xref: /freebsd/sys/powerpc/booke/pmap.c (revision eb6d21b4ca6d668cf89afd99eef7baeafa712197)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/types.h>
55 #include <sys/param.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/smp.h>
67 #include <sys/vmmeter.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_pager.h>
78 #include <vm/uma.h>
79 
80 #include <machine/bootinfo.h>
81 #include <machine/cpu.h>
82 #include <machine/pcb.h>
83 #include <machine/platform.h>
84 
85 #include <machine/tlb.h>
86 #include <machine/spr.h>
87 #include <machine/vmparam.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 #define DEBUG
96 #undef DEBUG
97 
98 #ifdef  DEBUG
99 #define debugf(fmt, args...) printf(fmt, ##args)
100 #else
101 #define debugf(fmt, args...)
102 #endif
103 
104 #define TODO			panic("%s: not implemented", __func__);
105 
106 #include "opt_sched.h"
107 #ifndef SCHED_4BSD
108 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
109 #endif
110 extern struct mtx sched_lock;
111 
112 extern int dumpsys_minidump;
113 
114 extern unsigned char _etext[];
115 extern unsigned char _end[];
116 
117 /* Kernel physical load address. */
118 extern uint32_t kernload;
119 vm_offset_t kernstart;
120 vm_size_t kernsize;
121 
122 /* Message buffer and tables. */
123 static vm_offset_t data_start;
124 static vm_size_t data_end;
125 
126 /* Phys/avail memory regions. */
127 static struct mem_region *availmem_regions;
128 static int availmem_regions_sz;
129 static struct mem_region *physmem_regions;
130 static int physmem_regions_sz;
131 
132 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
133 static vm_offset_t zero_page_va;
134 static struct mtx zero_page_mutex;
135 
136 static struct mtx tlbivax_mutex;
137 
138 /*
139  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
140  * by idle thred only, no lock required.
141  */
142 static vm_offset_t zero_page_idle_va;
143 
144 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
145 static vm_offset_t copy_page_src_va;
146 static vm_offset_t copy_page_dst_va;
147 static struct mtx copy_page_mutex;
148 
149 /**************************************************************************/
150 /* PMAP */
151 /**************************************************************************/
152 
153 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
154     vm_prot_t, boolean_t);
155 
156 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
157 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
158 
159 /*
160  * If user pmap is processed with mmu_booke_remove and the resident count
161  * drops to 0, there are no more pages to remove, so we need not continue.
162  */
163 #define PMAP_REMOVE_DONE(pmap) \
164 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
165 
166 extern void tlb_lock(uint32_t *);
167 extern void tlb_unlock(uint32_t *);
168 extern void tid_flush(tlbtid_t);
169 
170 /**************************************************************************/
171 /* TLB and TID handling */
172 /**************************************************************************/
173 
174 /* Translation ID busy table */
175 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
176 
177 /*
178  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
179  * core revisions and should be read from h/w registers during early config.
