMAP_UNLOCK(pmap);
625 			rw_wunlock(&pvh_global_lock);
626 			if (nosleep) {
627 				ptbl_free_pmap_ptbl(pmap, ptbl);
628 				for (j = 0; j < i; j++)
629 					vm_page_free(mtbl[j]);
630 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
631 				return (NULL);
632 			}
633 			VM_WAIT;
634 			rw_wlock(&pvh_global_lock);
635 			PMAP_LOCK(pmap);
636 		}
637 		mtbl[i] = m;
638 	}
639 
640 	/* Map allocated pages into kernel_pmap. */
641 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
642 
643 	/* Zero whole ptbl. */
644 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
645 
646 	/* Add pbuf to the pmap ptbl bufs list. */
647 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
648 
649 	return (ptbl);
650 }
651 
652 /* Free ptbl pages and invalidate pdir entry. */
653 static void
654 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
655 {
656 	pte_t *ptbl;
657 	vm_paddr_t pa;
658 	vm_offset_t va;
659 	vm_page_t m;
660 	int i;
661 
662 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
663 	    (pmap == kernel_pmap), pdir_idx);
664 
665 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
666 	    ("ptbl_free: invalid pdir_idx"));
667 
668 	ptbl = pmap->pm_pdir[pdir_idx];
669 
670 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
671 
672 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
673 
674 	/*
675 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
676 	 * don't attempt to look up the page tables we are releasing.
677 	 */
678 	mtx_lock_spin(&tlbivax_mutex);
679 	tlb_miss_lock();
680 
681 	pmap->pm_pdir[pdir_idx] = NULL;
682 
683 	tlb_miss_unlock();
684 	mtx_unlock_spin(&tlbivax_mutex);
685 
686 	for (i = 0; i < PTBL_PAGES; i++) {
687 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
688 		pa = pte_vatopa(mmu, kernel_pmap, va);
689 		m = PHYS_TO_VM_PAGE(pa);
690 		vm_page_free_zero(m);
691 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
692 		mmu_booke_kremove(mmu, va);
693 	}
694 
695 	ptbl_free_pmap_ptbl(pmap, ptbl);
696 }
697 
698 /*
699  * Decrement ptbl pages hold count and attempt to free ptbl pages.
700  * Called when removing pte entry from ptbl.
701  *
702  * Return 1 if ptbl pages were freed.
703  */
704 static int
705 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
706 {
707 	pte_t *ptbl;
708 	vm_paddr_t pa;
709 	vm_page_t m;
710 	int i;
711 
712 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
713 	    (pmap == kernel_pmap), pdir_idx);
714 
715 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
716 	    ("ptbl_unhold: invalid pdir_idx"));
717 	KASSERT((pmap != kernel_pmap),
718 	    ("ptbl_unhold: unholding kernel ptbl!"));
719 
720 	ptbl = pmap->pm_pdir[pdir_idx];
721 
722 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
723 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
724 	    ("ptbl_unhold: non kva ptbl"));
725 
726 	/* decrement hold count */
727 	for (i = 0; i < PTBL_PAGES; i++) {
728 		pa = pte_vatopa(mmu, kernel_pmap,
729 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
730 		m = PHYS_TO_VM_PAGE(pa);
731 		m->wire_count--;
732 	}
733 
734 	/*
735 	 * Free ptbl pages if there are no pte etries in this ptbl.
736 	 * wire_count has the same value for all ptbl pages, so check the last
737 	 * page.
738 	 */
739 	if (m->wire_count == 0) {
740 		ptbl_free(mmu, pmap, pdir_idx);
741 
742 		//debugf("ptbl_unhold: e (freed ptbl)\n");
743 		return (1);
744 	}
745 
746 	return (0);
747 }
748 
749 /*
750  * Increment hold count for ptbl pages. This routine is used when a new pte
751  * entry is being inserted into the ptbl.
752  */
753 static void
754 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
755 {
756 	vm_paddr_t pa;
757 	pte_t *ptbl;
758 	vm_page_t m;
759 	int i;
760 
761 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
762 	    pdir_idx);
763 
764 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
765 	    ("ptbl_hold: invalid pdir_idx"));
766 	KASSERT((pmap != kernel_pmap),
767 	    ("ptbl_hold: holding kernel ptbl!"));
768 
769 	ptbl = pmap->pm_pdir[pdir_idx];
770 
771 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
772 
773 	for (i = 0; i < PTBL_PAGES; i++) {
774 		pa = pte_vatopa(mmu, kernel_pmap,
775 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
776 		m = PHYS_TO_VM_PAGE(pa);
777 		m->wire_count++;
778 	}
779 }
780 
781 /* Allocate pv_entry structure. */
782 pv_entry_t
783 pv_alloc(void)
784 {
785 	pv_entry_t pv;
786 
787 	pv_entry_count++;
788 	if (pv_entry_count > pv_entry_high_water)
789 		pagedaemon_wakeup();
790 	pv = uma_zalloc(pvzone, M_NOWAIT);
791 
792 	return (pv);
793 }
794 
795 /* Free pv_entry structure. */
796 static __inline void
797 pv_free(pv_entry_t pve)
798 {
799 
800 	pv_entry_count--;
801 	uma_zfree(pvzone, pve);
802 }
803 
804 
805 /* Allocate and initialize pv_entry structure. */
806 static void
807 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
808 {
809 	pv_entry_t pve;
810 
811 	//int su = (pmap == kernel_pmap);
812 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
813 	//	(u_int32_t)pmap, va, (u_int32_t)m);
814 
815 	pve = pv_alloc();
816 	if (pve == NULL)
817 		panic("pv_insert: no pv entries!");
818 
819 	pve->pv_pmap = pmap;
820 	pve->pv_va = va;
821 
822 	/* add to pv_list */
823 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
824 	rw_assert(&pvh_global_lock, RA_WLOCKED);
825 
826 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
827 
828 	//debugf("pv_insert: e\n");
829 }
830 
831 /* Destroy pv entry. */
832 static void
833 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
834 {
835 	pv_entry_t pve;
836 
837 	//int su = (pmap == kernel_pmap);
838 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
839 
840 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
841 	rw_assert(&pvh_global_lock, RA_WLOCKED);
842 
843 	/* find pv entry */
844 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
845 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
846 			/* remove from pv_list */
847 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
848 			if (TAILQ_EMPTY(&m->md.pv_list))
849 				vm_page_aflag_clear(m, PGA_WRITEABLE);
850 
851 			/* free pv entry struct */
852 			pv_free(pve);
853 			break;
854 		}
855 	}
856 
857 	//debugf("pv_remove: e\n");
858 }
859 
860 /*
861  * Clean pte entry, try to free page table page if requested.
862  *
863  * Return 1 if ptbl pages were freed, otherwise return 0.
864  */
865 static int
866 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
867 {
868 	unsigned int pdir_idx = PDIR_IDX(va);
869 	unsigned int ptbl_idx = PTBL_IDX(va);
870 	vm_page_t m;
871 	pte_t *ptbl;
872 	pte_t *pte;
873 
874 	//int su = (pmap == kernel_pmap);
875 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
876 	//		su, (u_int32_t)pmap, va, flags);
877 
878 	ptbl = pmap->pm_pdir[pdir_idx];
879 	KASSERT(ptbl, ("pte_remove: null ptbl"));
880 
881 	pte = &ptbl[ptbl_idx];
882 
883 	if (pte == NULL || !PTE_ISVALID(pte))
884 		return (0);
885 
886 	if (PTE_ISWIRED(pte))
887 		pmap->pm_stats.wired_count--;
888 
889 	/* Handle managed entry. */
890 	if (PTE_ISMANAGED(pte)) {
891 		/* Get vm_page_t for mapped pte. */
892 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
893 
894 		if (PTE_ISMODIFIED(pte))
895 			vm_page_dirty(m);
896 
897 		if (PTE_ISREFERENCED(pte))
898 			vm_page_aflag_set(m, PGA_REFERENCED);
899 
900 		pv_remove(pmap, va, m);
901 	}
902 
903 	mtx_lock_spin(&tlbivax_mutex);
904 	tlb_miss_lock();
905 
906 	tlb0_flush_entry(va);
907 	*pte = 0;
908 
909 	tlb_miss_unlock();
910 	mtx_unlock_spin(&tlbivax_mutex);
911 
912 	pmap->pm_stats.resident_count--;
913 
914 	if (flags & PTBL_UNHOLD) {
915 		//debugf("pte_remove: e (unhold)\n");
916 		return (ptbl_unhold(mmu, pmap, pdir_idx));
917 	}
918 
919 	//debugf("pte_remove: e\n");
920 	return (0);
921 }
922 
923 /*
924  * Insert PTE for a given page and virtual address.
925  */
926 static int
927 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
928     boolean_t nosleep)
929 {
930 	unsigned int pdir_idx = PDIR_IDX(va);
931 	unsigned int ptbl_idx = PTBL_IDX(va);
932 	pte_t *ptbl, *pte;
933 
934 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
935 	    pmap == kernel_pmap, pmap, va);
936 
937 	/* Get the page table pointer. */
938 	ptbl = pmap->pm_pdir[pdir_idx];
939 
940 	if (ptbl == NULL) {
941 		/* Allocate page table pages. */
942 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
943 		if (ptbl == NULL) {
944 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
945 			return (ENOMEM);
946 		}
947 	} else {
948 		/*
949 		 * Check if there is valid mapping for requested
950 		 * va, if there is, remove it.
951 		 */
952 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
953 		if (PTE_ISVALID(pte)) {
954 			pte_remove(mmu, pmap, va, PTBL_HOLD);
955 		} else {
956 			/*
957 			 * pte is not used, increment hold count
958 			 * for ptbl pages.
959 			 */
960 			if (pmap != kernel_pmap)
961 				ptbl_hold(mmu, pmap, pdir_idx);
962 		}
963 	}
964 
965 	/*
966 	 * Insert pv_entry into pv_list for mapped page if part of managed
967 	 * memory.
968 	 */
969 	if ((m->oflags & VPO_UNMANAGED) == 0) {
970 		flags |= PTE_MANAGED;
971 
972 		/* Create and insert pv entry. */
973 		pv_insert(pmap, va, m);
974 	}
975 
976 	pmap->pm_stats.resident_count++;
977 
978 	mtx_lock_spin(&tlbivax_mutex);
979 	tlb_miss_lock();
980 
981 	tlb0_flush_entry(va);
982 	if (pmap->pm_pdir[pdir_idx] == NULL) {
983 		/*
984 		 * If we just allocated a new page table, hook it in
985 		 * the pdir.
