xref: /freebsd/sys/powerpc/booke/pmap.c (revision 2e1417489338b971e5fd599ff48b5f65df9e8d3b)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/types.h>
55 #include <sys/param.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/linker.h>
64 #include <sys/msgbuf.h>
65 #include <sys/lock.h>
66 #include <sys/mutex.h>
67 #include <sys/sched.h>
68 #include <sys/smp.h>
69 #include <sys/vmmeter.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_pager.h>
80 #include <vm/uma.h>
81 
82 #include <machine/cpu.h>
83 #include <machine/pcb.h>
84 #include <machine/platform.h>
85 
86 #include <machine/tlb.h>
87 #include <machine/spr.h>
88 #include <machine/vmparam.h>
89 #include <machine/md_var.h>
90 #include <machine/mmuvar.h>
91 #include <machine/pmap.h>
92 #include <machine/pte.h>
93 
94 #include "mmu_if.h"
95 
96 #ifdef  DEBUG
97 #define debugf(fmt, args...) printf(fmt, ##args)
98 #else
99 #define debugf(fmt, args...)
100 #endif
101 
102 #define TODO			panic("%s: not implemented", __func__);
103 
104 #include "opt_sched.h"
105 #ifndef SCHED_4BSD
106 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
107 #endif
108 extern struct mtx sched_lock;
109 
110 extern int dumpsys_minidump;
111 
112 extern unsigned char _etext[];
113 extern unsigned char _end[];
114 
115 extern uint32_t *bootinfo;
116 
117 #ifdef SMP
118 extern uint32_t kernload_ap;
119 #endif
120 
121 vm_paddr_t kernload;
122 vm_offset_t kernstart;
123 vm_size_t kernsize;
124 
125 /* Message buffer and tables. */
126 static vm_offset_t data_start;
127 static vm_size_t data_end;
128 
129 /* Phys/avail memory regions. */
130 static struct mem_region *availmem_regions;
131 static int availmem_regions_sz;
132 static struct mem_region *physmem_regions;
133 static int physmem_regions_sz;
134 
135 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
136 static vm_offset_t zero_page_va;
137 static struct mtx zero_page_mutex;
138 
139 static struct mtx tlbivax_mutex;
140 
141 /*
142  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
143  * by idle thred only, no lock required.
144  */
145 static vm_offset_t zero_page_idle_va;
146 
147 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
148 static vm_offset_t copy_page_src_va;
149 static vm_offset_t copy_page_dst_va;
150 static struct mtx copy_page_mutex;
151 
152 /**************************************************************************/
153 /* PMAP */
154 /**************************************************************************/
155 
156 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
157     vm_prot_t, boolean_t);
158 
159 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
160 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
161 
162 /*
163  * If user pmap is processed with mmu_booke_remove and the resident count
164  * drops to 0, there are no more pages to remove, so we need not continue.
165  */
166 #define PMAP_REMOVE_DONE(pmap) \
167 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
168 
169 extern void tid_flush(tlbtid_t);
170 
171 /**************************************************************************/
172 /* TLB and TID handling */
173 /**************************************************************************/
174 
175 /* Translation ID busy table */
176 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
177 
178 /*
179  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
180  * core revisions and should be read from h/w registers during early config.
181  */
182 uint32_t tlb0_entries;
183 uint32_t tlb0_ways;
184 uint32_t tlb0_entries_per_way;
185 
186 #define TLB0_ENTRIES		(tlb0_entries)
187 #define TLB0_WAYS		(tlb0_ways)
188 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
189 
190 #define TLB1_ENTRIES 16
191 
192 /* In-ram copy of the TLB1 */
193 static tlb_entry_t tlb1[TLB1_ENTRIES];
194 
195 /* Next free entry in the TLB1 */
196 static unsigned int tlb1_idx;
197 
198 static tlbtid_t tid_alloc(struct pmap *);
199 
200 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
201 
202 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
203 static void tlb1_write_entry(unsigned int);
204 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
205 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
206 
207 static vm_size_t tsize2size(unsigned int);
208 static unsigned int size2tsize(vm_size_t);
209 static unsigned int ilog2(unsigned int);
210 
211 static void set_mas4_defaults(void);
212 
213 static inline void tlb0_flush_entry(vm_offset_t);
214 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
215 
216 /**************************************************************************/
217 /* Page table management */
218 /**************************************************************************/
219 
220 /* Data for the pv entry allocation mechanism */
221 static uma_zone_t pvzone;
222 static struct vm_object pvzone_obj;
223 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
224 
225 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
226 
227 #ifndef PMAP_SHPGPERPROC
228 #define PMAP_SHPGPERPROC	200
229 #endif
230 
231 static void ptbl_init(void);
232 static struct ptbl_buf *ptbl_buf_alloc(void);
233 static void ptbl_buf_free(struct ptbl_buf *);
234 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
235 
236 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
237 static void ptbl_free(mmu_t, pmap_t, unsigned int);
238 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
239 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
240 
241 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
242 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
243 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
244 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
245 
246 static pv_entry_t pv_alloc(void);
247 static void pv_free(pv_entry_t);
248 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
249 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
250 
251 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
252 #define PTBL_BUFS		(128 * 16)
253 
254 struct ptbl_buf {
255 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
256 	vm_offset_t kva;		/* va of mapping */
257 };
258 
259 /* ptbl free list and a lock used for access synchronization. */
260 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
261 static struct mtx ptbl_buf_freelist_lock;
262 
263 /* Base address of kva space allocated fot ptbl bufs. */
264 static vm_offset_t ptbl_buf_pool_vabase;
265 
266 /* Pointer to ptbl_buf structures. */
267 static struct ptbl_buf *ptbl_bufs;
268 
269 void pmap_bootstrap_ap(volatile uint32_t *);
270 
271 /*
272  * Kernel MMU interface
273  */
274 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
275 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
276 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
277 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
278     vm_size_t, vm_offset_t);
279 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
280 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t, boolean_t);
282 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
283     vm_page_t, vm_prot_t);
284 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
285     vm_prot_t);
286 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
287 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
288     vm_prot_t);
289 static void		mmu_booke_init(mmu_t);
290 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
291 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
292 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
293 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
294 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
295     int);
296 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
297     vm_paddr_t *);
298 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
299     vm_object_t, vm_pindex_t, vm_size_t);
300 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
301 static void		mmu_booke_page_init(mmu_t, vm_page_t);
302 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
303 static void		mmu_booke_pinit(mmu_t, pmap_t);
304 static void		mmu_booke_pinit0(mmu_t, pmap_t);
305 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
306     vm_prot_t);
307 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
308 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
309 static void		mmu_booke_release(mmu_t, pmap_t);
310 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
311 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
312 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
313 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
314 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
315 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
316 static void		mmu_booke_activate(mmu_t, struct thread *);
317 static void		mmu_booke_deactivate(mmu_t, struct thread *);
318 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
319 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
320 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
321 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
322 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
323 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
324 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
325 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
326     vm_size_t);
327 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
328     vm_size_t, vm_size_t *);
329 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
330     vm_size_t, vm_offset_t);
331 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
332 
333 static mmu_method_t mmu_booke_methods[] = {
334 	/* pmap dispatcher interface */
335 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
336 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
337 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
338 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
339 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
340 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
341 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
342 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
343 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
344 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
345 	MMUMETHOD(mmu_init,		mmu_booke_init),
346 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
347 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
348 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
349 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
350 	MMUMETHOD(mmu_map,		mmu_booke_map),
351 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
352 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
353 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
354 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
355 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
356 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
357 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
358 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
359 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
360 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
361 	MMUMETHOD(mmu_release,		mmu_booke_release),
362 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
363 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
364 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
365 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
366 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
367 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
368 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
369 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
370 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
371 
372 	/* Internal interfaces */
373 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
374 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
375 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
376 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
377 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
378 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
379 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
380 
381 	/* dumpsys() support */
382 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
383 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
384 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
385 
386 	{ 0, 0 }
387 };
388 
389 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
390 
391 static inline void
392 tlb_miss_lock(void)
393 {
394 #ifdef SMP
395 	struct pcpu *pc;
396 
397 	if (!smp_started)
398 		return;
399 
400 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
401 		if (pc != pcpup) {
402 
403 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
404 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
405 
406 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
407 			    ("tlb_miss_lock: tried to lock self"));
408 
409 			tlb_lock(pc->pc_booke_tlb_lock);
410 
411 			CTR1(KTR_PMAP, "%s: locked", __func__);
412 		}
413 	}
414 #endif
415 }
416 
417 static inline void
418 tlb_miss_unlock(void)
419 {
420 #ifdef SMP
421 	struct pcpu *pc;
422 
423 	if (!smp_started)
424 		return;
425 
426 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
427 		if (pc != pcpup) {
428 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
429 			    __func__, pc->pc_cpuid);
430 
431 			tlb_unlock(pc->pc_booke_tlb_lock);
432 
433 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
434 		}
435 	}
436 #endif
437 }
438 
439 /* Return number of entries in TLB0. */
440 static __inline void
441 tlb0_get_tlbconf(void)
442 {
443 	uint32_t tlb0_cfg;
444 
445 	tlb0_cfg = mfspr(SPR_TLB0CFG);
446 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
447 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
448 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
449 }
450 
451 /* Initialize pool of kva ptbl buffers. */
452 static void
453 ptbl_init(void)
454 {
455 	int i;
456 
457 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
458 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
459 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
460 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
461 
462 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
463 	TAILQ_INIT(&ptbl_buf_freelist);
464 
465 	for (i = 0; i < PTBL_BUFS; i++) {
466 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
467 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
468 	}
469 }
470 
471 /* Get a ptbl_buf from the freelist. */
472 static struct ptbl_buf *
473 ptbl_buf_alloc(void)
474 {
475 	struct ptbl_buf *buf;
476 
477 	mtx_lock(&ptbl_buf_freelist_lock);
478 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
479 	if (buf != NULL)
480 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
481 	mtx_unlock(&ptbl_buf_freelist_lock);
482 
483 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
484 
485 	return (buf);
486 }
487 
488 /* Return ptbl buff to free pool. */
489 static void
490 ptbl_buf_free(struct ptbl_buf *buf)
491 {
492 
493 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
494 
495 	mtx_lock(&ptbl_buf_freelist_lock);
496 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
497 	mtx_unlock(&ptbl_buf_freelist_lock);
498 }
499 
500 /*
501  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
502  */
503 static void
504 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
505 {
506 	struct ptbl_buf *pbuf;
507 
508 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
509 
510 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
511 
512 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
513 		if (pbuf->kva == (vm_offset_t)ptbl) {
514 			/* Remove from pmap ptbl buf list. */
515 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
516 
517 			/* Free corresponding ptbl buf. */
518 			ptbl_buf_free(pbuf);
519 			break;
520 		}
521 }
522 
523 /* Allocate page table. */
524 static pte_t *
525 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
526 {
527 	vm_page_t mtbl[PTBL_PAGES];
528 	vm_page_t m;
529 	struct ptbl_buf *pbuf;
530 	unsigned int pidx;
531 	pte_t *ptbl;
532 	int i;
533 
534 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
535 	    (pmap == kernel_pmap), pdir_idx);
536 
537 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
538 	    ("ptbl_alloc: invalid pdir_idx"));
539 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
540 	    ("pte_alloc: valid ptbl entry exists!"));
541 
542 	pbuf = ptbl_buf_alloc();
543 	if (pbuf == NULL)
544 		panic("pte_alloc: couldn't alloc kernel virtual memory");
545 
546 	ptbl = (pte_t *)pbuf->kva;
547 
548 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
549 
550 	/* Allocate ptbl pages, this will sleep! */
551 	for (i = 0; i < PTBL_PAGES; i++) {
552 		pidx = (PTBL_PAGES * pdir_idx) + i;
553 		while ((m = vm_page_alloc(NULL, pidx,
554 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
555 
556 			PMAP_UNLOCK(pmap);
557 			vm_page_unlock_queues();
558 			VM_WAIT;
559 			vm_page_lock_queues();
560 			PMAP_LOCK(pmap);
561 		}
562 		mtbl[i] = m;
563 	}
564 
565 	/* Map allocated pages into kernel_pmap. */
566 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
567 
568 	/* Zero whole ptbl. */
569 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
570 
571 	/* Add pbuf to the pmap ptbl bufs list. */
572 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
573 
574 	return (ptbl);
575 }
576 
577 /* Free ptbl pages and invalidate pdir entry. */
578 static void
579 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
580 {
581 	pte_t *ptbl;
582 	vm_paddr_t pa;
583 	vm_offset_t va;
584 	vm_page_t m;
585 	int i;
586 
587 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
588 	    (pmap == kernel_pmap), pdir_idx);
589 
590 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
591 	    ("ptbl_free: invalid pdir_idx"));
592 
593 	ptbl = pmap->pm_pdir[pdir_idx];
594 
595 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
596 
597 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
598 
599 	/*
600 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
601 	 * don't attempt to look up the page tables we are releasing.