180  */
181 uint32_t tlb0_entries;
182 uint32_t tlb0_ways;
183 uint32_t tlb0_entries_per_way;
184 
185 #define TLB0_ENTRIES		(tlb0_entries)
186 #define TLB0_WAYS		(tlb0_ways)
187 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
188 
189 #define TLB1_ENTRIES 16
190 
191 /* In-ram copy of the TLB1 */
192 static tlb_entry_t tlb1[TLB1_ENTRIES];
193 
194 /* Next free entry in the TLB1 */
195 static unsigned int tlb1_idx;
196 
197 static tlbtid_t tid_alloc(struct pmap *);
198 
199 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
200 
201 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
202 static void tlb1_write_entry(unsigned int);
203 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
204 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_offset_t, vm_size_t);
205 
206 static vm_size_t tsize2size(unsigned int);
207 static unsigned int size2tsize(vm_size_t);
208 static unsigned int ilog2(unsigned int);
209 
210 static void set_mas4_defaults(void);
211 
212 static inline void tlb0_flush_entry(vm_offset_t);
213 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
214 
215 /**************************************************************************/
216 /* Page table management */
217 /**************************************************************************/
218 
219 /* Data for the pv entry allocation mechanism */
220 static uma_zone_t pvzone;
221 static struct vm_object pvzone_obj;
222 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
223 
224 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
225 
226 #ifndef PMAP_SHPGPERPROC
227 #define PMAP_SHPGPERPROC	200
228 #endif
229 
230 static void ptbl_init(void);
231 static struct ptbl_buf *ptbl_buf_alloc(void);
232 static void ptbl_buf_free(struct ptbl_buf *);
233 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
234 
235 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
236 static void ptbl_free(mmu_t, pmap_t, unsigned int);
237 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
238 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
239 
240 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
241 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
242 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
243 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
244 
245 static pv_entry_t pv_alloc(void);
246 static void pv_free(pv_entry_t);
247 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
248 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
249 
250 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
251 #define PTBL_BUFS		(128 * 16)
252 
253 struct ptbl_buf {
254 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
255 	vm_offset_t kva;		/* va of mapping */
256 };
257 
258 /* ptbl free list and a lock used for access synchronization. */
259 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
260 static struct mtx ptbl_buf_freelist_lock;
261 
262 /* Base address of kva space allocated fot ptbl bufs. */
263 static vm_offset_t ptbl_buf_pool_vabase;
264 
265 /* Pointer to ptbl_buf structures. */
266 static struct ptbl_buf *ptbl_bufs;
267 
268 void pmap_bootstrap_ap(volatile uint32_t *);
269 
270 /*
271  * Kernel MMU interface
272  */
273 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
274 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
275 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
276 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
277     vm_size_t, vm_offset_t);
278 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
279 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
280     vm_prot_t, boolean_t);
281 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
282     vm_page_t, vm_prot_t);
283 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
284     vm_prot_t);
285 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
286 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
287     vm_prot_t);
288 static void		mmu_booke_init(mmu_t);
289 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
290 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
291 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
292 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
293     int);
294 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t);
295 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
296     vm_object_t, vm_pindex_t, vm_size_t);
297 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
298 static void		mmu_booke_page_init(mmu_t, vm_page_t);
299 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
300 static void		mmu_booke_pinit(mmu_t, pmap_t);
301 static void		mmu_booke_pinit0(mmu_t, pmap_t);
302 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
303     vm_prot_t);
304 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
305 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
306 static void		mmu_booke_release(mmu_t, pmap_t);
307 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
308 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
309 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
310 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
311 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
312 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
313 static void		mmu_booke_activate(mmu_t, struct thread *);
314 static void		mmu_booke_deactivate(mmu_t, struct thread *);
315 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
316 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
317 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
318 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
319 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
320 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
321 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
322 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
323     vm_size_t);
324 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
325     vm_size_t, vm_size_t *);
326 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
327     vm_size_t, vm_offset_t);
328 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
329 
330 static mmu_method_t mmu_booke_methods[] = {
331 	/* pmap dispatcher interface */
332 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
333 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
334 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
335 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
336 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
337 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
338 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
339 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
340 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
341 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
342 	MMUMETHOD(mmu_init,		mmu_booke_init),
343 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
344 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
345 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
346 	MMUMETHOD(mmu_map,		mmu_booke_map),
347 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
348 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
349 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
350 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
351 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
352 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
