xref: /freebsd/sys/powerpc/booke/pmap.c (revision a321cc5dc908a14d42e57e2468923937f18c21fc)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include "opt_kstack_pages.h"
55 
56 #include <sys/param.h>
57 #include <sys/conf.h>
58 #include <sys/malloc.h>
59 #include <sys/ktr.h>
60 #include <sys/proc.h>
61 #include <sys/user.h>
62 #include <sys/queue.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/kerneldump.h>
66 #include <sys/linker.h>
67 #include <sys/msgbuf.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/vmmeter.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_pager.h>
84 #include <vm/uma.h>
85 
86 #include <machine/cpu.h>
87 #include <machine/pcb.h>
88 #include <machine/platform.h>
89 
90 #include <machine/tlb.h>
91 #include <machine/spr.h>
92 #include <machine/md_var.h>
93 #include <machine/mmuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/pte.h>
96 
97 #include "mmu_if.h"
98 
99 #define	SPARSE_MAPDEV
100 #ifdef  DEBUG
101 #define debugf(fmt, args...) printf(fmt, ##args)
102 #else
103 #define debugf(fmt, args...)
104 #endif
105 
106 #define TODO			panic("%s: not implemented", __func__);
107 
108 extern unsigned char _etext[];
109 extern unsigned char _end[];
110 
111 extern uint32_t *bootinfo;
112 
113 #ifdef SMP
114 extern uint32_t bp_ntlb1s;
115 #endif
116 
117 vm_paddr_t kernload;
118 vm_offset_t kernstart;
119 vm_size_t kernsize;
120 
121 /* Message buffer and tables. */
122 static vm_offset_t data_start;
123 static vm_size_t data_end;
124 
125 /* Phys/avail memory regions. */
126 static struct mem_region *availmem_regions;
127 static int availmem_regions_sz;
128 static struct mem_region *physmem_regions;
129 static int physmem_regions_sz;
130 
131 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
132 static vm_offset_t zero_page_va;
133 static struct mtx zero_page_mutex;
134 
135 static struct mtx tlbivax_mutex;
136 
137 /*
138  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
139  * by idle thred only, no lock required.
140  */
141 static vm_offset_t zero_page_idle_va;
142 
143 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
144 static vm_offset_t copy_page_src_va;
145 static vm_offset_t copy_page_dst_va;
146 static struct mtx copy_page_mutex;
147 
148 /**************************************************************************/
149 /* PMAP */
150 /**************************************************************************/
151 
152 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
153     vm_prot_t, u_int flags, int8_t psind);
154 
155 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
156 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
157 
158 /*
159  * If user pmap is processed with mmu_booke_remove and the resident count
160  * drops to 0, there are no more pages to remove, so we need not continue.
161  */
162 #define PMAP_REMOVE_DONE(pmap) \
163 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
164 
165 extern int elf32_nxstack;
166 
167 /**************************************************************************/
168 /* TLB and TID handling */
169 /**************************************************************************/
170 
171 /* Translation ID busy table */
172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 
174 /*
175  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
176  * core revisions and should be read from h/w registers during early config.
177  */
178 uint32_t tlb0_entries;
179 uint32_t tlb0_ways;
180 uint32_t tlb0_entries_per_way;
181 uint32_t tlb1_entries;
182 
183 #define TLB0_ENTRIES		(tlb0_entries)
184 #define TLB0_WAYS		(tlb0_ways)
185 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
186 
187 #define TLB1_ENTRIES (tlb1_entries)
188 #define TLB1_MAXENTRIES	64
189 
190 /* In-ram copy of the TLB1 */
191 static tlb_entry_t tlb1[TLB1_MAXENTRIES];
192 
193 /* Next free entry in the TLB1 */
194 static unsigned int tlb1_idx;
195 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
196 
197 static tlbtid_t tid_alloc(struct pmap *);
198 static void tid_flush(tlbtid_t tid);
199 
200 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
201 
202 static void tlb1_write_entry(unsigned int);
203 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
204 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
205 
206 static vm_size_t tsize2size(unsigned int);
207 static unsigned int size2tsize(vm_size_t);
208 static unsigned int ilog2(unsigned int);
209 
210 static void set_mas4_defaults(void);
211 
212 static inline void tlb0_flush_entry(vm_offset_t);
213 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
214 
215 /**************************************************************************/
216 /* Page table management */
217 /**************************************************************************/
218 
219 static struct rwlock_padalign pvh_global_lock;
220 
221 /* Data for the pv entry allocation mechanism */
222 static uma_zone_t pvzone;
223 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
224 
225 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
226 
227 #ifndef PMAP_SHPGPERPROC
228 #define PMAP_SHPGPERPROC	200
229 #endif
230 
231 static void ptbl_init(void);
232 static struct ptbl_buf *ptbl_buf_alloc(void);
233 static void ptbl_buf_free(struct ptbl_buf *);
234 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
235 
236 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
237 static void ptbl_free(mmu_t, pmap_t, unsigned int);
238 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
239 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
240 
241 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
242 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
243 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
244 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
245 static void kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr,
246 			     vm_offset_t pdir);
247 
248 static pv_entry_t pv_alloc(void);
249 static void pv_free(pv_entry_t);
250 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
251 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
252 
253 static void booke_pmap_init_qpages(void);
254 
255 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
256 #define PTBL_BUFS		(128 * 16)
257 
258 struct ptbl_buf {
259 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
260 	vm_offset_t kva;		/* va of mapping */
261 };
262 
263 /* ptbl free list and a lock used for access synchronization. */
264 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
265 static struct mtx ptbl_buf_freelist_lock;
266 
267 /* Base address of kva space allocated fot ptbl bufs. */
268 static vm_offset_t ptbl_buf_pool_vabase;
269 
270 /* Pointer to ptbl_buf structures. */
271 static struct ptbl_buf *ptbl_bufs;
272 
273 #ifdef SMP
274 void pmap_bootstrap_ap(volatile uint32_t *);
275 #endif
276 
277 /*
278  * Kernel MMU interface
279  */
280 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
281 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
282     vm_size_t, vm_offset_t);
283 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
284 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
285     vm_offset_t, vm_page_t *, vm_offset_t, int);
286 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
287     vm_prot_t, u_int flags, int8_t psind);
288 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
289     vm_page_t, vm_prot_t);
290 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
291     vm_prot_t);
292 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
293 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
294     vm_prot_t);
295 static void		mmu_booke_init(mmu_t);
296 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
297 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
298 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
299 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
300 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
301     int);
302 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
303     vm_paddr_t *);
304 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
305     vm_object_t, vm_pindex_t, vm_size_t);
306 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
307 static void		mmu_booke_page_init(mmu_t, vm_page_t);
308 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
309 static void		mmu_booke_pinit(mmu_t, pmap_t);
310 static void		mmu_booke_pinit0(mmu_t, pmap_t);
311 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
312     vm_prot_t);
313 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
314 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
315 static void		mmu_booke_release(mmu_t, pmap_t);
316 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
317 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
318 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
319 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
320 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
321 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
322 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
323 static void		mmu_booke_activate(mmu_t, struct thread *);
324 static void		mmu_booke_deactivate(mmu_t, struct thread *);
325 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
326 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
327 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
328 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
329 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
330 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
331 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
332 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
333 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
334 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
335     vm_size_t);
336 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
337     void **);
338 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
339     void *);
340 static void		mmu_booke_scan_init(mmu_t);
341 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
342 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
343 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
344     vm_size_t sz, vm_memattr_t mode);
345 
346 static mmu_method_t mmu_booke_methods[] = {
347 	/* pmap dispatcher interface */
348 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
349 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
350 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
351 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
352 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
353 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
354 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
355 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
356 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
357 	MMUMETHOD(mmu_init,		mmu_booke_init),
358 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
359 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
360 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
361 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
362 	MMUMETHOD(mmu_map,		mmu_booke_map),
363 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
364 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
365 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
366 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
367 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
368 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
369 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
370 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
371 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
372 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
373 	MMUMETHOD(mmu_release,		mmu_booke_release),
374 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
375 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
376 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
377 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
378 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
379 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
380 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
381 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
382 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
383 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
384 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
385 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
386 
387 	/* Internal interfaces */
388 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
389 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
390 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
391 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
392 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
393 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
394 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
395 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
396 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
397 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
398 
399 	/* dumpsys() support */
400 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
401 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
402 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
403 
404 	{ 0, 0 }
405 };
406 
407 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
408 
409 static __inline uint32_t
410 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
411 {
412 	uint32_t attrib;
413 	int i;
414 
415 	if (ma != VM_MEMATTR_DEFAULT) {
416 		switch (ma) {
417 		case VM_MEMATTR_UNCACHEABLE:
418 			return (MAS2_I | MAS2_G);
419 		case VM_MEMATTR_WRITE_COMBINING:
420 		case VM_MEMATTR_WRITE_BACK:
421 		case VM_MEMATTR_PREFETCHABLE:
422 			return (MAS2_I);
423 		case VM_MEMATTR_WRITE_THROUGH:
424 			return (MAS2_W | MAS2_M);
425 		case VM_MEMATTR_CACHEABLE:
426 			return (MAS2_M);
427 		}
428 	}
429 
430 	/*
431 	 * Assume the page is cache inhibited and access is guarded unless
432 	 * it's in our available memory array.
433 	 */
434 	attrib = _TLB_ENTRY_IO;
435 	for (i = 0; i < physmem_regions_sz; i++) {
436 		if ((pa >= physmem_regions[i].mr_start) &&
437 		    (pa < (physmem_regions[i].mr_start +
438 		     physmem_regions[i].mr_size))) {
439 			attrib = _TLB_ENTRY_MEM;
440 			break;
441 		}
442 	}
443 
444 	return (attrib);
445 }
446 
447 static inline void
448 tlb_miss_lock(void)
449 {
450 #ifdef SMP
451 	struct pcpu *pc;
452 
453 	if (!smp_started)
454 		return;
455 
456 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
457 		if (pc != pcpup) {
458 
459 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
460 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
461 
462 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
463 			    ("tlb_miss_lock: tried to lock self"));
464 
465 			tlb_lock(pc->pc_booke_tlb_lock);
466 
467 			CTR1(KTR_PMAP, "%s: locked", __func__);
468 		}
469 	}
470 #endif
471 }
472 
473 static inline void
474 tlb_miss_unlock(void)
475 {
476 #ifdef SMP
477 	struct pcpu *pc;
478 
479 	if (!smp_started)
480 		return;
481 
482 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
483 		if (pc != pcpup) {
484 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
485 			    __func__, pc->pc_cpuid);
486 
487 			tlb_unlock(pc->pc_booke_tlb_lock);
488 
489 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
490 		}
491 	}
492 #endif
493 }
494 
495 /* Return number of entries in TLB0. */
496 static __inline void
497 tlb0_get_tlbconf(void)
498 {
499 	uint32_t tlb0_cfg;
500 
501 	tlb0_cfg = mfspr(SPR_TLB0CFG);
502 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
503 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
504 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
505 }
506 
507 /* Return number of entries in TLB1. */
508 static __inline void
509 tlb1_get_tlbconf(void)
510 {
511 	uint32_t tlb1_cfg;
512 
513 	tlb1_cfg = mfspr(SPR_TLB1CFG);
514 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
515 }
516 
517 /**************************************************************************/
518 /* Page table related */
519 /**************************************************************************/
520 
521 /* Initialize pool of kva ptbl buffers. */
522 static void
523 ptbl_init(void)
524 {
525 	int i;
526 
527 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
528 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
529 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
530 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
531 
532 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
533 	TAILQ_INIT(&ptbl_buf_freelist);
534 
535 	for (i = 0; i < PTBL_BUFS; i++) {
536 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
537 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
538 	}
539 }
540 
541 /* Get a ptbl_buf from the freelist. */
542 static struct ptbl_buf *
543 ptbl_buf_alloc(void)
544 {
545 	struct ptbl_buf *buf;
546 
547 	mtx_lock(&ptbl_buf_freelist_lock);
548 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
549 	if (buf != NULL)
550 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
551 	mtx_unlock(&ptbl_buf_freelist_lock);
552 
553 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
554 
555 	return (buf);
556 }
557 
558 /* Return ptbl buff to free pool. */
559 static void
560 ptbl_buf_free(struct ptbl_buf *buf)
561 {
562 
563 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
564 
565 	mtx_lock(&ptbl_buf_freelist_lock);
566 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
567 	mtx_unlock(&ptbl_buf_freelist_lock);
568 }
569 
570 /*
571  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
572  */
573 static void
574 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
575 {
576 	struct ptbl_buf *pbuf;
577 
578 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
579 
580 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
581 
582 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
583 		if (pbuf->kva == (vm_offset_t)ptbl) {
584 			/* Remove from pmap ptbl buf list. */
585 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
586 
587 			/* Free corresponding ptbl buf. */
588 			ptbl_buf_free(pbuf);
589 			break;
590 		}
591 }
592 
593 /* Allocate page table. */
594 static pte_t *
595 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
596 {
597 	vm_page_t mtbl[PTBL_PAGES];
598 	vm_page_t m;
599 	struct ptbl_buf *pbuf;
600 	unsigned int pidx;
601 	pte_t *ptbl;
602 	int i, j;
603 
604 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
605 	    (pmap == kernel_pmap), pdir_idx);
606 
607 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
608 	    ("ptbl_alloc: invalid pdir_idx"));
609 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
610 	    ("pte_alloc: valid ptbl entry exists!"));
611 
612 	pbuf = ptbl_buf_alloc();
613 	if (pbuf == NULL)
614 		panic("pte_alloc: couldn't alloc kernel virtual memory");
615 
616 	ptbl = (pte_t *)pbuf->kva;
617 
618 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
619 
620 	/* Allocate ptbl pages, this will sleep! */
621 	for (i = 0; i < PTBL_PAGES; i++) {
622 		pidx = (PTBL_PAGES * pdir_idx) + i;
623 		while ((m = vm_page_alloc(NULL, pidx,
624 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
625 			PMAP_UNLOCK(pmap);
626 			rw_wunlock(&pvh_global_lock);
627 			if (nosleep) {
628 				ptbl_free_pmap_ptbl(pmap, ptbl);
629 				for (j = 0; j < i; j++)
630 					vm_page_free(mtbl[j]);
631 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
632 				return (NULL);
633 			}
634 			VM_WAIT;
635 			rw_wlock(&pvh_global_lock);
636 			PMAP_LOCK(pmap);
637 		}
638 		mtbl[i] = m;
639 	}
640 
641 	/* Map allocated pages into kernel_pmap. */
642 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
643 
644 	/* Zero whole ptbl. */
645 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
646 
647 	/* Add pbuf to the pmap ptbl bufs list. */
648 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
649 
650 	return (ptbl);
651 }
652 
653 /* Free ptbl pages and invalidate pdir entry. */
654 static void
655 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
656 {
657 	pte_t *ptbl;
658 	vm_paddr_t pa;
659 	vm_offset_t va;
660 	vm_page_t m;
661 	int i;
662 
663 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
664 	    (pmap == kernel_pmap), pdir_idx);
665 
666 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
667 	    ("ptbl_free: invalid pdir_idx"));
668 
669 	ptbl = pmap->pm_pdir[pdir_idx];
670 
671 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
672 
673 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
674 
675 	/*
676 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
677 	 * don't attempt to look up the page tables we are releasing.