602 	 */
603 	mtx_lock_spin(&tlbivax_mutex);
604 	tlb_miss_lock();
605 
606 	pmap->pm_pdir[pdir_idx] = NULL;
607 
608 	tlb_miss_unlock();
609 	mtx_unlock_spin(&tlbivax_mutex);
610 
611 	for (i = 0; i < PTBL_PAGES; i++) {
612 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
613 		pa = pte_vatopa(mmu, kernel_pmap, va);
614 		m = PHYS_TO_VM_PAGE(pa);
615 		vm_page_free_zero(m);
616 		atomic_subtract_int(&cnt.v_wire_count, 1);
617 		mmu_booke_kremove(mmu, va);
618 	}
619 
620 	ptbl_free_pmap_ptbl(pmap, ptbl);
621 }
622 
623 /*
624  * Decrement ptbl pages hold count and attempt to free ptbl pages.
625  * Called when removing pte entry from ptbl.
626  *
627  * Return 1 if ptbl pages were freed.
628  */
629 static int
630 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
631 {
632 	pte_t *ptbl;
633 	vm_paddr_t pa;
634 	vm_page_t m;
635 	int i;
636 
637 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
638 	    (pmap == kernel_pmap), pdir_idx);
639 
640 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
641 	    ("ptbl_unhold: invalid pdir_idx"));
642 	KASSERT((pmap != kernel_pmap),
643 	    ("ptbl_unhold: unholding kernel ptbl!"));
644 
645 	ptbl = pmap->pm_pdir[pdir_idx];
646 
647 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
648 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
649 	    ("ptbl_unhold: non kva ptbl"));
650 
651 	/* decrement hold count */
652 	for (i = 0; i < PTBL_PAGES; i++) {
653 		pa = pte_vatopa(mmu, kernel_pmap,
654 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
655 		m = PHYS_TO_VM_PAGE(pa);
656 		m->wire_count--;
657 	}
658 
659 	/*
660 	 * Free ptbl pages if there are no pte etries in this ptbl.
661 	 * wire_count has the same value for all ptbl pages, so check the last
662 	 * page.
663 	 */
664 	if (m->wire_count == 0) {
665 		ptbl_free(mmu, pmap, pdir_idx);
666 
667 		//debugf("ptbl_unhold: e (freed ptbl)\n");
668 		return (1);
669 	}
670 
671 	return (0);
672 }
673 
674 /*
675  * Increment hold count for ptbl pages. This routine is used when a new pte
676  * entry is being inserted into the ptbl.
677  */
678 static void
679 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
680 {
681 	vm_paddr_t pa;
682 	pte_t *ptbl;
683 	vm_page_t m;
684 	int i;
685 
686 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
687 	    pdir_idx);
688 
689 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
690 	    ("ptbl_hold: invalid pdir_idx"));
691 	KASSERT((pmap != kernel_pmap),
692 	    ("ptbl_hold: holding kernel ptbl!"));
693 
694 	ptbl = pmap->pm_pdir[pdir_idx];
695 
696 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
697 
698 	for (i = 0; i < PTBL_PAGES; i++) {
699 		pa = pte_vatopa(mmu, kernel_pmap,
700 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
701 		m = PHYS_TO_VM_PAGE(pa);
702 		m->wire_count++;
703 	}
704 }
705 
706 /* Allocate pv_entry structure. */
707 pv_entry_t
708 pv_alloc(void)
709 {
710 	pv_entry_t pv;
711 
712 	pv_entry_count++;
713 	if (pv_entry_count > pv_entry_high_water)
714 		pagedaemon_wakeup();
715 	pv = uma_zalloc(pvzone, M_NOWAIT);
716 
717 	return (pv);
718 }
719 
720 /* Free pv_entry structure. */
721 static __inline void
722 pv_free(pv_entry_t pve)
723 {
724 
725 	pv_entry_count--;
726 	uma_zfree(pvzone, pve);
727 }
728 
729 
730 /* Allocate and initialize pv_entry structure. */
731 static void
732 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
733 {
734 	pv_entry_t pve;
735 
736 	//int su = (pmap == kernel_pmap);
737 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
738 	//	(u_int32_t)pmap, va, (u_int32_t)m);
739 
740 	pve = pv_alloc();
741 	if (pve == NULL)
742 		panic("pv_insert: no pv entries!");
743 
744 	pve->pv_pmap = pmap;
745 	pve->pv_va = va;
746 
747 	/* add to pv_list */
748 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
749 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
750 
751 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
752 
753 	//debugf("pv_insert: e\n");
754 }
755 
756 /* Destroy pv entry. */
757 static void
758 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
759 {
760 	pv_entry_t pve;
761 
762 	//int su = (pmap == kernel_pmap);
763 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
764 
765 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
766 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
767 
768 	/* find pv entry */
769 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
770 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
771 			/* remove from pv_list */
772 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
773 			if (TAILQ_EMPTY(&m->md.pv_list))
774 				vm_page_aflag_clear(m, PGA_WRITEABLE);
775 
776 			/* free pv entry struct */
777 			pv_free(pve);
778 			break;
779 		}
780 	}
781 
782 	//debugf("pv_remove: e\n");
783 }
784 
785 /*
786  * Clean pte entry, try to free page table page if requested.
787  *
788  * Return 1 if ptbl pages were freed, otherwise return 0.
789  */
790 static int
791 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
792 {
793 	unsigned int pdir_idx = PDIR_IDX(va);
794 	unsigned int ptbl_idx = PTBL_IDX(va);
795 	vm_page_t m;
796 	pte_t *ptbl;
797 	pte_t *pte;
798 
799 	//int su = (pmap == kernel_pmap);
800 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
801 	//		su, (u_int32_t)pmap, va, flags);
802 
803 	ptbl = pmap->pm_pdir[pdir_idx];
804 	KASSERT(ptbl, ("pte_remove: null ptbl"));
805 
806 	pte = &ptbl[ptbl_idx];
807 
808 	if (pte == NULL || !PTE_ISVALID(pte))
809 		return (0);
810 
811 	if (PTE_ISWIRED(pte))
812 		pmap->pm_stats.wired_count--;
813 
814 	/* Handle managed entry. */
815 	if (PTE_ISMANAGED(pte)) {
816 		/* Get vm_page_t for mapped pte. */
817 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
818 
819 		if (PTE_ISMODIFIED(pte))
820 			vm_page_dirty(m);
821 
822 		if (PTE_ISREFERENCED(pte))
823 			vm_page_aflag_set(m, PGA_REFERENCED);
824 
825 		pv_remove(pmap, va, m);
826 	}
827 
828 	mtx_lock_spin(&tlbivax_mutex);
829 	tlb_miss_lock();
830 
831 	tlb0_flush_entry(va);
832 	pte->flags = 0;
833 	pte->rpn = 0;
834 
835 	tlb_miss_unlock();
836 	mtx_unlock_spin(&tlbivax_mutex);
837 
838 	pmap->pm_stats.resident_count--;
839 
840 	if (flags & PTBL_UNHOLD) {
841 		//debugf("pte_remove: e (unhold)\n");
842 		return (ptbl_unhold(mmu, pmap, pdir_idx));
843 	}
844 
845 	//debugf("pte_remove: e\n");
846 	return (0);
847 }
848 
849 /*
850  * Insert PTE for a given page and virtual address.
851  */
852 static void
853 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
854 {
855 	unsigned int pdir_idx = PDIR_IDX(va);
856 	unsigned int ptbl_idx = PTBL_IDX(va);
857 	pte_t *ptbl, *pte;
858 
859 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
860 	    pmap == kernel_pmap, pmap, va);
861 
862 	/* Get the page table pointer. */
863 	ptbl = pmap->pm_pdir[pdir_idx];
864 
865 	if (ptbl == NULL) {
866 		/* Allocate page table pages. */
867 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
868 	} else {
869 		/*
870 		 * Check if there is valid mapping for requested
871 		 * va, if there is, remove it.
872 		 */
873 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
874 		if (PTE_ISVALID(pte)) {
875 			pte_remove(mmu, pmap, va, PTBL_HOLD);
876 		} else {
877 			/*
878 			 * pte is not used, increment hold count
879 			 * for ptbl pages.
880 			 */
881 			if (pmap != kernel_pmap)
882 				ptbl_hold(mmu, pmap, pdir_idx);
883 		}
884 	}
885 
886 	/*
887 	 * Insert pv_entry into pv_list for mapped page if part of managed
888 	 * memory.
889 	 */
890 	if ((m->oflags & VPO_UNMANAGED) == 0) {
891 		flags |= PTE_MANAGED;
892 
893 		/* Create and insert pv entry. */
894 		pv_insert(pmap, va, m);
895 	}
896 
897 	pmap->pm_stats.resident_count++;
898 
899 	mtx_lock_spin(&tlbivax_mutex);
900 	tlb_miss_lock();
901 
902 	tlb0_flush_entry(va);
903 	if (pmap->pm_pdir[pdir_idx] == NULL) {
904 		/*
905 		 * If we just allocated a new page table, hook it in
906 		 * the pdir.
