xref: /freebsd/sys/powerpc/booke/pmap.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
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/param.h>
55 #include <sys/malloc.h>
56 #include <sys/ktr.h>
57 #include <sys/proc.h>
58 #include <sys/user.h>
59 #include <sys/queue.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/linker.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/rwlock.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/md_var.h>
89 #include <machine/mmuvar.h>
90 #include <machine/pmap.h>
91 #include <machine/pte.h>
92 
93 #include "mmu_if.h"
94 
95 #ifdef  DEBUG
96 #define debugf(fmt, args...) printf(fmt, ##args)
97 #else
98 #define debugf(fmt, args...)
99 #endif
100 
101 #define TODO			panic("%s: not implemented", __func__);
102 
103 extern struct mtx sched_lock;
104 
105 extern int dumpsys_minidump;
106 
107 extern unsigned char _etext[];
108 extern unsigned char _end[];
109 
110 extern uint32_t *bootinfo;
111 
112 #ifdef SMP
113 extern uint32_t bp_ntlb1s;
114 #endif
115 
116 vm_paddr_t ccsrbar_pa;
117 vm_paddr_t kernload;
118 vm_offset_t kernstart;
119 vm_size_t kernsize;
120 
121 /* Message buffer and tables. */
122 static vm_offset_t data_start;
123 static vm_size_t data_end;
124 
125 /* Phys/avail memory regions. */
126 static struct mem_region *availmem_regions;
127 static int availmem_regions_sz;
128 static struct mem_region *physmem_regions;
129 static int physmem_regions_sz;
130 
131 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
132 static vm_offset_t zero_page_va;
133 static struct mtx zero_page_mutex;
134 
135 static struct mtx tlbivax_mutex;
136 
137 /*
138  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
139  * by idle thred only, no lock required.
140  */
141 static vm_offset_t zero_page_idle_va;
142 
143 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
144 static vm_offset_t copy_page_src_va;
145 static vm_offset_t copy_page_dst_va;
146 static struct mtx copy_page_mutex;
147 
148 /**************************************************************************/
149 /* PMAP */
150 /**************************************************************************/
151 
152 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
153     vm_prot_t, boolean_t);
154 
155 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
156 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
157 
158 /*
159  * If user pmap is processed with mmu_booke_remove and the resident count
160  * drops to 0, there are no more pages to remove, so we need not continue.
161  */
162 #define PMAP_REMOVE_DONE(pmap) \
163 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
164 
165 extern void tid_flush(tlbtid_t);
166 
167 /**************************************************************************/
168 /* TLB and TID handling */
169 /**************************************************************************/
170 
171 /* Translation ID busy table */
172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 
174 /*
175  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
176  * core revisions and should be read from h/w registers during early config.
177  */
178 uint32_t tlb0_entries;
179 uint32_t tlb0_ways;
180 uint32_t tlb0_entries_per_way;
181 
182 #define TLB0_ENTRIES		(tlb0_entries)
183 #define TLB0_WAYS		(tlb0_ways)
184 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
185 
186 #define TLB1_ENTRIES 16
187 
188 /* In-ram copy of the TLB1 */
189 static tlb_entry_t tlb1[TLB1_ENTRIES];
190 
191 /* Next free entry in the TLB1 */
192 static unsigned int tlb1_idx;
193 
194 static tlbtid_t tid_alloc(struct pmap *);
195 
196 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
197 
198 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
199 static void tlb1_write_entry(unsigned int);
200 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
201 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
202 
203 static vm_size_t tsize2size(unsigned int);
204 static unsigned int size2tsize(vm_size_t);
205 static unsigned int ilog2(unsigned int);
206 
207 static void set_mas4_defaults(void);
208 
209 static inline void tlb0_flush_entry(vm_offset_t);
210 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
211 
212 /**************************************************************************/
213 /* Page table management */
214 /**************************************************************************/
215 
216 static struct rwlock_padalign pvh_global_lock;
217 
218 /* Data for the pv entry allocation mechanism */
219 static uma_zone_t pvzone;
220 static struct vm_object pvzone_obj;
221 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
222 
223 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
224 
225 #ifndef PMAP_SHPGPERPROC
226 #define PMAP_SHPGPERPROC	200
227 #endif
228 
229 static void ptbl_init(void);
230 static struct ptbl_buf *ptbl_buf_alloc(void);
231 static void ptbl_buf_free(struct ptbl_buf *);
232 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
233 
234 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
235 static void ptbl_free(mmu_t, pmap_t, unsigned int);
236 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
237 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
238 
239 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
240 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
241 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
242 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
243 
244 static pv_entry_t pv_alloc(void);
245 static void pv_free(pv_entry_t);
246 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
247 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
248 
249 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
250 #define PTBL_BUFS		(128 * 16)
251 
252 struct ptbl_buf {
253 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
254 	vm_offset_t kva;		/* va of mapping */
255 };
256 
257 /* ptbl free list and a lock used for access synchronization. */
258 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
259 static struct mtx ptbl_buf_freelist_lock;
260 
261 /* Base address of kva space allocated fot ptbl bufs. */
262 static vm_offset_t ptbl_buf_pool_vabase;
263 
264 /* Pointer to ptbl_buf structures. */
265 static struct ptbl_buf *ptbl_bufs;
266 
267 void pmap_bootstrap_ap(volatile uint32_t *);
268 
269 /*
270  * Kernel MMU interface
271  */
272 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
273 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
274 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
275 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
276     vm_size_t, vm_offset_t);
277 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
278 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
279     vm_prot_t, boolean_t);
280 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
281     vm_page_t, vm_prot_t);
282 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
283     vm_prot_t);
284 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
285 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
286     vm_prot_t);
287 static void		mmu_booke_init(mmu_t);
288 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
289 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
290 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
291 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
292 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
293     int);
294 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
295     vm_paddr_t *);
296 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
297     vm_object_t, vm_pindex_t, vm_size_t);
298 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
299 static void		mmu_booke_page_init(mmu_t, vm_page_t);
300 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
301 static void		mmu_booke_pinit(mmu_t, pmap_t);
302 static void		mmu_booke_pinit0(mmu_t, pmap_t);
303 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
304     vm_prot_t);
305 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
306 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
307 static void		mmu_booke_release(mmu_t, pmap_t);
308 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
309 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
310 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
311 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
312 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
313 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
314 static void		mmu_booke_activate(mmu_t, struct thread *);
315 static void		mmu_booke_deactivate(mmu_t, struct thread *);
316 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
317 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
318 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
319 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
320 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
321 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
322 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
323 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
324     vm_size_t);
325 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
326     vm_size_t, vm_size_t *);
327 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
328     vm_size_t, vm_offset_t);
329 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
330 
331 static mmu_method_t mmu_booke_methods[] = {
332 	/* pmap dispatcher interface */
333 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
334 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
335 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
336 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
337 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
338 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
339 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
340 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
341 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
342 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
343 	MMUMETHOD(mmu_init,		mmu_booke_init),
344 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
345 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
346 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
347 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
348 	MMUMETHOD(mmu_map,		mmu_booke_map),
349 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
350 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
351 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
352 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
353 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
354 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
355 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
356 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
357 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
358 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
359 	MMUMETHOD(mmu_release,		mmu_booke_release),
360 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
361 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
362 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
363 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
364 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
365 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
366 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
367 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
368 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
369 
370 	/* Internal interfaces */
371 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
372 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
373 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
374 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
375 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
376 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
377 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
378 
379 	/* dumpsys() support */
380 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
381 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
382 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
383 
384 	{ 0, 0 }
385 };
386 
387 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
388 
389 static inline void
390 tlb_miss_lock(void)
391 {
392 #ifdef SMP
393 	struct pcpu *pc;
394 
395 	if (!smp_started)
396 		return;
397 
398 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
399 		if (pc != pcpup) {
400 
401 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
402 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
403 
404 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
405 			    ("tlb_miss_lock: tried to lock self"));
406 
407 			tlb_lock(pc->pc_booke_tlb_lock);
408 
409 			CTR1(KTR_PMAP, "%s: locked", __func__);
410 		}
411 	}
412 #endif
413 }
414 
415 static inline void
416 tlb_miss_unlock(void)
417 {
418 #ifdef SMP
419 	struct pcpu *pc;
420 
421 	if (!smp_started)
422 		return;
423 
424 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
425 		if (pc != pcpup) {
426 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
427 			    __func__, pc->pc_cpuid);
428 
429 			tlb_unlock(pc->pc_booke_tlb_lock);
430 
431 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
432 		}
433 	}
434 #endif
435 }
436 
437 /* Return number of entries in TLB0. */
438 static __inline void
439 tlb0_get_tlbconf(void)
440 {
441 	uint32_t tlb0_cfg;
442 
443 	tlb0_cfg = mfspr(SPR_TLB0CFG);
444 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
445 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
446 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
447 }
448 
449 /* Initialize pool of kva ptbl buffers. */
450 static void
451 ptbl_init(void)
452 {
453 	int i;
454 
455 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
456 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
457 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
458 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
459 
460 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
461 	TAILQ_INIT(&ptbl_buf_freelist);
462 
463 	for (i = 0; i < PTBL_BUFS; i++) {
464 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
465 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
466 	}
467 }
468 
469 /* Get a ptbl_buf from the freelist. */
470 static struct ptbl_buf *
471 ptbl_buf_alloc(void)
472 {
473 	struct ptbl_buf *buf;
474 
475 	mtx_lock(&ptbl_buf_freelist_lock);
476 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
477 	if (buf != NULL)
478 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
479 	mtx_unlock(&ptbl_buf_freelist_lock);
480 
481 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
482 
483 	return (buf);
484 }
485 
486 /* Return ptbl buff to free pool. */
487 static void
488 ptbl_buf_free(struct ptbl_buf *buf)
489 {
490 
491 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
492 
493 	mtx_lock(&ptbl_buf_freelist_lock);
494 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
495 	mtx_unlock(&ptbl_buf_freelist_lock);
496 }
497 
498 /*
499  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
500  */
501 static void
502 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
503 {
504 	struct ptbl_buf *pbuf;
505 
506 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
507 
508 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
509 
510 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
511 		if (pbuf->kva == (vm_offset_t)ptbl) {
512 			/* Remove from pmap ptbl buf list. */
513 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
514 
515 			/* Free corresponding ptbl buf. */
516 			ptbl_buf_free(pbuf);
517 			break;
518 		}
519 }
520 
521 /* Allocate page table. */
522 static pte_t *
523 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
524 {
525 	vm_page_t mtbl[PTBL_PAGES];
526 	vm_page_t m;
527 	struct ptbl_buf *pbuf;
528 	unsigned int pidx;
529 	pte_t *ptbl;
530 	int i;
531 
532 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
533 	    (pmap == kernel_pmap), pdir_idx);
534 
535 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
536 	    ("ptbl_alloc: invalid pdir_idx"));
537 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
538 	    ("pte_alloc: valid ptbl entry exists!"));
539 
540 	pbuf = ptbl_buf_alloc();
541 	if (pbuf == NULL)
542 		panic("pte_alloc: couldn't alloc kernel virtual memory");
543 
544 	ptbl = (pte_t *)pbuf->kva;
545 
546 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
547 
548 	/* Allocate ptbl pages, this will sleep! */
549 	for (i = 0; i < PTBL_PAGES; i++) {
550 		pidx = (PTBL_PAGES * pdir_idx) + i;
551 		while ((m = vm_page_alloc(NULL, pidx,
552 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
553 
554 			PMAP_UNLOCK(pmap);
555 			rw_wunlock(&pvh_global_lock);
556 			VM_WAIT;
557 			rw_wlock(&pvh_global_lock);
558 			PMAP_LOCK(pmap);
559 		}
560 		mtbl[i] = m;
561 	}
562 
563 	/* Map allocated pages into kernel_pmap. */
564 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
565 
566 	/* Zero whole ptbl. */
567 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
568 
569 	/* Add pbuf to the pmap ptbl bufs list. */
570 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
571 
572 	return (ptbl);
573 }
574 
575 /* Free ptbl pages and invalidate pdir entry. */
576 static void
577 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
578 {
579 	pte_t *ptbl;
580 	vm_paddr_t pa;
581 	vm_offset_t va;
582 	vm_page_t m;
583 	int i;
584 
585 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
586 	    (pmap == kernel_pmap), pdir_idx);
587 
588 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
589 	    ("ptbl_free: invalid pdir_idx"));
590 
591 	ptbl = pmap->pm_pdir[pdir_idx];
592 
593 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
594 
595 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
596 
597 	/*
598 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
599 	 * don't attempt to look up the page tables we are releasing.