678 	 */
679 	mtx_lock_spin(&tlbivax_mutex);
680 	tlb_miss_lock();
681 
682 	pmap->pm_pdir[pdir_idx] = NULL;
683 
684 	tlb_miss_unlock();
685 	mtx_unlock_spin(&tlbivax_mutex);
686 
687 	for (i = 0; i < PTBL_PAGES; i++) {
688 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
689 		pa = pte_vatopa(mmu, kernel_pmap, va);
690 		m = PHYS_TO_VM_PAGE(pa);
691 		vm_page_free_zero(m);
692 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
693 		mmu_booke_kremove(mmu, va);
694 	}
695 
696 	ptbl_free_pmap_ptbl(pmap, ptbl);
697 }
698 
699 /*
700  * Decrement ptbl pages hold count and attempt to free ptbl pages.
701  * Called when removing pte entry from ptbl.
702  *
703  * Return 1 if ptbl pages were freed.
704  */
705 static int
706 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
707 {
708 	pte_t *ptbl;
709 	vm_paddr_t pa;
710 	vm_page_t m;
711 	int i;
712 
713 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
714 	    (pmap == kernel_pmap), pdir_idx);
715 
716 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
717 	    ("ptbl_unhold: invalid pdir_idx"));
718 	KASSERT((pmap != kernel_pmap),
719 	    ("ptbl_unhold: unholding kernel ptbl!"));
720 
721 	ptbl = pmap->pm_pdir[pdir_idx];
722 
723 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
724 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
725 	    ("ptbl_unhold: non kva ptbl"));
726 
727 	/* decrement hold count */
728 	for (i = 0; i < PTBL_PAGES; i++) {
729 		pa = pte_vatopa(mmu, kernel_pmap,
730 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
731 		m = PHYS_TO_VM_PAGE(pa);
732 		m->wire_count--;
733 	}
734 
735 	/*
736 	 * Free ptbl pages if there are no pte etries in this ptbl.
737 	 * wire_count has the same value for all ptbl pages, so check the last
738 	 * page.
739 	 */
740 	if (m->wire_count == 0) {
741 		ptbl_free(mmu, pmap, pdir_idx);
742 
743 		//debugf("ptbl_unhold: e (freed ptbl)\n");
744 		return (1);
745 	}
746 
747 	return (0);
748 }
749 
750 /*
751  * Increment hold count for ptbl pages. This routine is used when a new pte
752  * entry is being inserted into the ptbl.
753  */
754 static void
755 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
756 {
757 	vm_paddr_t pa;
758 	pte_t *ptbl;
759 	vm_page_t m;
760 	int i;
761 
762 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
763 	    pdir_idx);
764 
765 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
766 	    ("ptbl_hold: invalid pdir_idx"));
767 	KASSERT((pmap != kernel_pmap),
768 	    ("ptbl_hold: holding kernel ptbl!"));
769 
770 	ptbl = pmap->pm_pdir[pdir_idx];
771 
772 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
773 
774 	for (i = 0; i < PTBL_PAGES; i++) {
775 		pa = pte_vatopa(mmu, kernel_pmap,
776 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
777 		m = PHYS_TO_VM_PAGE(pa);
778 		m->wire_count++;
779 	}
780 }
781 
782 /* Allocate pv_entry structure. */
783 pv_entry_t
784 pv_alloc(void)
785 {
786 	pv_entry_t pv;
787 
788 	pv_entry_count++;
789 	if (pv_entry_count > pv_entry_high_water)
790 		pagedaemon_wakeup();
791 	pv = uma_zalloc(pvzone, M_NOWAIT);
792 
793 	return (pv);
794 }
795 
796 /* Free pv_entry structure. */
797 static __inline void
798 pv_free(pv_entry_t pve)
799 {
800 
801 	pv_entry_count--;
802 	uma_zfree(pvzone, pve);
803 }
804 
805 
806 /* Allocate and initialize pv_entry structure. */
807 static void
808 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
809 {
810 	pv_entry_t pve;
811 
812 	//int su = (pmap == kernel_pmap);
813 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
814 	//	(u_int32_t)pmap, va, (u_int32_t)m);
815 
816 	pve = pv_alloc();
817 	if (pve == NULL)
818 		panic("pv_insert: no pv entries!");
819 
820 	pve->pv_pmap = pmap;
821 	pve->pv_va = va;
822 
823 	/* add to pv_list */
824 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
825 	rw_assert(&pvh_global_lock, RA_WLOCKED);
826 
827 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
828 
829 	//debugf("pv_insert: e\n");
830 }
831 
832 /* Destroy pv entry. */
833 static void
834 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
835 {
836 	pv_entry_t pve;
837 
838 	//int su = (pmap == kernel_pmap);
839 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
840 
841 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
842 	rw_assert(&pvh_global_lock, RA_WLOCKED);
843 
844 	/* find pv entry */
845 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
846 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
847 			/* remove from pv_list */
848 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
849 			if (TAILQ_EMPTY(&m->md.pv_list))
850 				vm_page_aflag_clear(m, PGA_WRITEABLE);
851 
852 			/* free pv entry struct */
853 			pv_free(pve);
854 			break;
855 		}
856 	}
857 
858 	//debugf("pv_remove: e\n");
859 }
860 
861 /*
862  * Clean pte entry, try to free page table page if requested.
863  *
864  * Return 1 if ptbl pages were freed, otherwise return 0.
865  */
866 static int
867 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
868 {
869 	unsigned int pdir_idx = PDIR_IDX(va);
870 	unsigned int ptbl_idx = PTBL_IDX(va);
871 	vm_page_t m;
872 	pte_t *ptbl;
873 	pte_t *pte;
874 
875 	//int su = (pmap == kernel_pmap);
876 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
877 	//		su, (u_int32_t)pmap, va, flags);
878 
879 	ptbl = pmap->pm_pdir[pdir_idx];
880 	KASSERT(ptbl, ("pte_remove: null ptbl"));
881 
882 	pte = &ptbl[ptbl_idx];
883 
884 	if (pte == NULL || !PTE_ISVALID(pte))
885 		return (0);
886 
887 	if (PTE_ISWIRED(pte))
888 		pmap->pm_stats.wired_count--;
889 
890 	/* Handle managed entry. */
891 	if (PTE_ISMANAGED(pte)) {
892 		/* Get vm_page_t for mapped pte. */
893 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
894 
895 		if (PTE_ISMODIFIED(pte))
896 			vm_page_dirty(m);
897 
898 		if (PTE_ISREFERENCED(pte))
899 			vm_page_aflag_set(m, PGA_REFERENCED);
900 
901 		pv_remove(pmap, va, m);
902 	}
903 
904 	mtx_lock_spin(&tlbivax_mutex);
905 	tlb_miss_lock();
906 
907 	tlb0_flush_entry(va);
908 	*pte = 0;
909 
910 	tlb_miss_unlock();
911 	mtx_unlock_spin(&tlbivax_mutex);
912 
913 	pmap->pm_stats.resident_count--;
914 
915 	if (flags & PTBL_UNHOLD) {
916 		//debugf("pte_remove: e (unhold)\n");
917 		return (ptbl_unhold(mmu, pmap, pdir_idx));
918 	}
919 
920 	//debugf("pte_remove: e\n");
921 	return (0);
922 }
923 
924 /*
925  * Insert PTE for a given page and virtual address.
926  */
927 static int
928 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
929     boolean_t nosleep)
930 {
931 	unsigned int pdir_idx = PDIR_IDX(va);
932 	unsigned int ptbl_idx = PTBL_IDX(va);
933 	pte_t *ptbl, *pte;
934 
935 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
936 	    pmap == kernel_pmap, pmap, va);
937 
938 	/* Get the page table pointer. */
939 	ptbl = pmap->pm_pdir[pdir_idx];
940 
941 	if (ptbl == NULL) {
942 		/* Allocate page table pages. */
943 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
944 		if (ptbl == NULL) {
945 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
946 			return (ENOMEM);
947 		}
948 	} else {
949 		/*
950 		 * Check if there is valid mapping for requested
951 		 * va, if there is, remove it.
952 		 */
953 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
954 		if (PTE_ISVALID(pte)) {
955 			pte_remove(mmu, pmap, va, PTBL_HOLD);
956 		} else {
957 			/*
958 			 * pte is not used, increment hold count
959 			 * for ptbl pages.
960 			 */
961 			if (pmap != kernel_pmap)
962 				ptbl_hold(mmu, pmap, pdir_idx);
963 		}
964 	}
965 
966 	/*
967 	 * Insert pv_entry into pv_list for mapped page if part of managed
968 	 * memory.
969 	 */
970 	if ((m->oflags & VPO_UNMANAGED) == 0) {
971 		flags |= PTE_MANAGED;
972 
973 		/* Create and insert pv entry. */
974 		pv_insert(pmap, va, m);
975 	}
976 
977 	pmap->pm_stats.resident_count++;
978 
979 	mtx_lock_spin(&tlbivax_mutex);
980 	tlb_miss_lock();
981 
982 	tlb0_flush_entry(va);
983 	if (pmap->pm_pdir[pdir_idx] == NULL) {
984 		/*
985 		 * If we just allocated a new page table, hook it in
986 		 * the pdir.