986 		 */
987 		pmap->pm_pdir[pdir_idx] = ptbl;
988 	}
989 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
990 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
991 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
992 
993 	tlb_miss_unlock();
994 	mtx_unlock_spin(&tlbivax_mutex);
995 	return (0);
996 }
997 
998 /* Return the pa for the given pmap/va. */
999 static vm_paddr_t
1000 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1001 {
1002 	vm_paddr_t pa = 0;
1003 	pte_t *pte;
1004 
1005 	pte = pte_find(mmu, pmap, va);
1006 	if ((pte != NULL) && PTE_ISVALID(pte))
1007 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1008 	return (pa);
1009 }
1010 
1011 /* Get a pointer to a PTE in a page table. */
1012 static pte_t *
1013 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1014 {
1015 	unsigned int pdir_idx = PDIR_IDX(va);
1016 	unsigned int ptbl_idx = PTBL_IDX(va);
1017 
1018 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1019 
1020 	if (pmap->pm_pdir[pdir_idx])
1021 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1022 
1023 	return (NULL);
1024 }
1025 
1026 /* Set up kernel page tables. */
1027 static void
1028 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1029 {
1030 	int		i;
1031 	vm_offset_t	va;
1032 	pte_t		*pte;
1033 
1034 	/* Initialize kernel pdir */
1035 	for (i = 0; i < kernel_ptbls; i++)
1036 		kernel_pmap->pm_pdir[kptbl_min + i] =
1037 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1038 
1039 	/*
1040 	 * Fill in PTEs covering kernel code and data. They are not required
1041 	 * for address translation, as this area is covered by static TLB1
1042 	 * entries, but for pte_vatopa() to work correctly with kernel area
1043 	 * addresses.
1044 	 */
1045 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1046 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1047 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1048 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1049 		    PTE_VALID | PTE_PS_4KB;
1050 	}
1051 }
1052 
1053 /**************************************************************************/
1054 /* PMAP related */
1055 /**************************************************************************/
1056 
1057 /*
1058  * This is called during booke_init, before the system is really initialized.
1059  */
1060 static void
1061 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1062 {
1063 	vm_paddr_t phys_kernelend;
1064 	struct mem_region *mp, *mp1;
1065 	int cnt, i, j;
1066 	vm_paddr_t s, e, sz;
1067 	vm_paddr_t physsz, hwphyssz;
1068 	u_int phys_avail_count;
1069 	vm_size_t kstack0_sz;
1070 	vm_offset_t kernel_pdir, kstack0;
1071 	vm_paddr_t kstack0_phys;
1072 	void *dpcpu;
1073 
1074 	debugf("mmu_booke_bootstrap: entered\n");
1075 
1076 	/* Set interesting system properties */
1077 	hw_direct_map = 0;
1078 	elf32_nxstack = 1;
1079 
1080 	/* Initialize invalidation mutex */
1081 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1082 
1083 	/* Read TLB0 size and associativity. */
1084 	tlb0_get_tlbconf();
1085 
1086 	/*
1087 	 * Align kernel start and end address (kernel image).
1088 	 * Note that kernel end does not necessarily relate to kernsize.
1089 	 * kernsize is the size of the kernel that is actually mapped.
1090 	 */
1091 	kernstart = trunc_page(start);
1092 	data_start = round_page(kernelend);
1093 	data_end = data_start;
1094 
1095 	/*
1096 	 * Addresses of preloaded modules (like file systems) use
1097 	 * physical addresses. Make sure we relocate those into
1098 	 * virtual addresses.
1099 	 */
1100 	preload_addr_relocate = kernstart - kernload;
1101 
1102 	/* Allocate the dynamic per-cpu area. */
1103 	dpcpu = (void *)data_end;
1104 	data_end += DPCPU_SIZE;
1105 
1106 	/* Allocate space for the message buffer. */
1107 	msgbufp = (struct msgbuf *)data_end;
1108 	data_end += msgbufsize;
1109 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1110 	    data_end);
1111 
1112 	data_end = round_page(data_end);
1113 
1114 	/* Allocate space for ptbl_bufs. */
1115 	ptbl_bufs = (struct ptbl_buf *)data_end;
1116 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1117 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1118 	    data_end);
1119 
1120 	data_end = round_page(data_end);
1121 
1122 	/* Allocate PTE tables for kernel KVA. */
1123 	kernel_pdir = data_end;
1124 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1125 	    PDIR_SIZE - 1) / PDIR_SIZE;
1126 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1127 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1128 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1129 
1130 	debugf(" data_end: 0x%08x\n", data_end);
1131 	if (data_end - kernstart > kernsize) {
1132 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1133 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1134 	}
1135 	data_end = kernstart + kernsize;
1136 	debugf(" updated data_end: 0x%08x\n", data_end);
1137 
1138 	/*
1139 	 * Clear the structures - note we can only do it safely after the
1140 	 * possible additional TLB1 translations are in place (above) so that
1141 	 * all range up to the currently calculated 'data_end' is covered.
1142 	 */
1143 	dpcpu_init(dpcpu, 0);
1144 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1145 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1146 
1147 	/*******************************************************/
1148 	/* Set the start and end of kva. */
1149 	/*******************************************************/
1150 	virtual_avail = round_page(data_end);
1151 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1152 
1153 	/* Allocate KVA space for page zero/copy operations. */
1154 	zero_page_va = virtual_avail;
1155 	virtual_avail += PAGE_SIZE;
1156 	zero_page_idle_va = virtual_avail;
1157 	virtual_avail += PAGE_SIZE;
1158 	copy_page_src_va = virtual_avail;
1159 	virtual_avail += PAGE_SIZE;
1160 	copy_page_dst_va = virtual_avail;
1161 	virtual_avail += PAGE_SIZE;
1162 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1163 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1164 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1165 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1166 
1167 	/* Initialize page zero/copy mutexes. */
1168 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1169 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1170 
1171 	/* Allocate KVA space for ptbl bufs. */
1172 	ptbl_buf_pool_vabase = virtual_avail;
1173 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1174 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1175 	    ptbl_buf_pool_vabase, virtual_avail);
1176 
1177 	/* Calculate corresponding physical addresses for the kernel region. */
1178 	phys_kernelend = kernload + kernsize;
1179 	debugf("kernel image and allocated data:\n");
1180 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1181 	debugf(" kernstart   = 0x%08x\n", kernstart);
1182 	debugf(" kernsize    = 0x%08x\n", kernsize);
1183 
1184 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1185 		panic("mmu_booke_bootstrap: phys_avail too small");
1186 
1187 	/*
1188 	 * Remove kernel physical address range from avail regions list. Page
1189 	 * align all regions.  Non-page aligned memory isn't very interesting
1190 	 * to us.  Also, sort the entries for ascending addresses.
1191 	 */
1192 
1193 	/* Retrieve phys/avail mem regions */
1194 	mem_regions(&physmem_regions, &physmem_regions_sz,
1195 	    &availmem_regions, &availmem_regions_sz);
1196 	sz = 0;
1197 	cnt = availmem_regions_sz;
1198 	debugf("processing avail regions:\n");
1199 	for (mp = availmem_regions; mp->mr_size; mp++) {
1200 		s = mp->mr_start;
1201 		e = mp->mr_start + mp->mr_size;
1202 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1203 		/* Check whether this region holds all of the kernel. */
1204 		if (s < kernload && e > phys_kernelend) {
1205 			availmem_regions[cnt].mr_start = phys_kernelend;
1206 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1207 			e = kernload;
1208 		}
1209 		/* Look whether this regions starts within the kernel. */
1210 		if (s >= kernload && s < phys_kernelend) {
1211 			if (e <= phys_kernelend)
1212 				goto empty;
1213 			s = phys_kernelend;
1214 		}
1215 		/* Now look whether this region ends within the kernel. */
1216 		if (e > kernload && e <= phys_kernelend) {
1217 			if (s >= kernload)
1218 				goto empty;
1219 			e = kernload;
1220 		}
1221 		/* Now page align the start and size of the region. */
1222 		s = round_page(s);
1223 		e = trunc_page(e);
1224 		if (e < s)
1225 			e = s;
1226 		sz = e - s;
1227 		debugf("%09jx-%09jx = %jx\n",
1228 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1229 
1230 		/* Check whether some memory is left here. */
1231 		if (sz == 0) {
1232 		empty:
1233 			memmove(mp, mp + 1,
1234 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1235 			cnt--;
1236 			mp--;
1237 			continue;
1238 		}
1239 
1240 		/* Do an insertion sort. */
1241 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1242 			if (s < mp1->mr_start)
1243 				break;
1244 		if (mp1 < mp) {
1245 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1246 			mp1->mr_start = s;
1247 			mp1->mr_size = sz;
1248 		} else {
1249 			mp->mr_start = s;
1250 			mp->mr_size = sz;
1251 		}
1252 	}
1253 	availmem_regions_sz = cnt;
1254 
1255 	/*******************************************************/
1256 	/* Steal physical memory for kernel stack from the end */
1257 	/* of the first avail region                           */
1258 	/*******************************************************/
1259 	kstack0_sz = kstack_pages * PAGE_SIZE;
1260 	kstack0_phys = availmem_regions[0].mr_start +
1261 	    availmem_regions[0].mr_size;
1262 	kstack0_phys -= kstack0_sz;
1263 	availmem_regions[0].mr_size -= kstack0_sz;
1264 
1265 	/*******************************************************/
1266 	/* Fill in phys_avail table, based on availmem_regions */
1267 	/*******************************************************/
1268 	phys_avail_count = 0;
1269 	physsz = 0;
1270 	hwphyssz = 0;
1271 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1272 
1273 	debugf("fill in phys_avail:\n");
1274 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1275 
1276 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1277 		    (uintmax_t)availmem_regions[i].mr_start,
1278 		    (uintmax_t)availmem_regions[i].mr_start +
1279 		        availmem_regions[i].mr_size,
1280 		    (uintmax_t)availmem_regions[i].mr_size);
1281 
1282 		if (hwphyssz != 0 &&
1283 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1284 			debugf(" hw.physmem adjust\n");
1285 			if (physsz < hwphyssz) {
1286 				phys_avail[j] = availmem_regions[i].mr_start;
1287 				phys_avail[j + 1] =
1288 				    availmem_regions[i].mr_start +
1289 				    hwphyssz - physsz;
1290 				physsz = hwphyssz;
1291 				phys_avail_count++;
1292 			}
1293 			break;
1294 		}
1295 
1296 		phys_avail[j] = availmem_regions[i].mr_start;
1297 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1298 		    availmem_regions[i].mr_size;
1299 		phys_avail_count++;
1300 		physsz += availmem_regions[i].mr_size;
1301 	}
1302 	physmem = btoc(physsz);
1303 
1304 	/* Calculate the last available physical address. */
1305 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1306 		;
1307 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1308 
1309 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1310 	debugf("phys_avail_count = %d\n", phys_avail_count);
1311 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1312 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1313 
1314 	/*******************************************************/
1315 	/* Initialize (statically allocated) kernel pmap. */
1316 	/*******************************************************/
1317 	PMAP_LOCK_INIT(kernel_pmap);
1318 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1319 
1320 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1321 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1322 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1323 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1324 
1325 	kernel_pte_alloc(data_end, kernstart, kernel_pdir);
1326 	for (i = 0; i < MAXCPU; i++) {
1327 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1328 
1329 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1330 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1331 	}
1332 
1333 	/* Mark kernel_pmap active on all CPUs */
1334 	CPU_FILL(&kernel_pmap->pm_active);
1335 
1336  	/*
1337 	 * Initialize the global pv list lock.