907 		 */
908 		pmap->pm_pdir[pdir_idx] = ptbl;
909 	}
910 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
911 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
912 	pte->flags |= (PTE_VALID | flags);
913 
914 	tlb_miss_unlock();
915 	mtx_unlock_spin(&tlbivax_mutex);
916 }
917 
918 /* Return the pa for the given pmap/va. */
919 static vm_paddr_t
920 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
921 {
922 	vm_paddr_t pa = 0;
923 	pte_t *pte;
924 
925 	pte = pte_find(mmu, pmap, va);
926 	if ((pte != NULL) && PTE_ISVALID(pte))
927 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
928 	return (pa);
929 }
930 
931 /* Get a pointer to a PTE in a page table. */
932 static pte_t *
933 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
934 {
935 	unsigned int pdir_idx = PDIR_IDX(va);
936 	unsigned int ptbl_idx = PTBL_IDX(va);
937 
938 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
939 
940 	if (pmap->pm_pdir[pdir_idx])
941 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
942 
943 	return (NULL);
944 }
945 
946 /**************************************************************************/
947 /* PMAP related */
948 /**************************************************************************/
949 
950 /*
951  * This is called during booke_init, before the system is really initialized.
952  */
953 static void
954 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
955 {
956 	vm_offset_t phys_kernelend;
957 	struct mem_region *mp, *mp1;
958 	int cnt, i, j;
959 	u_int s, e, sz;
960 	u_int phys_avail_count;
961 	vm_size_t physsz, hwphyssz, kstack0_sz;
962 	vm_offset_t kernel_pdir, kstack0, va;
963 	vm_paddr_t kstack0_phys;
964 	void *dpcpu;
965 	pte_t *pte;
966 
967 	debugf("mmu_booke_bootstrap: entered\n");
968 
969 #ifdef SMP
970 	kernload_ap = kernload;
971 #endif
972 
973 
974 	/* Initialize invalidation mutex */
975 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
976 
977 	/* Read TLB0 size and associativity. */
978 	tlb0_get_tlbconf();
979 
980 	/*
981 	 * Align kernel start and end address (kernel image).
982 	 * Note that kernel end does not necessarily relate to kernsize.
983 	 * kernsize is the size of the kernel that is actually mapped.
984 	 */
985 	kernstart = trunc_page(start);
986 	data_start = round_page(kernelend);
987 	data_end = data_start;
988 
989 	/*
990 	 * Addresses of preloaded modules (like file systems) use
991 	 * physical addresses. Make sure we relocate those into
992 	 * virtual addresses.
993 	 */
994 	preload_addr_relocate = kernstart - kernload;
995 
996 	/* Allocate the dynamic per-cpu area. */
997 	dpcpu = (void *)data_end;
998 	data_end += DPCPU_SIZE;
999 
1000 	/* Allocate space for the message buffer. */
1001 	msgbufp = (struct msgbuf *)data_end;
1002 	data_end += msgbufsize;
1003 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1004 	    data_end);
1005 
1006 	data_end = round_page(data_end);
1007 
1008 	/* Allocate space for ptbl_bufs. */
1009 	ptbl_bufs = (struct ptbl_buf *)data_end;
1010 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1011 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1012 	    data_end);
1013 
1014 	data_end = round_page(data_end);
1015 
1016 	/* Allocate PTE tables for kernel KVA. */
1017 	kernel_pdir = data_end;
1018 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1019 	    PDIR_SIZE - 1) / PDIR_SIZE;
1020 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1021 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1022 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1023 
1024 	debugf(" data_end: 0x%08x\n", data_end);
1025 	if (data_end - kernstart > kernsize) {
1026 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1027 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1028 	}
1029 	data_end = kernstart + kernsize;
1030 	debugf(" updated data_end: 0x%08x\n", data_end);
1031 
1032 	/*
1033 	 * Clear the structures - note we can only do it safely after the
1034 	 * possible additional TLB1 translations are in place (above) so that
1035 	 * all range up to the currently calculated 'data_end' is covered.
1036 	 */
1037 	dpcpu_init(dpcpu, 0);
1038 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1039 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1040 
1041 	/*******************************************************/
1042 	/* Set the start and end of kva. */
1043 	/*******************************************************/
1044 	virtual_avail = round_page(data_end);
1045 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1046 
1047 	/* Allocate KVA space for page zero/copy operations. */
1048 	zero_page_va = virtual_avail;
1049 	virtual_avail += PAGE_SIZE;
1050 	zero_page_idle_va = virtual_avail;
1051 	virtual_avail += PAGE_SIZE;
1052 	copy_page_src_va = virtual_avail;
1053 	virtual_avail += PAGE_SIZE;
1054 	copy_page_dst_va = virtual_avail;
1055 	virtual_avail += PAGE_SIZE;
1056 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1057 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1058 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1059 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1060 
1061 	/* Initialize page zero/copy mutexes. */
1062 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1063 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1064 
1065 	/* Allocate KVA space for ptbl bufs. */
1066 	ptbl_buf_pool_vabase = virtual_avail;
1067 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1068 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1069 	    ptbl_buf_pool_vabase, virtual_avail);
1070 
1071 	/* Calculate corresponding physical addresses for the kernel region. */
1072 	phys_kernelend = kernload + kernsize;
1073 	debugf("kernel image and allocated data:\n");
1074 	debugf(" kernload    = 0x%08x\n", kernload);
1075 	debugf(" kernstart   = 0x%08x\n", kernstart);
1076 	debugf(" kernsize    = 0x%08x\n", kernsize);
1077 
1078 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1079 		panic("mmu_booke_bootstrap: phys_avail too small");
1080 
1081 	/*
1082 	 * Remove kernel physical address range from avail regions list. Page
1083 	 * align all regions.  Non-page aligned memory isn't very interesting
1084 	 * to us.  Also, sort the entries for ascending addresses.
1085 	 */
1086 
1087 	/* Retrieve phys/avail mem regions */
1088 	mem_regions(&physmem_regions, &physmem_regions_sz,
1089 	    &availmem_regions, &availmem_regions_sz);
1090 	sz = 0;
1091 	cnt = availmem_regions_sz;
1092 	debugf("processing avail regions:\n");
1093 	for (mp = availmem_regions; mp->mr_size; mp++) {
1094 		s = mp->mr_start;
1095 		e = mp->mr_start + mp->mr_size;
1096 		debugf(" %08x-%08x -> ", s, e);
1097 		/* Check whether this region holds all of the kernel. */
1098 		if (s < kernload && e > phys_kernelend) {
1099 			availmem_regions[cnt].mr_start = phys_kernelend;
1100 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1101 			e = kernload;
1102 		}
1103 		/* Look whether this regions starts within the kernel. */
1104 		if (s >= kernload && s < phys_kernelend) {
1105 			if (e <= phys_kernelend)
1106 				goto empty;
1107 			s = phys_kernelend;
1108 		}
1109 		/* Now look whether this region ends within the kernel. */
1110 		if (e > kernload && e <= phys_kernelend) {
1111 			if (s >= kernload)
1112 				goto empty;
1113 			e = kernload;
1114 		}
1115 		/* Now page align the start and size of the region. */
1116 		s = round_page(s);
1117 		e = trunc_page(e);
1118 		if (e < s)
1119 			e = s;
1120 		sz = e - s;
1121 		debugf("%08x-%08x = %x\n", s, e, sz);
1122 
1123 		/* Check whether some memory is left here. */
1124 		if (sz == 0) {
1125 		empty:
1126 			memmove(mp, mp + 1,
1127 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1128 			cnt--;
1129 			mp--;
1130 			continue;
1131 		}
1132 
1133 		/* Do an insertion sort. */
1134 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1135 			if (s < mp1->mr_start)
1136 				break;
1137 		if (mp1 < mp) {
1138 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1139 			mp1->mr_start = s;
1140 			mp1->mr_size = sz;
1141 		} else {
1142 			mp->mr_start = s;
1143 			mp->mr_size = sz;
1144 		}
1145 	}
1146 	availmem_regions_sz = cnt;
1147 
1148 	/*******************************************************/
1149 	/* Steal physical memory for kernel stack from the end */
1150 	/* of the first avail region                           */
1151 	/*******************************************************/
1152 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1153 	kstack0_phys = availmem_regions[0].mr_start +
1154 	    availmem_regions[0].mr_size;
1155 	kstack0_phys -= kstack0_sz;
1156 	availmem_regions[0].mr_size -= kstack0_sz;
1157 
1158 	/*******************************************************/
1159 	/* Fill in phys_avail table, based on availmem_regions */
1160 	/*******************************************************/
1161 	phys_avail_count = 0;
1162 	physsz = 0;
1163 	hwphyssz = 0;
1164 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1165 
1166 	debugf("fill in phys_avail:\n");
1167 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1168 
1169 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1170 		    availmem_regions[i].mr_start,
1171 		    availmem_regions[i].mr_start +
1172 		        availmem_regions[i].mr_size,
1173 		    availmem_regions[i].mr_size);
1174 
1175 		if (hwphyssz != 0 &&
1176 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1177 			debugf(" hw.physmem adjust\n");
1178 			if (physsz < hwphyssz) {
1179 				phys_avail[j] = availmem_regions[i].mr_start;
1180 				phys_avail[j + 1] =
1181 				    availmem_regions[i].mr_start +
1182 				    hwphyssz - physsz;
1183 				physsz = hwphyssz;
1184 				phys_avail_count++;
1185 			}
1186 			break;
1187 		}
1188 
1189 		phys_avail[j] = availmem_regions[i].mr_start;
1190 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1191 		    availmem_regions[i].mr_size;
1192 		phys_avail_count++;
1193 		physsz += availmem_regions[i].mr_size;
1194 	}
1195 	physmem = btoc(physsz);
1196 
1197 	/* Calculate the last available physical address. */
1198 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1199 		;
1200 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1201 
1202 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1203 	debugf("phys_avail_count = %d\n", phys_avail_count);
1204 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1205 	    physmem);
1206 
1207 	/*******************************************************/
1208 	/* Initialize (statically allocated) kernel pmap. */
1209 	/*******************************************************/
1210 	PMAP_LOCK_INIT(kernel_pmap);
1211 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1212 
1213 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1214 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1215 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1216 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1217 
1218 	/* Initialize kernel pdir */
1219 	for (i = 0; i < kernel_ptbls; i++)
1220 		kernel_pmap->pm_pdir[kptbl_min + i] =
1221 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1222 
1223 	for (i = 0; i < MAXCPU; i++) {
1224 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1225 
1226 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1227 		tidbusy[i][0] = kernel_pmap;
1228 	}
1229 
1230 	/*
1231 	 * Fill in PTEs covering kernel code and data. They are not required
1232 	 * for address translation, as this area is covered by static TLB1
1233 	 * entries, but for pte_vatopa() to work correctly with kernel area
1234 	 * addresses.