600 	 */
601 	mtx_lock_spin(&tlbivax_mutex);
602 	tlb_miss_lock();
603 
604 	pmap->pm_pdir[pdir_idx] = NULL;
605 
606 	tlb_miss_unlock();
607 	mtx_unlock_spin(&tlbivax_mutex);
608 
609 	for (i = 0; i < PTBL_PAGES; i++) {
610 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
611 		pa = pte_vatopa(mmu, kernel_pmap, va);
612 		m = PHYS_TO_VM_PAGE(pa);
613 		vm_page_free_zero(m);
614 		atomic_subtract_int(&cnt.v_wire_count, 1);
615 		mmu_booke_kremove(mmu, va);
616 	}
617 
618 	ptbl_free_pmap_ptbl(pmap, ptbl);
619 }
620 
621 /*
622  * Decrement ptbl pages hold count and attempt to free ptbl pages.
623  * Called when removing pte entry from ptbl.
624  *
625  * Return 1 if ptbl pages were freed.
626  */
627 static int
628 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
629 {
630 	pte_t *ptbl;
631 	vm_paddr_t pa;
632 	vm_page_t m;
633 	int i;
634 
635 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
636 	    (pmap == kernel_pmap), pdir_idx);
637 
638 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
639 	    ("ptbl_unhold: invalid pdir_idx"));
640 	KASSERT((pmap != kernel_pmap),
641 	    ("ptbl_unhold: unholding kernel ptbl!"));
642 
643 	ptbl = pmap->pm_pdir[pdir_idx];
644 
645 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
646 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
647 	    ("ptbl_unhold: non kva ptbl"));
648 
649 	/* decrement hold count */
650 	for (i = 0; i < PTBL_PAGES; i++) {
651 		pa = pte_vatopa(mmu, kernel_pmap,
652 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
653 		m = PHYS_TO_VM_PAGE(pa);
654 		m->wire_count--;
655 	}
656 
657 	/*
658 	 * Free ptbl pages if there are no pte etries in this ptbl.
659 	 * wire_count has the same value for all ptbl pages, so check the last
660 	 * page.
661 	 */
662 	if (m->wire_count == 0) {
663 		ptbl_free(mmu, pmap, pdir_idx);
664 
665 		//debugf("ptbl_unhold: e (freed ptbl)\n");
666 		return (1);
667 	}
668 
669 	return (0);
670 }
671 
672 /*
673  * Increment hold count for ptbl pages. This routine is used when a new pte
674  * entry is being inserted into the ptbl.
675  */
676 static void
677 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
678 {
679 	vm_paddr_t pa;
680 	pte_t *ptbl;
681 	vm_page_t m;
682 	int i;
683 
684 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
685 	    pdir_idx);
686 
687 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
688 	    ("ptbl_hold: invalid pdir_idx"));
689 	KASSERT((pmap != kernel_pmap),
690 	    ("ptbl_hold: holding kernel ptbl!"));
691 
692 	ptbl = pmap->pm_pdir[pdir_idx];
693 
694 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
695 
696 	for (i = 0; i < PTBL_PAGES; i++) {
697 		pa = pte_vatopa(mmu, kernel_pmap,
698 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
699 		m = PHYS_TO_VM_PAGE(pa);
700 		m->wire_count++;
701 	}
702 }
703 
704 /* Allocate pv_entry structure. */
705 pv_entry_t
706 pv_alloc(void)
707 {
708 	pv_entry_t pv;
709 
710 	pv_entry_count++;
711 	if (pv_entry_count > pv_entry_high_water)
712 		pagedaemon_wakeup();
713 	pv = uma_zalloc(pvzone, M_NOWAIT);
714 
715 	return (pv);
716 }
717 
718 /* Free pv_entry structure. */
719 static __inline void
720 pv_free(pv_entry_t pve)
721 {
722 
723 	pv_entry_count--;
724 	uma_zfree(pvzone, pve);
725 }
726 
727 
728 /* Allocate and initialize pv_entry structure. */
729 static void
730 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
731 {
732 	pv_entry_t pve;
733 
734 	//int su = (pmap == kernel_pmap);
735 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
736 	//	(u_int32_t)pmap, va, (u_int32_t)m);
737 
738 	pve = pv_alloc();
739 	if (pve == NULL)
740 		panic("pv_insert: no pv entries!");
741 
742 	pve->pv_pmap = pmap;
743 	pve->pv_va = va;
744 
745 	/* add to pv_list */
746 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
747 	rw_assert(&pvh_global_lock, RA_WLOCKED);
748 
749 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
750 
751 	//debugf("pv_insert: e\n");
752 }
753 
754 /* Destroy pv entry. */
755 static void
756 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
757 {
758 	pv_entry_t pve;
759 
760 	//int su = (pmap == kernel_pmap);
761 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
762 
763 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
764 	rw_assert(&pvh_global_lock, RA_WLOCKED);
765 
766 	/* find pv entry */
767 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
768 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
769 			/* remove from pv_list */
770 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
771 			if (TAILQ_EMPTY(&m->md.pv_list))
772 				vm_page_aflag_clear(m, PGA_WRITEABLE);
773 
774 			/* free pv entry struct */
775 			pv_free(pve);
776 			break;
777 		}
778 	}
779 
780 	//debugf("pv_remove: e\n");
781 }
782 
783 /*
784  * Clean pte entry, try to free page table page if requested.
785  *
786  * Return 1 if ptbl pages were freed, otherwise return 0.
787  */
788 static int
789 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
790 {
791 	unsigned int pdir_idx = PDIR_IDX(va);
792 	unsigned int ptbl_idx = PTBL_IDX(va);
793 	vm_page_t m;
794 	pte_t *ptbl;
795 	pte_t *pte;
796 
797 	//int su = (pmap == kernel_pmap);
798 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
799 	//		su, (u_int32_t)pmap, va, flags);
800 
801 	ptbl = pmap->pm_pdir[pdir_idx];
802 	KASSERT(ptbl, ("pte_remove: null ptbl"));
803 
804 	pte = &ptbl[ptbl_idx];
805 
806 	if (pte == NULL || !PTE_ISVALID(pte))
807 		return (0);
808 
809 	if (PTE_ISWIRED(pte))
810 		pmap->pm_stats.wired_count--;
811 
812 	/* Handle managed entry. */
813 	if (PTE_ISMANAGED(pte)) {
814 		/* Get vm_page_t for mapped pte. */
815 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
816 
817 		if (PTE_ISMODIFIED(pte))
818 			vm_page_dirty(m);
819 
820 		if (PTE_ISREFERENCED(pte))
821 			vm_page_aflag_set(m, PGA_REFERENCED);
822 
823 		pv_remove(pmap, va, m);
824 	}
825 
826 	mtx_lock_spin(&tlbivax_mutex);
827 	tlb_miss_lock();
828 
829 	tlb0_flush_entry(va);
830 	pte->flags = 0;
831 	pte->rpn = 0;
832 
833 	tlb_miss_unlock();
834 	mtx_unlock_spin(&tlbivax_mutex);
835 
836 	pmap->pm_stats.resident_count--;
837 
838 	if (flags & PTBL_UNHOLD) {
839 		//debugf("pte_remove: e (unhold)\n");
840 		return (ptbl_unhold(mmu, pmap, pdir_idx));
841 	}
842 
843 	//debugf("pte_remove: e\n");
844 	return (0);
845 }
846 
847 /*
848  * Insert PTE for a given page and virtual address.
849  */
850 static void
851 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
852 {
853 	unsigned int pdir_idx = PDIR_IDX(va);
854 	unsigned int ptbl_idx = PTBL_IDX(va);
855 	pte_t *ptbl, *pte;
856 
857 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
858 	    pmap == kernel_pmap, pmap, va);
859 
860 	/* Get the page table pointer. */
861 	ptbl = pmap->pm_pdir[pdir_idx];
862 
863 	if (ptbl == NULL) {
864 		/* Allocate page table pages. */
865 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
866 	} else {
867 		/*
868 		 * Check if there is valid mapping for requested
869 		 * va, if there is, remove it.
870 		 */
871 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
872 		if (PTE_ISVALID(pte)) {
873 			pte_remove(mmu, pmap, va, PTBL_HOLD);
874 		} else {
875 			/*
876 			 * pte is not used, increment hold count
877 			 * for ptbl pages.
878 			 */
879 			if (pmap != kernel_pmap)
880 				ptbl_hold(mmu, pmap, pdir_idx);
881 		}
882 	}
883 
884 	/*
885 	 * Insert pv_entry into pv_list for mapped page if part of managed
886 	 * memory.
887 	 */
888 	if ((m->oflags & VPO_UNMANAGED) == 0) {
889 		flags |= PTE_MANAGED;
890 
891 		/* Create and insert pv entry. */
892 		pv_insert(pmap, va, m);
893 	}
894 
895 	pmap->pm_stats.resident_count++;
896 
897 	mtx_lock_spin(&tlbivax_mutex);
898 	tlb_miss_lock();
899 
900 	tlb0_flush_entry(va);
901 	if (pmap->pm_pdir[pdir_idx] == NULL) {
902 		/*
903 		 * If we just allocated a new page table, hook it in
904 		 * the pdir.
905 		 */
906 		pmap->pm_pdir[pdir_idx] = ptbl;
907 	}
908 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
909 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
910 	pte->flags |= (PTE_VALID | flags);
911 
912 	tlb_miss_unlock();
913 	mtx_unlock_spin(&tlbivax_mutex);
914 }
915 
916 /* Return the pa for the given pmap/va. */
917 static vm_paddr_t
918 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
919 {
920 	vm_paddr_t pa = 0;
921 	pte_t *pte;
922 
923 	pte = pte_find(mmu, pmap, va);
924 	if ((pte != NULL) && PTE_ISVALID(pte))
925 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
926 	return (pa);
927 }
928 
929 /* Get a pointer to a PTE in a page table. */
930 static pte_t *
931 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
932 {
933 	unsigned int pdir_idx = PDIR_IDX(va);
934 	unsigned int ptbl_idx = PTBL_IDX(va);
935 
936 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
937 
938 	if (pmap->pm_pdir[pdir_idx])
939 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
940 
941 	return (NULL);
942 }
943 
944 /**************************************************************************/
945 /* PMAP related */
946 /**************************************************************************/
947 
948 /*
949  * This is called during booke_init, before the system is really initialized.