1338 	 */
1339 	rw_init(&pvh_global_lock, "pmap pv global");
1340 
1341 	/*******************************************************/
1342 	/* Final setup */
1343 	/*******************************************************/
1344 
1345 	/* Enter kstack0 into kernel map, provide guard page */
1346 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1347 	thread0.td_kstack = kstack0;
1348 	thread0.td_kstack_pages = kstack_pages;
1349 
1350 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1351 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1352 	    kstack0_phys, kstack0_phys + kstack0_sz);
1353 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1354 
1355 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1356 	for (i = 0; i < kstack_pages; i++) {
1357 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1358 		kstack0 += PAGE_SIZE;
1359 		kstack0_phys += PAGE_SIZE;
1360 	}
1361 
1362 	pmap_bootstrapped = 1;
1363 
1364 	debugf("virtual_avail = %08x\n", virtual_avail);
1365 	debugf("virtual_end   = %08x\n", virtual_end);
1366 
1367 	debugf("mmu_booke_bootstrap: exit\n");
1368 }
1369 
1370 #ifdef SMP
1371 void
1372 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1373 {
1374 	int i;
1375 
1376 	/*
1377 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1378 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1379 	 * these values directly to (re)program AP's TLB1 hardware.
1380 	 */
1381 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1382 		/* Skip invalid entries */
1383 		if (!(tlb1[i].mas1 & MAS1_VALID))
1384 			continue;
1385 
1386 		tlb1_write_entry(i);
1387 	}
1388 
1389 	set_mas4_defaults();
1390 }
1391 #endif
1392 
1393 static void
1394 booke_pmap_init_qpages(void)
1395 {
1396 	struct pcpu *pc;
1397 	int i;
1398 
1399 	CPU_FOREACH(i) {
1400 		pc = pcpu_find(i);
1401 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1402 		if (pc->pc_qmap_addr == 0)
1403 			panic("pmap_init_qpages: unable to allocate KVA");
1404 	}
1405 }
1406 
1407 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1408 
1409 /*
1410  * Get the physical page address for the given pmap/virtual address.
1411  */
1412 static vm_paddr_t
1413 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1414 {
1415 	vm_paddr_t pa;
1416 
1417 	PMAP_LOCK(pmap);
1418 	pa = pte_vatopa(mmu, pmap, va);
1419 	PMAP_UNLOCK(pmap);
1420 
1421 	return (pa);
1422 }
1423 
1424 /*
1425  * Extract the physical page address associated with the given
1426  * kernel virtual address.
1427  */
1428 static vm_paddr_t
1429 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1430 {
1431 	int i;
1432 
1433 	/* Check TLB1 mappings */
1434 	for (i = 0; i < tlb1_idx; i++) {
1435 		if (!(tlb1[i].mas1 & MAS1_VALID))
1436 			continue;
1437 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1438 			return (tlb1[i].phys + (va - tlb1[i].virt));
1439 	}
1440 
1441 	return (pte_vatopa(mmu, kernel_pmap, va));
1442 }
1443 
1444 /*
1445  * Initialize the pmap module.
1446  * Called by vm_init, to initialize any structures that the pmap
1447  * system needs to map virtual memory.
1448  */
1449 static void
1450 mmu_booke_init(mmu_t mmu)
1451 {
1452 	int shpgperproc = PMAP_SHPGPERPROC;
1453 
1454 	/*
1455 	 * Initialize the address space (zone) for the pv entries.  Set a
1456 	 * high water mark so that the system can recover from excessive
1457 	 * numbers of pv entries.
1458 	 */
1459 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1460 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1461 
1462 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1463 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1464 
1465 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1466 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1467 
1468 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1469 
1470 	/* Pre-fill pvzone with initial number of pv entries. */
1471 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1472 
1473 	/* Initialize ptbl allocation. */
1474 	ptbl_init();
1475 }
1476 
1477 /*
1478  * Map a list of wired pages into kernel virtual address space.  This is
1479  * intended for temporary mappings which do not need page modification or
1480  * references recorded.  Existing mappings in the region are overwritten.
1481  */
1482 static void
1483 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1484 {
1485 	vm_offset_t va;
1486 
1487 	va = sva;
1488 	while (count-- > 0) {
1489 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1490 		va += PAGE_SIZE;
1491 		m++;
1492 	}
1493 }
1494 
1495 /*
1496  * Remove page mappings from kernel virtual address space.  Intended for
1497  * temporary mappings entered by mmu_booke_qenter.
1498  */
1499 static void
1500 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1501 {
1502 	vm_offset_t va;
1503 
1504 	va = sva;
1505 	while (count-- > 0) {
1506 		mmu_booke_kremove(mmu, va);
1507 		va += PAGE_SIZE;
1508 	}
1509 }
1510 
1511 /*
1512  * Map a wired page into kernel virtual address space.
1513  */
1514 static void
1515 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1516 {
1517 
1518 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1519 }
1520 
1521 static void
1522 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1523 {
1524 	uint32_t flags;
1525 	pte_t *pte;
1526 
1527 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1528 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1529 
1530 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1531 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1532 	flags |= PTE_PS_4KB;
1533 
1534 	pte = pte_find(mmu, kernel_pmap, va);
1535 
1536 	mtx_lock_spin(&tlbivax_mutex);
1537 	tlb_miss_lock();
1538 
1539 	if (PTE_ISVALID(pte)) {
1540 
1541 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1542 
1543 		/* Flush entry from TLB0 */
1544 		tlb0_flush_entry(va);
1545 	}
1546 
1547 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1548 
1549 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1550 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1551 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1552 
1553 	/* Flush the real memory from the instruction cache. */
1554 	if ((flags & (PTE_I | PTE_G)) == 0)
1555 		__syncicache((void *)va, PAGE_SIZE);
1556 
1557 	tlb_miss_unlock();
1558 	mtx_unlock_spin(&tlbivax_mutex);
1559 }
1560 
1561 /*
1562  * Remove a page from kernel page table.
1563  */
1564 static void
1565 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1566 {
1567 	pte_t *pte;
1568 
1569 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
1570 
1571 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1572 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1573 	    ("mmu_booke_kremove: invalid va"));
1574 
1575 	pte = pte_find(mmu, kernel_pmap, va);
1576 
1577 	if (!PTE_ISVALID(pte)) {
1578 
1579 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1580 
1581 		return;
1582 	}
1583 
1584 	mtx_lock_spin(&tlbivax_mutex);
1585 	tlb_miss_lock();
1586 
1587 	/* Invalidate entry in TLB0, update PTE. */
1588 	tlb0_flush_entry(va);
1589 	*pte = 0;
1590 
1591 	tlb_miss_unlock();
1592 	mtx_unlock_spin(&tlbivax_mutex);
1593 }
1594 
1595 /*
1596  * Initialize pmap associated with process 0.
1597  */
1598 static void
1599 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1600 {
1601 
1602 	PMAP_LOCK_INIT(pmap);
1603 	mmu_booke_pinit(mmu, pmap);
1604 	PCPU_SET(curpmap, pmap);
1605 }
1606 
1607 /*
1608  * Initialize a preallocated and zeroed pmap structure,
1609  * such as one in a vmspace structure.
1610  */
1611 static void
1612 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1613 {
1614 	int i;
1615 
1616 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1617 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1618 
1619 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1620 
1621 	for (i = 0; i < MAXCPU; i++)
1622 		pmap->pm_tid[i] = TID_NONE;
1623 	CPU_ZERO(&kernel_pmap->pm_active);
1624 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1625 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1626 	TAILQ_INIT(&pmap->pm_ptbl_list);
1627 }
1628 
1629 /*
1630  * Release any resources held by the given physical map.
1631  * Called when a pmap initialized by mmu_booke_pinit is being released.
1632  * Should only be called if the map contains no valid mappings.
1633  */
1634 static void
1635 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1636 {
1637 
1638 	KASSERT(pmap->pm_stats.resident_count == 0,
1639 	    ("pmap_release: pmap resident count %ld != 0",
1640 	    pmap->pm_stats.resident_count));
1641 }
1642 
1643 /*
1644  * Insert the given physical page at the specified virtual address in the
1645  * target physical map with the protection requested. If specified the page
1646  * will be wired down.
1647  */
1648 static int
1649 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1650     vm_prot_t prot, u_int flags, int8_t psind)
1651 {
1652 	int error;
1653 
1654 	rw_wlock(&pvh_global_lock);
1655 	PMAP_LOCK(pmap);
1656 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1657 	rw_wunlock(&pvh_global_lock);
1658 	PMAP_UNLOCK(pmap);
1659 	return (error);
1660 }
1661 
1662 static int
1663 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1664     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1665 {
1666 	pte_t *pte;
1667 	vm_paddr_t pa;
1668 	uint32_t flags;
1669 	int error, su, sync;
1670 
1671 	pa = VM_PAGE_TO_PHYS(m);
1672 	su = (pmap == kernel_pmap);
1673 	sync = 0;
1674 
1675 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1676 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1677 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1678 	//		(u_int32_t)m, va, pa, prot, flags);
1679 
1680 	if (su) {
1681 		KASSERT(((va >= virtual_avail) &&
1682 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1683 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1684 	} else {
1685 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1686 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1687 	}
1688 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1689 		VM_OBJECT_ASSERT_LOCKED(m->object);
1690 
1691 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1692 
1693 	/*
1694 	 * If there is an existing mapping, and the physical address has not
1695 	 * changed, must be protection or wiring change.
1696 	 */
1697 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1698 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1699 
1700 		/*
1701 		 * Before actually updating pte->flags we calculate and
1702 		 * prepare its new value in a helper var.
1703 		 */
1704 		flags = *pte;
1705 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1706 
1707 		/* Wiring change, just update stats. */
1708 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1709 			if (!PTE_ISWIRED(pte)) {
1710 				flags |= PTE_WIRED;
1711 				pmap->pm_stats.wired_count++;
1712 			}
1713 		} else {
1714 			if (PTE_ISWIRED(pte)) {
1715 				flags &= ~PTE_WIRED;
1716 				pmap->pm_stats.wired_count--;
1717 			}
1718 		}
1719 
1720 		if (prot & VM_PROT_WRITE) {
1721 			/* Add write permissions. */
1722 			flags |= PTE_SW;
1723 			if (!su)
1724 				flags |= PTE_UW;
1725 
1726 			if ((flags & PTE_MANAGED) != 0)
1727 				vm_page_aflag_set(m, PGA_WRITEABLE);
1728 		} else {
1729 			/* Handle modified pages, sense modify status. */
1730 
1731 			/*
1732 			 * The PTE_MODIFIED flag could be set by underlying
1733 			 * TLB misses since we last read it (above), possibly
1734 			 * other CPUs could update it so we check in the PTE
1735 			 * directly rather than rely on that saved local flags
1736 			 * copy.