987 		 */
988 		pmap->pm_pdir[pdir_idx] = ptbl;
989 	}
990 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
991 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
992 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
993 
994 	tlb_miss_unlock();
995 	mtx_unlock_spin(&tlbivax_mutex);
996 	return (0);
997 }
998 
999 /* Return the pa for the given pmap/va. */
1000 static vm_paddr_t
1001 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1002 {
1003 	vm_paddr_t pa = 0;
1004 	pte_t *pte;
1005 
1006 	pte = pte_find(mmu, pmap, va);
1007 	if ((pte != NULL) && PTE_ISVALID(pte))
1008 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1009 	return (pa);
1010 }
1011 
1012 /* Get a pointer to a PTE in a page table. */
1013 static pte_t *
1014 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1015 {
1016 	unsigned int pdir_idx = PDIR_IDX(va);
1017 	unsigned int ptbl_idx = PTBL_IDX(va);
1018 
1019 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1020 
1021 	if (pmap->pm_pdir[pdir_idx])
1022 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1023 
1024 	return (NULL);
1025 }
1026 
1027 /* Set up kernel page tables. */
1028 static void
1029 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1030 {
1031 	int		i;
1032 	vm_offset_t	va;
1033 	pte_t		*pte;
1034 
1035 	/* Initialize kernel pdir */
1036 	for (i = 0; i < kernel_ptbls; i++)
1037 		kernel_pmap->pm_pdir[kptbl_min + i] =
1038 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1039 
1040 	/*
1041 	 * Fill in PTEs covering kernel code and data. They are not required
1042 	 * for address translation, as this area is covered by static TLB1
1043 	 * entries, but for pte_vatopa() to work correctly with kernel area
1044 	 * addresses.
1045 	 */
1046 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1047 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1048 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1049 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1050 		    PTE_VALID | PTE_PS_4KB;
1051 	}
1052 }
1053 
1054 /**************************************************************************/
1055 /* PMAP related */
1056 /**************************************************************************/
1057 
1058 /*
1059  * This is called during booke_init, before the system is really initialized.
1060  */
1061 static void
1062 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1063 {
1064 	vm_paddr_t phys_kernelend;
1065 	struct mem_region *mp, *mp1;
1066 	int cnt, i, j;
1067 	vm_paddr_t s, e, sz;
1068 	vm_paddr_t physsz, hwphyssz;
1069 	u_int phys_avail_count;
1070 	vm_size_t kstack0_sz;
1071 	vm_offset_t kernel_pdir, kstack0;
1072 	vm_paddr_t kstack0_phys;
1073 	void *dpcpu;
1074 
1075 	debugf("mmu_booke_bootstrap: entered\n");
1076 
1077 	/* Set interesting system properties */
1078 	hw_direct_map = 0;
1079 	elf32_nxstack = 1;
1080 
1081 	/* Initialize invalidation mutex */
1082 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1083 
1084 	/* Read TLB0 size and associativity. */
1085 	tlb0_get_tlbconf();
1086 
1087 	/*
1088 	 * Align kernel start and end address (kernel image).
1089 	 * Note that kernel end does not necessarily relate to kernsize.
1090 	 * kernsize is the size of the kernel that is actually mapped.
1091 	 */
1092 	kernstart = trunc_page(start);
1093 	data_start = round_page(kernelend);
1094 	data_end = data_start;
1095 
1096 	/*
1097 	 * Addresses of preloaded modules (like file systems) use
1098 	 * physical addresses. Make sure we relocate those into
1099 	 * virtual addresses.
1100 	 */
1101 	preload_addr_relocate = kernstart - kernload;
1102 
1103 	/* Allocate the dynamic per-cpu area. */
1104 	dpcpu = (void *)data_end;
1105 	data_end += DPCPU_SIZE;
1106 
1107 	/* Allocate space for the message buffer. */
1108 	msgbufp = (struct msgbuf *)data_end;
1109 	data_end += msgbufsize;
1110 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1111 	    data_end);
1112 
1113 	data_end = round_page(data_end);
1114 
1115 	/* Allocate space for ptbl_bufs. */
1116 	ptbl_bufs = (struct ptbl_buf *)data_end;
1117 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1118 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1119 	    data_end);
1120 
1121 	data_end = round_page(data_end);
1122 
1123 	/* Allocate PTE tables for kernel KVA. */
1124 	kernel_pdir = data_end;
1125 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1126 	    PDIR_SIZE - 1) / PDIR_SIZE;
1127 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1128 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1129 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1130 
1131 	debugf(" data_end: 0x%08x\n", data_end);
1132 	if (data_end - kernstart > kernsize) {
1133 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1134 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1135 	}
1136 	data_end = kernstart + kernsize;
1137 	debugf(" updated data_end: 0x%08x\n", data_end);
1138 
1139 	/*
1140 	 * Clear the structures - note we can only do it safely after the
1141 	 * possible additional TLB1 translations are in place (above) so that
1142 	 * all range up to the currently calculated 'data_end' is covered.
1143 	 */
1144 	dpcpu_init(dpcpu, 0);
1145 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1146 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1147 
1148 	/*******************************************************/
1149 	/* Set the start and end of kva. */
1150 	/*******************************************************/
1151 	virtual_avail = round_page(data_end);
1152 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1153 
1154 	/* Allocate KVA space for page zero/copy operations. */
1155 	zero_page_va = virtual_avail;
1156 	virtual_avail += PAGE_SIZE;
1157 	zero_page_idle_va = virtual_avail;
1158 	virtual_avail += PAGE_SIZE;
1159 	copy_page_src_va = virtual_avail;
1160 	virtual_avail += PAGE_SIZE;
1161 	copy_page_dst_va = virtual_avail;
1162 	virtual_avail += PAGE_SIZE;
1163 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1164 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1165 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1166 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1167 
1168 	/* Initialize page zero/copy mutexes. */
1169 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1170 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1171 
1172 	/* Allocate KVA space for ptbl bufs. */
1173 	ptbl_buf_pool_vabase = virtual_avail;
1174 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1175 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1176 	    ptbl_buf_pool_vabase, virtual_avail);
1177 
1178 	/* Calculate corresponding physical addresses for the kernel region. */
1179 	phys_kernelend = kernload + kernsize;
1180 	debugf("kernel image and allocated data:\n");
1181 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1182 	debugf(" kernstart   = 0x%08x\n", kernstart);
1183 	debugf(" kernsize    = 0x%08x\n", kernsize);
1184 
1185 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1186 		panic("mmu_booke_bootstrap: phys_avail too small");
1187 
1188 	/*
1189 	 * Remove kernel physical address range from avail regions list. Page
1190 	 * align all regions.  Non-page aligned memory isn't very interesting
1191 	 * to us.  Also, sort the entries for ascending addresses.
1192 	 */
1193 
1194 	/* Retrieve phys/avail mem regions */
1195 	mem_regions(&physmem_regions, &physmem_regions_sz,
1196 	    &availmem_regions, &availmem_regions_sz);
1197 	sz = 0;
1198 	cnt = availmem_regions_sz;
1199 	debugf("processing avail regions:\n");
1200 	for (mp = availmem_regions; mp->mr_size; mp++) {
1201 		s = mp->mr_start;
1202 		e = mp->mr_start + mp->mr_size;
1203 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1204 		/* Check whether this region holds all of the kernel. */
1205 		if (s < kernload && e > phys_kernelend) {
1206 			availmem_regions[cnt].mr_start = phys_kernelend;
1207 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1208 			e = kernload;
1209 		}
1210 		/* Look whether this regions starts within the kernel. */
1211 		if (s >= kernload && s < phys_kernelend) {
1212 			if (e <= phys_kernelend)
1213 				goto empty;
1214 			s = phys_kernelend;
1215 		}
1216 		/* Now look whether this region ends within the kernel. */
1217 		if (e > kernload && e <= phys_kernelend) {
1218 			if (s >= kernload)
1219 				goto empty;
1220 			e = kernload;
1221 		}
1222 		/* Now page align the start and size of the region. */
1223 		s = round_page(s);
1224 		e = trunc_page(e);
1225 		if (e < s)
1226 			e = s;
1227 		sz = e - s;
1228 		debugf("%09jx-%09jx = %jx\n",
1229 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1230 
1231 		/* Check whether some memory is left here. */
1232 		if (sz == 0) {
1233 		empty:
1234 			memmove(mp, mp + 1,
1235 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1236 			cnt--;
1237 			mp--;
1238 			continue;
1239 		}
1240 
1241 		/* Do an insertion sort. */
1242 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1243 			if (s < mp1->mr_start)
1244 				break;
1245 		if (mp1 < mp) {
1246 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1247 			mp1->mr_start = s;
1248 			mp1->mr_size = sz;
1249 		} else {
1250 			mp->mr_start = s;
1251 			mp->mr_size = sz;
1252 		}
1253 	}
1254 	availmem_regions_sz = cnt;
1255 
1256 	/*******************************************************/
1257 	/* Steal physical memory for kernel stack from the end */
1258 	/* of the first avail region                           */
1259 	/*******************************************************/
1260 	kstack0_sz = kstack_pages * PAGE_SIZE;
1261 	kstack0_phys = availmem_regions[0].mr_start +
1262 	    availmem_regions[0].mr_size;
1263 	kstack0_phys -= kstack0_sz;
1264 	availmem_regions[0].mr_size -= kstack0_sz;
1265 
1266 	/*******************************************************/
1267 	/* Fill in phys_avail table, based on availmem_regions */
1268 	/*******************************************************/
1269 	phys_avail_count = 0;
1270 	physsz = 0;
1271 	hwphyssz = 0;
1272 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1273 
1274 	debugf("fill in phys_avail:\n");
1275 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1276 
1277 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1278 		    (uintmax_t)availmem_regions[i].mr_start,
1279 		    (uintmax_t)availmem_regions[i].mr_start +
1280 		        availmem_regions[i].mr_size,
1281 		    (uintmax_t)availmem_regions[i].mr_size);
1282 
1283 		if (hwphyssz != 0 &&
1284 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1285 			debugf(" hw.physmem adjust\n");
1286 			if (physsz < hwphyssz) {
1287 				phys_avail[j] = availmem_regions[i].mr_start;
1288 				phys_avail[j + 1] =
1289 				    availmem_regions[i].mr_start +
1290 				    hwphyssz - physsz;
1291 				physsz = hwphyssz;
1292 				phys_avail_count++;
1293 			}
1294 			break;
1295 		}
1296 
1297 		phys_avail[j] = availmem_regions[i].mr_start;
1298 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1299 		    availmem_regions[i].mr_size;
1300 		phys_avail_count++;
1301 		physsz += availmem_regions[i].mr_size;
1302 	}
1303 	physmem = btoc(physsz);
1304 
1305 	/* Calculate the last available physical address. */
1306 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1307 		;
1308 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1309 
1310 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1311 	debugf("phys_avail_count = %d\n", phys_avail_count);
1312 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1313 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1314 
1315 	/*******************************************************/
1316 	/* Initialize (statically allocated) kernel pmap. */
1317 	/*******************************************************/
1318 	PMAP_LOCK_INIT(kernel_pmap);
1319 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1320 
1321 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1322 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1323 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1324 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1325 
1326 	kernel_pte_alloc(data_end, kernstart, kernel_pdir);
1327 	for (i = 0; i < MAXCPU; i++) {
1328 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1329 
1330 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1331 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1332 	}
1333 
1334 	/* Mark kernel_pmap active on all CPUs */
1335 	CPU_FILL(&kernel_pmap->pm_active);
1336 
1337  	/*
1338 	 * Initialize the global pv list lock.
1339 	 */
1340 	rw_init(&pvh_global_lock, "pmap pv global");
1341 
1342 	/*******************************************************/
1343 	/* Final setup */
1344 	/*******************************************************/
1345 
1346 	/* Enter kstack0 into kernel map, provide guard page */
1347 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1348 	thread0.td_kstack = kstack0;
1349 	thread0.td_kstack_pages = kstack_pages;
1350 
1351 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1352 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1353 	    kstack0_phys, kstack0_phys + kstack0_sz);
1354 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1355 
1356 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1357 	for (i = 0; i < kstack_pages; i++) {
1358 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1359 		kstack0 += PAGE_SIZE;
1360 		kstack0_phys += PAGE_SIZE;
1361 	}
1362 
1363 	pmap_bootstrapped = 1;
1364 
1365 	debugf("virtual_avail = %08x\n", virtual_avail);
1366 	debugf("virtual_end   = %08x\n", virtual_end);
1367 
1368 	debugf("mmu_booke_bootstrap: exit\n");
1369 }
1370 
1371 #ifdef SMP
1372 void
1373 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1374 {
1375 	int i;
1376 
1377 	/*
1378 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1379 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1380 	 * these values directly to (re)program AP's TLB1 hardware.
1381 	 */
1382 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1383 		/* Skip invalid entries */
1384 		if (!(tlb1[i].mas1 & MAS1_VALID))
1385 			continue;
1386 
1387 		tlb1_write_entry(i);
1388 	}
1389 
1390 	set_mas4_defaults();
1391 }
1392 #endif
1393 
1394 static void
1395 booke_pmap_init_qpages(void)
1396 {
1397 	struct pcpu *pc;
1398 	int i;
1399 
1400 	CPU_FOREACH(i) {
1401 		pc = pcpu_find(i);
1402 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1403 		if (pc->pc_qmap_addr == 0)
1404 			panic("pmap_init_qpages: unable to allocate KVA");
1405 	}
1406 }
1407 
1408 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1409 
1410 /*
1411  * Get the physical page address for the given pmap/virtual address.