1235 	 */
1236 	for (va = KERNBASE; va < data_end; va += PAGE_SIZE) {
1237 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1238 		pte->rpn = kernload + (va - KERNBASE);
1239 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1240 		    PTE_VALID;
1241 	}
1242 	/* Mark kernel_pmap active on all CPUs */
1243 	CPU_FILL(&kernel_pmap->pm_active);
1244 
1245 	/*******************************************************/
1246 	/* Final setup */
1247 	/*******************************************************/
1248 
1249 	/* Enter kstack0 into kernel map, provide guard page */
1250 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1251 	thread0.td_kstack = kstack0;
1252 	thread0.td_kstack_pages = KSTACK_PAGES;
1253 
1254 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1255 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1256 	    kstack0_phys, kstack0_phys + kstack0_sz);
1257 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1258 
1259 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1260 	for (i = 0; i < KSTACK_PAGES; i++) {
1261 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1262 		kstack0 += PAGE_SIZE;
1263 		kstack0_phys += PAGE_SIZE;
1264 	}
1265 
1266 	debugf("virtual_avail = %08x\n", virtual_avail);
1267 	debugf("virtual_end   = %08x\n", virtual_end);
1268 
1269 	debugf("mmu_booke_bootstrap: exit\n");
1270 }
1271 
1272 void
1273 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1274 {
1275 	int i;
1276 
1277 	/*
1278 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1279 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1280 	 * these values directly to (re)program AP's TLB1 hardware.
1281 	 */
1282 	for (i = 0; i < tlb1_idx; i ++) {
1283 		/* Skip invalid entries */
1284 		if (!(tlb1[i].mas1 & MAS1_VALID))
1285 			continue;
1286 
1287 		tlb1_write_entry(i);
1288 	}
1289 
1290 	set_mas4_defaults();
1291 }
1292 
1293 /*
1294  * Get the physical page address for the given pmap/virtual address.
1295  */
1296 static vm_paddr_t
1297 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1298 {
1299 	vm_paddr_t pa;
1300 
1301 	PMAP_LOCK(pmap);
1302 	pa = pte_vatopa(mmu, pmap, va);
1303 	PMAP_UNLOCK(pmap);
1304 
1305 	return (pa);
1306 }
1307 
1308 /*
1309  * Extract the physical page address associated with the given
1310  * kernel virtual address.
1311  */
1312 static vm_paddr_t
1313 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1314 {
1315 
1316 	return (pte_vatopa(mmu, kernel_pmap, va));
1317 }
1318 
1319 /*
1320  * Initialize the pmap module.
1321  * Called by vm_init, to initialize any structures that the pmap
1322  * system needs to map virtual memory.
1323  */
1324 static void
1325 mmu_booke_init(mmu_t mmu)
1326 {
1327 	int shpgperproc = PMAP_SHPGPERPROC;
1328 
1329 	/*
1330 	 * Initialize the address space (zone) for the pv entries.  Set a
1331 	 * high water mark so that the system can recover from excessive
1332 	 * numbers of pv entries.
1333 	 */
1334 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1335 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1336 
1337 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1338 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1339 
1340 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1341 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1342 
1343 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1344 
1345 	/* Pre-fill pvzone with initial number of pv entries. */
1346 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1347 
1348 	/* Initialize ptbl allocation. */
1349 	ptbl_init();
1350 }
1351 
1352 /*
1353  * Map a list of wired pages into kernel virtual address space.  This is
1354  * intended for temporary mappings which do not need page modification or
1355  * references recorded.  Existing mappings in the region are overwritten.
1356  */
1357 static void
1358 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1359 {
1360 	vm_offset_t va;
1361 
1362 	va = sva;
1363 	while (count-- > 0) {
1364 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1365 		va += PAGE_SIZE;
1366 		m++;
1367 	}
1368 }
1369 
1370 /*
1371  * Remove page mappings from kernel virtual address space.  Intended for
1372  * temporary mappings entered by mmu_booke_qenter.
1373  */
1374 static void
1375 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1376 {
1377 	vm_offset_t va;
1378 
1379 	va = sva;
1380 	while (count-- > 0) {
1381 		mmu_booke_kremove(mmu, va);
1382 		va += PAGE_SIZE;
1383 	}
1384 }
1385 
1386 /*
1387  * Map a wired page into kernel virtual address space.
1388  */
1389 static void
1390 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1391 {
1392 	unsigned int pdir_idx = PDIR_IDX(va);
1393 	unsigned int ptbl_idx = PTBL_IDX(va);
1394 	uint32_t flags;
1395 	pte_t *pte;
1396 
1397 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1398 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1399 
1400 	flags = 0;
1401 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1402 	flags |= PTE_M;
1403 
1404 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1405 
1406 	mtx_lock_spin(&tlbivax_mutex);
1407 	tlb_miss_lock();
1408 
1409 	if (PTE_ISVALID(pte)) {
1410 
1411 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1412 
1413 		/* Flush entry from TLB0 */
1414 		tlb0_flush_entry(va);
1415 	}
1416 
1417 	pte->rpn = pa & ~PTE_PA_MASK;
1418 	pte->flags = flags;
1419 
1420 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1421 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1422 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1423 
1424 	/* Flush the real memory from the instruction cache. */
1425 	if ((flags & (PTE_I | PTE_G)) == 0) {
1426 		__syncicache((void *)va, PAGE_SIZE);
1427 	}
1428 
1429 	tlb_miss_unlock();
1430 	mtx_unlock_spin(&tlbivax_mutex);
1431 }
1432 
1433 /*
1434  * Remove a page from kernel page table.
1435  */
1436 static void
1437 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1438 {
1439 	unsigned int pdir_idx = PDIR_IDX(va);
1440 	unsigned int ptbl_idx = PTBL_IDX(va);
1441 	pte_t *pte;
1442 
1443 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1444 
1445 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1446 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1447 	    ("mmu_booke_kremove: invalid va"));
1448 
1449 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1450 
1451 	if (!PTE_ISVALID(pte)) {
1452 
1453 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1454 
1455 		return;
1456 	}
1457 
1458 	mtx_lock_spin(&tlbivax_mutex);
1459 	tlb_miss_lock();
1460 
1461 	/* Invalidate entry in TLB0, update PTE. */
1462 	tlb0_flush_entry(va);
1463 	pte->flags = 0;
1464 	pte->rpn = 0;
1465 
1466 	tlb_miss_unlock();
1467 	mtx_unlock_spin(&tlbivax_mutex);
1468 }
1469 
1470 /*
1471  * Initialize pmap associated with process 0.
1472  */
1473 static void
1474 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1475 {
1476 
1477 	mmu_booke_pinit(mmu, pmap);
1478 	PCPU_SET(curpmap, pmap);
1479 }
1480 
1481 /*
1482  * Initialize a preallocated and zeroed pmap structure,
1483  * such as one in a vmspace structure.
1484  */
1485 static void
1486 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1487 {
1488 	int i;
1489 
1490 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1491 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1492 
1493 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1494 
1495 	PMAP_LOCK_INIT(pmap);
1496 	for (i = 0; i < MAXCPU; i++)
1497 		pmap->pm_tid[i] = TID_NONE;
1498 	CPU_ZERO(&kernel_pmap->pm_active);
1499 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1500 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1501 	TAILQ_INIT(&pmap->pm_ptbl_list);
1502 }
1503 
1504 /*
1505  * Release any resources held by the given physical map.
1506  * Called when a pmap initialized by mmu_booke_pinit is being released.
1507  * Should only be called if the map contains no valid mappings.
1508  */
1509 static void
1510 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1511 {
1512 
1513 	KASSERT(pmap->pm_stats.resident_count == 0,
1514 	    ("pmap_release: pmap resident count %ld != 0",
1515 	    pmap->pm_stats.resident_count));
1516 
1517 	PMAP_LOCK_DESTROY(pmap);
1518 }
1519 
1520 /*
1521  * Insert the given physical page at the specified virtual address in the
1522  * target physical map with the protection requested. If specified the page
1523  * will be wired down.
1524  */
1525 static void
1526 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1527     vm_prot_t prot, boolean_t wired)
1528 {
1529 
1530 	vm_page_lock_queues();
1531 	PMAP_LOCK(pmap);
1532 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1533 	vm_page_unlock_queues();
1534 	PMAP_UNLOCK(pmap);
1535 }
1536 
1537 static void
1538 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1539     vm_prot_t prot, boolean_t wired)
1540 {
1541 	pte_t *pte;
1542 	vm_paddr_t pa;
1543 	uint32_t flags;
1544 	int su, sync;
1545 
1546 	pa = VM_PAGE_TO_PHYS(m);
1547 	su = (pmap == kernel_pmap);
1548 	sync = 0;
1549 
1550 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1551 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1552 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1553 	//		(u_int32_t)m, va, pa, prot, wired);
1554 
1555 	if (su) {
1556 		KASSERT(((va >= virtual_avail) &&
1557 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1558 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1559 	} else {
1560 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1561 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1562 	}
1563 	KASSERT((m->oflags & (VPO_UNMANAGED | VPO_BUSY)) != 0 ||
1564 	    VM_OBJECT_LOCKED(m->object),
1565 	    ("mmu_booke_enter_locked: page %p is not busy", m));
1566 
1567 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1568 
1569 	/*
1570 	 * If there is an existing mapping, and the physical address has not
1571 	 * changed, must be protection or wiring change.
1572 	 */
1573 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1574 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1575 
1576 		/*
1577 		 * Before actually updating pte->flags we calculate and
1578 		 * prepare its new value in a helper var.
1579 		 */
1580 		flags = pte->flags;
1581 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1582 
1583 		/* Wiring change, just update stats. */
1584 		if (wired) {
1585 			if (!PTE_ISWIRED(pte)) {
1586 				flags |= PTE_WIRED;
1587 				pmap->pm_stats.wired_count++;
1588 			}
1589 		} else {
1590 			if (PTE_ISWIRED(pte)) {
1591 				flags &= ~PTE_WIRED;
1592 				pmap->pm_stats.wired_count--;
1593 			}
1594 		}
1595 
1596 		if (prot & VM_PROT_WRITE) {
1597 			/* Add write permissions. */
1598 			flags |= PTE_SW;
1599 			if (!su)
1600 				flags |= PTE_UW;
1601 
1602 			if ((flags & PTE_MANAGED) != 0)
1603 				vm_page_aflag_set(m, PGA_WRITEABLE);
1604 		} else {
1605 			/* Handle modified pages, sense modify status. */
1606 
1607 			/*
1608 			 * The PTE_MODIFIED flag could be set by underlying
1609 			 * TLB misses since we last read it (above), possibly
1610 			 * other CPUs could update it so we check in the PTE
1611 			 * directly rather than rely on that saved local flags
1612 			 * copy.
1613 			 */
1614 			if (PTE_ISMODIFIED(pte))
1615 				vm_page_dirty(m);
1616 		}
1617 
1618 		if (prot & VM_PROT_EXECUTE) {
1619 			flags |= PTE_SX;
1620 			if (!su)
1621 				flags |= PTE_UX;
1622 
1623 			/*
1624 			 * Check existing flags for execute permissions: if we
1625 			 * are turning execute permissions on, icache should
1626 			 * be flushed.
1627 			 */
1628 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1629 				sync++;
1630 		}
1631 
1632 		flags &= ~PTE_REFERENCED;
1633 
1634 		/*
1635 		 * The new flags value is all calculated -- only now actually
1636 		 * update the PTE.