1737 			 */
1738 			if (PTE_ISMODIFIED(pte))
1739 				vm_page_dirty(m);
1740 		}
1741 
1742 		if (prot & VM_PROT_EXECUTE) {
1743 			flags |= PTE_SX;
1744 			if (!su)
1745 				flags |= PTE_UX;
1746 
1747 			/*
1748 			 * Check existing flags for execute permissions: if we
1749 			 * are turning execute permissions on, icache should
1750 			 * be flushed.
1751 			 */
1752 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1753 				sync++;
1754 		}
1755 
1756 		flags &= ~PTE_REFERENCED;
1757 
1758 		/*
1759 		 * The new flags value is all calculated -- only now actually
1760 		 * update the PTE.
1761 		 */
1762 		mtx_lock_spin(&tlbivax_mutex);
1763 		tlb_miss_lock();
1764 
1765 		tlb0_flush_entry(va);
1766 		*pte &= ~PTE_FLAGS_MASK;
1767 		*pte |= flags;
1768 
1769 		tlb_miss_unlock();
1770 		mtx_unlock_spin(&tlbivax_mutex);
1771 
1772 	} else {
1773 		/*
1774 		 * If there is an existing mapping, but it's for a different
1775 		 * physical address, pte_enter() will delete the old mapping.
1776 		 */
1777 		//if ((pte != NULL) && PTE_ISVALID(pte))
1778 		//	debugf("mmu_booke_enter_locked: replace\n");
1779 		//else
1780 		//	debugf("mmu_booke_enter_locked: new\n");
1781 
1782 		/* Now set up the flags and install the new mapping. */
1783 		flags = (PTE_SR | PTE_VALID);
1784 		flags |= PTE_M;
1785 
1786 		if (!su)
1787 			flags |= PTE_UR;
1788 
1789 		if (prot & VM_PROT_WRITE) {
1790 			flags |= PTE_SW;
1791 			if (!su)
1792 				flags |= PTE_UW;
1793 
1794 			if ((m->oflags & VPO_UNMANAGED) == 0)
1795 				vm_page_aflag_set(m, PGA_WRITEABLE);
1796 		}
1797 
1798 		if (prot & VM_PROT_EXECUTE) {
1799 			flags |= PTE_SX;
1800 			if (!su)
1801 				flags |= PTE_UX;
1802 		}
1803 
1804 		/* If its wired update stats. */
1805 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1806 			flags |= PTE_WIRED;
1807 
1808 		error = pte_enter(mmu, pmap, m, va, flags,
1809 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1810 		if (error != 0)
1811 			return (KERN_RESOURCE_SHORTAGE);
1812 
1813 		if ((flags & PMAP_ENTER_WIRED) != 0)
1814 			pmap->pm_stats.wired_count++;
1815 
1816 		/* Flush the real memory from the instruction cache. */
1817 		if (prot & VM_PROT_EXECUTE)
1818 			sync++;
1819 	}
1820 
1821 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1822 		__syncicache((void *)va, PAGE_SIZE);
1823 		sync = 0;
1824 	}
1825 
1826 	return (KERN_SUCCESS);
1827 }
1828 
1829 /*
1830  * Maps a sequence of resident pages belonging to the same object.
1831  * The sequence begins with the given page m_start.  This page is
1832  * mapped at the given virtual address start.  Each subsequent page is
1833  * mapped at a virtual address that is offset from start by the same
1834  * amount as the page is offset from m_start within the object.  The
1835  * last page in the sequence is the page with the largest offset from
1836  * m_start that can be mapped at a virtual address less than the given
1837  * virtual address end.  Not every virtual page between start and end
1838  * is mapped; only those for which a resident page exists with the
1839  * corresponding offset from m_start are mapped.
1840  */
1841 static void
1842 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1843     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1844 {
1845 	vm_page_t m;
1846 	vm_pindex_t diff, psize;
1847 
1848 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1849 
1850 	psize = atop(end - start);
1851 	m = m_start;
1852 	rw_wlock(&pvh_global_lock);
1853 	PMAP_LOCK(pmap);
1854 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1855 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1856 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1857 		    PMAP_ENTER_NOSLEEP, 0);
1858 		m = TAILQ_NEXT(m, listq);
1859 	}
1860 	rw_wunlock(&pvh_global_lock);
1861 	PMAP_UNLOCK(pmap);
1862 }
1863 
1864 static void
1865 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1866     vm_prot_t prot)
1867 {
1868 
1869 	rw_wlock(&pvh_global_lock);
1870 	PMAP_LOCK(pmap);
1871 	mmu_booke_enter_locked(mmu, pmap, va, m,
1872 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1873 	    0);
1874 	rw_wunlock(&pvh_global_lock);
1875 	PMAP_UNLOCK(pmap);
1876 }
1877 
1878 /*
1879  * Remove the given range of addresses from the specified map.
1880  *
1881  * It is assumed that the start and end are properly rounded to the page size.
1882  */
1883 static void
1884 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1885 {
1886 	pte_t *pte;
1887 	uint8_t hold_flag;
1888 
1889 	int su = (pmap == kernel_pmap);
1890 
1891 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1892 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1893 
1894 	if (su) {
1895 		KASSERT(((va >= virtual_avail) &&
1896 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1897 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1898 	} else {
1899 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1900 		    ("mmu_booke_remove: user pmap, non user va"));
1901 	}
1902 
1903 	if (PMAP_REMOVE_DONE(pmap)) {
1904 		//debugf("mmu_booke_remove: e (empty)\n");
1905 		return;
1906 	}
1907 
1908 	hold_flag = PTBL_HOLD_FLAG(pmap);
1909 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1910 
1911 	rw_wlock(&pvh_global_lock);
1912 	PMAP_LOCK(pmap);
1913 	for (; va < endva; va += PAGE_SIZE) {
1914 		pte = pte_find(mmu, pmap, va);
1915 		if ((pte != NULL) && PTE_ISVALID(pte))
1916 			pte_remove(mmu, pmap, va, hold_flag);
1917 	}
1918 	PMAP_UNLOCK(pmap);
1919 	rw_wunlock(&pvh_global_lock);
1920 
1921 	//debugf("mmu_booke_remove: e\n");
1922 }
1923 
1924 /*
1925  * Remove physical page from all pmaps in which it resides.
1926  */
1927 static void
1928 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1929 {
1930 	pv_entry_t pv, pvn;
1931 	uint8_t hold_flag;
1932 
1933 	rw_wlock(&pvh_global_lock);
1934 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1935 		pvn = TAILQ_NEXT(pv, pv_link);
1936 
1937 		PMAP_LOCK(pv->pv_pmap);
1938 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1939 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1940 		PMAP_UNLOCK(pv->pv_pmap);
1941 	}
1942 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1943 	rw_wunlock(&pvh_global_lock);
1944 }
1945 
1946 /*
1947  * Map a range of physical addresses into kernel virtual address space.
1948  */
1949 static vm_offset_t
1950 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1951     vm_paddr_t pa_end, int prot)
1952 {
1953 	vm_offset_t sva = *virt;
1954 	vm_offset_t va = sva;
1955 
1956 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1957 	//		sva, pa_start, pa_end);
1958 
1959 	while (pa_start < pa_end) {
1960 		mmu_booke_kenter(mmu, va, pa_start);
1961 		va += PAGE_SIZE;
1962 		pa_start += PAGE_SIZE;
1963 	}
1964 	*virt = va;
1965 
1966 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1967 	return (sva);
1968 }
1969 
1970 /*
1971  * The pmap must be activated before it's address space can be accessed in any
1972  * way.
1973  */
1974 static void
1975 mmu_booke_activate(mmu_t mmu, struct thread *td)
1976 {
1977 	pmap_t pmap;
1978 	u_int cpuid;
1979 
1980 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1981 
1982 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1983 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1984 
1985 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1986 
1987 	sched_pin();
1988 
1989 	cpuid = PCPU_GET(cpuid);
1990 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1991 	PCPU_SET(curpmap, pmap);
1992 
1993 	if (pmap->pm_tid[cpuid] == TID_NONE)
1994 		tid_alloc(pmap);
1995 
1996 	/* Load PID0 register with pmap tid value. */
1997 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1998 	__asm __volatile("isync");
1999 
2000 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2001 
2002 	sched_unpin();
2003 
2004 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2005 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2006 }
2007 
2008 /*
2009  * Deactivate the specified process's address space.
2010  */
2011 static void
2012 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2013 {
2014 	pmap_t pmap;
2015 
2016 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2017 
2018 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2019 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2020 
2021 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2022 
2023 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2024 	PCPU_SET(curpmap, NULL);
2025 }
2026 
2027 /*
2028  * Copy the range specified by src_addr/len
2029  * from the source map to the range dst_addr/len
2030  * in the destination map.
2031  *
2032  * This routine is only advisory and need not do anything.
2033  */
2034 static void
2035 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2036     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2037 {
2038 
2039 }
2040 
2041 /*
2042  * Set the physical protection on the specified range of this map as requested.
2043  */
2044 static void
2045 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2046     vm_prot_t prot)
2047 {
2048 	vm_offset_t va;
2049 	vm_page_t m;
2050 	pte_t *pte;
2051 
2052 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2053 		mmu_booke_remove(mmu, pmap, sva, eva);
2054 		return;
2055 	}
2056 
2057 	if (prot & VM_PROT_WRITE)
2058 		return;
2059 
2060 	PMAP_LOCK(pmap);
2061 	for (va = sva; va < eva; va += PAGE_SIZE) {
2062 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2063 			if (PTE_ISVALID(pte)) {
2064 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2065 
2066 				mtx_lock_spin(&tlbivax_mutex);
2067 				tlb_miss_lock();
2068 
2069 				/* Handle modified pages. */
2070 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2071 					vm_page_dirty(m);
2072 
2073 				tlb0_flush_entry(va);
2074 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2075 
2076 				tlb_miss_unlock();
2077 				mtx_unlock_spin(&tlbivax_mutex);
2078 			}
2079 		}
2080 	}
2081 	PMAP_UNLOCK(pmap);
2082 }
2083 
2084 /*
2085  * Clear the write and modified bits in each of the given page's mappings.