1412  */
1413 static vm_paddr_t
1414 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1415 {
1416 	vm_paddr_t pa;
1417 
1418 	PMAP_LOCK(pmap);
1419 	pa = pte_vatopa(mmu, pmap, va);
1420 	PMAP_UNLOCK(pmap);
1421 
1422 	return (pa);
1423 }
1424 
1425 /*
1426  * Extract the physical page address associated with the given
1427  * kernel virtual address.
1428  */
1429 static vm_paddr_t
1430 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1431 {
1432 	int i;
1433 
1434 	/* Check TLB1 mappings */
1435 	for (i = 0; i < tlb1_idx; i++) {
1436 		if (!(tlb1[i].mas1 & MAS1_VALID))
1437 			continue;
1438 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1439 			return (tlb1[i].phys + (va - tlb1[i].virt));
1440 	}
1441 
1442 	return (pte_vatopa(mmu, kernel_pmap, va));
1443 }
1444 
1445 /*
1446  * Initialize the pmap module.
1447  * Called by vm_init, to initialize any structures that the pmap
1448  * system needs to map virtual memory.
1449  */
1450 static void
1451 mmu_booke_init(mmu_t mmu)
1452 {
1453 	int shpgperproc = PMAP_SHPGPERPROC;
1454 
1455 	/*
1456 	 * Initialize the address space (zone) for the pv entries.  Set a
1457 	 * high water mark so that the system can recover from excessive
1458 	 * numbers of pv entries.
1459 	 */
1460 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1461 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1462 
1463 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1464 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1465 
1466 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1467 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1468 
1469 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1470 
1471 	/* Pre-fill pvzone with initial number of pv entries. */
1472 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1473 
1474 	/* Initialize ptbl allocation. */
1475 	ptbl_init();
1476 }
1477 
1478 /*
1479  * Map a list of wired pages into kernel virtual address space.  This is
1480  * intended for temporary mappings which do not need page modification or
1481  * references recorded.  Existing mappings in the region are overwritten.
1482  */
1483 static void
1484 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1485 {
1486 	vm_offset_t va;
1487 
1488 	va = sva;
1489 	while (count-- > 0) {
1490 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1491 		va += PAGE_SIZE;
1492 		m++;
1493 	}
1494 }
1495 
1496 /*
1497  * Remove page mappings from kernel virtual address space.  Intended for
1498  * temporary mappings entered by mmu_booke_qenter.
1499  */
1500 static void
1501 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1502 {
1503 	vm_offset_t va;
1504 
1505 	va = sva;
1506 	while (count-- > 0) {
1507 		mmu_booke_kremove(mmu, va);
1508 		va += PAGE_SIZE;
1509 	}
1510 }
1511 
1512 /*
1513  * Map a wired page into kernel virtual address space.
1514  */
1515 static void
1516 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1517 {
1518 
1519 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1520 }
1521 
1522 static void
1523 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1524 {
1525 	uint32_t flags;
1526 	pte_t *pte;
1527 
1528 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1529 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1530 
1531 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1532 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1533 	flags |= PTE_PS_4KB;
1534 
1535 	pte = pte_find(mmu, kernel_pmap, va);
1536 
1537 	mtx_lock_spin(&tlbivax_mutex);
1538 	tlb_miss_lock();
1539 
1540 	if (PTE_ISVALID(pte)) {
1541 
1542 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1543 
1544 		/* Flush entry from TLB0 */
1545 		tlb0_flush_entry(va);
1546 	}
1547 
1548 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1549 
1550 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1551 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1552 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1553 
1554 	/* Flush the real memory from the instruction cache. */
1555 	if ((flags & (PTE_I | PTE_G)) == 0)
1556 		__syncicache((void *)va, PAGE_SIZE);
1557 
1558 	tlb_miss_unlock();
1559 	mtx_unlock_spin(&tlbivax_mutex);
1560 }
1561 
1562 /*
1563  * Remove a page from kernel page table.
1564  */
1565 static void
1566 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1567 {
1568 	pte_t *pte;
1569 
1570 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
1571 
1572 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1573 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1574 	    ("mmu_booke_kremove: invalid va"));
1575 
1576 	pte = pte_find(mmu, kernel_pmap, va);
1577 
1578 	if (!PTE_ISVALID(pte)) {
1579 
1580 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1581 
1582 		return;
1583 	}
1584 
1585 	mtx_lock_spin(&tlbivax_mutex);
1586 	tlb_miss_lock();
1587 
1588 	/* Invalidate entry in TLB0, update PTE. */
1589 	tlb0_flush_entry(va);
1590 	*pte = 0;
1591 
1592 	tlb_miss_unlock();
1593 	mtx_unlock_spin(&tlbivax_mutex);
1594 }
1595 
1596 /*
1597  * Initialize pmap associated with process 0.
1598  */
1599 static void
1600 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1601 {
1602 
1603 	PMAP_LOCK_INIT(pmap);
1604 	mmu_booke_pinit(mmu, pmap);
1605 	PCPU_SET(curpmap, pmap);
1606 }
1607 
1608 /*
1609  * Initialize a preallocated and zeroed pmap structure,
1610  * such as one in a vmspace structure.
1611  */
1612 static void
1613 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1614 {
1615 	int i;
1616 
1617 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1618 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1619 
1620 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1621 
1622 	for (i = 0; i < MAXCPU; i++)
1623 		pmap->pm_tid[i] = TID_NONE;
1624 	CPU_ZERO(&kernel_pmap->pm_active);
1625 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1626 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1627 	TAILQ_INIT(&pmap->pm_ptbl_list);
1628 }
1629 
1630 /*
1631  * Release any resources held by the given physical map.
1632  * Called when a pmap initialized by mmu_booke_pinit is being released.
1633  * Should only be called if the map contains no valid mappings.
1634  */
1635 static void
1636 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1637 {
1638 
1639 	KASSERT(pmap->pm_stats.resident_count == 0,
1640 	    ("pmap_release: pmap resident count %ld != 0",
1641 	    pmap->pm_stats.resident_count));
1642 }
1643 
1644 /*
1645  * Insert the given physical page at the specified virtual address in the
1646  * target physical map with the protection requested. If specified the page
1647  * will be wired down.
1648  */
1649 static int
1650 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1651     vm_prot_t prot, u_int flags, int8_t psind)
1652 {
1653 	int error;
1654 
1655 	rw_wlock(&pvh_global_lock);
1656 	PMAP_LOCK(pmap);
1657 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1658 	rw_wunlock(&pvh_global_lock);
1659 	PMAP_UNLOCK(pmap);
1660 	return (error);
1661 }
1662 
1663 static int
1664 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1665     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1666 {
1667 	pte_t *pte;
1668 	vm_paddr_t pa;
1669 	uint32_t flags;
1670 	int error, su, sync;
1671 
1672 	pa = VM_PAGE_TO_PHYS(m);
1673 	su = (pmap == kernel_pmap);
1674 	sync = 0;
1675 
1676 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1677 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1678 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1679 	//		(u_int32_t)m, va, pa, prot, flags);
1680 
1681 	if (su) {
1682 		KASSERT(((va >= virtual_avail) &&
1683 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1684 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1685 	} else {
1686 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1687 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1688 	}
1689 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1690 		VM_OBJECT_ASSERT_LOCKED(m->object);
1691 
1692 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1693 
1694 	/*
1695 	 * If there is an existing mapping, and the physical address has not
1696 	 * changed, must be protection or wiring change.
1697 	 */
1698 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1699 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1700 
1701 		/*
1702 		 * Before actually updating pte->flags we calculate and
1703 		 * prepare its new value in a helper var.
1704 		 */
1705 		flags = *pte;
1706 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1707 
1708 		/* Wiring change, just update stats. */
1709 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1710 			if (!PTE_ISWIRED(pte)) {
1711 				flags |= PTE_WIRED;
1712 				pmap->pm_stats.wired_count++;
1713 			}
1714 		} else {
1715 			if (PTE_ISWIRED(pte)) {
1716 				flags &= ~PTE_WIRED;
1717 				pmap->pm_stats.wired_count--;
1718 			}
1719 		}
1720 
1721 		if (prot & VM_PROT_WRITE) {
1722 			/* Add write permissions. */
1723 			flags |= PTE_SW;
1724 			if (!su)
1725 				flags |= PTE_UW;
1726 
1727 			if ((flags & PTE_MANAGED) != 0)
1728 				vm_page_aflag_set(m, PGA_WRITEABLE);
1729 		} else {
1730 			/* Handle modified pages, sense modify status. */
1731 
1732 			/*
1733 			 * The PTE_MODIFIED flag could be set by underlying
1734 			 * TLB misses since we last read it (above), possibly
1735 			 * other CPUs could update it so we check in the PTE
1736 			 * directly rather than rely on that saved local flags
1737 			 * copy.
1738 			 */
1739 			if (PTE_ISMODIFIED(pte))
1740 				vm_page_dirty(m);
1741 		}
1742 
1743 		if (prot & VM_PROT_EXECUTE) {
1744 			flags |= PTE_SX;
1745 			if (!su)
1746 				flags |= PTE_UX;
1747 
1748 			/*
1749 			 * Check existing flags for execute permissions: if we
1750 			 * are turning execute permissions on, icache should
1751 			 * be flushed.
1752 			 */
1753 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1754 				sync++;
1755 		}
1756 
1757 		flags &= ~PTE_REFERENCED;
1758 
1759 		/*
1760 		 * The new flags value is all calculated -- only now actually
1761 		 * update the PTE.
1762 		 */
1763 		mtx_lock_spin(&tlbivax_mutex);
1764 		tlb_miss_lock();
1765 
1766 		tlb0_flush_entry(va);
1767 		*pte &= ~PTE_FLAGS_MASK;
1768 		*pte |= flags;
1769 
1770 		tlb_miss_unlock();
1771 		mtx_unlock_spin(&tlbivax_mutex);
1772 
1773 	} else {
1774 		/*
1775 		 * If there is an existing mapping, but it's for a different
1776 		 * physical address, pte_enter() will delete the old mapping.
1777 		 */
1778 		//if ((pte != NULL) && PTE_ISVALID(pte))
1779 		//	debugf("mmu_booke_enter_locked: replace\n");
1780 		//else
1781 		//	debugf("mmu_booke_enter_locked: new\n");
1782 
1783 		/* Now set up the flags and install the new mapping. */
1784 		flags = (PTE_SR | PTE_VALID);
1785 		flags |= PTE_M;
1786 
1787 		if (!su)
1788 			flags |= PTE_UR;
1789 
1790 		if (prot & VM_PROT_WRITE) {
1791 			flags |= PTE_SW;
1792 			if (!su)
1793 				flags |= PTE_UW;
1794 
1795 			if ((m->oflags & VPO_UNMANAGED) == 0)
1796 				vm_page_aflag_set(m, PGA_WRITEABLE);
1797 		}
1798 
1799 		if (prot & VM_PROT_EXECUTE) {
1800 			flags |= PTE_SX;
1801 			if (!su)
1802 				flags |= PTE_UX;
1803 		}
1804 
1805 		/* If its wired update stats. */
1806 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1807 			flags |= PTE_WIRED;
1808 
1809 		error = pte_enter(mmu, pmap, m, va, flags,
1810 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1811 		if (error != 0)
1812 			return (KERN_RESOURCE_SHORTAGE);
1813 
1814 		if ((flags & PMAP_ENTER_WIRED) != 0)
1815 			pmap->pm_stats.wired_count++;
1816 
1817 		/* Flush the real memory from the instruction cache. */
1818 		if (prot & VM_PROT_EXECUTE)
1819 			sync++;
1820 	}
1821 
1822 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1823 		__syncicache((void *)va, PAGE_SIZE);
1824 		sync = 0;
1825 	}
1826 
1827 	return (KERN_SUCCESS);
1828 }
1829 
1830 /*
1831  * Maps a sequence of resident pages belonging to the same object.
1832  * The sequence begins with the given page m_start.  This page is
1833  * mapped at the given virtual address start.  Each subsequent page is
1834  * mapped at a virtual address that is offset from start by the same
1835  * amount as the page is offset from m_start within the object.  The
1836  * last page in the sequence is the page with the largest offset from
1837  * m_start that can be mapped at a virtual address less than the given
1838  * virtual address end.  Not every virtual page between start and end
1839  * is mapped; only those for which a resident page exists with the
1840  * corresponding offset from m_start are mapped.