950  */
951 static void
952 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
953 {
954 	vm_offset_t phys_kernelend;
955 	struct mem_region *mp, *mp1;
956 	int cnt, i, j;
957 	u_int s, e, sz;
958 	u_int phys_avail_count;
959 	vm_size_t physsz, hwphyssz, kstack0_sz;
960 	vm_offset_t kernel_pdir, kstack0, va;
961 	vm_paddr_t kstack0_phys;
962 	void *dpcpu;
963 	pte_t *pte;
964 
965 	debugf("mmu_booke_bootstrap: entered\n");
966 
967 	/* Initialize invalidation mutex */
968 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
969 
970 	/* Read TLB0 size and associativity. */
971 	tlb0_get_tlbconf();
972 
973 	/*
974 	 * Align kernel start and end address (kernel image).
975 	 * Note that kernel end does not necessarily relate to kernsize.
976 	 * kernsize is the size of the kernel that is actually mapped.
977 	 * Also note that "start - 1" is deliberate. With SMP, the
978 	 * entry point is exactly a page from the actual load address.
979 	 * As such, trunc_page() has no effect and we're off by a page.
980 	 * Since we always have the ELF header between the load address
981 	 * and the entry point, we can safely subtract 1 to compensate.
982 	 */
983 	kernstart = trunc_page(start - 1);
984 	data_start = round_page(kernelend);
985 	data_end = data_start;
986 
987 	/*
988 	 * Addresses of preloaded modules (like file systems) use
989 	 * physical addresses. Make sure we relocate those into
990 	 * virtual addresses.
991 	 */
992 	preload_addr_relocate = kernstart - kernload;
993 
994 	/* Allocate the dynamic per-cpu area. */
995 	dpcpu = (void *)data_end;
996 	data_end += DPCPU_SIZE;
997 
998 	/* Allocate space for the message buffer. */
999 	msgbufp = (struct msgbuf *)data_end;
1000 	data_end += msgbufsize;
1001 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1002 	    data_end);
1003 
1004 	data_end = round_page(data_end);
1005 
1006 	/* Allocate space for ptbl_bufs. */
1007 	ptbl_bufs = (struct ptbl_buf *)data_end;
1008 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1009 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1010 	    data_end);
1011 
1012 	data_end = round_page(data_end);
1013 
1014 	/* Allocate PTE tables for kernel KVA. */
1015 	kernel_pdir = data_end;
1016 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1017 	    PDIR_SIZE - 1) / PDIR_SIZE;
1018 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1019 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1020 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1021 
1022 	debugf(" data_end: 0x%08x\n", data_end);
1023 	if (data_end - kernstart > kernsize) {
1024 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1025 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1026 	}
1027 	data_end = kernstart + kernsize;
1028 	debugf(" updated data_end: 0x%08x\n", data_end);
1029 
1030 	/*
1031 	 * Clear the structures - note we can only do it safely after the
1032 	 * possible additional TLB1 translations are in place (above) so that
1033 	 * all range up to the currently calculated 'data_end' is covered.
1034 	 */
1035 	dpcpu_init(dpcpu, 0);
1036 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1037 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1038 
1039 	/*******************************************************/
1040 	/* Set the start and end of kva. */
1041 	/*******************************************************/
1042 	virtual_avail = round_page(data_end);
1043 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1044 
1045 	/* Allocate KVA space for page zero/copy operations. */
1046 	zero_page_va = virtual_avail;
1047 	virtual_avail += PAGE_SIZE;
1048 	zero_page_idle_va = virtual_avail;
1049 	virtual_avail += PAGE_SIZE;
1050 	copy_page_src_va = virtual_avail;
1051 	virtual_avail += PAGE_SIZE;
1052 	copy_page_dst_va = virtual_avail;
1053 	virtual_avail += PAGE_SIZE;
1054 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1055 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1056 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1057 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1058 
1059 	/* Initialize page zero/copy mutexes. */
1060 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1061 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1062 
1063 	/* Allocate KVA space for ptbl bufs. */
1064 	ptbl_buf_pool_vabase = virtual_avail;
1065 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1066 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1067 	    ptbl_buf_pool_vabase, virtual_avail);
1068 
1069 	/* Calculate corresponding physical addresses for the kernel region. */
1070 	phys_kernelend = kernload + kernsize;
1071 	debugf("kernel image and allocated data:\n");
1072 	debugf(" kernload    = 0x%08x\n", kernload);
1073 	debugf(" kernstart   = 0x%08x\n", kernstart);
1074 	debugf(" kernsize    = 0x%08x\n", kernsize);
1075 
1076 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1077 		panic("mmu_booke_bootstrap: phys_avail too small");
1078 
1079 	/*
1080 	 * Remove kernel physical address range from avail regions list. Page
1081 	 * align all regions.  Non-page aligned memory isn't very interesting
1082 	 * to us.  Also, sort the entries for ascending addresses.
1083 	 */
1084 
1085 	/* Retrieve phys/avail mem regions */
1086 	mem_regions(&physmem_regions, &physmem_regions_sz,
1087 	    &availmem_regions, &availmem_regions_sz);
1088 	sz = 0;
1089 	cnt = availmem_regions_sz;
1090 	debugf("processing avail regions:\n");
1091 	for (mp = availmem_regions; mp->mr_size; mp++) {
1092 		s = mp->mr_start;
1093 		e = mp->mr_start + mp->mr_size;
1094 		debugf(" %08x-%08x -> ", s, e);
1095 		/* Check whether this region holds all of the kernel. */
1096 		if (s < kernload && e > phys_kernelend) {
1097 			availmem_regions[cnt].mr_start = phys_kernelend;
1098 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1099 			e = kernload;
1100 		}
1101 		/* Look whether this regions starts within the kernel. */
1102 		if (s >= kernload && s < phys_kernelend) {
1103 			if (e <= phys_kernelend)
1104 				goto empty;
1105 			s = phys_kernelend;
1106 		}
1107 		/* Now look whether this region ends within the kernel. */
1108 		if (e > kernload && e <= phys_kernelend) {
1109 			if (s >= kernload)
1110 				goto empty;
1111 			e = kernload;
1112 		}
1113 		/* Now page align the start and size of the region. */
1114 		s = round_page(s);
1115 		e = trunc_page(e);
1116 		if (e < s)
1117 			e = s;
1118 		sz = e - s;
1119 		debugf("%08x-%08x = %x\n", s, e, sz);
1120 
1121 		/* Check whether some memory is left here. */
1122 		if (sz == 0) {
1123 		empty:
1124 			memmove(mp, mp + 1,
1125 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1126 			cnt--;
1127 			mp--;
1128 			continue;
1129 		}
1130 
1131 		/* Do an insertion sort. */
1132 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1133 			if (s < mp1->mr_start)
1134 				break;
1135 		if (mp1 < mp) {
1136 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1137 			mp1->mr_start = s;
1138 			mp1->mr_size = sz;
1139 		} else {
1140 			mp->mr_start = s;
1141 			mp->mr_size = sz;
1142 		}
1143 	}
1144 	availmem_regions_sz = cnt;
1145 
1146 	/*******************************************************/
1147 	/* Steal physical memory for kernel stack from the end */
1148 	/* of the first avail region                           */
1149 	/*******************************************************/
1150 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1151 	kstack0_phys = availmem_regions[0].mr_start +
1152 	    availmem_regions[0].mr_size;
1153 	kstack0_phys -= kstack0_sz;
1154 	availmem_regions[0].mr_size -= kstack0_sz;
1155 
1156 	/*******************************************************/
1157 	/* Fill in phys_avail table, based on availmem_regions */
1158 	/*******************************************************/
1159 	phys_avail_count = 0;
1160 	physsz = 0;
1161 	hwphyssz = 0;
1162 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1163 
1164 	debugf("fill in phys_avail:\n");
1165 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1166 
1167 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1168 		    availmem_regions[i].mr_start,
1169 		    availmem_regions[i].mr_start +
1170 		        availmem_regions[i].mr_size,
1171 		    availmem_regions[i].mr_size);
1172 
1173 		if (hwphyssz != 0 &&
1174 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1175 			debugf(" hw.physmem adjust\n");
1176 			if (physsz < hwphyssz) {
1177 				phys_avail[j] = availmem_regions[i].mr_start;
1178 				phys_avail[j + 1] =
1179 				    availmem_regions[i].mr_start +
1180 				    hwphyssz - physsz;
1181 				physsz = hwphyssz;
1182 				phys_avail_count++;
1183 			}
1184 			break;
1185 		}
1186 
1187 		phys_avail[j] = availmem_regions[i].mr_start;
1188 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1189 		    availmem_regions[i].mr_size;
1190 		phys_avail_count++;
1191 		physsz += availmem_regions[i].mr_size;
1192 	}
1193 	physmem = btoc(physsz);
1194 
1195 	/* Calculate the last available physical address. */
1196 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1197 		;
1198 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1199 
1200 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1201 	debugf("phys_avail_count = %d\n", phys_avail_count);
1202 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1203 	    physmem);
1204 
1205 	/*******************************************************/
1206 	/* Initialize (statically allocated) kernel pmap. */
1207 	/*******************************************************/
1208 	PMAP_LOCK_INIT(kernel_pmap);
1209 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1210 
1211 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1212 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1213 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1214 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1215 
1216 	/* Initialize kernel pdir */
1217 	for (i = 0; i < kernel_ptbls; i++)
1218 		kernel_pmap->pm_pdir[kptbl_min + i] =
1219 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1220 
1221 	for (i = 0; i < MAXCPU; i++) {
1222 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1223 
1224 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1225 		tidbusy[i][0] = kernel_pmap;
1226 	}
1227 
1228 	/*
1229 	 * Fill in PTEs covering kernel code and data. They are not required
1230 	 * for address translation, as this area is covered by static TLB1
1231 	 * entries, but for pte_vatopa() to work correctly with kernel area
1232 	 * addresses.
1233 	 */
1234 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1235 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1236 		pte->rpn = kernload + (va - kernstart);
1237 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1238 		    PTE_VALID;
1239 	}
1240 	/* Mark kernel_pmap active on all CPUs */
1241 	CPU_FILL(&kernel_pmap->pm_active);
1242 
1243  	/*
1244 	 * Initialize the global pv list lock.