1637 		 */
1638 		mtx_lock_spin(&tlbivax_mutex);
1639 		tlb_miss_lock();
1640 
1641 		tlb0_flush_entry(va);
1642 		pte->flags = flags;
1643 
1644 		tlb_miss_unlock();
1645 		mtx_unlock_spin(&tlbivax_mutex);
1646 
1647 	} else {
1648 		/*
1649 		 * If there is an existing mapping, but it's for a different
1650 		 * physical address, pte_enter() will delete the old mapping.
1651 		 */
1652 		//if ((pte != NULL) && PTE_ISVALID(pte))
1653 		//	debugf("mmu_booke_enter_locked: replace\n");
1654 		//else
1655 		//	debugf("mmu_booke_enter_locked: new\n");
1656 
1657 		/* Now set up the flags and install the new mapping. */
1658 		flags = (PTE_SR | PTE_VALID);
1659 		flags |= PTE_M;
1660 
1661 		if (!su)
1662 			flags |= PTE_UR;
1663 
1664 		if (prot & VM_PROT_WRITE) {
1665 			flags |= PTE_SW;
1666 			if (!su)
1667 				flags |= PTE_UW;
1668 
1669 			if ((m->oflags & VPO_UNMANAGED) == 0)
1670 				vm_page_aflag_set(m, PGA_WRITEABLE);
1671 		}
1672 
1673 		if (prot & VM_PROT_EXECUTE) {
1674 			flags |= PTE_SX;
1675 			if (!su)
1676 				flags |= PTE_UX;
1677 		}
1678 
1679 		/* If its wired update stats. */
1680 		if (wired) {
1681 			pmap->pm_stats.wired_count++;
1682 			flags |= PTE_WIRED;
1683 		}
1684 
1685 		pte_enter(mmu, pmap, m, va, flags);
1686 
1687 		/* Flush the real memory from the instruction cache. */
1688 		if (prot & VM_PROT_EXECUTE)
1689 			sync++;
1690 	}
1691 
1692 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1693 		__syncicache((void *)va, PAGE_SIZE);
1694 		sync = 0;
1695 	}
1696 }
1697 
1698 /*
1699  * Maps a sequence of resident pages belonging to the same object.
1700  * The sequence begins with the given page m_start.  This page is
1701  * mapped at the given virtual address start.  Each subsequent page is
1702  * mapped at a virtual address that is offset from start by the same
1703  * amount as the page is offset from m_start within the object.  The
1704  * last page in the sequence is the page with the largest offset from
1705  * m_start that can be mapped at a virtual address less than the given
1706  * virtual address end.  Not every virtual page between start and end
1707  * is mapped; only those for which a resident page exists with the
1708  * corresponding offset from m_start are mapped.
1709  */
1710 static void
1711 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1712     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1713 {
1714 	vm_page_t m;
1715 	vm_pindex_t diff, psize;
1716 
1717 	psize = atop(end - start);
1718 	m = m_start;
1719 	vm_page_lock_queues();
1720 	PMAP_LOCK(pmap);
1721 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1722 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1723 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1724 		m = TAILQ_NEXT(m, listq);
1725 	}
1726 	vm_page_unlock_queues();
1727 	PMAP_UNLOCK(pmap);
1728 }
1729 
1730 static void
1731 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1732     vm_prot_t prot)
1733 {
1734 
1735 	vm_page_lock_queues();
1736 	PMAP_LOCK(pmap);
1737 	mmu_booke_enter_locked(mmu, pmap, va, m,
1738 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1739 	vm_page_unlock_queues();
1740 	PMAP_UNLOCK(pmap);
1741 }
1742 
1743 /*
1744  * Remove the given range of addresses from the specified map.
1745  *
1746  * It is assumed that the start and end are properly rounded to the page size.
1747  */
1748 static void
1749 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1750 {
1751 	pte_t *pte;
1752 	uint8_t hold_flag;
1753 
1754 	int su = (pmap == kernel_pmap);
1755 
1756 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1757 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1758 
1759 	if (su) {
1760 		KASSERT(((va >= virtual_avail) &&
1761 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1762 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1763 	} else {
1764 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1765 		    ("mmu_booke_remove: user pmap, non user va"));
1766 	}
1767 
1768 	if (PMAP_REMOVE_DONE(pmap)) {
1769 		//debugf("mmu_booke_remove: e (empty)\n");
1770 		return;
1771 	}
1772 
1773 	hold_flag = PTBL_HOLD_FLAG(pmap);
1774 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1775 
1776 	vm_page_lock_queues();
1777 	PMAP_LOCK(pmap);
1778 	for (; va < endva; va += PAGE_SIZE) {
1779 		pte = pte_find(mmu, pmap, va);
1780 		if ((pte != NULL) && PTE_ISVALID(pte))
1781 			pte_remove(mmu, pmap, va, hold_flag);
1782 	}
1783 	PMAP_UNLOCK(pmap);
1784 	vm_page_unlock_queues();
1785 
1786 	//debugf("mmu_booke_remove: e\n");
1787 }
1788 
1789 /*
1790  * Remove physical page from all pmaps in which it resides.
1791  */
1792 static void
1793 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1794 {
1795 	pv_entry_t pv, pvn;
1796 	uint8_t hold_flag;
1797 
1798 	vm_page_lock_queues();
1799 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1800 		pvn = TAILQ_NEXT(pv, pv_link);
1801 
1802 		PMAP_LOCK(pv->pv_pmap);
1803 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1804 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1805 		PMAP_UNLOCK(pv->pv_pmap);
1806 	}
1807 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1808 	vm_page_unlock_queues();
1809 }
1810 
1811 /*
1812  * Map a range of physical addresses into kernel virtual address space.
1813  */
1814 static vm_offset_t
1815 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1816     vm_offset_t pa_end, int prot)
1817 {
1818 	vm_offset_t sva = *virt;
1819 	vm_offset_t va = sva;
1820 
1821 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1822 	//		sva, pa_start, pa_end);
1823 
1824 	while (pa_start < pa_end) {
1825 		mmu_booke_kenter(mmu, va, pa_start);
1826 		va += PAGE_SIZE;
1827 		pa_start += PAGE_SIZE;
1828 	}
1829 	*virt = va;
1830 
1831 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1832 	return (sva);
1833 }
1834 
1835 /*
1836  * The pmap must be activated before it's address space can be accessed in any
1837  * way.
1838  */
1839 static void
1840 mmu_booke_activate(mmu_t mmu, struct thread *td)
1841 {
1842 	pmap_t pmap;
1843 	u_int cpuid;
1844 
1845 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1846 
1847 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1848 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1849 
1850 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1851 
1852 	mtx_lock_spin(&sched_lock);
1853 
1854 	cpuid = PCPU_GET(cpuid);
1855 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1856 	PCPU_SET(curpmap, pmap);
1857 
1858 	if (pmap->pm_tid[cpuid] == TID_NONE)
1859 		tid_alloc(pmap);
1860 
1861 	/* Load PID0 register with pmap tid value. */
1862 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1863 	__asm __volatile("isync");
1864 
1865 	mtx_unlock_spin(&sched_lock);
1866 
1867 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1868 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1869 }
1870 
1871 /*
1872  * Deactivate the specified process's address space.
1873  */
1874 static void
1875 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1876 {
1877 	pmap_t pmap;
1878 
1879 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1880 
1881 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1882 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1883 
1884 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1885 	PCPU_SET(curpmap, NULL);
1886 }
1887 
1888 /*
1889  * Copy the range specified by src_addr/len
1890  * from the source map to the range dst_addr/len
1891  * in the destination map.
1892  *
1893  * This routine is only advisory and need not do anything.
1894  */
1895 static void
1896 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1897     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1898 {
1899 
1900 }
1901 
1902 /*
1903  * Set the physical protection on the specified range of this map as requested.
1904  */
1905 static void
1906 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1907     vm_prot_t prot)
1908 {
1909 	vm_offset_t va;
1910 	vm_page_t m;
1911 	pte_t *pte;
1912 
1913 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1914 		mmu_booke_remove(mmu, pmap, sva, eva);
1915 		return;
1916 	}
1917 
1918 	if (prot & VM_PROT_WRITE)
1919 		return;
1920 
1921 	PMAP_LOCK(pmap);
1922 	for (va = sva; va < eva; va += PAGE_SIZE) {
1923 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1924 			if (PTE_ISVALID(pte)) {
1925 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1926 
1927 				mtx_lock_spin(&tlbivax_mutex);
1928 				tlb_miss_lock();
1929 
1930 				/* Handle modified pages. */
1931 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1932 					vm_page_dirty(m);
1933 
1934 				tlb0_flush_entry(va);
1935 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1936 
1937 				tlb_miss_unlock();
1938 				mtx_unlock_spin(&tlbivax_mutex);
1939 			}
1940 		}
1941 	}
1942 	PMAP_UNLOCK(pmap);
1943 }
1944 
1945 /*
1946  * Clear the write and modified bits in each of the given page's mappings.
1947  */
1948 static void
1949 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1950 {
1951 	pv_entry_t pv;
1952 	pte_t *pte;
1953 
1954 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1955 	    ("mmu_booke_remove_write: page %p is not managed", m));
1956 
1957 	/*
1958 	 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be set by
1959 	 * another thread while the object is locked.  Thus, if PGA_WRITEABLE
1960 	 * is clear, no page table entries need updating.
1961 	 */
1962 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1963 	if ((m->oflags & VPO_BUSY) == 0 &&
1964 	    (m->aflags & PGA_WRITEABLE) == 0)
1965 		return;
1966 	vm_page_lock_queues();
1967 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1968 		PMAP_LOCK(pv->pv_pmap);
1969 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1970 			if (PTE_ISVALID(pte)) {
1971 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1972 
1973 				mtx_lock_spin(&tlbivax_mutex);
1974 				tlb_miss_lock();
1975 
1976 				/* Handle modified pages. */
1977 				if (PTE_ISMODIFIED(pte))
1978 					vm_page_dirty(m);
1979 
1980 				/* Flush mapping from TLB0. */
1981 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1982 
1983 				tlb_miss_unlock();
1984 				mtx_unlock_spin(&tlbivax_mutex);
1985 			}
1986 		}
1987 		PMAP_UNLOCK(pv->pv_pmap);
1988 	}
1989 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1990 	vm_page_unlock_queues();
1991 }
1992 
1993 static void
1994 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1995 {
1996 	pte_t *pte;
1997 	pmap_t pmap;
1998 	vm_page_t m;
1999 	vm_offset_t addr;
2000 	vm_paddr_t pa;
2001 	int active, valid;
2002 
2003 	va = trunc_page(va);
2004 	sz = round_page(sz);
2005 
2006 	vm_page_lock_queues();
2007 	pmap = PCPU_GET(curpmap);
2008 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2009 	while (sz > 0) {
2010 		PMAP_LOCK(pm);
2011 		pte = pte_find(mmu, pm, va);
2012 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2013 		if (valid)
2014 			pa = PTE_PA(pte);
2015 		PMAP_UNLOCK(pm);
2016 		if (valid) {
2017 			if (!active) {
2018 				/* Create a mapping in the active pmap. */
2019 				addr = 0;
2020 				m = PHYS_TO_VM_PAGE(pa);
2021 				PMAP_LOCK(pmap);
2022 				pte_enter(mmu, pmap, m, addr,
2023 				    PTE_SR | PTE_VALID | PTE_UR);
2024 				__syncicache((void *)addr, PAGE_SIZE);
2025 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2026 				PMAP_UNLOCK(pmap);
2027 			} else
2028 				__syncicache((void *)va, PAGE_SIZE);
2029 		}
2030 		va += PAGE_SIZE;
2031 		sz -= PAGE_SIZE;
2032 	}
2033 	vm_page_unlock_queues();
2034 }
2035 
2036 /*
2037  * Atomically extract and hold the physical page with the given
2038  * pmap and virtual address pair if that mapping permits the given
2039  * protection.