1841  */
1842 static void
1843 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1844     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1845 {
1846 	vm_page_t m;
1847 	vm_pindex_t diff, psize;
1848 
1849 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1850 
1851 	psize = atop(end - start);
1852 	m = m_start;
1853 	rw_wlock(&pvh_global_lock);
1854 	PMAP_LOCK(pmap);
1855 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1856 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1857 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1858 		    PMAP_ENTER_NOSLEEP, 0);
1859 		m = TAILQ_NEXT(m, listq);
1860 	}
1861 	rw_wunlock(&pvh_global_lock);
1862 	PMAP_UNLOCK(pmap);
1863 }
1864 
1865 static void
1866 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1867     vm_prot_t prot)
1868 {
1869 
1870 	rw_wlock(&pvh_global_lock);
1871 	PMAP_LOCK(pmap);
1872 	mmu_booke_enter_locked(mmu, pmap, va, m,
1873 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1874 	    0);
1875 	rw_wunlock(&pvh_global_lock);
1876 	PMAP_UNLOCK(pmap);
1877 }
1878 
1879 /*
1880  * Remove the given range of addresses from the specified map.
1881  *
1882  * It is assumed that the start and end are properly rounded to the page size.
1883  */
1884 static void
1885 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1886 {
1887 	pte_t *pte;
1888 	uint8_t hold_flag;
1889 
1890 	int su = (pmap == kernel_pmap);
1891 
1892 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1893 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1894 
1895 	if (su) {
1896 		KASSERT(((va >= virtual_avail) &&
1897 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1898 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1899 	} else {
1900 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1901 		    ("mmu_booke_remove: user pmap, non user va"));
1902 	}
1903 
1904 	if (PMAP_REMOVE_DONE(pmap)) {
1905 		//debugf("mmu_booke_remove: e (empty)\n");
1906 		return;
1907 	}
1908 
1909 	hold_flag = PTBL_HOLD_FLAG(pmap);
1910 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1911 
1912 	rw_wlock(&pvh_global_lock);
1913 	PMAP_LOCK(pmap);
1914 	for (; va < endva; va += PAGE_SIZE) {
1915 		pte = pte_find(mmu, pmap, va);
1916 		if ((pte != NULL) && PTE_ISVALID(pte))
1917 			pte_remove(mmu, pmap, va, hold_flag);
1918 	}
1919 	PMAP_UNLOCK(pmap);
1920 	rw_wunlock(&pvh_global_lock);
1921 
1922 	//debugf("mmu_booke_remove: e\n");
1923 }
1924 
1925 /*
1926  * Remove physical page from all pmaps in which it resides.
1927  */
1928 static void
1929 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1930 {
1931 	pv_entry_t pv, pvn;
1932 	uint8_t hold_flag;
1933 
1934 	rw_wlock(&pvh_global_lock);
1935 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1936 		pvn = TAILQ_NEXT(pv, pv_link);
1937 
1938 		PMAP_LOCK(pv->pv_pmap);
1939 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1940 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1941 		PMAP_UNLOCK(pv->pv_pmap);
1942 	}
1943 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1944 	rw_wunlock(&pvh_global_lock);
1945 }
1946 
1947 /*
1948  * Map a range of physical addresses into kernel virtual address space.
1949  */
1950 static vm_offset_t
1951 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1952     vm_paddr_t pa_end, int prot)
1953 {
1954 	vm_offset_t sva = *virt;
1955 	vm_offset_t va = sva;
1956 
1957 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1958 	//		sva, pa_start, pa_end);
1959 
1960 	while (pa_start < pa_end) {
1961 		mmu_booke_kenter(mmu, va, pa_start);
1962 		va += PAGE_SIZE;
1963 		pa_start += PAGE_SIZE;
1964 	}
1965 	*virt = va;
1966 
1967 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1968 	return (sva);
1969 }
1970 
1971 /*
1972  * The pmap must be activated before it's address space can be accessed in any
1973  * way.
1974  */
1975 static void
1976 mmu_booke_activate(mmu_t mmu, struct thread *td)
1977 {
1978 	pmap_t pmap;
1979 	u_int cpuid;
1980 
1981 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1982 
1983 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1984 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1985 
1986 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1987 
1988 	sched_pin();
1989 
1990 	cpuid = PCPU_GET(cpuid);
1991 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1992 	PCPU_SET(curpmap, pmap);
1993 
1994 	if (pmap->pm_tid[cpuid] == TID_NONE)
1995 		tid_alloc(pmap);
1996 
1997 	/* Load PID0 register with pmap tid value. */
1998 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1999 	__asm __volatile("isync");
2000 
2001 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2002 
2003 	sched_unpin();
2004 
2005 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2006 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2007 }
2008 
2009 /*
2010  * Deactivate the specified process's address space.
2011  */
2012 static void
2013 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2014 {
2015 	pmap_t pmap;
2016 
2017 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2018 
2019 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2020 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2021 
2022 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2023 
2024 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2025 	PCPU_SET(curpmap, NULL);
2026 }
2027 
2028 /*
2029  * Copy the range specified by src_addr/len
2030  * from the source map to the range dst_addr/len
2031  * in the destination map.
2032  *
2033  * This routine is only advisory and need not do anything.
2034  */
2035 static void
2036 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2037     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2038 {
2039 
2040 }
2041 
2042 /*
2043  * Set the physical protection on the specified range of this map as requested.
2044  */
2045 static void
2046 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2047     vm_prot_t prot)
2048 {
2049 	vm_offset_t va;
2050 	vm_page_t m;
2051 	pte_t *pte;
2052 
2053 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2054 		mmu_booke_remove(mmu, pmap, sva, eva);
2055 		return;
2056 	}
2057 
2058 	if (prot & VM_PROT_WRITE)
2059 		return;
2060 
2061 	PMAP_LOCK(pmap);
2062 	for (va = sva; va < eva; va += PAGE_SIZE) {
2063 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2064 			if (PTE_ISVALID(pte)) {
2065 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2066 
2067 				mtx_lock_spin(&tlbivax_mutex);
2068 				tlb_miss_lock();
2069 
2070 				/* Handle modified pages. */
2071 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2072 					vm_page_dirty(m);
2073 
2074 				tlb0_flush_entry(va);
2075 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2076 
2077 				tlb_miss_unlock();
2078 				mtx_unlock_spin(&tlbivax_mutex);
2079 			}
2080 		}
2081 	}
2082 	PMAP_UNLOCK(pmap);
2083 }
2084 
2085 /*
2086  * Clear the write and modified bits in each of the given page's mappings.
2087  */
2088 static void
2089 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2090 {
2091 	pv_entry_t pv;
2092 	pte_t *pte;
2093 
2094 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2095 	    ("mmu_booke_remove_write: page %p is not managed", m));
2096 
2097 	/*
2098 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2099 	 * set by another thread while the object is locked.  Thus,
2100 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2101 	 */
2102 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2103 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2104 		return;
2105 	rw_wlock(&pvh_global_lock);
2106 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2107 		PMAP_LOCK(pv->pv_pmap);
2108 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2109 			if (PTE_ISVALID(pte)) {
2110 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2111 
2112 				mtx_lock_spin(&tlbivax_mutex);
2113 				tlb_miss_lock();
2114 
2115 				/* Handle modified pages. */
2116 				if (PTE_ISMODIFIED(pte))
2117 					vm_page_dirty(m);
2118 
2119 				/* Flush mapping from TLB0. */
2120 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2121 
2122 				tlb_miss_unlock();
2123 				mtx_unlock_spin(&tlbivax_mutex);
2124 			}
2125 		}
2126 		PMAP_UNLOCK(pv->pv_pmap);
2127 	}
2128 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2129 	rw_wunlock(&pvh_global_lock);
2130 }
2131 
2132 static void
2133 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2134 {
2135 	pte_t *pte;
2136 	pmap_t pmap;
2137 	vm_page_t m;
2138 	vm_offset_t addr;
2139 	vm_paddr_t pa = 0;
2140 	int active, valid;
2141 
2142 	va = trunc_page(va);
2143 	sz = round_page(sz);
2144 
2145 	rw_wlock(&pvh_global_lock);
2146 	pmap = PCPU_GET(curpmap);
2147 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2148 	while (sz > 0) {
2149 		PMAP_LOCK(pm);
2150 		pte = pte_find(mmu, pm, va);
2151 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2152 		if (valid)
2153 			pa = PTE_PA(pte);
2154 		PMAP_UNLOCK(pm);
2155 		if (valid) {
2156 			if (!active) {
2157 				/* Create a mapping in the active pmap. */
2158 				addr = 0;
2159 				m = PHYS_TO_VM_PAGE(pa);
2160 				PMAP_LOCK(pmap);
2161 				pte_enter(mmu, pmap, m, addr,
2162 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2163 				__syncicache((void *)addr, PAGE_SIZE);
2164 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2165 				PMAP_UNLOCK(pmap);
2166 			} else
2167 				__syncicache((void *)va, PAGE_SIZE);
2168 		}
2169 		va += PAGE_SIZE;
2170 		sz -= PAGE_SIZE;
2171 	}
2172 	rw_wunlock(&pvh_global_lock);
2173 }
2174 
2175 /*
2176  * Atomically extract and hold the physical page with the given
2177  * pmap and virtual address pair if that mapping permits the given
2178  * protection.
2179  */
2180 static vm_page_t
2181 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2182     vm_prot_t prot)
2183 {
2184 	pte_t *pte;
2185 	vm_page_t m;
2186 	uint32_t pte_wbit;
2187 	vm_paddr_t pa;
2188 
2189 	m = NULL;
2190 	pa = 0;
2191 	PMAP_LOCK(pmap);
2192 retry:
2193 	pte = pte_find(mmu, pmap, va);
2194 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2195 		if (pmap == kernel_pmap)
2196 			pte_wbit = PTE_SW;
2197 		else
2198 			pte_wbit = PTE_UW;
2199 
2200 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2201 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2202 				goto retry;
2203 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2204 			vm_page_hold(m);
2205 		}
2206 	}
2207 
2208 	PA_UNLOCK_COND(pa);
2209 	PMAP_UNLOCK(pmap);
2210 	return (m);
2211 }
2212 
2213 /*
2214  * Initialize a vm_page's machine-dependent fields.
2215  */
2216 static void
2217 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2218 {
2219 
2220 	TAILQ_INIT(&m->md.pv_list);
2221 }
2222 
2223 /*
2224  * mmu_booke_zero_page_area zeros the specified hardware page by
2225  * mapping it into virtual memory and using bzero to clear
2226  * its contents.
2227  *
2228  * off and size must reside within a single page.
2229  */
2230 static void
2231 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2232 {
2233 	vm_offset_t va;
2234 
2235 	/* XXX KASSERT off and size are within a single page? */
2236 
2237 	mtx_lock(&zero_page_mutex);
2238 	va = zero_page_va;
2239 
2240 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2241 	bzero((caddr_t)va + off, size);
2242 	mmu_booke_kremove(mmu, va);
2243 
2244 	mtx_unlock(&zero_page_mutex);
2245 }
2246 
2247 /*
2248  * mmu_booke_zero_page zeros the specified hardware page.
2249  */
2250 static void
2251 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2252 {
2253 	vm_offset_t off, va;
2254 
2255 	mtx_lock(&zero_page_mutex);
2256 	va = zero_page_va;
2257 
2258 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2259 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2260 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2261 	mmu_booke_kremove(mmu, va);
2262 
2263 	mtx_unlock(&zero_page_mutex);
2264 }
2265 
2266 /*
2267  * mmu_booke_copy_page copies the specified (machine independent) page by
2268  * mapping the page into virtual memory and using memcopy to copy the page,
2269  * one machine dependent page at a time.
2270  */
2271 static void
2272 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2273 {
2274 	vm_offset_t sva, dva;
2275 
2276 	sva = copy_page_src_va;
2277 	dva = copy_page_dst_va;
2278 
2279 	mtx_lock(&copy_page_mutex);
2280 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2281 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2282 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2283 	mmu_booke_kremove(mmu, dva);
2284 	mmu_booke_kremove(mmu, sva);
2285 	mtx_unlock(&copy_page_mutex);
2286 }
2287 
2288 static inline void
2289 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2290     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2291 {
2292 	void *a_cp, *b_cp;
2293 	vm_offset_t a_pg_offset, b_pg_offset;
2294 	int cnt;
2295 
2296 	mtx_lock(&copy_page_mutex);
2297 	while (xfersize > 0) {
2298 		a_pg_offset = a_offset & PAGE_MASK;
2299 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2300 		mmu_booke_kenter(mmu, copy_page_src_va,
2301 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2302 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2303 		b_pg_offset = b_offset & PAGE_MASK;
2304 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2305 		mmu_booke_kenter(mmu, copy_page_dst_va,
2306 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2307 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2308 		bcopy(a_cp, b_cp, cnt);
2309 		mmu_booke_kremove(mmu, copy_page_dst_va);
2310 		mmu_booke_kremove(mmu, copy_page_src_va);
2311 		a_offset += cnt;
2312 		b_offset += cnt;
2313 		xfersize -= cnt;
2314 	}
2315 	mtx_unlock(&copy_page_mutex);
2316 }
2317 
2318 /*
2319  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2320  * into virtual memory and using bzero to clear its contents. This is intended
2321  * to be called from the vm_pagezero process only and outside of Giant. No
2322  * lock is required.