1245 	 */
1246 	rw_init(&pvh_global_lock, "pmap pv global");
1247 
1248 	/*******************************************************/
1249 	/* Final setup */
1250 	/*******************************************************/
1251 
1252 	/* Enter kstack0 into kernel map, provide guard page */
1253 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1254 	thread0.td_kstack = kstack0;
1255 	thread0.td_kstack_pages = KSTACK_PAGES;
1256 
1257 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1258 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1259 	    kstack0_phys, kstack0_phys + kstack0_sz);
1260 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1261 
1262 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1263 	for (i = 0; i < KSTACK_PAGES; i++) {
1264 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1265 		kstack0 += PAGE_SIZE;
1266 		kstack0_phys += PAGE_SIZE;
1267 	}
1268 
1269 	debugf("virtual_avail = %08x\n", virtual_avail);
1270 	debugf("virtual_end   = %08x\n", virtual_end);
1271 
1272 	debugf("mmu_booke_bootstrap: exit\n");
1273 }
1274 
1275 void
1276 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1277 {
1278 	int i;
1279 
1280 	/*
1281 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1282 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1283 	 * these values directly to (re)program AP's TLB1 hardware.
1284 	 */
1285 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1286 		/* Skip invalid entries */
1287 		if (!(tlb1[i].mas1 & MAS1_VALID))
1288 			continue;
1289 
1290 		tlb1_write_entry(i);
1291 	}
1292 
1293 	set_mas4_defaults();
1294 }
1295 
1296 /*
1297  * Get the physical page address for the given pmap/virtual address.
1298  */
1299 static vm_paddr_t
1300 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1301 {
1302 	vm_paddr_t pa;
1303 
1304 	PMAP_LOCK(pmap);
1305 	pa = pte_vatopa(mmu, pmap, va);
1306 	PMAP_UNLOCK(pmap);
1307 
1308 	return (pa);
1309 }
1310 
1311 /*
1312  * Extract the physical page address associated with the given
1313  * kernel virtual address.
1314  */
1315 static vm_paddr_t
1316 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1317 {
1318 
1319 	return (pte_vatopa(mmu, kernel_pmap, va));
1320 }
1321 
1322 /*
1323  * Initialize the pmap module.
1324  * Called by vm_init, to initialize any structures that the pmap
1325  * system needs to map virtual memory.
1326  */
1327 static void
1328 mmu_booke_init(mmu_t mmu)
1329 {
1330 	int shpgperproc = PMAP_SHPGPERPROC;
1331 
1332 	/*
1333 	 * Initialize the address space (zone) for the pv entries.  Set a
1334 	 * high water mark so that the system can recover from excessive
1335 	 * numbers of pv entries.
1336 	 */
1337 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1338 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1339 
1340 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1341 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1342 
1343 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1344 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1345 
1346 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1347 
1348 	/* Pre-fill pvzone with initial number of pv entries. */
1349 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1350 
1351 	/* Initialize ptbl allocation. */
1352 	ptbl_init();
1353 }
1354 
1355 /*
1356  * Map a list of wired pages into kernel virtual address space.  This is
1357  * intended for temporary mappings which do not need page modification or
1358  * references recorded.  Existing mappings in the region are overwritten.
1359  */
1360 static void
1361 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1362 {
1363 	vm_offset_t va;
1364 
1365 	va = sva;
1366 	while (count-- > 0) {
1367 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1368 		va += PAGE_SIZE;
1369 		m++;
1370 	}
1371 }
1372 
1373 /*
1374  * Remove page mappings from kernel virtual address space.  Intended for
1375  * temporary mappings entered by mmu_booke_qenter.
1376  */
1377 static void
1378 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1379 {
1380 	vm_offset_t va;
1381 
1382 	va = sva;
1383 	while (count-- > 0) {
1384 		mmu_booke_kremove(mmu, va);
1385 		va += PAGE_SIZE;
1386 	}
1387 }
1388 
1389 /*
1390  * Map a wired page into kernel virtual address space.
1391  */
1392 static void
1393 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1394 {
1395 	unsigned int pdir_idx = PDIR_IDX(va);
1396 	unsigned int ptbl_idx = PTBL_IDX(va);
1397 	uint32_t flags;
1398 	pte_t *pte;
1399 
1400 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1401 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1402 
1403 	flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1404 
1405 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1406 
1407 	mtx_lock_spin(&tlbivax_mutex);
1408 	tlb_miss_lock();
1409 
1410 	if (PTE_ISVALID(pte)) {
1411 
1412 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1413 
1414 		/* Flush entry from TLB0 */
1415 		tlb0_flush_entry(va);
1416 	}
1417 
1418 	pte->rpn = pa & ~PTE_PA_MASK;
1419 	pte->flags = flags;
1420 
1421 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1422 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1423 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1424 
1425 	/* Flush the real memory from the instruction cache. */
1426 	if ((flags & (PTE_I | PTE_G)) == 0) {
1427 		__syncicache((void *)va, PAGE_SIZE);
1428 	}
1429 
1430 	tlb_miss_unlock();
1431 	mtx_unlock_spin(&tlbivax_mutex);
1432 }
1433 
1434 /*
1435  * Remove a page from kernel page table.
1436  */
1437 static void
1438 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1439 {
1440 	unsigned int pdir_idx = PDIR_IDX(va);
1441 	unsigned int ptbl_idx = PTBL_IDX(va);
1442 	pte_t *pte;
1443 
1444 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1445 
1446 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1447 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1448 	    ("mmu_booke_kremove: invalid va"));
1449 
1450 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1451 
1452 	if (!PTE_ISVALID(pte)) {
1453 
1454 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1455 
1456 		return;
1457 	}
1458 
1459 	mtx_lock_spin(&tlbivax_mutex);
1460 	tlb_miss_lock();
1461 
1462 	/* Invalidate entry in TLB0, update PTE. */
1463 	tlb0_flush_entry(va);
1464 	pte->flags = 0;
1465 	pte->rpn = 0;
1466 
1467 	tlb_miss_unlock();
1468 	mtx_unlock_spin(&tlbivax_mutex);
1469 }
1470 
1471 /*
1472  * Initialize pmap associated with process 0.
1473  */
1474 static void
1475 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1476 {
1477 
1478 	mmu_booke_pinit(mmu, pmap);
1479 	PCPU_SET(curpmap, pmap);
1480 }
1481 
1482 /*
1483  * Initialize a preallocated and zeroed pmap structure,
1484  * such as one in a vmspace structure.
1485  */
1486 static void
1487 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1488 {
1489 	int i;
1490 
1491 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1492 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1493 
1494 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1495 
1496 	PMAP_LOCK_INIT(pmap);
1497 	for (i = 0; i < MAXCPU; i++)
1498 		pmap->pm_tid[i] = TID_NONE;
1499 	CPU_ZERO(&kernel_pmap->pm_active);
1500 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1501 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1502 	TAILQ_INIT(&pmap->pm_ptbl_list);
1503 }
1504 
1505 /*
1506  * Release any resources held by the given physical map.
1507  * Called when a pmap initialized by mmu_booke_pinit is being released.
1508  * Should only be called if the map contains no valid mappings.
1509  */
1510 static void
1511 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1512 {
1513 
1514 	KASSERT(pmap->pm_stats.resident_count == 0,
1515 	    ("pmap_release: pmap resident count %ld != 0",
1516 	    pmap->pm_stats.resident_count));
1517 
1518 	PMAP_LOCK_DESTROY(pmap);
1519 }
1520 
1521 /*
1522  * Insert the given physical page at the specified virtual address in the
1523  * target physical map with the protection requested. If specified the page
1524  * will be wired down.
1525  */
1526 static void
1527 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1528     vm_prot_t prot, boolean_t wired)
1529 {
1530 
1531 	rw_wlock(&pvh_global_lock);
1532 	PMAP_LOCK(pmap);
1533 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1534 	rw_wunlock(&pvh_global_lock);
1535 	PMAP_UNLOCK(pmap);
1536 }
1537 
1538 static void
1539 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1540     vm_prot_t prot, boolean_t wired)
1541 {
1542 	pte_t *pte;
1543 	vm_paddr_t pa;
1544 	uint32_t flags;
1545 	int su, sync;
1546 
1547 	pa = VM_PAGE_TO_PHYS(m);
1548 	su = (pmap == kernel_pmap);
1549 	sync = 0;
1550 
1551 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1552 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1553 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1554 	//		(u_int32_t)m, va, pa, prot, wired);
1555 
1556 	if (su) {
1557 		KASSERT(((va >= virtual_avail) &&
1558 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1559 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1560 	} else {
1561 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1562 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1563 	}
1564 	KASSERT((m->oflags & (VPO_UNMANAGED | VPO_BUSY)) != 0 ||
1565 	    VM_OBJECT_LOCKED(m->object),
1566 	    ("mmu_booke_enter_locked: page %p is not busy", m));
1567 
1568 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1569 
1570 	/*
1571 	 * If there is an existing mapping, and the physical address has not
1572 	 * changed, must be protection or wiring change.
1573 	 */
1574 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1575 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1576 
1577 		/*
1578 		 * Before actually updating pte->flags we calculate and
1579 		 * prepare its new value in a helper var.
1580 		 */
1581 		flags = pte->flags;
1582 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1583 
1584 		/* Wiring change, just update stats. */
1585 		if (wired) {
1586 			if (!PTE_ISWIRED(pte)) {
1587 				flags |= PTE_WIRED;
1588 				pmap->pm_stats.wired_count++;
1589 			}
1590 		} else {
1591 			if (PTE_ISWIRED(pte)) {
1592 				flags &= ~PTE_WIRED;
1593 				pmap->pm_stats.wired_count--;
1594 			}
1595 		}
1596 
1597 		if (prot & VM_PROT_WRITE) {
1598 			/* Add write permissions. */
1599 			flags |= PTE_SW;
1600 			if (!su)
1601 				flags |= PTE_UW;
1602 
1603 			if ((flags & PTE_MANAGED) != 0)
1604 				vm_page_aflag_set(m, PGA_WRITEABLE);
1605 		} else {
1606 			/* Handle modified pages, sense modify status. */
1607 
1608 			/*
1609 			 * The PTE_MODIFIED flag could be set by underlying
1610 			 * TLB misses since we last read it (above), possibly
1611 			 * other CPUs could update it so we check in the PTE
1612 			 * directly rather than rely on that saved local flags
1613 			 * copy.
1614 			 */
1615 			if (PTE_ISMODIFIED(pte))
1616 				vm_page_dirty(m);
1617 		}
1618 
1619 		if (prot & VM_PROT_EXECUTE) {
1620 			flags |= PTE_SX;
1621 			if (!su)
1622 				flags |= PTE_UX;
1623 
1624 			/*
1625 			 * Check existing flags for execute permissions: if we
1626 			 * are turning execute permissions on, icache should
1627 			 * be flushed.
1628 			 */
1629 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1630 				sync++;
1631 		}
1632 
1633 		flags &= ~PTE_REFERENCED;
1634 
1635 		/*
1636 		 * The new flags value is all calculated -- only now actually
1637 		 * update the PTE.
1638 		 */
1639 		mtx_lock_spin(&tlbivax_mutex);
1640 		tlb_miss_lock();
1641 
1642 		tlb0_flush_entry(va);
1643 		pte->flags = flags;
1644 
1645 		tlb_miss_unlock();
1646 		mtx_unlock_spin(&tlbivax_mutex);
1647 
1648 	} else {
1649 		/*
1650 		 * If there is an existing mapping, but it's for a different
1651 		 * physical address, pte_enter() will delete the old mapping.