2040  */
2041 static vm_page_t
2042 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2043     vm_prot_t prot)
2044 {
2045 	pte_t *pte;
2046 	vm_page_t m;
2047 	uint32_t pte_wbit;
2048 	vm_paddr_t pa;
2049 
2050 	m = NULL;
2051 	pa = 0;
2052 	PMAP_LOCK(pmap);
2053 retry:
2054 	pte = pte_find(mmu, pmap, va);
2055 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2056 		if (pmap == kernel_pmap)
2057 			pte_wbit = PTE_SW;
2058 		else
2059 			pte_wbit = PTE_UW;
2060 
2061 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2062 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2063 				goto retry;
2064 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2065 			vm_page_hold(m);
2066 		}
2067 	}
2068 
2069 	PA_UNLOCK_COND(pa);
2070 	PMAP_UNLOCK(pmap);
2071 	return (m);
2072 }
2073 
2074 /*
2075  * Initialize a vm_page's machine-dependent fields.
2076  */
2077 static void
2078 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2079 {
2080 
2081 	TAILQ_INIT(&m->md.pv_list);
2082 }
2083 
2084 /*
2085  * mmu_booke_zero_page_area zeros the specified hardware page by
2086  * mapping it into virtual memory and using bzero to clear
2087  * its contents.
2088  *
2089  * off and size must reside within a single page.
2090  */
2091 static void
2092 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2093 {
2094 	vm_offset_t va;
2095 
2096 	/* XXX KASSERT off and size are within a single page? */
2097 
2098 	mtx_lock(&zero_page_mutex);
2099 	va = zero_page_va;
2100 
2101 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2102 	bzero((caddr_t)va + off, size);
2103 	mmu_booke_kremove(mmu, va);
2104 
2105 	mtx_unlock(&zero_page_mutex);
2106 }
2107 
2108 /*
2109  * mmu_booke_zero_page zeros the specified hardware page.
2110  */
2111 static void
2112 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2113 {
2114 
2115 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2116 }
2117 
2118 /*
2119  * mmu_booke_copy_page copies the specified (machine independent) page by
2120  * mapping the page into virtual memory and using memcopy to copy the page,
2121  * one machine dependent page at a time.
2122  */
2123 static void
2124 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2125 {
2126 	vm_offset_t sva, dva;
2127 
2128 	sva = copy_page_src_va;
2129 	dva = copy_page_dst_va;
2130 
2131 	mtx_lock(&copy_page_mutex);
2132 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2133 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2134 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2135 	mmu_booke_kremove(mmu, dva);
2136 	mmu_booke_kremove(mmu, sva);
2137 	mtx_unlock(&copy_page_mutex);
2138 }
2139 
2140 /*
2141  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2142  * into virtual memory and using bzero to clear its contents. This is intended
2143  * to be called from the vm_pagezero process only and outside of Giant. No
2144  * lock is required.
2145  */
2146 static void
2147 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2148 {
2149 	vm_offset_t va;
2150 
2151 	va = zero_page_idle_va;
2152 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2153 	bzero((caddr_t)va, PAGE_SIZE);
2154 	mmu_booke_kremove(mmu, va);
2155 }
2156 
2157 /*
2158  * Return whether or not the specified physical page was modified
2159  * in any of physical maps.
2160  */
2161 static boolean_t
2162 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2163 {
2164 	pte_t *pte;
2165 	pv_entry_t pv;
2166 	boolean_t rv;
2167 
2168 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2169 	    ("mmu_booke_is_modified: page %p is not managed", m));
2170 	rv = FALSE;
2171 
2172 	/*
2173 	 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be
2174 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2175 	 * is clear, no PTEs can be modified.
2176 	 */
2177 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2178 	if ((m->oflags & VPO_BUSY) == 0 &&
2179 	    (m->aflags & PGA_WRITEABLE) == 0)
2180 		return (rv);
2181 	vm_page_lock_queues();
2182 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2183 		PMAP_LOCK(pv->pv_pmap);
2184 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2185 		    PTE_ISVALID(pte)) {
2186 			if (PTE_ISMODIFIED(pte))
2187 				rv = TRUE;
2188 		}
2189 		PMAP_UNLOCK(pv->pv_pmap);
2190 		if (rv)
2191 			break;
2192 	}
2193 	vm_page_unlock_queues();
2194 	return (rv);
2195 }
2196 
2197 /*
2198  * Return whether or not the specified virtual address is eligible
2199  * for prefault.
2200  */
2201 static boolean_t
2202 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2203 {
2204 
2205 	return (FALSE);
2206 }
2207 
2208 /*
2209  * Return whether or not the specified physical page was referenced
2210  * in any physical maps.
2211  */
2212 static boolean_t
2213 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2214 {
2215 	pte_t *pte;
2216 	pv_entry_t pv;
2217 	boolean_t rv;
2218 
2219 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2220 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2221 	rv = FALSE;
2222 	vm_page_lock_queues();
2223 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2224 		PMAP_LOCK(pv->pv_pmap);
2225 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2226 		    PTE_ISVALID(pte)) {
2227 			if (PTE_ISREFERENCED(pte))
2228 				rv = TRUE;
2229 		}
2230 		PMAP_UNLOCK(pv->pv_pmap);
2231 		if (rv)
2232 			break;
2233 	}
2234 	vm_page_unlock_queues();
2235 	return (rv);
2236 }
2237 
2238 /*
2239  * Clear the modify bits on the specified physical page.
2240  */
2241 static void
2242 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2243 {
2244 	pte_t *pte;
2245 	pv_entry_t pv;
2246 
2247 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2248 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2249 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2250 	KASSERT((m->oflags & VPO_BUSY) == 0,
2251 	    ("mmu_booke_clear_modify: page %p is busy", m));
2252 
2253 	/*
2254 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2255 	 * If the object containing the page is locked and the page is not
2256 	 * VPO_BUSY, then PG_AWRITEABLE cannot be concurrently set.
2257 	 */
2258 	if ((m->aflags & PGA_WRITEABLE) == 0)
2259 		return;
2260 	vm_page_lock_queues();
2261 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2262 		PMAP_LOCK(pv->pv_pmap);
2263 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2264 		    PTE_ISVALID(pte)) {
2265 			mtx_lock_spin(&tlbivax_mutex);
2266 			tlb_miss_lock();
2267 
2268 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2269 				tlb0_flush_entry(pv->pv_va);
2270 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2271 				    PTE_REFERENCED);
2272 			}
2273 
2274 			tlb_miss_unlock();
2275 			mtx_unlock_spin(&tlbivax_mutex);
2276 		}
2277 		PMAP_UNLOCK(pv->pv_pmap);
2278 	}
2279 	vm_page_unlock_queues();
2280 }
2281 
2282 /*
2283  * Return a count of reference bits for a page, clearing those bits.
2284  * It is not necessary for every reference bit to be cleared, but it
2285  * is necessary that 0 only be returned when there are truly no
2286  * reference bits set.
2287  *
2288  * XXX: The exact number of bits to check and clear is a matter that
2289  * should be tested and standardized at some point in the future for
2290  * optimal aging of shared pages.
2291  */
2292 static int
2293 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2294 {
2295 	pte_t *pte;
2296 	pv_entry_t pv;
2297 	int count;
2298 
2299 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2300 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2301 	count = 0;
2302 	vm_page_lock_queues();
2303 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2304 		PMAP_LOCK(pv->pv_pmap);
2305 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2306 		    PTE_ISVALID(pte)) {
2307 			if (PTE_ISREFERENCED(pte)) {
2308 				mtx_lock_spin(&tlbivax_mutex);
2309 				tlb_miss_lock();
2310 
2311 				tlb0_flush_entry(pv->pv_va);
2312 				pte->flags &= ~PTE_REFERENCED;
2313 
2314 				tlb_miss_unlock();
2315 				mtx_unlock_spin(&tlbivax_mutex);
2316 
2317 				if (++count > 4) {
2318 					PMAP_UNLOCK(pv->pv_pmap);
2319 					break;
2320 				}
2321 			}
2322 		}
2323 		PMAP_UNLOCK(pv->pv_pmap);
2324 	}
2325 	vm_page_unlock_queues();
2326 	return (count);
2327 }
2328 
2329 /*
2330  * Clear the reference bit on the specified physical page.
2331  */
2332 static void
2333 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2334 {
2335 	pte_t *pte;
2336 	pv_entry_t pv;
2337 
2338 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2339 	    ("mmu_booke_clear_reference: page %p is not managed", m));
2340 	vm_page_lock_queues();
2341 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2342 		PMAP_LOCK(pv->pv_pmap);
2343 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2344 		    PTE_ISVALID(pte)) {
2345 			if (PTE_ISREFERENCED(pte)) {
2346 				mtx_lock_spin(&tlbivax_mutex);
2347 				tlb_miss_lock();
2348 
2349 				tlb0_flush_entry(pv->pv_va);
2350 				pte->flags &= ~PTE_REFERENCED;
2351 
2352 				tlb_miss_unlock();
2353 				mtx_unlock_spin(&tlbivax_mutex);
2354 			}
2355 		}
2356 		PMAP_UNLOCK(pv->pv_pmap);
2357 	}
2358 	vm_page_unlock_queues();
2359 }
2360 
2361 /*
2362  * Change wiring attribute for a map/virtual-address pair.
2363  */
2364 static void
2365 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2366 {
2367 	pte_t *pte;
2368 
2369 	PMAP_LOCK(pmap);
2370 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2371 		if (wired) {
2372 			if (!PTE_ISWIRED(pte)) {
2373 				pte->flags |= PTE_WIRED;
2374 				pmap->pm_stats.wired_count++;
2375 			}
2376 		} else {
2377 			if (PTE_ISWIRED(pte)) {
2378 				pte->flags &= ~PTE_WIRED;
2379 				pmap->pm_stats.wired_count--;
2380 			}
2381 		}
2382 	}
2383 	PMAP_UNLOCK(pmap);
2384 }
2385 
2386 /*
2387  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2388  * page.  This count may be changed upwards or downwards in the future; it is
2389  * only necessary that true be returned for a small subset of pmaps for proper
2390  * page aging.