2323  */
2324 static void
2325 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2326 {
2327 	vm_offset_t va;
2328 
2329 	va = zero_page_idle_va;
2330 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2331 	bzero((caddr_t)va, PAGE_SIZE);
2332 	mmu_booke_kremove(mmu, va);
2333 }
2334 
2335 static vm_offset_t
2336 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2337 {
2338 	vm_paddr_t paddr;
2339 	vm_offset_t qaddr;
2340 	uint32_t flags;
2341 	pte_t *pte;
2342 
2343 	paddr = VM_PAGE_TO_PHYS(m);
2344 
2345 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2346 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2347 	flags |= PTE_PS_4KB;
2348 
2349 	critical_enter();
2350 	qaddr = PCPU_GET(qmap_addr);
2351 
2352 	pte = pte_find(mmu, kernel_pmap, qaddr);
2353 
2354 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2355 
2356 	/*
2357 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2358  	 * not be present in other TLBs.  Is there a better instruction
2359 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2360 	 */
2361 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2362 	__asm __volatile("isync; msync");
2363 
2364 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
2365 
2366 	/* Flush the real memory from the instruction cache. */
2367 	if ((flags & (PTE_I | PTE_G)) == 0)
2368 		__syncicache((void *)qaddr, PAGE_SIZE);
2369 
2370 	return (qaddr);
2371 }
2372 
2373 static void
2374 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2375 {
2376 	pte_t *pte;
2377 
2378 	pte = pte_find(mmu, kernel_pmap, addr);
2379 
2380 	KASSERT(PCPU_GET(qmap_addr) == addr,
2381 	    ("mmu_booke_quick_remove_page: invalid address"));
2382 	KASSERT(*pte != 0,
2383 	    ("mmu_booke_quick_remove_page: PTE not in use"));
2384 
2385 	*pte = 0;
2386 	critical_exit();
2387 }
2388 
2389 /*
2390  * Return whether or not the specified physical page was modified
2391  * in any of physical maps.
2392  */
2393 static boolean_t
2394 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2395 {
2396 	pte_t *pte;
2397 	pv_entry_t pv;
2398 	boolean_t rv;
2399 
2400 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2401 	    ("mmu_booke_is_modified: page %p is not managed", m));
2402 	rv = FALSE;
2403 
2404 	/*
2405 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2406 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2407 	 * is clear, no PTEs can be modified.
2408 	 */
2409 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2410 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2411 		return (rv);
2412 	rw_wlock(&pvh_global_lock);
2413 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2414 		PMAP_LOCK(pv->pv_pmap);
2415 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2416 		    PTE_ISVALID(pte)) {
2417 			if (PTE_ISMODIFIED(pte))
2418 				rv = TRUE;
2419 		}
2420 		PMAP_UNLOCK(pv->pv_pmap);
2421 		if (rv)
2422 			break;
2423 	}
2424 	rw_wunlock(&pvh_global_lock);
2425 	return (rv);
2426 }
2427 
2428 /*
2429  * Return whether or not the specified virtual address is eligible
2430  * for prefault.
2431  */
2432 static boolean_t
2433 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2434 {
2435 
2436 	return (FALSE);
2437 }
2438 
2439 /*
2440  * Return whether or not the specified physical page was referenced
2441  * in any physical maps.
2442  */
2443 static boolean_t
2444 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2445 {
2446 	pte_t *pte;
2447 	pv_entry_t pv;
2448 	boolean_t rv;
2449 
2450 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2451 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2452 	rv = FALSE;
2453 	rw_wlock(&pvh_global_lock);
2454 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2455 		PMAP_LOCK(pv->pv_pmap);
2456 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2457 		    PTE_ISVALID(pte)) {
2458 			if (PTE_ISREFERENCED(pte))
2459 				rv = TRUE;
2460 		}
2461 		PMAP_UNLOCK(pv->pv_pmap);
2462 		if (rv)
2463 			break;
2464 	}
2465 	rw_wunlock(&pvh_global_lock);
2466 	return (rv);
2467 }
2468 
2469 /*
2470  * Clear the modify bits on the specified physical page.
2471  */
2472 static void
2473 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2474 {
2475 	pte_t *pte;
2476 	pv_entry_t pv;
2477 
2478 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2479 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2480 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2481 	KASSERT(!vm_page_xbusied(m),
2482 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2483 
2484 	/*
2485 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2486 	 * If the object containing the page is locked and the page is not
2487 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2488 	 */
2489 	if ((m->aflags & PGA_WRITEABLE) == 0)
2490 		return;
2491 	rw_wlock(&pvh_global_lock);
2492 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2493 		PMAP_LOCK(pv->pv_pmap);
2494 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2495 		    PTE_ISVALID(pte)) {
2496 			mtx_lock_spin(&tlbivax_mutex);
2497 			tlb_miss_lock();
2498 
2499 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2500 				tlb0_flush_entry(pv->pv_va);
2501 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2502 				    PTE_REFERENCED);
2503 			}
2504 
2505 			tlb_miss_unlock();
2506 			mtx_unlock_spin(&tlbivax_mutex);
2507 		}
2508 		PMAP_UNLOCK(pv->pv_pmap);
2509 	}
2510 	rw_wunlock(&pvh_global_lock);
2511 }
2512 
2513 /*
2514  * Return a count of reference bits for a page, clearing those bits.
2515  * It is not necessary for every reference bit to be cleared, but it
2516  * is necessary that 0 only be returned when there are truly no
2517  * reference bits set.
2518  *
2519  * XXX: The exact number of bits to check and clear is a matter that
2520  * should be tested and standardized at some point in the future for
2521  * optimal aging of shared pages.
2522  */
2523 static int
2524 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2525 {
2526 	pte_t *pte;
2527 	pv_entry_t pv;
2528 	int count;
2529 
2530 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2531 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2532 	count = 0;
2533 	rw_wlock(&pvh_global_lock);
2534 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2535 		PMAP_LOCK(pv->pv_pmap);
2536 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2537 		    PTE_ISVALID(pte)) {
2538 			if (PTE_ISREFERENCED(pte)) {
2539 				mtx_lock_spin(&tlbivax_mutex);
2540 				tlb_miss_lock();
2541 
2542 				tlb0_flush_entry(pv->pv_va);
2543 				*pte &= ~PTE_REFERENCED;
2544 
2545 				tlb_miss_unlock();
2546 				mtx_unlock_spin(&tlbivax_mutex);
2547 
2548 				if (++count > 4) {
2549 					PMAP_UNLOCK(pv->pv_pmap);
2550 					break;
2551 				}
2552 			}
2553 		}
2554 		PMAP_UNLOCK(pv->pv_pmap);
2555 	}
2556 	rw_wunlock(&pvh_global_lock);
2557 	return (count);
2558 }
2559 
2560 /*
2561  * Clear the wired attribute from the mappings for the specified range of
2562  * addresses in the given pmap.  Every valid mapping within that range must
2563  * have the wired attribute set.  In contrast, invalid mappings cannot have
2564  * the wired attribute set, so they are ignored.
2565  *
2566  * The wired attribute of the page table entry is not a hardware feature, so
2567  * there is no need to invalidate any TLB entries.
2568  */
2569 static void
2570 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2571 {
2572 	vm_offset_t va;
2573 	pte_t *pte;
2574 
2575 	PMAP_LOCK(pmap);
2576 	for (va = sva; va < eva; va += PAGE_SIZE) {
2577 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2578 		    PTE_ISVALID(pte)) {
2579 			if (!PTE_ISWIRED(pte))
2580 				panic("mmu_booke_unwire: pte %p isn't wired",
2581 				    pte);
2582 			*pte &= ~PTE_WIRED;
2583 			pmap->pm_stats.wired_count--;
2584 		}
2585 	}
2586 	PMAP_UNLOCK(pmap);
2587 
2588 }
2589 
2590 /*
2591  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2592  * page.  This count may be changed upwards or downwards in the future; it is
2593  * only necessary that true be returned for a small subset of pmaps for proper
2594  * page aging.
2595  */
2596 static boolean_t
2597 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2598 {
2599 	pv_entry_t pv;
2600 	int loops;
2601 	boolean_t rv;
2602 
2603 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2604 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2605 	loops = 0;
2606 	rv = FALSE;
2607 	rw_wlock(&pvh_global_lock);
2608 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2609 		if (pv->pv_pmap == pmap) {
2610 			rv = TRUE;
2611 			break;
2612 		}
2613 		if (++loops >= 16)
2614 			break;
2615 	}
2616 	rw_wunlock(&pvh_global_lock);
2617 	return (rv);
2618 }
2619 
2620 /*
2621  * Return the number of managed mappings to the given physical page that are
2622  * wired.
2623  */
2624 static int
2625 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2626 {
2627 	pv_entry_t pv;
2628 	pte_t *pte;
2629 	int count = 0;
2630 
2631 	if ((m->oflags & VPO_UNMANAGED) != 0)
2632 		return (count);
2633 	rw_wlock(&pvh_global_lock);
2634 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2635 		PMAP_LOCK(pv->pv_pmap);
2636 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2637 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2638 				count++;
2639 		PMAP_UNLOCK(pv->pv_pmap);
2640 	}
2641 	rw_wunlock(&pvh_global_lock);
2642 	return (count);
2643 }
2644 
2645 static int
2646 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2647 {
2648 	int i;
2649 	vm_offset_t va;
2650 
2651 	/*
2652 	 * This currently does not work for entries that
2653 	 * overlap TLB1 entries.
2654 	 */
2655 	for (i = 0; i < tlb1_idx; i ++) {
2656 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2657 			return (0);
2658 	}
2659 
2660 	return (EFAULT);
2661 }
2662 
2663 void
2664 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2665 {
2666 	vm_paddr_t ppa;
2667 	vm_offset_t ofs;
2668 	vm_size_t gran;
2669 
2670 	/* Minidumps are based on virtual memory addresses. */
2671 	if (do_minidump) {
2672 		*va = (void *)(vm_offset_t)pa;
2673 		return;
2674 	}
2675 
2676 	/* Raw physical memory dumps don't have a virtual address. */
2677 	/* We always map a 256MB page at 256M. */
2678 	gran = 256 * 1024 * 1024;
2679 	ppa = pa & ~(gran - 1);
2680 	ofs = pa - ppa;
2681 	*va = (void *)gran;
2682 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2683 
2684 	if (sz > (gran - ofs))
2685 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2686 		    _TLB_ENTRY_IO);
2687 }
2688 
2689 void
2690 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2691 {
2692 	vm_paddr_t ppa;
2693 	vm_offset_t ofs;
2694 	vm_size_t gran;
2695 
2696 	/* Minidumps are based on virtual memory addresses. */
2697 	/* Nothing to do... */
2698 	if (do_minidump)
2699 		return;
2700 
2701 	/* Raw physical memory dumps don't have a virtual address. */
2702 	tlb1_idx--;
2703 	tlb1[tlb1_idx].mas1 = 0;
2704 	tlb1[tlb1_idx].mas2 = 0;
2705 	tlb1[tlb1_idx].mas3 = 0;
2706 	tlb1_write_entry(tlb1_idx);
2707 
2708 	gran = 256 * 1024 * 1024;
2709 	ppa = pa & ~(gran - 1);
2710 	ofs = pa - ppa;
2711 	if (sz > (gran - ofs)) {
2712 		tlb1_idx--;
2713 		tlb1[tlb1_idx].mas1 = 0;
2714 		tlb1[tlb1_idx].mas2 = 0;
2715 		tlb1[tlb1_idx].mas3 = 0;
2716 		tlb1_write_entry(tlb1_idx);
2717 	}
2718 }
2719 
2720 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2721 
2722 void
2723 mmu_booke_scan_init(mmu_t mmu)
2724 {
2725 	vm_offset_t va;
2726 	pte_t *pte;
2727 	int i;
2728 
2729 	if (!do_minidump) {
2730 		/* Initialize phys. segments for dumpsys(). */
2731 		memset(&dump_map, 0, sizeof(dump_map));
2732 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2733 		    &availmem_regions_sz);
2734 		for (i = 0; i < physmem_regions_sz; i++) {
2735 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2736 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2737 		}
2738 		return;
2739 	}
2740 
2741 	/* Virtual segments for minidumps: */
2742 	memset(&dump_map, 0, sizeof(dump_map));
2743 
2744 	/* 1st: kernel .data and .bss. */
2745 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2746 	dump_map[0].pa_size =
2747 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2748 
2749 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2750 	dump_map[1].pa_start = data_start;
2751 	dump_map[1].pa_size = data_end - data_start;
2752 
2753 	/* 3rd: kernel VM. */
2754 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2755 	/* Find start of next chunk (from va). */
2756 	while (va < virtual_end) {
2757 		/* Don't dump the buffer cache. */
2758 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2759 			va = kmi.buffer_eva;
2760 			continue;
2761 		}
2762 		pte = pte_find(mmu, kernel_pmap, va);
2763 		if (pte != NULL && PTE_ISVALID(pte))
2764 			break;
2765 		va += PAGE_SIZE;
2766 	}
2767 	if (va < virtual_end) {
2768 		dump_map[2].pa_start = va;
2769 		va += PAGE_SIZE;
2770 		/* Find last page in chunk. */
2771 		while (va < virtual_end) {
2772 			/* Don't run into the buffer cache. */
2773 			if (va == kmi.buffer_sva)
2774 				break;
2775 			pte = pte_find(mmu, kernel_pmap, va);
2776 			if (pte == NULL || !PTE_ISVALID(pte))
2777 				break;
2778 			va += PAGE_SIZE;
2779 		}
2780 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2781 	}
2782 }
2783 
2784 /*
2785  * Map a set of physical memory pages into the kernel virtual address space.