1652 		 */
1653 		//if ((pte != NULL) && PTE_ISVALID(pte))
1654 		//	debugf("mmu_booke_enter_locked: replace\n");
1655 		//else
1656 		//	debugf("mmu_booke_enter_locked: new\n");
1657 
1658 		/* Now set up the flags and install the new mapping. */
1659 		flags = (PTE_SR | PTE_VALID);
1660 		flags |= PTE_M;
1661 
1662 		if (!su)
1663 			flags |= PTE_UR;
1664 
1665 		if (prot & VM_PROT_WRITE) {
1666 			flags |= PTE_SW;
1667 			if (!su)
1668 				flags |= PTE_UW;
1669 
1670 			if ((m->oflags & VPO_UNMANAGED) == 0)
1671 				vm_page_aflag_set(m, PGA_WRITEABLE);
1672 		}
1673 
1674 		if (prot & VM_PROT_EXECUTE) {
1675 			flags |= PTE_SX;
1676 			if (!su)
1677 				flags |= PTE_UX;
1678 		}
1679 
1680 		/* If its wired update stats. */
1681 		if (wired) {
1682 			pmap->pm_stats.wired_count++;
1683 			flags |= PTE_WIRED;
1684 		}
1685 
1686 		pte_enter(mmu, pmap, m, va, flags);
1687 
1688 		/* Flush the real memory from the instruction cache. */
1689 		if (prot & VM_PROT_EXECUTE)
1690 			sync++;
1691 	}
1692 
1693 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1694 		__syncicache((void *)va, PAGE_SIZE);
1695 		sync = 0;
1696 	}
1697 }
1698 
1699 /*
1700  * Maps a sequence of resident pages belonging to the same object.
1701  * The sequence begins with the given page m_start.  This page is
1702  * mapped at the given virtual address start.  Each subsequent page is
1703  * mapped at a virtual address that is offset from start by the same
1704  * amount as the page is offset from m_start within the object.  The
1705  * last page in the sequence is the page with the largest offset from
1706  * m_start that can be mapped at a virtual address less than the given
1707  * virtual address end.  Not every virtual page between start and end
1708  * is mapped; only those for which a resident page exists with the
1709  * corresponding offset from m_start are mapped.
1710  */
1711 static void
1712 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1713     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1714 {
1715 	vm_page_t m;
1716 	vm_pindex_t diff, psize;
1717 
1718 	psize = atop(end - start);
1719 	m = m_start;
1720 	rw_wlock(&pvh_global_lock);
1721 	PMAP_LOCK(pmap);
1722 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1723 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1724 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1725 		m = TAILQ_NEXT(m, listq);
1726 	}
1727 	rw_wunlock(&pvh_global_lock);
1728 	PMAP_UNLOCK(pmap);
1729 }
1730 
1731 static void
1732 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1733     vm_prot_t prot)
1734 {
1735 
1736 	rw_wlock(&pvh_global_lock);
1737 	PMAP_LOCK(pmap);
1738 	mmu_booke_enter_locked(mmu, pmap, va, m,
1739 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1740 	rw_wunlock(&pvh_global_lock);
1741 	PMAP_UNLOCK(pmap);
1742 }
1743 
1744 /*
1745  * Remove the given range of addresses from the specified map.
1746  *
1747  * It is assumed that the start and end are properly rounded to the page size.
1748  */
1749 static void
1750 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1751 {
1752 	pte_t *pte;
1753 	uint8_t hold_flag;
1754 
1755 	int su = (pmap == kernel_pmap);
1756 
1757 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1758 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1759 
1760 	if (su) {
1761 		KASSERT(((va >= virtual_avail) &&
1762 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1763 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1764 	} else {
1765 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1766 		    ("mmu_booke_remove: user pmap, non user va"));
1767 	}
1768 
1769 	if (PMAP_REMOVE_DONE(pmap)) {
1770 		//debugf("mmu_booke_remove: e (empty)\n");
1771 		return;
1772 	}
1773 
1774 	hold_flag = PTBL_HOLD_FLAG(pmap);
1775 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1776 
1777 	rw_wlock(&pvh_global_lock);
1778 	PMAP_LOCK(pmap);
1779 	for (; va < endva; va += PAGE_SIZE) {
1780 		pte = pte_find(mmu, pmap, va);
1781 		if ((pte != NULL) && PTE_ISVALID(pte))
1782 			pte_remove(mmu, pmap, va, hold_flag);
1783 	}
1784 	PMAP_UNLOCK(pmap);
1785 	rw_wunlock(&pvh_global_lock);
1786 
1787 	//debugf("mmu_booke_remove: e\n");
1788 }
1789 
1790 /*
1791  * Remove physical page from all pmaps in which it resides.
1792  */
1793 static void
1794 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1795 {
1796 	pv_entry_t pv, pvn;
1797 	uint8_t hold_flag;
1798 
1799 	rw_wlock(&pvh_global_lock);
1800 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1801 		pvn = TAILQ_NEXT(pv, pv_link);
1802 
1803 		PMAP_LOCK(pv->pv_pmap);
1804 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1805 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1806 		PMAP_UNLOCK(pv->pv_pmap);
1807 	}
1808 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1809 	rw_wunlock(&pvh_global_lock);
1810 }
1811 
1812 /*
1813  * Map a range of physical addresses into kernel virtual address space.
1814  */
1815 static vm_offset_t
1816 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1817     vm_paddr_t pa_end, int prot)
1818 {
1819 	vm_offset_t sva = *virt;
1820 	vm_offset_t va = sva;
1821 
1822 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1823 	//		sva, pa_start, pa_end);
1824 
1825 	while (pa_start < pa_end) {
1826 		mmu_booke_kenter(mmu, va, pa_start);
1827 		va += PAGE_SIZE;
1828 		pa_start += PAGE_SIZE;
1829 	}
1830 	*virt = va;
1831 
1832 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1833 	return (sva);
1834 }
1835 
1836 /*
1837  * The pmap must be activated before it's address space can be accessed in any
1838  * way.
1839  */
1840 static void
1841 mmu_booke_activate(mmu_t mmu, struct thread *td)
1842 {
1843 	pmap_t pmap;
1844 	u_int cpuid;
1845 
1846 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1847 
1848 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1849 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1850 
1851 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1852 
1853 	mtx_lock_spin(&sched_lock);
1854 
1855 	cpuid = PCPU_GET(cpuid);
1856 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1857 	PCPU_SET(curpmap, pmap);
1858 
1859 	if (pmap->pm_tid[cpuid] == TID_NONE)
1860 		tid_alloc(pmap);
1861 
1862 	/* Load PID0 register with pmap tid value. */
1863 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1864 	__asm __volatile("isync");
1865 
1866 	mtx_unlock_spin(&sched_lock);
1867 
1868 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1869 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1870 }
1871 
1872 /*
1873  * Deactivate the specified process's address space.
1874  */
1875 static void
1876 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1877 {
1878 	pmap_t pmap;
1879 
1880 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1881 
1882 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1883 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1884 
1885 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1886 	PCPU_SET(curpmap, NULL);
1887 }
1888 
1889 /*
1890  * Copy the range specified by src_addr/len
1891  * from the source map to the range dst_addr/len
1892  * in the destination map.
1893  *
1894  * This routine is only advisory and need not do anything.
1895  */
1896 static void
1897 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1898     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1899 {
1900 
1901 }
1902 
1903 /*
1904  * Set the physical protection on the specified range of this map as requested.
1905  */
1906 static void
1907 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1908     vm_prot_t prot)
1909 {
1910 	vm_offset_t va;
1911 	vm_page_t m;
1912 	pte_t *pte;
1913 
1914 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1915 		mmu_booke_remove(mmu, pmap, sva, eva);
1916 		return;
1917 	}
1918 
1919 	if (prot & VM_PROT_WRITE)
1920 		return;
1921 
1922 	PMAP_LOCK(pmap);
1923 	for (va = sva; va < eva; va += PAGE_SIZE) {
1924 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1925 			if (PTE_ISVALID(pte)) {
1926 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1927 
1928 				mtx_lock_spin(&tlbivax_mutex);
1929 				tlb_miss_lock();
1930 
1931 				/* Handle modified pages. */
1932 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1933 					vm_page_dirty(m);
1934 
1935 				tlb0_flush_entry(va);
1936 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1937 
1938 				tlb_miss_unlock();
1939 				mtx_unlock_spin(&tlbivax_mutex);
1940 			}
1941 		}
1942 	}
1943 	PMAP_UNLOCK(pmap);
1944 }
1945 
1946 /*
1947  * Clear the write and modified bits in each of the given page's mappings.
1948  */
1949 static void
1950 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1951 {
1952 	pv_entry_t pv;
1953 	pte_t *pte;
1954 
1955 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1956 	    ("mmu_booke_remove_write: page %p is not managed", m));
1957 
1958 	/*
1959 	 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be set by
1960 	 * another thread while the object is locked.  Thus, if PGA_WRITEABLE
1961 	 * is clear, no page table entries need updating.
1962 	 */
1963 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1964 	if ((m->oflags & VPO_BUSY) == 0 &&
1965 	    (m->aflags & PGA_WRITEABLE) == 0)
1966 		return;
1967 	rw_wlock(&pvh_global_lock);
1968 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1969 		PMAP_LOCK(pv->pv_pmap);
1970 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1971 			if (PTE_ISVALID(pte)) {
1972 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1973 
1974 				mtx_lock_spin(&tlbivax_mutex);
1975 				tlb_miss_lock();
1976 
1977 				/* Handle modified pages. */
1978 				if (PTE_ISMODIFIED(pte))
1979 					vm_page_dirty(m);
1980 
1981 				/* Flush mapping from TLB0. */
1982 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1983 
1984 				tlb_miss_unlock();
1985 				mtx_unlock_spin(&tlbivax_mutex);
1986 			}
1987 		}
1988 		PMAP_UNLOCK(pv->pv_pmap);
1989 	}
1990 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1991 	rw_wunlock(&pvh_global_lock);
1992 }
1993 
1994 static void
1995 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1996 {
1997 	pte_t *pte;
1998 	pmap_t pmap;
1999 	vm_page_t m;
2000 	vm_offset_t addr;
2001 	vm_paddr_t pa;
2002 	int active, valid;
2003 
2004 	va = trunc_page(va);
2005 	sz = round_page(sz);
2006 
2007 	rw_wlock(&pvh_global_lock);
2008 	pmap = PCPU_GET(curpmap);
2009 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2010 	while (sz > 0) {
2011 		PMAP_LOCK(pm);
2012 		pte = pte_find(mmu, pm, va);
2013 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2014 		if (valid)
2015 			pa = PTE_PA(pte);
2016 		PMAP_UNLOCK(pm);
2017 		if (valid) {
2018 			if (!active) {
2019 				/* Create a mapping in the active pmap. */
2020 				addr = 0;
2021 				m = PHYS_TO_VM_PAGE(pa);
2022 				PMAP_LOCK(pmap);
2023 				pte_enter(mmu, pmap, m, addr,
2024 				    PTE_SR | PTE_VALID | PTE_UR);
2025 				__syncicache((void *)addr, PAGE_SIZE);
2026 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2027 				PMAP_UNLOCK(pmap);
2028 			} else
2029 				__syncicache((void *)va, PAGE_SIZE);
2030 		}
2031 		va += PAGE_SIZE;
2032 		sz -= PAGE_SIZE;
2033 	}
2034 	rw_wunlock(&pvh_global_lock);
2035 }
2036 
2037 /*
2038  * Atomically extract and hold the physical page with the given
2039  * pmap and virtual address pair if that mapping permits the given
2040  * protection.