2391  */
2392 static boolean_t
2393 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2394 {
2395 	pv_entry_t pv;
2396 	int loops;
2397 	boolean_t rv;
2398 
2399 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2400 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2401 	loops = 0;
2402 	rv = FALSE;
2403 	vm_page_lock_queues();
2404 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2405 		if (pv->pv_pmap == pmap) {
2406 			rv = TRUE;
2407 			break;
2408 		}
2409 		if (++loops >= 16)
2410 			break;
2411 	}
2412 	vm_page_unlock_queues();
2413 	return (rv);
2414 }
2415 
2416 /*
2417  * Return the number of managed mappings to the given physical page that are
2418  * wired.
2419  */
2420 static int
2421 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2422 {
2423 	pv_entry_t pv;
2424 	pte_t *pte;
2425 	int count = 0;
2426 
2427 	if ((m->oflags & VPO_UNMANAGED) != 0)
2428 		return (count);
2429 	vm_page_lock_queues();
2430 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2431 		PMAP_LOCK(pv->pv_pmap);
2432 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2433 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2434 				count++;
2435 		PMAP_UNLOCK(pv->pv_pmap);
2436 	}
2437 	vm_page_unlock_queues();
2438 	return (count);
2439 }
2440 
2441 static int
2442 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2443 {
2444 	int i;
2445 	vm_offset_t va;
2446 
2447 	/*
2448 	 * This currently does not work for entries that
2449 	 * overlap TLB1 entries.
2450 	 */
2451 	for (i = 0; i < tlb1_idx; i ++) {
2452 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2453 			return (0);
2454 	}
2455 
2456 	return (EFAULT);
2457 }
2458 
2459 vm_offset_t
2460 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2461     vm_size_t *sz)
2462 {
2463 	vm_paddr_t pa, ppa;
2464 	vm_offset_t va;
2465 	vm_size_t gran;
2466 
2467 	/* Raw physical memory dumps don't have a virtual address. */
2468 	if (md->md_vaddr == ~0UL) {
2469 		/* We always map a 256MB page at 256M. */
2470 		gran = 256 * 1024 * 1024;
2471 		pa = md->md_paddr + ofs;
2472 		ppa = pa & ~(gran - 1);
2473 		ofs = pa - ppa;
2474 		va = gran;
2475 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2476 		if (*sz > (gran - ofs))
2477 			*sz = gran - ofs;
2478 		return (va + ofs);
2479 	}
2480 
2481 	/* Minidumps are based on virtual memory addresses. */
2482 	va = md->md_vaddr + ofs;
2483 	if (va >= kernstart + kernsize) {
2484 		gran = PAGE_SIZE - (va & PAGE_MASK);
2485 		if (*sz > gran)
2486 			*sz = gran;
2487 	}
2488 	return (va);
2489 }
2490 
2491 void
2492 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2493     vm_offset_t va)
2494 {
2495 
2496 	/* Raw physical memory dumps don't have a virtual address. */
2497 	if (md->md_vaddr == ~0UL) {
2498 		tlb1_idx--;
2499 		tlb1[tlb1_idx].mas1 = 0;
2500 		tlb1[tlb1_idx].mas2 = 0;
2501 		tlb1[tlb1_idx].mas3 = 0;
2502 		tlb1_write_entry(tlb1_idx);
2503 		return;
2504 	}
2505 
2506 	/* Minidumps are based on virtual memory addresses. */
2507 	/* Nothing to do... */
2508 }
2509 
2510 struct pmap_md *
2511 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2512 {
2513 	static struct pmap_md md;
2514 	pte_t *pte;
2515 	vm_offset_t va;
2516 
2517 	if (dumpsys_minidump) {
2518 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2519 		if (prev == NULL) {
2520 			/* 1st: kernel .data and .bss. */
2521 			md.md_index = 1;
2522 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2523 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2524 			return (&md);
2525 		}
2526 		switch (prev->md_index) {
2527 		case 1:
2528 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2529 			md.md_index = 2;
2530 			md.md_vaddr = data_start;
2531 			md.md_size = data_end - data_start;
2532 			break;
2533 		case 2:
2534 			/* 3rd: kernel VM. */
2535 			va = prev->md_vaddr + prev->md_size;
2536 			/* Find start of next chunk (from va). */
2537 			while (va < virtual_end) {
2538 				/* Don't dump the buffer cache. */
2539 				if (va >= kmi.buffer_sva &&
2540 				    va < kmi.buffer_eva) {
2541 					va = kmi.buffer_eva;
2542 					continue;
2543 				}
2544 				pte = pte_find(mmu, kernel_pmap, va);
2545 				if (pte != NULL && PTE_ISVALID(pte))
2546 					break;
2547 				va += PAGE_SIZE;
2548 			}
2549 			if (va < virtual_end) {
2550 				md.md_vaddr = va;
2551 				va += PAGE_SIZE;
2552 				/* Find last page in chunk. */
2553 				while (va < virtual_end) {
2554 					/* Don't run into the buffer cache. */
2555 					if (va == kmi.buffer_sva)
2556 						break;
2557 					pte = pte_find(mmu, kernel_pmap, va);
2558 					if (pte == NULL || !PTE_ISVALID(pte))
2559 						break;
2560 					va += PAGE_SIZE;
2561 				}
2562 				md.md_size = va - md.md_vaddr;
2563 				break;
2564 			}
2565 			md.md_index = 3;
2566 			/* FALLTHROUGH */
2567 		default:
2568 			return (NULL);
2569 		}
2570 	} else { /* minidumps */
2571 		mem_regions(&physmem_regions, &physmem_regions_sz,
2572 		    &availmem_regions, &availmem_regions_sz);
2573 
2574 		if (prev == NULL) {
2575 			/* first physical chunk. */
2576 			md.md_paddr = physmem_regions[0].mr_start;
2577 			md.md_size = physmem_regions[0].mr_size;
2578 			md.md_vaddr = ~0UL;
2579 			md.md_index = 1;
2580 		} else if (md.md_index < physmem_regions_sz) {
2581 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2582 			md.md_size = physmem_regions[md.md_index].mr_size;
2583 			md.md_vaddr = ~0UL;
2584 			md.md_index++;
2585 		} else {
2586 			/* There's no next physical chunk. */
2587 			return (NULL);
2588 		}
2589 	}
2590 
2591 	return (&md);
2592 }
2593 
2594 /*
2595  * Map a set of physical memory pages into the kernel virtual address space.
2596  * Return a pointer to where it is mapped. This routine is intended to be used
2597  * for mapping device memory, NOT real memory.
2598  */
2599 static void *
2600 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2601 {
2602 	void *res;
2603 	uintptr_t va;
2604 	vm_size_t sz;
2605 
2606 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2607 	res = (void *)va;
2608 
2609 	do {
2610 		sz = 1 << (ilog2(size) & ~1);
2611 		if (bootverbose)
2612 			printf("Wiring VA=%x to PA=%x (size=%x), "
2613 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2614 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2615 		size -= sz;
2616 		pa += sz;
2617 		va += sz;
2618 	} while (size > 0);
2619 
2620 	return (res);
2621 }
2622 
2623 /*
2624  * 'Unmap' a range mapped by mmu_booke_mapdev().
2625  */
2626 static void
2627 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2628 {
2629 	vm_offset_t base, offset;
2630 
2631 	/*
2632 	 * Unmap only if this is inside kernel virtual space.
2633 	 */
2634 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2635 		base = trunc_page(va);
2636 		offset = va & PAGE_MASK;
2637 		size = roundup(offset + size, PAGE_SIZE);
2638 		kmem_free(kernel_map, base, size);
2639 	}
2640 }
2641 
2642 /*
2643  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2644  * specified pmap. This eliminates the blast of soft faults on process startup
2645  * and immediately after an mmap.
2646  */
2647 static void
2648 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2649     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2650 {
2651 
2652 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2653 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2654 	    ("mmu_booke_object_init_pt: non-device object"));
2655 }
2656 
2657 /*
2658  * Perform the pmap work for mincore.
2659  */
2660 static int
2661 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2662     vm_paddr_t *locked_pa)
2663 {
2664 
2665 	TODO;
2666 	return (0);
2667 }
2668 
2669 /**************************************************************************/
2670 /* TID handling */
2671 /**************************************************************************/
2672 
2673 /*
2674  * Allocate a TID. If necessary, steal one from someone else.
2675  * The new TID is flushed from the TLB before returning.
2676  */
2677 static tlbtid_t
2678 tid_alloc(pmap_t pmap)
2679 {
2680 	tlbtid_t tid;
2681 	int thiscpu;
2682 
2683 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2684 
2685 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2686 
2687 	thiscpu = PCPU_GET(cpuid);
2688 
2689 	tid = PCPU_GET(tid_next);
2690 	if (tid > TID_MAX)
2691 		tid = TID_MIN;
2692 	PCPU_SET(tid_next, tid + 1);
2693 
2694 	/* If we are stealing TID then clear the relevant pmap's field */
2695 	if (tidbusy[thiscpu][tid] != NULL) {
2696 
2697 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2698 
2699 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2700 
2701 		/* Flush all entries from TLB0 matching this TID. */
2702 		tid_flush(tid);
2703 	}
2704 
2705 	tidbusy[thiscpu][tid] = pmap;
2706 	pmap->pm_tid[thiscpu] = tid;
2707 	__asm __volatile("msync; isync");
2708 
2709 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2710 	    PCPU_GET(tid_next));
2711 
2712 	return (tid);
2713 }
2714 
2715 /**************************************************************************/
2716 /* TLB0 handling */
2717 /**************************************************************************/
2718 
2719 static void
2720 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2721     uint32_t mas7)
2722 {
2723 	int as;
2724 	char desc[3];
2725 	tlbtid_t tid;
2726 	vm_size_t size;
2727 	unsigned int tsize;
2728 
2729 	desc[2] = '\0';
2730 	if (mas1 & MAS1_VALID)
2731 		desc[0] = 'V';
2732 	else
2733 		desc[0] = ' ';
2734 
2735 	if (mas1 & MAS1_IPROT)
2736 		desc[1] = 'P';
2737 	else
2738 		desc[1] = ' ';
2739 
2740 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2741 	tid = MAS1_GETTID(mas1);
2742 
2743 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2744 	size = 0;
2745 	if (tsize)
2746 		size = tsize2size(tsize);
2747 
2748 	debugf("%3d: (%s) [AS=%d] "
2749 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2750 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2751 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2752 }
2753 
2754 /* Convert TLB0 va and way number to tlb0[] table index. */
2755 static inline unsigned int
2756 tlb0_tableidx(vm_offset_t va, unsigned int way)
2757 {
2758 	unsigned int idx;
2759 
2760 	idx = (way * TLB0_ENTRIES_PER_WAY);
2761 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2762 	return (idx);
2763 }
2764 
2765 /*
2766  * Invalidate TLB0 entry.