353 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
354 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
355 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
356 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
357 	MMUMETHOD(mmu_release,		mmu_booke_release),
358 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
359 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
360 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
361 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
362 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
363 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
364 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
365 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
366 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
367 
368 	/* Internal interfaces */
369 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
370 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
371 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
372 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
373 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
374 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
375 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
376 
377 	/* dumpsys() support */
378 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
379 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
380 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
381 
382 	{ 0, 0 }
383 };
384 
385 static mmu_def_t booke_mmu = {
386 	MMU_TYPE_BOOKE,
387 	mmu_booke_methods,
388 	0
389 };
390 MMU_DEF(booke_mmu);
391 
392 static inline void
393 tlb_miss_lock(void)
394 {
395 #ifdef SMP
396 	struct pcpu *pc;
397 
398 	if (!smp_started)
399 		return;
400 
401 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
402 		if (pc != pcpup) {
403 
404 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
405 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
406 
407 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
408 			    ("tlb_miss_lock: tried to lock self"));
409 
410 			tlb_lock(pc->pc_booke_tlb_lock);
411 
412 			CTR1(KTR_PMAP, "%s: locked", __func__);
413 		}
414 	}
415 #endif
416 }
417 
418 static inline void
419 tlb_miss_unlock(void)
420 {
421 #ifdef SMP
422 	struct pcpu *pc;
423 
424 	if (!smp_started)
425 		return;
426 
427 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
428 		if (pc != pcpup) {
429 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
430 			    __func__, pc->pc_cpuid);
431 
432 			tlb_unlock(pc->pc_booke_tlb_lock);
433 
434 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
435 		}
436 	}
437 #endif
438 }
439 
440 /* Return number of entries in TLB0. */
441 static __inline void
442 tlb0_get_tlbconf(void)
443 {
444 	uint32_t tlb0_cfg;
445 
446 	tlb0_cfg = mfspr(SPR_TLB0CFG);
447 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
448 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
449 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
450 }
451 
452 /* Initialize pool of kva ptbl buffers. */
453 static void
454 ptbl_init(void)
455 {
456 	int i;
457 
458 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
459 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
460 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
461 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
462 
463 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
464 	TAILQ_INIT(&ptbl_buf_freelist);
465 
466 	for (i = 0; i < PTBL_BUFS; i++) {
467 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
468 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
469 	}
470 }
471 
472 /* Get a ptbl_buf from the freelist. */
473 static struct ptbl_buf *
474 ptbl_buf_alloc(void)
475 {
476 	struct ptbl_buf *buf;
477 
478 	mtx_lock(&ptbl_buf_freelist_lock);
479 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
480 	if (buf != NULL)
481 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
482 	mtx_unlock(&ptbl_buf_freelist_lock);
483 
484 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
485 
486 	return (buf);
487 }
488 
489 /* Return ptbl buff to free pool. */
490 static void
491 ptbl_buf_free(struct ptbl_buf *buf)
492 {
493 
494 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
495 
496 	mtx_lock(&ptbl_buf_freelist_lock);
497 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
498 	mtx_unlock(&ptbl_buf_freelist_lock);
499 }
500 
501 /*
502  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
503  */
504 static void
505 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
506 {
507 	struct ptbl_buf *pbuf;
508 
509 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
510 
511 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
512 
513 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
514 		if (pbuf->kva == (vm_offset_t)ptbl) {
515 			/* Remove from pmap ptbl buf list. */
516 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
517 
518 			/* Free corresponding ptbl buf. */
519 			ptbl_buf_free(pbuf);
520 			break;
521 		}
522 }
523 
524 /* Allocate page table. */
525 static pte_t *
526 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
527 {
528 	vm_page_t mtbl[PTBL_PAGES];
529 	vm_page_t m;
530 	struct ptbl_buf *pbuf;
531 	unsigned int pidx;
532 	pte_t *ptbl;
533 	int i;
534 
535 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
536 	    (pmap == kernel_pmap), pdir_idx);
537 
538 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
539 	    ("ptbl_alloc: invalid pdir_idx"));
540 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
541 	    ("pte_alloc: valid ptbl entry exists!"));
542 
543 	pbuf = ptbl_buf_alloc();
544 	if (pbuf == NULL)
545 		panic("pte_alloc: couldn't alloc kernel virtual memory");
546 
547 	ptbl = (pte_t *)pbuf->kva;
548 
549 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
550 
551 	/* Allocate ptbl pages, this will sleep! */
552 	for (i = 0; i < PTBL_PAGES; i++) {
553 		pidx = (PTBL_PAGES * pdir_idx) + i;
554 		while ((m = vm_page_alloc(NULL, pidx,
555 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
556 
557 			PMAP_UNLOCK(pmap);
558 			vm_page_unlock_queues();
559 			VM_WAIT;
560 			vm_page_lock_queues();
561 			PMAP_LOCK(pmap);
562 		}
563 		mtbl[i] = m;
564 	}
565 
566 	/* Map allocated pages into kernel_pmap. */
567 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
568 
569 	/* Zero whole ptbl. */
570 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
571 
572 	/* Add pbuf to the pmap ptbl bufs list. */
573 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
574 
575 	return (ptbl);
576 }
577 
578 /* Free ptbl pages and invalidate pdir entry. */
579 static void
580 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
581 {
582 	pte_t *ptbl;
583 	vm_paddr_t pa;
584 	vm_offset_t va;
585 	vm_page_t m;
586 	int i;
587 
588 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
589 	    (pmap == kernel_pmap), pdir_idx);
590 
591 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
592 	    ("ptbl_free: invalid pdir_idx"));
593 
594 	ptbl = pmap->pm_pdir[pdir_idx];
595 
596 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
597 
598 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
599 
600 	/*
601 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
602 	 * don't attempt to look up the page tables we are releasing.