2041  */
2042 static vm_page_t
2043 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2044     vm_prot_t prot)
2045 {
2046 	pte_t *pte;
2047 	vm_page_t m;
2048 	uint32_t pte_wbit;
2049 	vm_paddr_t pa;
2050 
2051 	m = NULL;
2052 	pa = 0;
2053 	PMAP_LOCK(pmap);
2054 retry:
2055 	pte = pte_find(mmu, pmap, va);
2056 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2057 		if (pmap == kernel_pmap)
2058 			pte_wbit = PTE_SW;
2059 		else
2060 			pte_wbit = PTE_UW;
2061 
2062 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2063 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2064 				goto retry;
2065 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2066 			vm_page_hold(m);
2067 		}
2068 	}
2069 
2070 	PA_UNLOCK_COND(pa);
2071 	PMAP_UNLOCK(pmap);
2072 	return (m);
2073 }
2074 
2075 /*
2076  * Initialize a vm_page's machine-dependent fields.
2077  */
2078 static void
2079 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2080 {
2081 
2082 	TAILQ_INIT(&m->md.pv_list);
2083 }
2084 
2085 /*
2086  * mmu_booke_zero_page_area zeros the specified hardware page by
2087  * mapping it into virtual memory and using bzero to clear
2088  * its contents.
2089  *
2090  * off and size must reside within a single page.
2091  */
2092 static void
2093 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2094 {
2095 	vm_offset_t va;
2096 
2097 	/* XXX KASSERT off and size are within a single page? */
2098 
2099 	mtx_lock(&zero_page_mutex);
2100 	va = zero_page_va;
2101 
2102 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2103 	bzero((caddr_t)va + off, size);
2104 	mmu_booke_kremove(mmu, va);
2105 
2106 	mtx_unlock(&zero_page_mutex);
2107 }
2108 
2109 /*
2110  * mmu_booke_zero_page zeros the specified hardware page.
2111  */
2112 static void
2113 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2114 {
2115 
2116 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2117 }
2118 
2119 /*
2120  * mmu_booke_copy_page copies the specified (machine independent) page by
2121  * mapping the page into virtual memory and using memcopy to copy the page,
2122  * one machine dependent page at a time.
2123  */
2124 static void
2125 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2126 {
2127 	vm_offset_t sva, dva;
2128 
2129 	sva = copy_page_src_va;
2130 	dva = copy_page_dst_va;
2131 
2132 	mtx_lock(&copy_page_mutex);
2133 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2134 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2135 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2136 	mmu_booke_kremove(mmu, dva);
2137 	mmu_booke_kremove(mmu, sva);
2138 	mtx_unlock(&copy_page_mutex);
2139 }
2140 
2141 /*
2142  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2143  * into virtual memory and using bzero to clear its contents. This is intended
2144  * to be called from the vm_pagezero process only and outside of Giant. No
2145  * lock is required.
2146  */
2147 static void
2148 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2149 {
2150 	vm_offset_t va;
2151 
2152 	va = zero_page_idle_va;
2153 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2154 	bzero((caddr_t)va, PAGE_SIZE);
2155 	mmu_booke_kremove(mmu, va);
2156 }
2157 
2158 /*
2159  * Return whether or not the specified physical page was modified
2160  * in any of physical maps.
2161  */
2162 static boolean_t
2163 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2164 {
2165 	pte_t *pte;
2166 	pv_entry_t pv;
2167 	boolean_t rv;
2168 
2169 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2170 	    ("mmu_booke_is_modified: page %p is not managed", m));
2171 	rv = FALSE;
2172 
2173 	/*
2174 	 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be
2175 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2176 	 * is clear, no PTEs can be modified.
2177 	 */
2178 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2179 	if ((m->oflags & VPO_BUSY) == 0 &&
2180 	    (m->aflags & PGA_WRITEABLE) == 0)
2181 		return (rv);
2182 	rw_wlock(&pvh_global_lock);
2183 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2184 		PMAP_LOCK(pv->pv_pmap);
2185 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2186 		    PTE_ISVALID(pte)) {
2187 			if (PTE_ISMODIFIED(pte))
2188 				rv = TRUE;
2189 		}
2190 		PMAP_UNLOCK(pv->pv_pmap);
2191 		if (rv)
2192 			break;
2193 	}
2194 	rw_wunlock(&pvh_global_lock);
2195 	return (rv);
2196 }
2197 
2198 /*
2199  * Return whether or not the specified virtual address is eligible
2200  * for prefault.
2201  */
2202 static boolean_t
2203 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2204 {
2205 
2206 	return (FALSE);
2207 }
2208 
2209 /*
2210  * Return whether or not the specified physical page was referenced
2211  * in any physical maps.
2212  */
2213 static boolean_t
2214 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2215 {
2216 	pte_t *pte;
2217 	pv_entry_t pv;
2218 	boolean_t rv;
2219 
2220 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2221 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2222 	rv = FALSE;
2223 	rw_wlock(&pvh_global_lock);
2224 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2225 		PMAP_LOCK(pv->pv_pmap);
2226 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2227 		    PTE_ISVALID(pte)) {
2228 			if (PTE_ISREFERENCED(pte))
2229 				rv = TRUE;
2230 		}
2231 		PMAP_UNLOCK(pv->pv_pmap);
2232 		if (rv)
2233 			break;
2234 	}
2235 	rw_wunlock(&pvh_global_lock);
2236 	return (rv);
2237 }
2238 
2239 /*
2240  * Clear the modify bits on the specified physical page.
2241  */
2242 static void
2243 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2244 {
2245 	pte_t *pte;
2246 	pv_entry_t pv;
2247 
2248 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2249 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2250 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2251 	KASSERT((m->oflags & VPO_BUSY) == 0,
2252 	    ("mmu_booke_clear_modify: page %p is busy", m));
2253 
2254 	/*
2255 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2256 	 * If the object containing the page is locked and the page is not
2257 	 * VPO_BUSY, then PG_AWRITEABLE cannot be concurrently set.
2258 	 */
2259 	if ((m->aflags & PGA_WRITEABLE) == 0)
2260 		return;
2261 	rw_wlock(&pvh_global_lock);
2262 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2263 		PMAP_LOCK(pv->pv_pmap);
2264 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2265 		    PTE_ISVALID(pte)) {
2266 			mtx_lock_spin(&tlbivax_mutex);
2267 			tlb_miss_lock();
2268 
2269 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2270 				tlb0_flush_entry(pv->pv_va);
2271 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2272 				    PTE_REFERENCED);
2273 			}
2274 
2275 			tlb_miss_unlock();
2276 			mtx_unlock_spin(&tlbivax_mutex);
2277 		}
2278 		PMAP_UNLOCK(pv->pv_pmap);
2279 	}
2280 	rw_wunlock(&pvh_global_lock);
2281 }
2282 
2283 /*
2284  * Return a count of reference bits for a page, clearing those bits.
2285  * It is not necessary for every reference bit to be cleared, but it
2286  * is necessary that 0 only be returned when there are truly no
2287  * reference bits set.
2288  *
2289  * XXX: The exact number of bits to check and clear is a matter that
2290  * should be tested and standardized at some point in the future for
2291  * optimal aging of shared pages.
2292  */
2293 static int
2294 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2295 {
2296 	pte_t *pte;
2297 	pv_entry_t pv;
2298 	int count;
2299 
2300 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2301 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2302 	count = 0;
2303 	rw_wlock(&pvh_global_lock);
2304 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2305 		PMAP_LOCK(pv->pv_pmap);
2306 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2307 		    PTE_ISVALID(pte)) {
2308 			if (PTE_ISREFERENCED(pte)) {
2309 				mtx_lock_spin(&tlbivax_mutex);
2310 				tlb_miss_lock();
2311 
2312 				tlb0_flush_entry(pv->pv_va);
2313 				pte->flags &= ~PTE_REFERENCED;
2314 
2315 				tlb_miss_unlock();
2316 				mtx_unlock_spin(&tlbivax_mutex);
2317 
2318 				if (++count > 4) {
2319 					PMAP_UNLOCK(pv->pv_pmap);
2320 					break;
2321 				}
2322 			}
2323 		}
2324 		PMAP_UNLOCK(pv->pv_pmap);
2325 	}
2326 	rw_wunlock(&pvh_global_lock);
2327 	return (count);
2328 }
2329 
2330 /*
2331  * Clear the reference bit on the specified physical page.
2332  */
2333 static void
2334 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2335 {
2336 	pte_t *pte;
2337 	pv_entry_t pv;
2338 
2339 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2340 	    ("mmu_booke_clear_reference: page %p is not managed", m));
2341 	rw_wlock(&pvh_global_lock);
2342 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2343 		PMAP_LOCK(pv->pv_pmap);
2344 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2345 		    PTE_ISVALID(pte)) {
2346 			if (PTE_ISREFERENCED(pte)) {
2347 				mtx_lock_spin(&tlbivax_mutex);
2348 				tlb_miss_lock();
2349 
2350 				tlb0_flush_entry(pv->pv_va);
2351 				pte->flags &= ~PTE_REFERENCED;
2352 
2353 				tlb_miss_unlock();
2354 				mtx_unlock_spin(&tlbivax_mutex);
2355 			}
2356 		}
2357 		PMAP_UNLOCK(pv->pv_pmap);
2358 	}
2359 	rw_wunlock(&pvh_global_lock);
2360 }
2361 
2362 /*
2363  * Change wiring attribute for a map/virtual-address pair.
2364  */
2365 static void
2366 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2367 {
2368 	pte_t *pte;
2369 
2370 	PMAP_LOCK(pmap);
2371 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2372 		if (wired) {
2373 			if (!PTE_ISWIRED(pte)) {
2374 				pte->flags |= PTE_WIRED;
2375 				pmap->pm_stats.wired_count++;
2376 			}
2377 		} else {
2378 			if (PTE_ISWIRED(pte)) {
2379 				pte->flags &= ~PTE_WIRED;
2380 				pmap->pm_stats.wired_count--;
2381 			}
2382 		}
2383 	}
2384 	PMAP_UNLOCK(pmap);
2385 }
2386 
2387 /*
2388  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2389  * page.  This count may be changed upwards or downwards in the future; it is
2390  * only necessary that true be returned for a small subset of pmaps for proper
2391  * page aging.
2392  */
2393 static boolean_t
2394 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2395 {
2396 	pv_entry_t pv;
2397 	int loops;
2398 	boolean_t rv;
2399 
2400 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2401 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2402 	loops = 0;
2403 	rv = FALSE;
2404 	rw_wlock(&pvh_global_lock);
2405 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2406 		if (pv->pv_pmap == pmap) {
2407 			rv = TRUE;
2408 			break;
2409 		}
2410 		if (++loops >= 16)
2411 			break;
2412 	}
2413 	rw_wunlock(&pvh_global_lock);
2414 	return (rv);
2415 }
2416 
2417 /*
2418  * Return the number of managed mappings to the given physical page that are
2419  * wired.