2767  */
2768 static inline void
2769 tlb0_flush_entry(vm_offset_t va)
2770 {
2771 
2772 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2773 
2774 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2775 
2776 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2777 	__asm __volatile("isync; msync");
2778 	__asm __volatile("tlbsync; msync");
2779 
2780 	CTR1(KTR_PMAP, "%s: e", __func__);
2781 }
2782 
2783 /* Print out contents of the MAS registers for each TLB0 entry */
2784 void
2785 tlb0_print_tlbentries(void)
2786 {
2787 	uint32_t mas0, mas1, mas2, mas3, mas7;
2788 	int entryidx, way, idx;
2789 
2790 	debugf("TLB0 entries:\n");
2791 	for (way = 0; way < TLB0_WAYS; way ++)
2792 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2793 
2794 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2795 			mtspr(SPR_MAS0, mas0);
2796 			__asm __volatile("isync");
2797 
2798 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2799 			mtspr(SPR_MAS2, mas2);
2800 
2801 			__asm __volatile("isync; tlbre");
2802 
2803 			mas1 = mfspr(SPR_MAS1);
2804 			mas2 = mfspr(SPR_MAS2);
2805 			mas3 = mfspr(SPR_MAS3);
2806 			mas7 = mfspr(SPR_MAS7);
2807 
2808 			idx = tlb0_tableidx(mas2, way);
2809 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2810 		}
2811 }
2812 
2813 /**************************************************************************/
2814 /* TLB1 handling */
2815 /**************************************************************************/
2816 
2817 /*
2818  * TLB1 mapping notes:
2819  *
2820  * TLB1[0]	CCSRBAR
2821  * TLB1[1]	Kernel text and data.
2822  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2823  *		windows, other devices mappings.
2824  */
2825 
2826 /*
2827  * Write given entry to TLB1 hardware.
2828  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2829  */
2830 static void
2831 tlb1_write_entry(unsigned int idx)
2832 {
2833 	uint32_t mas0, mas7;
2834 
2835 	//debugf("tlb1_write_entry: s\n");
2836 
2837 	/* Clear high order RPN bits */
2838 	mas7 = 0;
2839 
2840 	/* Select entry */
2841 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2842 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2843 
2844 	mtspr(SPR_MAS0, mas0);
2845 	__asm __volatile("isync");
2846 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2847 	__asm __volatile("isync");
2848 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2849 	__asm __volatile("isync");
2850 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2851 	__asm __volatile("isync");
2852 	mtspr(SPR_MAS7, mas7);
2853 	__asm __volatile("isync; tlbwe; isync; msync");
2854 
2855 	//debugf("tlb1_write_entry: e\n");
2856 }
2857 
2858 /*
2859  * Return the largest uint value log such that 2^log <= num.
2860  */
2861 static unsigned int
2862 ilog2(unsigned int num)
2863 {
2864 	int lz;
2865 
2866 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2867 	return (31 - lz);
2868 }
2869 
2870 /*
2871  * Convert TLB TSIZE value to mapped region size.
2872  */
2873 static vm_size_t
2874 tsize2size(unsigned int tsize)
2875 {
2876 
2877 	/*
2878 	 * size = 4^tsize KB
2879 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2880 	 */
2881 
2882 	return ((1 << (2 * tsize)) * 1024);
2883 }
2884 
2885 /*
2886  * Convert region size (must be power of 4) to TLB TSIZE value.
2887  */
2888 static unsigned int
2889 size2tsize(vm_size_t size)
2890 {
2891 
2892 	return (ilog2(size) / 2 - 5);
2893 }
2894 
2895 /*
2896  * Register permanent kernel mapping in TLB1.
2897  *
2898  * Entries are created starting from index 0 (current free entry is
2899  * kept in tlb1_idx) and are not supposed to be invalidated.
2900  */
2901 static int
2902 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2903     uint32_t flags)
2904 {
2905 	uint32_t ts, tid;
2906 	int tsize;
2907 
2908 	if (tlb1_idx >= TLB1_ENTRIES) {
2909 		printf("tlb1_set_entry: TLB1 full!\n");
2910 		return (-1);
2911 	}
2912 
2913 	/* Convert size to TSIZE */
2914 	tsize = size2tsize(size);
2915 
2916 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2917 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2918 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2919 
2920 	/* XXX LOCK tlb1[] */
2921 
2922 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2923 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2924 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2925 
2926 	/* Set supervisor RWX permission bits */
2927 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2928 
2929 	tlb1_write_entry(tlb1_idx++);
2930 
2931 	/* XXX UNLOCK tlb1[] */
2932 
2933 	/*
2934 	 * XXX in general TLB1 updates should be propagated between CPUs,
2935 	 * since current design assumes to have the same TLB1 set-up on all
2936 	 * cores.
2937 	 */
2938 	return (0);
2939 }
2940 
2941 /*
2942  * Map in contiguous RAM region into the TLB1 using maximum of
2943  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2944  *
2945  * If necessary round up last entry size and return total size
2946  * used by all allocated entries.
2947  */
2948 vm_size_t
2949 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
2950 {
2951 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
2952 	vm_size_t mapped, pgsz, base, mask;
2953 	int idx, nents;
2954 
2955 	/* Round up to the next 1M */
2956 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
2957 
2958 	mapped = 0;
2959 	idx = 0;
2960 	base = va;
2961 	pgsz = 64*1024*1024;
2962 	while (mapped < size) {
2963 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
2964 			while (pgsz > (size - mapped))
2965 				pgsz >>= 2;
2966 			pgs[idx++] = pgsz;
2967 			mapped += pgsz;
2968 		}
2969 
2970 		/* We under-map. Correct for this. */
2971 		if (mapped < size) {
2972 			while (pgs[idx - 1] == pgsz) {
2973 				idx--;
2974 				mapped -= pgsz;
2975 			}
2976 			/* XXX We may increase beyond out starting point. */
2977 			pgsz <<= 2;
2978 			pgs[idx++] = pgsz;
2979 			mapped += pgsz;
2980 		}
2981 	}
2982 
2983 	nents = idx;
2984 	mask = pgs[0] - 1;
2985 	/* Align address to the boundary */
2986 	if (va & mask) {
2987 		va = (va + mask) & ~mask;
2988 		pa = (pa + mask) & ~mask;
2989 	}
2990 
2991 	for (idx = 0; idx < nents; idx++) {
2992 		pgsz = pgs[idx];
2993 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
2994 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
2995 		pa += pgsz;
2996 		va += pgsz;
2997 	}
2998 
2999 	mapped = (va - base);
3000 	debugf("mapped size 0x%08x (wasted space 0x%08x)\n",
3001 	    mapped, mapped - size);
3002 	return (mapped);
3003 }
3004 
3005 /*
3006  * TLB1 initialization routine, to be called after the very first
3007  * assembler level setup done in locore.S.
3008  */
3009 void
3010 tlb1_init(vm_offset_t ccsrbar)
3011 {
3012 	uint32_t mas0, mas1, mas3;
3013 	uint32_t tsz;
3014 	u_int i;
3015 
3016 	if (bootinfo != NULL && bootinfo[0] != 1) {
3017 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3018 	} else
3019 		tlb1_idx = 1;
3020 
3021 	/* The first entry/entries are used to map the kernel. */
3022 	for (i = 0; i < tlb1_idx; i++) {
3023 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3024 		mtspr(SPR_MAS0, mas0);
3025 		__asm __volatile("isync; tlbre");
3026 
3027 		mas1 = mfspr(SPR_MAS1);
3028 		if ((mas1 & MAS1_VALID) == 0)
3029 			continue;
3030 
3031 		mas3 = mfspr(SPR_MAS3);
3032 
3033 		tlb1[i].mas1 = mas1;
3034 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3035 		tlb1[i].mas3 = mas3;
3036 
3037 		if (i == 0)
3038 			kernload = mas3 & MAS3_RPN;
3039 
3040 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3041 		kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
3042 	}
3043 
3044 	/* Map in CCSRBAR. */
3045 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
3046 
3047 	/* Setup TLB miss defaults */
3048 	set_mas4_defaults();
3049 }
3050 
3051 /*
3052  * Setup MAS4 defaults.
3053  * These values are loaded to MAS0-2 on a TLB miss.
3054  */
3055 static void
3056 set_mas4_defaults(void)
3057 {
3058 	uint32_t mas4;
3059 
3060 	/* Defaults: TLB0, PID0, TSIZED=4K */
3061 	mas4 = MAS4_TLBSELD0;
3062 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3063 #ifdef SMP
3064 	mas4 |= MAS4_MD;
3065 #endif
3066 	mtspr(SPR_MAS4, mas4);
3067 	__asm __volatile("isync");
3068 }
3069 
3070 /*
3071  * Print out contents of the MAS registers for each TLB1 entry
3072  */
3073 void
3074 tlb1_print_tlbentries(void)
3075 {
3076 	uint32_t mas0, mas1, mas2, mas3, mas7;
3077 	int i;
3078 
3079 	debugf("TLB1 entries:\n");
3080 	for (i = 0; i < TLB1_ENTRIES; i++) {
3081 
3082 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3083 		mtspr(SPR_MAS0, mas0);
3084 
3085 		__asm __volatile("isync; tlbre");
3086 
3087 		mas1 = mfspr(SPR_MAS1);
3088 		mas2 = mfspr(SPR_MAS2);
3089 		mas3 = mfspr(SPR_MAS3);
3090 		mas7 = mfspr(SPR_MAS7);
3091 
3092 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3093 	}
3094 }
3095 
3096 /*
3097  * Print out contents of the in-ram tlb1 table.
3098  */
3099 void
3100 tlb1_print_entries(void)
3101 {
3102 	int i;
3103 
3104 	debugf("tlb1[] table entries:\n");
3105 	for (i = 0; i < TLB1_ENTRIES; i++)
3106 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3107 }
3108 
3109 /*
3110  * Return 0 if the physical IO range is encompassed by one of the
3111  * the TLB1 entries, otherwise return related error code.
3112  */
3113 static int
3114 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3115 {
3116 	uint32_t prot;
3117 	vm_paddr_t pa_start;
3118 	vm_paddr_t pa_end;
3119 	unsigned int entry_tsize;
3120 	vm_size_t entry_size;
3121 
3122 	*va = (vm_offset_t)NULL;
3123 
3124 	/* Skip invalid entries */
3125 	if (!(tlb1[i].mas1 & MAS1_VALID))
3126 		return (EINVAL);
3127 
3128 	/*
3129 	 * The entry must be cache-inhibited, guarded, and r/w
3130 	 * so it can function as an i/o page
3131 	 */
3132 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3133 	if (prot != (MAS2_I | MAS2_G))
3134 		return (EPERM);
3135 
3136 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3137 	if (prot != (MAS3_SR | MAS3_SW))
3138 		return (EPERM);
3139 
3140 	/* The address should be within the entry range. */
3141 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3142 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3143 
3144 	entry_size = tsize2size(entry_tsize);
3145 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3146 	pa_end = pa_start + entry_size - 1;
3147 
3148 	if ((pa < pa_start) || ((pa + size) > pa_end))
3149 		return (ERANGE);
3150 
3151 	/* Return virtual address of this mapping. */
3152 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3153 	return (0);
3154 }
3155