2786  * Return a pointer to where it is mapped. This routine is intended to be used
2787  * for mapping device memory, NOT real memory.
2788  */
2789 static void *
2790 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2791 {
2792 
2793 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2794 }
2795 
2796 static void *
2797 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2798 {
2799 	void *res;
2800 	uintptr_t va, tmpva;
2801 	vm_size_t sz;
2802 	int i;
2803 
2804 	/*
2805 	 * Check if this is premapped in TLB1. Note: this should probably also
2806 	 * check whether a sequence of TLB1 entries exist that match the
2807 	 * requirement, but now only checks the easy case.
2808 	 */
2809 	if (ma == VM_MEMATTR_DEFAULT) {
2810 		for (i = 0; i < tlb1_idx; i++) {
2811 			if (!(tlb1[i].mas1 & MAS1_VALID))
2812 				continue;
2813 			if (pa >= tlb1[i].phys &&
2814 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2815 				return (void *)(tlb1[i].virt +
2816 				    (vm_offset_t)(pa - tlb1[i].phys));
2817 		}
2818 	}
2819 
2820 	size = roundup(size, PAGE_SIZE);
2821 
2822 	/*
2823 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
2824 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
2825 	 */
2826 #ifdef SPARSE_MAPDEV
2827 	/*
2828 	 * With a sparse mapdev, align to the largest starting region.  This
2829 	 * could feasibly be optimized for a 'best-fit' alignment, but that
2830 	 * calculation could be very costly.
2831 	 */
2832 	do {
2833 	    tmpva = tlb1_map_base;
2834 	    va = roundup(tlb1_map_base, 1 << flsl(size));
2835 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
2836 #else
2837 	va = atomic_fetchadd_int(&tlb1_map_base, size);
2838 #endif
2839 	res = (void *)va;
2840 
2841 	do {
2842 		sz = 1 << (ilog2(size) & ~1);
2843 		if (va % sz != 0) {
2844 			do {
2845 				sz >>= 2;
2846 			} while (va % sz != 0);
2847 		}
2848 		if (bootverbose)
2849 			printf("Wiring VA=%x to PA=%jx (size=%x), "
2850 			    "using TLB1[%d]\n", va, (uintmax_t)pa, sz, tlb1_idx);
2851 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2852 		size -= sz;
2853 		pa += sz;
2854 		va += sz;
2855 	} while (size > 0);
2856 
2857 	return (res);
2858 }
2859 
2860 /*
2861  * 'Unmap' a range mapped by mmu_booke_mapdev().
2862  */
2863 static void
2864 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2865 {
2866 #ifdef SUPPORTS_SHRINKING_TLB1
2867 	vm_offset_t base, offset;
2868 
2869 	/*
2870 	 * Unmap only if this is inside kernel virtual space.
2871 	 */
2872 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2873 		base = trunc_page(va);
2874 		offset = va & PAGE_MASK;
2875 		size = roundup(offset + size, PAGE_SIZE);
2876 		kva_free(base, size);
2877 	}
2878 #endif
2879 }
2880 
2881 /*
2882  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2883  * specified pmap. This eliminates the blast of soft faults on process startup
2884  * and immediately after an mmap.
2885  */
2886 static void
2887 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2888     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2889 {
2890 
2891 	VM_OBJECT_ASSERT_WLOCKED(object);
2892 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2893 	    ("mmu_booke_object_init_pt: non-device object"));
2894 }
2895 
2896 /*
2897  * Perform the pmap work for mincore.
2898  */
2899 static int
2900 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2901     vm_paddr_t *locked_pa)
2902 {
2903 
2904 	/* XXX: this should be implemented at some point */
2905 	return (0);
2906 }
2907 
2908 static int
2909 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
2910     vm_memattr_t mode)
2911 {
2912 	vm_offset_t va;
2913 	pte_t *pte;
2914 	int i, j;
2915 
2916 	/* Check TLB1 mappings */
2917 	for (i = 0; i < tlb1_idx; i++) {
2918 		if (!(tlb1[i].mas1 & MAS1_VALID))
2919 			continue;
2920 		if (addr >= tlb1[i].virt && addr < tlb1[i].virt + tlb1[i].size)
2921 			break;
2922 	}
2923 	if (i < tlb1_idx) {
2924 		/* Only allow full mappings to be modified for now. */
2925 		/* Validate the range. */
2926 		for (j = i, va = addr; va < addr + sz; va += tlb1[j].size, j++) {
2927 			if (va != tlb1[j].virt || (sz - (va - addr) < tlb1[j].size))
2928 				return (EINVAL);
2929 		}
2930 		for (va = addr; va < addr + sz; va += tlb1[i].size, i++) {
2931 			tlb1[i].mas2 &= ~MAS2_WIMGE_MASK;
2932 			tlb1[i].mas2 |= tlb_calc_wimg(tlb1[i].phys, mode);
2933 
2934 			/*
2935 			 * Write it out to the TLB.  Should really re-sync with other
2936 			 * cores.
2937 			 */
2938 			tlb1_write_entry(i);
2939 		}
2940 		return (0);
2941 	}
2942 
2943 	/* Not in TLB1, try through pmap */
2944 	/* First validate the range. */
2945 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2946 		pte = pte_find(mmu, kernel_pmap, va);
2947 		if (pte == NULL || !PTE_ISVALID(pte))
2948 			return (EINVAL);
2949 	}
2950 
2951 	mtx_lock_spin(&tlbivax_mutex);
2952 	tlb_miss_lock();
2953 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2954 		pte = pte_find(mmu, kernel_pmap, va);
2955 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2956 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode << PTE_MAS2_SHIFT);
2957 		tlb0_flush_entry(va);
2958 	}
2959 	tlb_miss_unlock();
2960 	mtx_unlock_spin(&tlbivax_mutex);
2961 
2962 	return (pte_vatopa(mmu, kernel_pmap, va));
2963 }
2964 
2965 /**************************************************************************/
2966 /* TID handling */
2967 /**************************************************************************/
2968 
2969 /*
2970  * Allocate a TID. If necessary, steal one from someone else.
2971  * The new TID is flushed from the TLB before returning.
2972  */
2973 static tlbtid_t
2974 tid_alloc(pmap_t pmap)
2975 {
2976 	tlbtid_t tid;
2977 	int thiscpu;
2978 
2979 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2980 
2981 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2982 
2983 	thiscpu = PCPU_GET(cpuid);
2984 
2985 	tid = PCPU_GET(tid_next);
2986 	if (tid > TID_MAX)
2987 		tid = TID_MIN;
2988 	PCPU_SET(tid_next, tid + 1);
2989 
2990 	/* If we are stealing TID then clear the relevant pmap's field */
2991 	if (tidbusy[thiscpu][tid] != NULL) {
2992 
2993 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2994 
2995 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2996 
2997 		/* Flush all entries from TLB0 matching this TID. */
2998 		tid_flush(tid);
2999 	}
3000 
3001 	tidbusy[thiscpu][tid] = pmap;
3002 	pmap->pm_tid[thiscpu] = tid;
3003 	__asm __volatile("msync; isync");
3004 
3005 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3006 	    PCPU_GET(tid_next));
3007 
3008 	return (tid);
3009 }
3010 
3011 /**************************************************************************/
3012 /* TLB0 handling */
3013 /**************************************************************************/
3014 
3015 static void
3016 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3017     uint32_t mas7)
3018 {
3019 	int as;
3020 	char desc[3];
3021 	tlbtid_t tid;
3022 	vm_size_t size;
3023 	unsigned int tsize;
3024 
3025 	desc[2] = '\0';
3026 	if (mas1 & MAS1_VALID)
3027 		desc[0] = 'V';
3028 	else
3029 		desc[0] = ' ';
3030 
3031 	if (mas1 & MAS1_IPROT)
3032 		desc[1] = 'P';
3033 	else
3034 		desc[1] = ' ';
3035 
3036 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3037 	tid = MAS1_GETTID(mas1);
3038 
3039 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3040 	size = 0;
3041 	if (tsize)
3042 		size = tsize2size(tsize);
3043 
3044 	debugf("%3d: (%s) [AS=%d] "
3045 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3046 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3047 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3048 }
3049 
3050 /* Convert TLB0 va and way number to tlb0[] table index. */
3051 static inline unsigned int
3052 tlb0_tableidx(vm_offset_t va, unsigned int way)
3053 {
3054 	unsigned int idx;
3055 
3056 	idx = (way * TLB0_ENTRIES_PER_WAY);
3057 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3058 	return (idx);
3059 }
3060 
3061 /*
3062  * Invalidate TLB0 entry.
3063  */
3064 static inline void
3065 tlb0_flush_entry(vm_offset_t va)
3066 {
3067 
3068 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3069 
3070 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3071 
3072 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3073 	__asm __volatile("isync; msync");
3074 	__asm __volatile("tlbsync; msync");
3075 
3076 	CTR1(KTR_PMAP, "%s: e", __func__);
3077 }
3078 
3079 /* Print out contents of the MAS registers for each TLB0 entry */
3080 void
3081 tlb0_print_tlbentries(void)
3082 {
3083 	uint32_t mas0, mas1, mas2, mas3, mas7;
3084 	int entryidx, way, idx;
3085 
3086 	debugf("TLB0 entries:\n");
3087 	for (way = 0; way < TLB0_WAYS; way ++)
3088 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3089 
3090 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3091 			mtspr(SPR_MAS0, mas0);
3092 			__asm __volatile("isync");
3093 
3094 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3095 			mtspr(SPR_MAS2, mas2);
3096 
3097 			__asm __volatile("isync; tlbre");
3098 
3099 			mas1 = mfspr(SPR_MAS1);
3100 			mas2 = mfspr(SPR_MAS2);
3101 			mas3 = mfspr(SPR_MAS3);
3102 			mas7 = mfspr(SPR_MAS7);
3103 
3104 			idx = tlb0_tableidx(mas2, way);
3105 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3106 		}
3107 }
3108 
3109 /**************************************************************************/
3110 /* TLB1 handling */
3111 /**************************************************************************/
3112 
3113 /*
3114  * TLB1 mapping notes:
3115  *
3116  * TLB1[0]	Kernel text and data.
3117  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3118  *		windows, other devices mappings.
3119  */
3120 
3121 /*
3122  * Write given entry to TLB1 hardware.