2420  */
2421 static int
2422 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2423 {
2424 	pv_entry_t pv;
2425 	pte_t *pte;
2426 	int count = 0;
2427 
2428 	if ((m->oflags & VPO_UNMANAGED) != 0)
2429 		return (count);
2430 	rw_wlock(&pvh_global_lock);
2431 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2432 		PMAP_LOCK(pv->pv_pmap);
2433 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2434 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2435 				count++;
2436 		PMAP_UNLOCK(pv->pv_pmap);
2437 	}
2438 	rw_wunlock(&pvh_global_lock);
2439 	return (count);
2440 }
2441 
2442 static int
2443 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2444 {
2445 	int i;
2446 	vm_offset_t va;
2447 
2448 	/*
2449 	 * This currently does not work for entries that
2450 	 * overlap TLB1 entries.
2451 	 */
2452 	for (i = 0; i < tlb1_idx; i ++) {
2453 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2454 			return (0);
2455 	}
2456 
2457 	return (EFAULT);
2458 }
2459 
2460 vm_offset_t
2461 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2462     vm_size_t *sz)
2463 {
2464 	vm_paddr_t pa, ppa;
2465 	vm_offset_t va;
2466 	vm_size_t gran;
2467 
2468 	/* Raw physical memory dumps don't have a virtual address. */
2469 	if (md->md_vaddr == ~0UL) {
2470 		/* We always map a 256MB page at 256M. */
2471 		gran = 256 * 1024 * 1024;
2472 		pa = md->md_paddr + ofs;
2473 		ppa = pa & ~(gran - 1);
2474 		ofs = pa - ppa;
2475 		va = gran;
2476 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2477 		if (*sz > (gran - ofs))
2478 			*sz = gran - ofs;
2479 		return (va + ofs);
2480 	}
2481 
2482 	/* Minidumps are based on virtual memory addresses. */
2483 	va = md->md_vaddr + ofs;
2484 	if (va >= kernstart + kernsize) {
2485 		gran = PAGE_SIZE - (va & PAGE_MASK);
2486 		if (*sz > gran)
2487 			*sz = gran;
2488 	}
2489 	return (va);
2490 }
2491 
2492 void
2493 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2494     vm_offset_t va)
2495 {
2496 
2497 	/* Raw physical memory dumps don't have a virtual address. */
2498 	if (md->md_vaddr == ~0UL) {
2499 		tlb1_idx--;
2500 		tlb1[tlb1_idx].mas1 = 0;
2501 		tlb1[tlb1_idx].mas2 = 0;
2502 		tlb1[tlb1_idx].mas3 = 0;
2503 		tlb1_write_entry(tlb1_idx);
2504 		return;
2505 	}
2506 
2507 	/* Minidumps are based on virtual memory addresses. */
2508 	/* Nothing to do... */
2509 }
2510 
2511 struct pmap_md *
2512 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2513 {
2514 	static struct pmap_md md;
2515 	pte_t *pte;
2516 	vm_offset_t va;
2517 
2518 	if (dumpsys_minidump) {
2519 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2520 		if (prev == NULL) {
2521 			/* 1st: kernel .data and .bss. */
2522 			md.md_index = 1;
2523 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2524 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2525 			return (&md);
2526 		}
2527 		switch (prev->md_index) {
2528 		case 1:
2529 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2530 			md.md_index = 2;
2531 			md.md_vaddr = data_start;
2532 			md.md_size = data_end - data_start;
2533 			break;
2534 		case 2:
2535 			/* 3rd: kernel VM. */
2536 			va = prev->md_vaddr + prev->md_size;
2537 			/* Find start of next chunk (from va). */
2538 			while (va < virtual_end) {
2539 				/* Don't dump the buffer cache. */
2540 				if (va >= kmi.buffer_sva &&
2541 				    va < kmi.buffer_eva) {
2542 					va = kmi.buffer_eva;
2543 					continue;
2544 				}
2545 				pte = pte_find(mmu, kernel_pmap, va);
2546 				if (pte != NULL && PTE_ISVALID(pte))
2547 					break;
2548 				va += PAGE_SIZE;
2549 			}
2550 			if (va < virtual_end) {
2551 				md.md_vaddr = va;
2552 				va += PAGE_SIZE;
2553 				/* Find last page in chunk. */
2554 				while (va < virtual_end) {
2555 					/* Don't run into the buffer cache. */
2556 					if (va == kmi.buffer_sva)
2557 						break;
2558 					pte = pte_find(mmu, kernel_pmap, va);
2559 					if (pte == NULL || !PTE_ISVALID(pte))
2560 						break;
2561 					va += PAGE_SIZE;
2562 				}
2563 				md.md_size = va - md.md_vaddr;
2564 				break;
2565 			}
2566 			md.md_index = 3;
2567 			/* FALLTHROUGH */
2568 		default:
2569 			return (NULL);
2570 		}
2571 	} else { /* minidumps */
2572 		mem_regions(&physmem_regions, &physmem_regions_sz,
2573 		    &availmem_regions, &availmem_regions_sz);
2574 
2575 		if (prev == NULL) {
2576 			/* first physical chunk. */
2577 			md.md_paddr = physmem_regions[0].mr_start;
2578 			md.md_size = physmem_regions[0].mr_size;
2579 			md.md_vaddr = ~0UL;
2580 			md.md_index = 1;
2581 		} else if (md.md_index < physmem_regions_sz) {
2582 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2583 			md.md_size = physmem_regions[md.md_index].mr_size;
2584 			md.md_vaddr = ~0UL;
2585 			md.md_index++;
2586 		} else {
2587 			/* There's no next physical chunk. */
2588 			return (NULL);
2589 		}
2590 	}
2591 
2592 	return (&md);
2593 }
2594 
2595 /*
2596  * Map a set of physical memory pages into the kernel virtual address space.
2597  * Return a pointer to where it is mapped. This routine is intended to be used
2598  * for mapping device memory, NOT real memory.
2599  */
2600 static void *
2601 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2602 {
2603 	void *res;
2604 	uintptr_t va;
2605 	vm_size_t sz;
2606 
2607 	/*
2608 	 * CCSR is premapped. Note that (pa + size - 1) is there to make sure
2609 	 * we don't wrap around. Devices on the local bus typically extend all
2610 	 * the way up to and including 0xffffffff. In that case (pa + size)
2611 	 * would be 0. This creates a false positive (i.e. we think it's
2612 	 * within the CCSR) and not create a mapping.
2613 	 */
2614 	if (pa >= ccsrbar_pa && (pa + size - 1) < (ccsrbar_pa + CCSRBAR_SIZE)) {
2615 		va = CCSRBAR_VA + (pa - ccsrbar_pa);
2616 		return ((void *)va);
2617 	}
2618 
2619 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2620 	res = (void *)va;
2621 
2622 	do {
2623 		sz = 1 << (ilog2(size) & ~1);
2624 		if (bootverbose)
2625 			printf("Wiring VA=%x to PA=%x (size=%x), "
2626 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2627 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2628 		size -= sz;
2629 		pa += sz;
2630 		va += sz;
2631 	} while (size > 0);
2632 
2633 	return (res);
2634 }
2635 
2636 /*
2637  * 'Unmap' a range mapped by mmu_booke_mapdev().
2638  */
2639 static void
2640 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2641 {
2642 	vm_offset_t base, offset;
2643 
2644 	/*
2645 	 * Unmap only if this is inside kernel virtual space.
2646 	 */
2647 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2648 		base = trunc_page(va);
2649 		offset = va & PAGE_MASK;
2650 		size = roundup(offset + size, PAGE_SIZE);
2651 		kmem_free(kernel_map, base, size);
2652 	}
2653 }
2654 
2655 /*
2656  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2657  * specified pmap. This eliminates the blast of soft faults on process startup
2658  * and immediately after an mmap.
2659  */
2660 static void
2661 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2662     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2663 {
2664 
2665 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2666 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2667 	    ("mmu_booke_object_init_pt: non-device object"));
2668 }
2669 
2670 /*
2671  * Perform the pmap work for mincore.
2672  */
2673 static int
2674 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2675     vm_paddr_t *locked_pa)
2676 {
2677 
2678 	TODO;
2679 	return (0);
2680 }
2681 
2682 /**************************************************************************/
2683 /* TID handling */
2684 /**************************************************************************/
2685 
2686 /*
2687  * Allocate a TID. If necessary, steal one from someone else.
2688  * The new TID is flushed from the TLB before returning.
2689  */
2690 static tlbtid_t
2691 tid_alloc(pmap_t pmap)
2692 {
2693 	tlbtid_t tid;
2694 	int thiscpu;
2695 
2696 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2697 
2698 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2699 
2700 	thiscpu = PCPU_GET(cpuid);
2701 
2702 	tid = PCPU_GET(tid_next);
2703 	if (tid > TID_MAX)
2704 		tid = TID_MIN;
2705 	PCPU_SET(tid_next, tid + 1);
2706 
2707 	/* If we are stealing TID then clear the relevant pmap's field */
2708 	if (tidbusy[thiscpu][tid] != NULL) {
2709 
2710 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2711 
2712 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2713 
2714 		/* Flush all entries from TLB0 matching this TID. */
2715 		tid_flush(tid);
2716 	}
2717 
2718 	tidbusy[thiscpu][tid] = pmap;
2719 	pmap->pm_tid[thiscpu] = tid;
2720 	__asm __volatile("msync; isync");
2721 
2722 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2723 	    PCPU_GET(tid_next));
2724 
2725 	return (tid);
2726 }
2727 
2728 /**************************************************************************/
2729 /* TLB0 handling */
2730 /**************************************************************************/
2731 
2732 static void
2733 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2734     uint32_t mas7)
2735 {
2736 	int as;
2737 	char desc[3];
2738 	tlbtid_t tid;
2739 	vm_size_t size;
2740 	unsigned int tsize;
2741 
2742 	desc[2] = '\0';
2743 	if (mas1 & MAS1_VALID)
2744 		desc[0] = 'V';
2745 	else
2746 		desc[0] = ' ';
2747 
2748 	if (mas1 & MAS1_IPROT)
2749 		desc[1] = 'P';
2750 	else
2751 		desc[1] = ' ';
2752 
2753 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2754 	tid = MAS1_GETTID(mas1);
2755 
2756 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2757 	size = 0;
2758 	if (tsize)
2759 		size = tsize2size(tsize);
2760 
2761 	debugf("%3d: (%s) [AS=%d] "
2762 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2763 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2764 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2765 }
2766 
2767 /* Convert TLB0 va and way number to tlb0[] table index. */
2768 static inline unsigned int
2769 tlb0_tableidx(vm_offset_t va, unsigned int way)
2770 {
2771 	unsigned int idx;
2772 
2773 	idx = (way * TLB0_ENTRIES_PER_WAY);
2774 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2775 	return (idx);
2776 }
2777 
2778 /*
2779  * Invalidate TLB0 entry.