2086  */
2087 static void
2088 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2089 {
2090 	pv_entry_t pv;
2091 	pte_t *pte;
2092 
2093 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2094 	    ("mmu_booke_remove_write: page %p is not managed", m));
2095 
2096 	/*
2097 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2098 	 * set by another thread while the object is locked.  Thus,
2099 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2100 	 */
2101 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2102 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2103 		return;
2104 	rw_wlock(&pvh_global_lock);
2105 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2106 		PMAP_LOCK(pv->pv_pmap);
2107 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2108 			if (PTE_ISVALID(pte)) {
2109 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2110 
2111 				mtx_lock_spin(&tlbivax_mutex);
2112 				tlb_miss_lock();
2113 
2114 				/* Handle modified pages. */
2115 				if (PTE_ISMODIFIED(pte))
2116 					vm_page_dirty(m);
2117 
2118 				/* Flush mapping from TLB0. */
2119 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2120 
2121 				tlb_miss_unlock();
2122 				mtx_unlock_spin(&tlbivax_mutex);
2123 			}
2124 		}
2125 		PMAP_UNLOCK(pv->pv_pmap);
2126 	}
2127 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2128 	rw_wunlock(&pvh_global_lock);
2129 }
2130 
2131 static void
2132 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2133 {
2134 	pte_t *pte;
2135 	pmap_t pmap;
2136 	vm_page_t m;
2137 	vm_offset_t addr;
2138 	vm_paddr_t pa = 0;
2139 	int active, valid;
2140 
2141 	va = trunc_page(va);
2142 	sz = round_page(sz);
2143 
2144 	rw_wlock(&pvh_global_lock);
2145 	pmap = PCPU_GET(curpmap);
2146 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2147 	while (sz > 0) {
2148 		PMAP_LOCK(pm);
2149 		pte = pte_find(mmu, pm, va);
2150 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2151 		if (valid)
2152 			pa = PTE_PA(pte);
2153 		PMAP_UNLOCK(pm);
2154 		if (valid) {
2155 			if (!active) {
2156 				/* Create a mapping in the active pmap. */
2157 				addr = 0;
2158 				m = PHYS_TO_VM_PAGE(pa);
2159 				PMAP_LOCK(pmap);
2160 				pte_enter(mmu, pmap, m, addr,
2161 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2162 				__syncicache((void *)addr, PAGE_SIZE);
2163 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2164 				PMAP_UNLOCK(pmap);
2165 			} else
2166 				__syncicache((void *)va, PAGE_SIZE);
2167 		}
2168 		va += PAGE_SIZE;
2169 		sz -= PAGE_SIZE;
2170 	}
2171 	rw_wunlock(&pvh_global_lock);
2172 }
2173 
2174 /*
2175  * Atomically extract and hold the physical page with the given
2176  * pmap and virtual address pair if that mapping permits the given
2177  * protection.
2178  */
2179 static vm_page_t
2180 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2181     vm_prot_t prot)
2182 {
2183 	pte_t *pte;
2184 	vm_page_t m;
2185 	uint32_t pte_wbit;
2186 	vm_paddr_t pa;
2187 
2188 	m = NULL;
2189 	pa = 0;
2190 	PMAP_LOCK(pmap);
2191 retry:
2192 	pte = pte_find(mmu, pmap, va);
2193 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2194 		if (pmap == kernel_pmap)
2195 			pte_wbit = PTE_SW;
2196 		else
2197 			pte_wbit = PTE_UW;
2198 
2199 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2200 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2201 				goto retry;
2202 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2203 			vm_page_hold(m);
2204 		}
2205 	}
2206 
2207 	PA_UNLOCK_COND(pa);
2208 	PMAP_UNLOCK(pmap);
2209 	return (m);
2210 }
2211 
2212 /*
2213  * Initialize a vm_page's machine-dependent fields.
2214  */
2215 static void
2216 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2217 {
2218 
2219 	TAILQ_INIT(&m->md.pv_list);
2220 }
2221 
2222 /*
2223  * mmu_booke_zero_page_area zeros the specified hardware page by
2224  * mapping it into virtual memory and using bzero to clear
2225  * its contents.
2226  *
2227  * off and size must reside within a single page.
2228  */
2229 static void
2230 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2231 {
2232 	vm_offset_t va;
2233 
2234 	/* XXX KASSERT off and size are within a single page? */
2235 
2236 	mtx_lock(&zero_page_mutex);
2237 	va = zero_page_va;
2238 
2239 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2240 	bzero((caddr_t)va + off, size);
2241 	mmu_booke_kremove(mmu, va);
2242 
2243 	mtx_unlock(&zero_page_mutex);
2244 }
2245 
2246 /*
2247  * mmu_booke_zero_page zeros the specified hardware page.
2248  */
2249 static void
2250 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2251 {
2252 	vm_offset_t off, va;
2253 
2254 	mtx_lock(&zero_page_mutex);
2255 	va = zero_page_va;
2256 
2257 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2258 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2259 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2260 	mmu_booke_kremove(mmu, va);
2261 
2262 	mtx_unlock(&zero_page_mutex);
2263 }
2264 
2265 /*
2266  * mmu_booke_copy_page copies the specified (machine independent) page by
2267  * mapping the page into virtual memory and using memcopy to copy the page,
2268  * one machine dependent page at a time.
2269  */
2270 static void
2271 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2272 {
2273 	vm_offset_t sva, dva;
2274 
2275 	sva = copy_page_src_va;
2276 	dva = copy_page_dst_va;
2277 
2278 	mtx_lock(&copy_page_mutex);
2279 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2280 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2281 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2282 	mmu_booke_kremove(mmu, dva);
2283 	mmu_booke_kremove(mmu, sva);
2284 	mtx_unlock(&copy_page_mutex);
2285 }
2286 
2287 static inline void
2288 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2289     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2290 {
2291 	void *a_cp, *b_cp;
2292 	vm_offset_t a_pg_offset, b_pg_offset;
2293 	int cnt;
2294 
2295 	mtx_lock(&copy_page_mutex);
2296 	while (xfersize > 0) {
2297 		a_pg_offset = a_offset & PAGE_MASK;
2298 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2299 		mmu_booke_kenter(mmu, copy_page_src_va,
2300 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2301 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2302 		b_pg_offset = b_offset & PAGE_MASK;
2303 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2304 		mmu_booke_kenter(mmu, copy_page_dst_va,
2305 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2306 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2307 		bcopy(a_cp, b_cp, cnt);
2308 		mmu_booke_kremove(mmu, copy_page_dst_va);
2309 		mmu_booke_kremove(mmu, copy_page_src_va);
2310 		a_offset += cnt;
2311 		b_offset += cnt;
2312 		xfersize -= cnt;
2313 	}
2314 	mtx_unlock(&copy_page_mutex);
2315 }
2316 
2317 /*
2318  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2319  * into virtual memory and using bzero to clear its contents. This is intended
2320  * to be called from the vm_pagezero process only and outside of Giant. No
2321  * lock is required.
2322  */
2323 static void
2324 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2325 {
2326 	vm_offset_t va;
2327 
2328 	va = zero_page_idle_va;
2329 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2330 	bzero((caddr_t)va, PAGE_SIZE);
2331 	mmu_booke_kremove(mmu, va);
2332 }
2333 
2334 static vm_offset_t
2335 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2336 {
2337 	vm_paddr_t paddr;
2338 	vm_offset_t qaddr;
2339 	uint32_t flags;
2340 	pte_t *pte;
2341 
2342 	paddr = VM_PAGE_TO_PHYS(m);
2343 
2344 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2345 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2346 	flags |= PTE_PS_4KB;
2347 
2348 	critical_enter();
2349 	qaddr = PCPU_GET(qmap_addr);
2350 
2351 	pte = pte_find(mmu, kernel_pmap, qaddr);
2352 
2353 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2354 
2355 	/*
2356 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2357  	 * not be present in other TLBs.  Is there a better instruction
2358 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2359 	 */
2360 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2361 	__asm __volatile("isync; msync");
2362 
2363 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
2364 
2365 	/* Flush the real memory from the instruction cache. */
2366 	if ((flags & (PTE_I | PTE_G)) == 0)
2367 		__syncicache((void *)qaddr, PAGE_SIZE);
2368 
2369 	return (qaddr);
2370 }
2371 
2372 static void
2373 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2374 {
2375 	pte_t *pte;
2376 
2377 	pte = pte_find(mmu, kernel_pmap, addr);
2378 
2379 	KASSERT(PCPU_GET(qmap_addr) == addr,
2380 	    ("mmu_booke_quick_remove_page: invalid address"));
2381 	KASSERT(*pte != 0,
2382 	    ("mmu_booke_quick_remove_page: PTE not in use"));
2383 
2384 	*pte = 0;
2385 	critical_exit();
2386 }
2387 
2388 /*
2389  * Return whether or not the specified physical page was modified
2390  * in any of physical maps.
2391  */
2392 static boolean_t
2393 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2394 {
2395 	pte_t *pte;
2396 	pv_entry_t pv;
2397 	boolean_t rv;
2398 
2399 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2400 	    ("mmu_booke_is_modified: page %p is not managed", m));
2401 	rv = FALSE;
2402 
2403 	/*
2404 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2405 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2406 	 * is clear, no PTEs can be modified.
2407 	 */
2408 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2409 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2410 		return (rv);
2411 	rw_wlock(&pvh_global_lock);
2412 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2413 		PMAP_LOCK(pv->pv_pmap);
2414 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2415 		    PTE_ISVALID(pte)) {
2416 			if (PTE_ISMODIFIED(pte))
2417 				rv = TRUE;
2418 		}
2419 		PMAP_UNLOCK(pv->pv_pmap);
2420 		if (rv)
2421 			break;
2422 	}
2423 	rw_wunlock(&pvh_global_lock);
2424 	return (rv);
2425 }
2426 
2427 /*
2428  * Return whether or not the specified virtual address is eligible
2429  * for prefault.
2430  */
2431 static boolean_t
2432 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2433 {
2434 
2435 	return (FALSE);
2436 }
2437 
2438 /*
2439  * Return whether or not the specified physical page was referenced
2440  * in any physical maps.
2441  */
2442 static boolean_t
2443 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2444 {
2445 	pte_t *pte;
2446 	pv_entry_t pv;
2447 	boolean_t rv;
2448 
2449 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2450 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2451 	rv = FALSE;
2452 	rw_wlock(&pvh_global_lock);
2453 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2454 		PMAP_LOCK(pv->pv_pmap);
2455 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2456 		    PTE_ISVALID(pte)) {
2457 			if (PTE_ISREFERENCED(pte))
2458 				rv = TRUE;
2459 		}
2460 		PMAP_UNLOCK(pv->pv_pmap);
2461 		if (rv)
2462 			break;
2463 	}
2464 	rw_wunlock(&pvh_global_lock);
2465 	return (rv);
2466 }
2467 
2468 /*
2469  * Clear the modify bits on the specified physical page.