3123  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
3124  */
3125 static void
3126 tlb1_write_entry(unsigned int idx)
3127 {
3128 	uint32_t mas0;
3129 
3130 	//debugf("tlb1_write_entry: s\n");
3131 
3132 	/* Select entry */
3133 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3134 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
3135 
3136 	mtspr(SPR_MAS0, mas0);
3137 	__asm __volatile("isync");
3138 	mtspr(SPR_MAS1, tlb1[idx].mas1);
3139 	__asm __volatile("isync");
3140 	mtspr(SPR_MAS2, tlb1[idx].mas2);
3141 	__asm __volatile("isync");
3142 	mtspr(SPR_MAS3, tlb1[idx].mas3);
3143 	__asm __volatile("isync");
3144 	switch ((mfpvr() >> 16) & 0xFFFF) {
3145 	case FSL_E500mc:
3146 	case FSL_E5500:
3147 		mtspr(SPR_MAS8, 0);
3148 		__asm __volatile("isync");
3149 		/* FALLTHROUGH */
3150 	case FSL_E500v2:
3151 		mtspr(SPR_MAS7, tlb1[idx].mas7);
3152 		__asm __volatile("isync");
3153 		break;
3154 	default:
3155 		break;
3156 	}
3157 
3158 	__asm __volatile("tlbwe; isync; msync");
3159 
3160 	//debugf("tlb1_write_entry: e\n");
3161 }
3162 
3163 /*
3164  * Return the largest uint value log such that 2^log <= num.
3165  */
3166 static unsigned int
3167 ilog2(unsigned int num)
3168 {
3169 	int lz;
3170 
3171 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3172 	return (31 - lz);
3173 }
3174 
3175 /*
3176  * Convert TLB TSIZE value to mapped region size.
3177  */
3178 static vm_size_t
3179 tsize2size(unsigned int tsize)
3180 {
3181 
3182 	/*
3183 	 * size = 4^tsize KB
3184 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3185 	 */
3186 
3187 	return ((1 << (2 * tsize)) * 1024);
3188 }
3189 
3190 /*
3191  * Convert region size (must be power of 4) to TLB TSIZE value.
3192  */
3193 static unsigned int
3194 size2tsize(vm_size_t size)
3195 {
3196 
3197 	return (ilog2(size) / 2 - 5);
3198 }
3199 
3200 /*
3201  * Register permanent kernel mapping in TLB1.
3202  *
3203  * Entries are created starting from index 0 (current free entry is
3204  * kept in tlb1_idx) and are not supposed to be invalidated.
3205  */
3206 int
3207 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3208     uint32_t flags)
3209 {
3210 	uint32_t ts, tid;
3211 	int tsize, index;
3212 
3213 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3214 	if (index >= TLB1_ENTRIES) {
3215 		printf("tlb1_set_entry: TLB1 full!\n");
3216 		return (-1);
3217 	}
3218 
3219 	/* Convert size to TSIZE */
3220 	tsize = size2tsize(size);
3221 
3222 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3223 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3224 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3225 
3226 	/*
3227 	 * Atomicity is preserved by the atomic increment above since nothing
3228 	 * is ever removed from tlb1.
3229 	 */
3230 
3231 	tlb1[index].phys = pa;
3232 	tlb1[index].virt = va;
3233 	tlb1[index].size = size;
3234 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3235 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3236 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3237 
3238 	/* Set supervisor RWX permission bits */
3239 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3240 	tlb1[index].mas7 = (pa >> 32) & MAS7_RPN;
3241 
3242 	tlb1_write_entry(index);
3243 
3244 	/*
3245 	 * XXX in general TLB1 updates should be propagated between CPUs,
3246 	 * since current design assumes to have the same TLB1 set-up on all
3247 	 * cores.
3248 	 */
3249 	return (0);
3250 }
3251 
3252 /*
3253  * Map in contiguous RAM region into the TLB1 using maximum of
3254  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3255  *
3256  * If necessary round up last entry size and return total size
3257  * used by all allocated entries.
3258  */
3259 vm_size_t
3260 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3261 {
3262 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3263 	vm_size_t mapped, pgsz, base, mask;
3264 	int idx, nents;
3265 
3266 	/* Round up to the next 1M */
3267 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3268 
3269 	mapped = 0;
3270 	idx = 0;
3271 	base = va;
3272 	pgsz = 64*1024*1024;
3273 	while (mapped < size) {
3274 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3275 			while (pgsz > (size - mapped))
3276 				pgsz >>= 2;
3277 			pgs[idx++] = pgsz;
3278 			mapped += pgsz;
3279 		}
3280 
3281 		/* We under-map. Correct for this. */
3282 		if (mapped < size) {
3283 			while (pgs[idx - 1] == pgsz) {
3284 				idx--;
3285 				mapped -= pgsz;
3286 			}
3287 			/* XXX We may increase beyond out starting point. */
3288 			pgsz <<= 2;
3289 			pgs[idx++] = pgsz;
3290 			mapped += pgsz;
3291 		}
3292 	}
3293 
3294 	nents = idx;
3295 	mask = pgs[0] - 1;
3296 	/* Align address to the boundary */
3297 	if (va & mask) {
3298 		va = (va + mask) & ~mask;
3299 		pa = (pa + mask) & ~mask;
3300 	}
3301 
3302 	for (idx = 0; idx < nents; idx++) {
3303 		pgsz = pgs[idx];
3304 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
3305 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3306 		pa += pgsz;
3307 		va += pgsz;
3308 	}
3309 
3310 	mapped = (va - base);
3311 #ifdef __powerpc64__
3312 	printf("mapped size 0x%016lx (wasted space 0x%16lx)\n",
3313 #else
3314 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3315 #endif
3316 	    mapped, mapped - size);
3317 	return (mapped);
3318 }
3319 
3320 /*
3321  * TLB1 initialization routine, to be called after the very first
3322  * assembler level setup done in locore.S.
3323  */
3324 void
3325 tlb1_init()
3326 {
3327 	uint32_t mas0, mas1, mas2, mas3, mas7;
3328 	uint32_t tsz;
3329 	int i;
3330 
3331 	tlb1_idx = 1;
3332 
3333 	tlb1_get_tlbconf();
3334 
3335 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3336 	mtspr(SPR_MAS0, mas0);
3337 	__asm __volatile("isync; tlbre");
3338 
3339 	mas1 = mfspr(SPR_MAS1);
3340 	mas2 = mfspr(SPR_MAS2);
3341 	mas3 = mfspr(SPR_MAS3);
3342 	mas7 = mfspr(SPR_MAS7);
3343 
3344 	tlb1[0].mas1 = mas1;
3345 	tlb1[0].mas2 = mfspr(SPR_MAS2);
3346 	tlb1[0].mas3 = mas3;
3347 	tlb1[0].mas7 = mas7;
3348 	tlb1[0].virt = mas2 & MAS2_EPN_MASK;
3349 	tlb1[0].phys =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
3350 	    (mas3 & MAS3_RPN);
3351 
3352 	kernload = tlb1[0].phys;
3353 
3354 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3355 	tlb1[0].size = (tsz > 0) ? tsize2size(tsz) : 0;
3356 	kernsize += tlb1[0].size;
3357 
3358 #ifdef SMP
3359 	bp_ntlb1s = tlb1_idx;
3360 #endif
3361 
3362 	/* Purge the remaining entries */
3363 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3364 		tlb1_write_entry(i);
3365 
3366 	/* Setup TLB miss defaults */
3367 	set_mas4_defaults();
3368 }
3369 
3370 vm_offset_t
3371 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3372 {
3373 	vm_paddr_t pa_base;
3374 	vm_offset_t va, sz;
3375 	int i;
3376 
3377 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3378 
3379 	for (i = 0; i < tlb1_idx; i++) {
3380 		if (!(tlb1[i].mas1 & MAS1_VALID))
3381 			continue;
3382 		if (pa >= tlb1[i].phys && (pa + size) <=
3383 		    (tlb1[i].phys + tlb1[i].size))
3384 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3385 	}
3386 
3387 	pa_base = rounddown(pa, PAGE_SIZE);
3388 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3389 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3390 	va = tlb1_map_base + (pa - pa_base);
3391 
3392 	do {
3393 		sz = 1 << (ilog2(size) & ~1);
3394 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3395 		size -= sz;
3396 		pa_base += sz;
3397 		tlb1_map_base += sz;
3398 	} while (size > 0);
3399 
3400 #ifdef SMP
3401 	bp_ntlb1s = tlb1_idx;
3402 #endif
3403 
3404 	return (va);
3405 }
3406 
3407 /*
3408  * Setup MAS4 defaults.
3409  * These values are loaded to MAS0-2 on a TLB miss.
3410  */
3411 static void
3412 set_mas4_defaults(void)
3413 {
3414 	uint32_t mas4;
3415 
3416 	/* Defaults: TLB0, PID0, TSIZED=4K */
3417 	mas4 = MAS4_TLBSELD0;
3418 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3419 #ifdef SMP
3420 	mas4 |= MAS4_MD;
3421 #endif
3422 	mtspr(SPR_MAS4, mas4);
3423 	__asm __volatile("isync");
3424 }
3425 
3426 /*
3427  * Print out contents of the MAS registers for each TLB1 entry
3428  */
3429 void
3430 tlb1_print_tlbentries(void)
3431 {
3432 	uint32_t mas0, mas1, mas2, mas3, mas7;
3433 	int i;
3434 
3435 	debugf("TLB1 entries:\n");
3436 	for (i = 0; i < TLB1_ENTRIES; i++) {
3437 
3438 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3439 		mtspr(SPR_MAS0, mas0);
3440 
3441 		__asm __volatile("isync; tlbre");
3442 
3443 		mas1 = mfspr(SPR_MAS1);
3444 		mas2 = mfspr(SPR_MAS2);
3445 		mas3 = mfspr(SPR_MAS3);
3446 		mas7 = mfspr(SPR_MAS7);
3447 
3448 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3449 	}
3450 }
3451 
3452 /*
3453  * Print out contents of the in-ram tlb1 table.
3454  */
3455 void
3456 tlb1_print_entries(void)
3457 {
3458 	int i;
3459 
3460 	debugf("tlb1[] table entries:\n");
3461 	for (i = 0; i < TLB1_ENTRIES; i++)
3462 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3,
3463 		    tlb1[i].mas7);
3464 }
3465 
3466 /*
3467  * Return 0 if the physical IO range is encompassed by one of the
3468  * the TLB1 entries, otherwise return related error code.
3469  */
3470 static int
3471 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3472 {
3473 	uint32_t prot;
3474 	vm_paddr_t pa_start;
3475 	vm_paddr_t pa_end;
3476 	unsigned int entry_tsize;
3477 	vm_size_t entry_size;
3478 
3479 	*va = (vm_offset_t)NULL;
3480 
3481 	/* Skip invalid entries */
3482 	if (!(tlb1[i].mas1 & MAS1_VALID))
3483 		return (EINVAL);
3484 
3485 	/*
3486 	 * The entry must be cache-inhibited, guarded, and r/w
3487 	 * so it can function as an i/o page
3488 	 */
3489 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3490 	if (prot != (MAS2_I | MAS2_G))
3491 		return (EPERM);
3492 
3493 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3494 	if (prot != (MAS3_SR | MAS3_SW))
3495 		return (EPERM);
3496 
3497 	/* The address should be within the entry range. */
3498 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3499 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3500 
3501 	entry_size = tsize2size(entry_tsize);
3502 	pa_start = (((vm_paddr_t)tlb1[i].mas7 & MAS7_RPN) << 32) |
3503 	    (tlb1[i].mas3 & MAS3_RPN);
3504 	pa_end = pa_start + entry_size;
3505 
3506 	if ((pa < pa_start) || ((pa + size) > pa_end))
3507 		return (ERANGE);
3508 
3509 	/* Return virtual address of this mapping. */
3510 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3511 	return (0);
3512 }
3513 
3514 /*
3515  * Invalidate all TLB0 entries which match the given TID. Note this is
3516  * dedicated for cases when invalidations should NOT be propagated to other
3517  * CPUs.
3518  */
3519 static void
3520 tid_flush(tlbtid_t tid)
3521 {
3522 	register_t msr;
3523 	uint32_t mas0, mas1, mas2;
3524 	int entry, way;
3525 
3526 
3527 	/* Don't evict kernel translations */
3528 	if (tid == TID_KERNEL)
3529 		return;
3530 
3531 	msr = mfmsr();
3532 	__asm __volatile("wrteei 0");
3533 
3534 	for (way = 0; way < TLB0_WAYS; way++)
3535 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
3536 
3537 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3538 			mtspr(SPR_MAS0, mas0);
3539 			__asm __volatile("isync");
3540 
3541 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
3542 			mtspr(SPR_MAS2, mas2);
3543 
3544 			__asm __volatile("isync; tlbre");
3545 
3546 			mas1 = mfspr(SPR_MAS1);
3547 
3548 			if (!(mas1 & MAS1_VALID))
3549 				continue;
3550 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
3551 				continue;
3552 			mas1 &= ~MAS1_VALID;
3553 			mtspr(SPR_MAS1, mas1);
3554 			__asm __volatile("isync; tlbwe; isync; msync");
3555 		}
3556 	mtmsr(msr);
3557 }
3558