2780  */
2781 static inline void
2782 tlb0_flush_entry(vm_offset_t va)
2783 {
2784 
2785 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2786 
2787 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2788 
2789 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2790 	__asm __volatile("isync; msync");
2791 	__asm __volatile("tlbsync; msync");
2792 
2793 	CTR1(KTR_PMAP, "%s: e", __func__);
2794 }
2795 
2796 /* Print out contents of the MAS registers for each TLB0 entry */
2797 void
2798 tlb0_print_tlbentries(void)
2799 {
2800 	uint32_t mas0, mas1, mas2, mas3, mas7;
2801 	int entryidx, way, idx;
2802 
2803 	debugf("TLB0 entries:\n");
2804 	for (way = 0; way < TLB0_WAYS; way ++)
2805 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2806 
2807 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2808 			mtspr(SPR_MAS0, mas0);
2809 			__asm __volatile("isync");
2810 
2811 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2812 			mtspr(SPR_MAS2, mas2);
2813 
2814 			__asm __volatile("isync; tlbre");
2815 
2816 			mas1 = mfspr(SPR_MAS1);
2817 			mas2 = mfspr(SPR_MAS2);
2818 			mas3 = mfspr(SPR_MAS3);
2819 			mas7 = mfspr(SPR_MAS7);
2820 
2821 			idx = tlb0_tableidx(mas2, way);
2822 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2823 		}
2824 }
2825 
2826 /**************************************************************************/
2827 /* TLB1 handling */
2828 /**************************************************************************/
2829 
2830 /*
2831  * TLB1 mapping notes:
2832  *
2833  * TLB1[0]	CCSRBAR
2834  * TLB1[1]	Kernel text and data.
2835  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2836  *		windows, other devices mappings.
2837  */
2838 
2839 /*
2840  * Write given entry to TLB1 hardware.
2841  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2842  */
2843 static void
2844 tlb1_write_entry(unsigned int idx)
2845 {
2846 	uint32_t mas0, mas7;
2847 
2848 	//debugf("tlb1_write_entry: s\n");
2849 
2850 	/* Clear high order RPN bits */
2851 	mas7 = 0;
2852 
2853 	/* Select entry */
2854 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2855 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2856 
2857 	mtspr(SPR_MAS0, mas0);
2858 	__asm __volatile("isync");
2859 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2860 	__asm __volatile("isync");
2861 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2862 	__asm __volatile("isync");
2863 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2864 	__asm __volatile("isync");
2865 	mtspr(SPR_MAS7, mas7);
2866 	__asm __volatile("isync; tlbwe; isync; msync");
2867 
2868 	//debugf("tlb1_write_entry: e\n");
2869 }
2870 
2871 /*
2872  * Return the largest uint value log such that 2^log <= num.
2873  */
2874 static unsigned int
2875 ilog2(unsigned int num)
2876 {
2877 	int lz;
2878 
2879 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2880 	return (31 - lz);
2881 }
2882 
2883 /*
2884  * Convert TLB TSIZE value to mapped region size.
2885  */
2886 static vm_size_t
2887 tsize2size(unsigned int tsize)
2888 {
2889 
2890 	/*
2891 	 * size = 4^tsize KB
2892 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2893 	 */
2894 
2895 	return ((1 << (2 * tsize)) * 1024);
2896 }
2897 
2898 /*
2899  * Convert region size (must be power of 4) to TLB TSIZE value.
2900  */
2901 static unsigned int
2902 size2tsize(vm_size_t size)
2903 {
2904 
2905 	return (ilog2(size) / 2 - 5);
2906 }
2907 
2908 /*
2909  * Register permanent kernel mapping in TLB1.
2910  *
2911  * Entries are created starting from index 0 (current free entry is
2912  * kept in tlb1_idx) and are not supposed to be invalidated.
2913  */
2914 static int
2915 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2916     uint32_t flags)
2917 {
2918 	uint32_t ts, tid;
2919 	int tsize;
2920 
2921 	if (tlb1_idx >= TLB1_ENTRIES) {
2922 		printf("tlb1_set_entry: TLB1 full!\n");
2923 		return (-1);
2924 	}
2925 
2926 	/* Convert size to TSIZE */
2927 	tsize = size2tsize(size);
2928 
2929 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2930 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2931 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2932 
2933 	/* XXX LOCK tlb1[] */
2934 
2935 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2936 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2937 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2938 
2939 	/* Set supervisor RWX permission bits */
2940 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2941 
2942 	tlb1_write_entry(tlb1_idx++);
2943 
2944 	/* XXX UNLOCK tlb1[] */
2945 
2946 	/*
2947 	 * XXX in general TLB1 updates should be propagated between CPUs,
2948 	 * since current design assumes to have the same TLB1 set-up on all
2949 	 * cores.
2950 	 */
2951 	return (0);
2952 }
2953 
2954 /*
2955  * Map in contiguous RAM region into the TLB1 using maximum of
2956  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2957  *
2958  * If necessary round up last entry size and return total size
2959  * used by all allocated entries.
2960  */
2961 vm_size_t
2962 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
2963 {
2964 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
2965 	vm_size_t mapped, pgsz, base, mask;
2966 	int idx, nents;
2967 
2968 	/* Round up to the next 1M */
2969 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
2970 
2971 	mapped = 0;
2972 	idx = 0;
2973 	base = va;
2974 	pgsz = 64*1024*1024;
2975 	while (mapped < size) {
2976 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
2977 			while (pgsz > (size - mapped))
2978 				pgsz >>= 2;
2979 			pgs[idx++] = pgsz;
2980 			mapped += pgsz;
2981 		}
2982 
2983 		/* We under-map. Correct for this. */
2984 		if (mapped < size) {
2985 			while (pgs[idx - 1] == pgsz) {
2986 				idx--;
2987 				mapped -= pgsz;
2988 			}
2989 			/* XXX We may increase beyond out starting point. */
2990 			pgsz <<= 2;
2991 			pgs[idx++] = pgsz;
2992 			mapped += pgsz;
2993 		}
2994 	}
2995 
2996 	nents = idx;
2997 	mask = pgs[0] - 1;
2998 	/* Align address to the boundary */
2999 	if (va & mask) {
3000 		va = (va + mask) & ~mask;
3001 		pa = (pa + mask) & ~mask;
3002 	}
3003 
3004 	for (idx = 0; idx < nents; idx++) {
3005 		pgsz = pgs[idx];
3006 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3007 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3008 		pa += pgsz;
3009 		va += pgsz;
3010 	}
3011 
3012 	mapped = (va - base);
3013 	debugf("mapped size 0x%08x (wasted space 0x%08x)\n",
3014 	    mapped, mapped - size);
3015 	return (mapped);
3016 }
3017 
3018 /*
3019  * TLB1 initialization routine, to be called after the very first
3020  * assembler level setup done in locore.S.
3021  */
3022 void
3023 tlb1_init(vm_offset_t ccsrbar)
3024 {
3025 	uint32_t mas0, mas1, mas3;
3026 	uint32_t tsz;
3027 	u_int i;
3028 
3029 	ccsrbar_pa = ccsrbar;
3030 
3031 	if (bootinfo != NULL && bootinfo[0] != 1) {
3032 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3033 	} else
3034 		tlb1_idx = 1;
3035 
3036 	/* The first entry/entries are used to map the kernel. */
3037 	for (i = 0; i < tlb1_idx; i++) {
3038 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3039 		mtspr(SPR_MAS0, mas0);
3040 		__asm __volatile("isync; tlbre");
3041 
3042 		mas1 = mfspr(SPR_MAS1);
3043 		if ((mas1 & MAS1_VALID) == 0)
3044 			continue;
3045 
3046 		mas3 = mfspr(SPR_MAS3);
3047 
3048 		tlb1[i].mas1 = mas1;
3049 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3050 		tlb1[i].mas3 = mas3;
3051 
3052 		if (i == 0)
3053 			kernload = mas3 & MAS3_RPN;
3054 
3055 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3056 		kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
3057 	}
3058 
3059 	/* Map in CCSRBAR. */
3060 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
3061 
3062 #ifdef SMP
3063 	bp_ntlb1s = tlb1_idx;
3064 #endif
3065 
3066 	/* Purge the remaining entries */
3067 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3068 		tlb1_write_entry(i);
3069 
3070 	/* Setup TLB miss defaults */
3071 	set_mas4_defaults();
3072 }
3073 
3074 /*
3075  * Setup MAS4 defaults.
3076  * These values are loaded to MAS0-2 on a TLB miss.
3077  */
3078 static void
3079 set_mas4_defaults(void)
3080 {
3081 	uint32_t mas4;
3082 
3083 	/* Defaults: TLB0, PID0, TSIZED=4K */
3084 	mas4 = MAS4_TLBSELD0;
3085 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3086 #ifdef SMP
3087 	mas4 |= MAS4_MD;
3088 #endif
3089 	mtspr(SPR_MAS4, mas4);
3090 	__asm __volatile("isync");
3091 }
3092 
3093 /*
3094  * Print out contents of the MAS registers for each TLB1 entry
3095  */
3096 void
3097 tlb1_print_tlbentries(void)
3098 {
3099 	uint32_t mas0, mas1, mas2, mas3, mas7;
3100 	int i;
3101 
3102 	debugf("TLB1 entries:\n");
3103 	for (i = 0; i < TLB1_ENTRIES; i++) {
3104 
3105 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3106 		mtspr(SPR_MAS0, mas0);
3107 
3108 		__asm __volatile("isync; tlbre");
3109 
3110 		mas1 = mfspr(SPR_MAS1);
3111 		mas2 = mfspr(SPR_MAS2);
3112 		mas3 = mfspr(SPR_MAS3);
3113 		mas7 = mfspr(SPR_MAS7);
3114 
3115 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3116 	}
3117 }
3118 
3119 /*
3120  * Print out contents of the in-ram tlb1 table.
3121  */
3122 void
3123 tlb1_print_entries(void)
3124 {
3125 	int i;
3126 
3127 	debugf("tlb1[] table entries:\n");
3128 	for (i = 0; i < TLB1_ENTRIES; i++)
3129 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3130 }
3131 
3132 /*
3133  * Return 0 if the physical IO range is encompassed by one of the
3134  * the TLB1 entries, otherwise return related error code.
3135  */
3136 static int
3137 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3138 {
3139 	uint32_t prot;
3140 	vm_paddr_t pa_start;
3141 	vm_paddr_t pa_end;
3142 	unsigned int entry_tsize;
3143 	vm_size_t entry_size;
3144 
3145 	*va = (vm_offset_t)NULL;
3146 
3147 	/* Skip invalid entries */
3148 	if (!(tlb1[i].mas1 & MAS1_VALID))
3149 		return (EINVAL);
3150 
3151 	/*
3152 	 * The entry must be cache-inhibited, guarded, and r/w
3153 	 * so it can function as an i/o page
3154 	 */
3155 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3156 	if (prot != (MAS2_I | MAS2_G))
3157 		return (EPERM);
3158 
3159 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3160 	if (prot != (MAS3_SR | MAS3_SW))
3161 		return (EPERM);
3162 
3163 	/* The address should be within the entry range. */
3164 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3165 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3166 
3167 	entry_size = tsize2size(entry_tsize);
3168 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3169 	pa_end = pa_start + entry_size - 1;
3170 
3171 	if ((pa < pa_start) || ((pa + size) > pa_end))
3172 		return (ERANGE);
3173 
3174 	/* Return virtual address of this mapping. */
3175 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3176 	return (0);
3177 }
3178