2470  */
2471 static void
2472 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2473 {
2474 	pte_t *pte;
2475 	pv_entry_t pv;
2476 
2477 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2478 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2479 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2480 	KASSERT(!vm_page_xbusied(m),
2481 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2482 
2483 	/*
2484 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2485 	 * If the object containing the page is locked and the page is not
2486 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2487 	 */
2488 	if ((m->aflags & PGA_WRITEABLE) == 0)
2489 		return;
2490 	rw_wlock(&pvh_global_lock);
2491 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2492 		PMAP_LOCK(pv->pv_pmap);
2493 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2494 		    PTE_ISVALID(pte)) {
2495 			mtx_lock_spin(&tlbivax_mutex);
2496 			tlb_miss_lock();
2497 
2498 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2499 				tlb0_flush_entry(pv->pv_va);
2500 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2501 				    PTE_REFERENCED);
2502 			}
2503 
2504 			tlb_miss_unlock();
2505 			mtx_unlock_spin(&tlbivax_mutex);
2506 		}
2507 		PMAP_UNLOCK(pv->pv_pmap);
2508 	}
2509 	rw_wunlock(&pvh_global_lock);
2510 }
2511 
2512 /*
2513  * Return a count of reference bits for a page, clearing those bits.
2514  * It is not necessary for every reference bit to be cleared, but it
2515  * is necessary that 0 only be returned when there are truly no
2516  * reference bits set.
2517  *
2518  * XXX: The exact number of bits to check and clear is a matter that
2519  * should be tested and standardized at some point in the future for
2520  * optimal aging of shared pages.
2521  */
2522 static int
2523 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2524 {
2525 	pte_t *pte;
2526 	pv_entry_t pv;
2527 	int count;
2528 
2529 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2530 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2531 	count = 0;
2532 	rw_wlock(&pvh_global_lock);
2533 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2534 		PMAP_LOCK(pv->pv_pmap);
2535 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2536 		    PTE_ISVALID(pte)) {
2537 			if (PTE_ISREFERENCED(pte)) {
2538 				mtx_lock_spin(&tlbivax_mutex);
2539 				tlb_miss_lock();
2540 
2541 				tlb0_flush_entry(pv->pv_va);
2542 				*pte &= ~PTE_REFERENCED;
2543 
2544 				tlb_miss_unlock();
2545 				mtx_unlock_spin(&tlbivax_mutex);
2546 
2547 				if (++count > 4) {
2548 					PMAP_UNLOCK(pv->pv_pmap);
2549 					break;
2550 				}
2551 			}
2552 		}
2553 		PMAP_UNLOCK(pv->pv_pmap);
2554 	}
2555 	rw_wunlock(&pvh_global_lock);
2556 	return (count);
2557 }
2558 
2559 /*
2560  * Clear the wired attribute from the mappings for the specified range of
2561  * addresses in the given pmap.  Every valid mapping within that range must
2562  * have the wired attribute set.  In contrast, invalid mappings cannot have
2563  * the wired attribute set, so they are ignored.
2564  *
2565  * The wired attribute of the page table entry is not a hardware feature, so
2566  * there is no need to invalidate any TLB entries.
2567  */
2568 static void
2569 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2570 {
2571 	vm_offset_t va;
2572 	pte_t *pte;
2573 
2574 	PMAP_LOCK(pmap);
2575 	for (va = sva; va < eva; va += PAGE_SIZE) {
2576 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2577 		    PTE_ISVALID(pte)) {
2578 			if (!PTE_ISWIRED(pte))
2579 				panic("mmu_booke_unwire: pte %p isn't wired",
2580 				    pte);
2581 			*pte &= ~PTE_WIRED;
2582 			pmap->pm_stats.wired_count--;
2583 		}
2584 	}
2585 	PMAP_UNLOCK(pmap);
2586 
2587 }
2588 
2589 /*
2590  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2591  * page.  This count may be changed upwards or downwards in the future; it is
2592  * only necessary that true be returned for a small subset of pmaps for proper
2593  * page aging.
2594  */
2595 static boolean_t
2596 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2597 {
2598 	pv_entry_t pv;
2599 	int loops;
2600 	boolean_t rv;
2601 
2602 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2603 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2604 	loops = 0;
2605 	rv = FALSE;
2606 	rw_wlock(&pvh_global_lock);
2607 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2608 		if (pv->pv_pmap == pmap) {
2609 			rv = TRUE;
2610 			break;
2611 		}
2612 		if (++loops >= 16)
2613 			break;
2614 	}
2615 	rw_wunlock(&pvh_global_lock);
2616 	return (rv);
2617 }
2618 
2619 /*
2620  * Return the number of managed mappings to the given physical page that are
2621  * wired.
2622  */
2623 static int
2624 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2625 {
2626 	pv_entry_t pv;
2627 	pte_t *pte;
2628 	int count = 0;
2629 
2630 	if ((m->oflags & VPO_UNMANAGED) != 0)
2631 		return (count);
2632 	rw_wlock(&pvh_global_lock);
2633 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2634 		PMAP_LOCK(pv->pv_pmap);
2635 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2636 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2637 				count++;
2638 		PMAP_UNLOCK(pv->pv_pmap);
2639 	}
2640 	rw_wunlock(&pvh_global_lock);
2641 	return (count);
2642 }
2643 
2644 static int
2645 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2646 {
2647 	int i;
2648 	vm_offset_t va;
2649 
2650 	/*
2651 	 * This currently does not work for entries that
2652 	 * overlap TLB1 entries.
2653 	 */
2654 	for (i = 0; i < tlb1_idx; i ++) {
2655 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2656 			return (0);
2657 	}
2658 
2659 	return (EFAULT);
2660 }
2661 
2662 void
2663 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2664 {
2665 	vm_paddr_t ppa;
2666 	vm_offset_t ofs;
2667 	vm_size_t gran;
2668 
2669 	/* Minidumps are based on virtual memory addresses. */
2670 	if (do_minidump) {
2671 		*va = (void *)(vm_offset_t)pa;
2672 		return;
2673 	}
2674 
2675 	/* Raw physical memory dumps don't have a virtual address. */
2676 	/* We always map a 256MB page at 256M. */
2677 	gran = 256 * 1024 * 1024;
2678 	ppa = pa & ~(gran - 1);
2679 	ofs = pa - ppa;
2680 	*va = (void *)gran;
2681 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2682 
2683 	if (sz > (gran - ofs))
2684 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2685 		    _TLB_ENTRY_IO);
2686 }
2687 
2688 void
2689 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2690 {
2691 	vm_paddr_t ppa;
2692 	vm_offset_t ofs;
2693 	vm_size_t gran;
2694 
2695 	/* Minidumps are based on virtual memory addresses. */
2696 	/* Nothing to do... */
2697 	if (do_minidump)
2698 		return;
2699 
2700 	/* Raw physical memory dumps don't have a virtual address. */
2701 	tlb1_idx--;
2702 	tlb1[tlb1_idx].mas1 = 0;
2703 	tlb1[tlb1_idx].mas2 = 0;
2704 	tlb1[tlb1_idx].mas3 = 0;
2705 	tlb1_write_entry(tlb1_idx);
2706 
2707 	gran = 256 * 1024 * 1024;
2708 	ppa = pa & ~(gran - 1);
2709 	ofs = pa - ppa;
2710 	if (sz > (gran - ofs)) {
2711 		tlb1_idx--;
2712 		tlb1[tlb1_idx].mas1 = 0;
2713 		tlb1[tlb1_idx].mas2 = 0;
2714 		tlb1[tlb1_idx].mas3 = 0;
2715 		tlb1_write_entry(tlb1_idx);
2716 	}
2717 }
2718 
2719 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2720 
2721 void
2722 mmu_booke_scan_init(mmu_t mmu)
2723 {
2724 	vm_offset_t va;
2725 	pte_t *pte;
2726 	int i;
2727 
2728 	if (!do_minidump) {
2729 		/* Initialize phys. segments for dumpsys(). */
2730 		memset(&dump_map, 0, sizeof(dump_map));
2731 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2732 		    &availmem_regions_sz);
2733 		for (i = 0; i < physmem_regions_sz; i++) {
2734 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2735 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2736 		}
2737 		return;
2738 	}
2739 
2740 	/* Virtual segments for minidumps: */
2741 	memset(&dump_map, 0, sizeof(dump_map));
2742 
2743 	/* 1st: kernel .data and .bss. */
2744 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2745 	dump_map[0].pa_size =
2746 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2747 
2748 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2749 	dump_map[1].pa_start = data_start;
2750 	dump_map[1].pa_size = data_end - data_start;
2751 
2752 	/* 3rd: kernel VM. */
2753 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2754 	/* Find start of next chunk (from va). */
2755 	while (va < virtual_end) {
2756 		/* Don't dump the buffer cache. */
2757 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2758 			va = kmi.buffer_eva;
2759 			continue;
2760 		}
2761 		pte = pte_find(mmu, kernel_pmap, va);
2762 		if (pte != NULL && PTE_ISVALID(pte))
2763 			break;
2764 		va += PAGE_SIZE;
2765 	}
2766 	if (va < virtual_end) {
2767 		dump_map[2].pa_start = va;
2768 		va += PAGE_SIZE;
2769 		/* Find last page in chunk. */
2770 		while (va < virtual_end) {
2771 			/* Don't run into the buffer cache. */
2772 			if (va == kmi.buffer_sva)
2773 				break;
2774 			pte = pte_find(mmu, kernel_pmap, va);
2775 			if (pte == NULL || !PTE_ISVALID(pte))
2776 				break;
2777 			va += PAGE_SIZE;
2778 		}
2779 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2780 	}
2781 }
2782 
2783 /*
2784  * Map a set of physical memory pages into the kernel virtual address space.
2785  * Return a pointer to where it is mapped. This routine is intended to be used
2786  * for mapping device memory, NOT real memory.
2787  */
2788 static void *
2789 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2790 {
2791 
2792 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2793 }
2794 
2795 static void *
2796 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2797 {
2798 	void *res;
2799 	uintptr_t va;
2800 	vm_size_t sz;
2801 	int i;
2802 
2803 	/*
2804 	 * Check if this is premapped in TLB1. Note: this should probably also
2805 	 * check whether a sequence of TLB1 entries exist that match the
2806 	 * requirement, but now only checks the easy case.
2807 	 */
2808 	if (ma == VM_MEMATTR_DEFAULT) {
2809 		for (i = 0; i < tlb1_idx; i++) {
2810 			if (!(tlb1[i].mas1 & MAS1_VALID))
2811 				continue;
2812 			if (pa >= tlb1[i].phys &&
2813 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2814 				return (void *)(tlb1[i].virt +
2815 				    (vm_offset_t)(pa - tlb1[i].phys));
2816 		}
2817 	}
2818 
2819 	size = roundup(size, PAGE_SIZE);
2820 
2821 	/*
2822 	 * We leave a hole for device direct mapping between the maximum user
2823 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2824 	 * devices are in there, just map them 1:1. If not, map them to the
2825 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2826 	 * addresses should be pulled from an allocator, but since we do not
2827 	 * ever free TLB1 entries, it is safe just to increment a counter.
2828 	 * Note that there isn't a lot of address space here (128 MB) and it
2829 	 * is not at all difficult to imagine running out, since that is a 4:1
2830 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2831 	 * mapped there.
2832 	 */
2833 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2834 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2835 		va = pa;
2836 	else
2837 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2838 	res = (void *)va;
2839 
2840 	do {
2841 		sz = 1 << (ilog2(size) & ~1);
2842 		if (va % sz != 0) {
2843 			do {
2844 				sz >>= 2;
2845 			} while (va % sz != 0);
2846 		}
2847 		if (bootverbose)
2848 			printf("Wiring VA=%x to PA=%jx (size=%x), "
2849 			    "using TLB1[%d]\n", va, (uintmax_t)pa, sz, tlb1_idx);
2850 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2851 		size -= sz;
2852 		pa += sz;
2853 		va += sz;
2854 	} while (size > 0);
2855 
2856 	return (res);
2857 }
2858 
2859 /*
2860  * 'Unmap' a range mapped by mmu_booke_mapdev().
2861  */
2862 static void
2863 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2864 {
2865 #ifdef SUPPORTS_SHRINKING_TLB1
2866 	vm_offset_t base, offset;
2867 
2868 	/*
2869 	 * Unmap only if this is inside kernel virtual space.
2870 	 */
2871 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2872 		base = trunc_page(va);
2873 		offset = va & PAGE_MASK;
2874 		size = roundup(offset + size, PAGE_SIZE);
2875 		kva_free(base, size);
2876 	}
2877 #endif
2878 }
2879 
2880 /*
2881  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2882  * specified pmap. This eliminates the blast of soft faults on process startup
2883  * and immediately after an mmap.
2884  */
2885 static void
2886 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2887     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2888 {
2889 
2890 	VM_OBJECT_ASSERT_WLOCKED(object);
2891 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2892 	    ("mmu_booke_object_init_pt: non-device object"));
2893 }
2894 
2895 /*
2896  * Perform the pmap work for mincore.
2897  */
2898 static int
2899 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2900     vm_paddr_t *locked_pa)
2901 {
2902 
2903 	/* XXX: this should be implemented at some point */
2904 	return (0);
2905 }
2906 
2907 static int
2908 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
2909     vm_memattr_t mode)
2910 {
2911 	vm_offset_t va;
2912 	pte_t *pte;
2913 	int i, j;
2914 
2915 	/* Check TLB1 mappings */
2916 	for (i = 0; i < tlb1_idx; i++) {
2917 		if (!(tlb1[i].mas1 & MAS1_VALID))
2918 			continue;
2919 		if (addr >= tlb1[i].virt && addr < tlb1[i].virt + tlb1[i].size)
2920 			break;
2921 	}
2922 	if (i < tlb1_idx) {
2923 		/* Only allow full mappings to be modified for now. */
2924 		/* Validate the range. */
2925 		for (j = i, va = addr; va < addr + sz; va += tlb1[j].size, j++) {
2926 			if (va != tlb1[j].virt || (sz - (va - addr) < tlb1[j].size))
2927 				return (EINVAL);
2928 		}
2929 		for (va = addr; va < addr + sz; va += tlb1[i].size, i++) {
2930 			tlb1[i].mas2 &= ~MAS2_WIMGE_MASK;
2931 			tlb1[i].mas2 |= tlb_calc_wimg(tlb1[i].phys, mode);
2932 
2933 			/*
2934 			 * Write it out to the TLB.  Should really re-sync with other
2935 			 * cores.
2936 			 */
2937 			tlb1_write_entry(i);
2938 		}
2939 		return (0);
2940 	}
2941 
2942 	/* Not in TLB1, try through pmap */
2943 	/* First validate the range. */
2944 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2945 		pte = pte_find(mmu, kernel_pmap, va);
2946 		if (pte == NULL || !PTE_ISVALID(pte))
2947 			return (EINVAL);
2948 	}
2949 
2950 	mtx_lock_spin(&tlbivax_mutex);
2951 	tlb_miss_lock();
2952 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2953 		pte = pte_find(mmu, kernel_pmap, va);
2954 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2955 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode << PTE_MAS2_SHIFT);
2956 		tlb0_flush_entry(va);
2957 	}
2958 	tlb_miss_unlock();
2959 	mtx_unlock_spin(&tlbivax_mutex);
2960 
2961 	return (pte_vatopa(mmu, kernel_pmap, va));
2962 }
2963 
2964 /**************************************************************************/
2965 /* TID handling */
2966 /**************************************************************************/
2967 
2968 /*
2969  * Allocate a TID. If necessary, steal one from someone else.
2970  * The new TID is flushed from the TLB before returning.
2971  */
2972 static tlbtid_t
2973 tid_alloc(pmap_t pmap)
2974 {
2975 	tlbtid_t tid;
2976 	int thiscpu;
2977 
2978 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2979 
2980 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2981 
2982 	thiscpu = PCPU_GET(cpuid);
2983 
2984 	tid = PCPU_GET(tid_next);
2985 	if (tid > TID_MAX)
2986 		tid = TID_MIN;
2987 	PCPU_SET(tid_next, tid + 1);
2988 
2989 	/* If we are stealing TID then clear the relevant pmap's field */
2990 	if (tidbusy[thiscpu][tid] != NULL) {
2991 
2992 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2993 
2994 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2995 
2996 		/* Flush all entries from TLB0 matching this TID. */
2997 		tid_flush(tid);
2998 	}
2999 
3000 	tidbusy[thiscpu][tid] = pmap;
3001 	pmap->pm_tid[thiscpu] = tid;
3002 	__asm __volatile("msync; isync");
3003 
3004 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3005 	    PCPU_GET(tid_next));
3006 
3007 	return (tid);
3008 }
3009 
3010 /**************************************************************************/
3011 /* TLB0 handling */
3012 /**************************************************************************/
3013 
3014 static void
3015 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3016     uint32_t mas7)
3017 {
3018 	int as;
3019 	char desc[3];
3020 	tlbtid_t tid;
3021 	vm_size_t size;
3022 	unsigned int tsize;
3023 
3024 	desc[2] = '\0';
3025 	if (mas1 & MAS1_VALID)
3026 		desc[0] = 'V';
3027 	else
3028 		desc[0] = ' ';
3029 
3030 	if (mas1 & MAS1_IPROT)
3031 		desc[1] = 'P';
3032 	else
3033 		desc[1] = ' ';
3034 
3035 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3036 	tid = MAS1_GETTID(mas1);
3037 
3038 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3039 	size = 0;
3040 	if (tsize)
3041 		size = tsize2size(tsize);
3042 
3043 	debugf("%3d: (%s) [AS=%d] "
3044 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3045 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3046 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3047 }
3048 
3049 /* Convert TLB0 va and way number to tlb0[] table index. */
3050 static inline unsigned int
3051 tlb0_tableidx(vm_offset_t va, unsigned int way)
3052 {
3053 	unsigned int idx;
3054 
3055 	idx = (way * TLB0_ENTRIES_PER_WAY);
3056 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3057 	return (idx);
3058 }
3059 
3060 /*
3061  * Invalidate TLB0 entry.
3062  */
3063 static inline void
3064 tlb0_flush_entry(vm_offset_t va)
3065 {
3066 
3067 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3068 
3069 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3070 
3071 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3072 	__asm __volatile("isync; msync");
3073 	__asm __volatile("tlbsync; msync");
3074 
3075 	CTR1(KTR_PMAP, "%s: e", __func__);
3076 }
3077 
3078 /* Print out contents of the MAS registers for each TLB0 entry */
3079 void
3080 tlb0_print_tlbentries(void)
3081 {
3082 	uint32_t mas0, mas1, mas2, mas3, mas7;
3083 	int entryidx, way, idx;
3084 
3085 	debugf("TLB0 entries:\n");
3086 	for (way = 0; way < TLB0_WAYS; way ++)
3087 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3088 
3089 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3090 			mtspr(SPR_MAS0, mas0);
3091 			__asm __volatile("isync");
3092 
3093 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3094 			mtspr(SPR_MAS2, mas2);
3095 
3096 			__asm __volatile("isync; tlbre");
3097 
3098 			mas1 = mfspr(SPR_MAS1);
3099 			mas2 = mfspr(SPR_MAS2);
3100 			mas3 = mfspr(SPR_MAS3);
3101 			mas7 = mfspr(SPR_MAS7);
3102 
3103 			idx = tlb0_tableidx(mas2, way);
3104 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3105 		}
3106 }
3107 
3108 /**************************************************************************/
3109 /* TLB1 handling */
3110 /**************************************************************************/
3111 
3112 /*
3113  * TLB1 mapping notes:
3114  *
3115  * TLB1[0]	Kernel text and data.
3116  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3117  *		windows, other devices mappings.
3118  */
3119 
3120 /*
3121  * Write given entry to TLB1 hardware.
3122  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
3123  */
3124 static void
3125 tlb1_write_entry(unsigned int idx)
3126 {
3127 	uint32_t mas0;
3128 
3129 	//debugf("tlb1_write_entry: s\n");
3130 
3131 	/* Select entry */
3132 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3133 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
3134 
3135 	mtspr(SPR_MAS0, mas0);
3136 	__asm __volatile("isync");
3137 	mtspr(SPR_MAS1, tlb1[idx].mas1);
3138 	__asm __volatile("isync");
3139 	mtspr(SPR_MAS2, tlb1[idx].mas2);
3140 	__asm __volatile("isync");
3141 	mtspr(SPR_MAS3, tlb1[idx].mas3);
3142 	__asm __volatile("isync");
3143 	switch ((mfpvr() >> 16) & 0xFFFF) {
3144 	case FSL_E500mc:
3145 	case FSL_E5500:
3146 		mtspr(SPR_MAS8, 0);
3147 		__asm __volatile("isync");
3148 		/* FALLTHROUGH */
3149 	case FSL_E500v2:
3150 		mtspr(SPR_MAS7, tlb1[idx].mas7);
3151 		__asm __volatile("isync");
3152 		break;
3153 	default:
3154 		break;
3155 	}
3156 
3157 	__asm __volatile("tlbwe; isync; msync");
3158 
3159 	//debugf("tlb1_write_entry: e\n");
3160 }
3161 
3162 /*
3163  * Return the largest uint value log such that 2^log <= num.
3164  */
3165 static unsigned int
3166 ilog2(unsigned int num)
3167 {
3168 	int lz;
3169 
3170 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3171 	return (31 - lz);
3172 }
3173 
3174 /*
3175  * Convert TLB TSIZE value to mapped region size.
3176  */
3177 static vm_size_t
3178 tsize2size(unsigned int tsize)
3179 {
3180 
3181 	/*
3182 	 * size = 4^tsize KB
3183 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3184 	 */
3185 
3186 	return ((1 << (2 * tsize)) * 1024);
3187 }
3188 
3189 /*
3190  * Convert region size (must be power of 4) to TLB TSIZE value.
3191  */
3192 static unsigned int
3193 size2tsize(vm_size_t size)
3194 {
3195 
3196 	return (ilog2(size) / 2 - 5);
3197 }
3198 
3199 /*
3200  * Register permanent kernel mapping in TLB1.
3201  *
3202  * Entries are created starting from index 0 (current free entry is
3203  * kept in tlb1_idx) and are not supposed to be invalidated.
3204  */
3205 int
3206 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3207     uint32_t flags)
3208 {
3209 	uint32_t ts, tid;
3210 	int tsize, index;
3211 
3212 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3213 	if (index >= TLB1_ENTRIES) {
3214 		printf("tlb1_set_entry: TLB1 full!\n");
3215 		return (-1);
3216 	}
3217 
3218 	/* Convert size to TSIZE */
3219 	tsize = size2tsize(size);
3220 
3221 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3222 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3223 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3224 
3225 	/*
3226 	 * Atomicity is preserved by the atomic increment above since nothing
3227 	 * is ever removed from tlb1.
3228 	 */
3229 
3230 	tlb1[index].phys = pa;
3231 	tlb1[index].virt = va;
3232 	tlb1[index].size = size;
3233 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3234 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3235 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3236 
3237 	/* Set supervisor RWX permission bits */
3238 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3239 	tlb1[index].mas7 = (pa >> 32) & MAS7_RPN;
3240 
3241 	tlb1_write_entry(index);
3242 
3243 	/*
3244 	 * XXX in general TLB1 updates should be propagated between CPUs,
3245 	 * since current design assumes to have the same TLB1 set-up on all
3246 	 * cores.
3247 	 */
3248 	return (0);
3249 }
3250 
3251 /*
3252  * Map in contiguous RAM region into the TLB1 using maximum of
3253  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3254  *
3255  * If necessary round up last entry size and return total size
3256  * used by all allocated entries.
3257  */
3258 vm_size_t
3259 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3260 {
3261 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3262 	vm_size_t mapped, pgsz, base, mask;
3263 	int idx, nents;
3264 
3265 	/* Round up to the next 1M */
3266 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3267 
3268 	mapped = 0;
3269 	idx = 0;
3270 	base = va;
3271 	pgsz = 64*1024*1024;
3272 	while (mapped < size) {
3273 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3274 			while (pgsz > (size - mapped))
3275 				pgsz >>= 2;
3276 			pgs[idx++] = pgsz;
3277 			mapped += pgsz;
3278 		}
3279 
3280 		/* We under-map. Correct for this. */
3281 		if (mapped < size) {
3282 			while (pgs[idx - 1] == pgsz) {
3283 				idx--;
3284 				mapped -= pgsz;
3285 			}
3286 			/* XXX We may increase beyond out starting point. */
3287 			pgsz <<= 2;
3288 			pgs[idx++] = pgsz;
3289 			mapped += pgsz;
3290 		}
3291 	}
3292 
3293 	nents = idx;
3294 	mask = pgs[0] - 1;
3295 	/* Align address to the boundary */
3296 	if (va & mask) {
3297 		va = (va + mask) & ~mask;
3298 		pa = (pa + mask) & ~mask;
3299 	}
3300 
3301 	for (idx = 0; idx < nents; idx++) {
3302 		pgsz = pgs[idx];
3303 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
3304 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3305 		pa += pgsz;
3306 		va += pgsz;
3307 	}
3308 
3309 	mapped = (va - base);
3310 #ifdef __powerpc64__
3311 	printf("mapped size 0x%016lx (wasted space 0x%16lx)\n",
3312 #else
3313 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3314 #endif
3315 	    mapped, mapped - size);
3316 	return (mapped);
3317 }
3318 
3319 /*
3320  * TLB1 initialization routine, to be called after the very first
3321  * assembler level setup done in locore.S.
3322  */
3323 void
3324 tlb1_init()
3325 {
3326 	uint32_t mas0, mas1, mas2, mas3, mas7;
3327 	uint32_t tsz;
3328 	int i;
3329 
3330 	tlb1_idx = 1;
3331 
3332 	tlb1_get_tlbconf();
3333 
3334 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3335 	mtspr(SPR_MAS0, mas0);
3336 	__asm __volatile("isync; tlbre");
3337 
3338 	mas1 = mfspr(SPR_MAS1);
3339 	mas2 = mfspr(SPR_MAS2);
3340 	mas3 = mfspr(SPR_MAS3);
3341 	mas7 = mfspr(SPR_MAS7);
3342 
3343 	tlb1[0].mas1 = mas1;
3344 	tlb1[0].mas2 = mfspr(SPR_MAS2);
3345 	tlb1[0].mas3 = mas3;
3346 	tlb1[0].mas7 = mas7;
3347 	tlb1[0].virt = mas2 & MAS2_EPN_MASK;
3348 	tlb1[0].phys =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
3349 	    (mas3 & MAS3_RPN);
3350 
3351 	kernload = tlb1[0].phys;
3352 
3353 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3354 	tlb1[0].size = (tsz > 0) ? tsize2size(tsz) : 0;
3355 	kernsize += tlb1[0].size;
3356 
3357 #ifdef SMP
3358 	bp_ntlb1s = tlb1_idx;
3359 #endif
3360 
3361 	/* Purge the remaining entries */
3362 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3363 		tlb1_write_entry(i);
3364 
3365 	/* Setup TLB miss defaults */
3366 	set_mas4_defaults();
3367 }
3368 
3369 vm_offset_t
3370 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3371 {
3372 	vm_paddr_t pa_base;
3373 	vm_offset_t va, sz;
3374 	int i;
3375 
3376 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3377 
3378 	for (i = 0; i < tlb1_idx; i++) {
3379 		if (!(tlb1[i].mas1 & MAS1_VALID))
3380 			continue;
3381 		if (pa >= tlb1[i].phys && (pa + size) <=
3382 		    (tlb1[i].phys + tlb1[i].size))
3383 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3384 	}
3385 
3386 	pa_base = rounddown(pa, PAGE_SIZE);
3387 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3388 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3389 	va = tlb1_map_base + (pa - pa_base);
3390 
3391 	do {
3392 		sz = 1 << (ilog2(size) & ~1);
3393 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3394 		size -= sz;
3395 		pa_base += sz;
3396 		tlb1_map_base += sz;
3397 	} while (size > 0);
3398 
3399 #ifdef SMP
3400 	bp_ntlb1s = tlb1_idx;
3401 #endif
3402 
3403 	return (va);
3404 }
3405 
3406 /*
3407  * Setup MAS4 defaults.
3408  * These values are loaded to MAS0-2 on a TLB miss.
3409  */
3410 static void
3411 set_mas4_defaults(void)
3412 {
3413 	uint32_t mas4;
3414 
3415 	/* Defaults: TLB0, PID0, TSIZED=4K */
3416 	mas4 = MAS4_TLBSELD0;
3417 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3418 #ifdef SMP
3419 	mas4 |= MAS4_MD;
3420 #endif
3421 	mtspr(SPR_MAS4, mas4);
3422 	__asm __volatile("isync");
3423 }
3424 
3425 /*
3426  * Print out contents of the MAS registers for each TLB1 entry
3427  */
3428 void
3429 tlb1_print_tlbentries(void)
3430 {
3431 	uint32_t mas0, mas1, mas2, mas3, mas7;
3432 	int i;
3433 
3434 	debugf("TLB1 entries:\n");
3435 	for (i = 0; i < TLB1_ENTRIES; i++) {
3436 
3437 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3438 		mtspr(SPR_MAS0, mas0);
3439 
3440 		__asm __volatile("isync; tlbre");
3441 
3442 		mas1 = mfspr(SPR_MAS1);
3443 		mas2 = mfspr(SPR_MAS2);
3444 		mas3 = mfspr(SPR_MAS3);
3445 		mas7 = mfspr(SPR_MAS7);
3446 
3447 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3448 	}
3449 }
3450 
3451 /*
3452  * Print out contents of the in-ram tlb1 table.
3453  */
3454 void
3455 tlb1_print_entries(void)
3456 {
3457 	int i;
3458 
3459 	debugf("tlb1[] table entries:\n");
3460 	for (i = 0; i < TLB1_ENTRIES; i++)
3461 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3,
3462 		    tlb1[i].mas7);
3463 }
3464 
3465 /*
3466  * Return 0 if the physical IO range is encompassed by one of the
3467  * the TLB1 entries, otherwise return related error code.
3468  */
3469 static int
3470 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3471 {
3472 	uint32_t prot;
3473 	vm_paddr_t pa_start;
3474 	vm_paddr_t pa_end;
3475 	unsigned int entry_tsize;
3476 	vm_size_t entry_size;
3477 
3478 	*va = (vm_offset_t)NULL;
3479 
3480 	/* Skip invalid entries */
3481 	if (!(tlb1[i].mas1 & MAS1_VALID))
3482 		return (EINVAL);
3483 
3484 	/*
3485 	 * The entry must be cache-inhibited, guarded, and r/w
3486 	 * so it can function as an i/o page
3487 	 */
3488 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3489 	if (prot != (MAS2_I | MAS2_G))
3490 		return (EPERM);
3491 
3492 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3493 	if (prot != (MAS3_SR | MAS3_SW))
3494 		return (EPERM);
3495 
3496 	/* The address should be within the entry range. */
3497 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3498 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3499 
3500 	entry_size = tsize2size(entry_tsize);
3501 	pa_start = (((vm_paddr_t)tlb1[i].mas7 & MAS7_RPN) << 32) |
3502 	    (tlb1[i].mas3 & MAS3_RPN);
3503 	pa_end = pa_start + entry_size;
3504 
3505 	if ((pa < pa_start) || ((pa + size) > pa_end))
3506 		return (ERANGE);
3507 
3508 	/* Return virtual address of this mapping. */
3509 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3510 	return (0);
3511 }
3512 
3513 /*
3514  * Invalidate all TLB0 entries which match the given TID. Note this is
3515  * dedicated for cases when invalidations should NOT be propagated to other
3516  * CPUs.
3517  */
3518 static void
3519 tid_flush(tlbtid_t tid)
3520 {
3521 	register_t msr;
3522 	uint32_t mas0, mas1, mas2;
3523 	int entry, way;
3524 
3525 
3526 	/* Don't evict kernel translations */
3527 	if (tid == TID_KERNEL)
3528 		return;
3529 
3530 	msr = mfmsr();
3531 	__asm __volatile("wrteei 0");
3532 
3533 	for (way = 0; way < TLB0_WAYS; way++)
3534 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
3535 
3536 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3537 			mtspr(SPR_MAS0, mas0);
3538 			__asm __volatile("isync");
3539 
3540 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
3541 			mtspr(SPR_MAS2, mas2);
3542 
3543 			__asm __volatile("isync; tlbre");
3544 
3545 			mas1 = mfspr(SPR_MAS1);
3546 
3547 			if (!(mas1 & MAS1_VALID))
3548 				continue;
3549 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
3550 				continue;
3551 			mas1 &= ~MAS1_VALID;
3552 			mtspr(SPR_MAS1, mas1);
3553 			__asm __volatile("isync; tlbwe; isync; msync");
3554 		}
3555 	mtmsr(msr);
3556 }
3557