603 	 */
604 	mtx_lock_spin(&tlbivax_mutex);
605 	tlb_miss_lock();
606 
607 	pmap->pm_pdir[pdir_idx] = NULL;
608 
609 	tlb_miss_unlock();
610 	mtx_unlock_spin(&tlbivax_mutex);
611 
612 	for (i = 0; i < PTBL_PAGES; i++) {
613 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
614 		pa = pte_vatopa(mmu, kernel_pmap, va);
615 		m = PHYS_TO_VM_PAGE(pa);
616 		vm_page_free_zero(m);
617 		atomic_subtract_int(&cnt.v_wire_count, 1);
618 		mmu_booke_kremove(mmu, va);
619 	}
620 
621 	ptbl_free_pmap_ptbl(pmap, ptbl);
622 }
623 
624 /*
625  * Decrement ptbl pages hold count and attempt to free ptbl pages.
626  * Called when removing pte entry from ptbl.
627  *
628  * Return 1 if ptbl pages were freed.
629  */
630 static int
631 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
632 {
633 	pte_t *ptbl;
634 	vm_paddr_t pa;
635 	vm_page_t m;
636 	int i;
637 
638 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
639 	    (pmap == kernel_pmap), pdir_idx);
640 
641 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
642 	    ("ptbl_unhold: invalid pdir_idx"));
643 	KASSERT((pmap != kernel_pmap),
644 	    ("ptbl_unhold: unholding kernel ptbl!"));
645 
646 	ptbl = pmap->pm_pdir[pdir_idx];
647 
648 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
649 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
650 	    ("ptbl_unhold: non kva ptbl"));
651 
652 	/* decrement hold count */
653 	for (i = 0; i < PTBL_PAGES; i++) {
654 		pa = pte_vatopa(mmu, kernel_pmap,
655 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
656 		m = PHYS_TO_VM_PAGE(pa);
657 		m->wire_count--;
658 	}
659 
660 	/*
661 	 * Free ptbl pages if there are no pte etries in this ptbl.
662 	 * wire_count has the same value for all ptbl pages, so check the last
663 	 * page.
664 	 */
665 	if (m->wire_count == 0) {
666 		ptbl_free(mmu, pmap, pdir_idx);
667 
668 		//debugf("ptbl_unhold: e (freed ptbl)\n");
669 		return (1);
670 	}
671 
672 	return (0);
673 }
674 
675 /*
676  * Increment hold count for ptbl pages. This routine is used when a new pte
677  * entry is being inserted into the ptbl.
678  */
679 static void
680 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
681 {
682 	vm_paddr_t pa;
683 	pte_t *ptbl;
684 	vm_page_t m;
685 	int i;
686 
687 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
688 	    pdir_idx);
689 
690 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
691 	    ("ptbl_hold: invalid pdir_idx"));
692 	KASSERT((pmap != kernel_pmap),
693 	    ("ptbl_hold: holding kernel ptbl!"));
694 
695 	ptbl = pmap->pm_pdir[pdir_idx];
696 
697 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
698 
699 	for (i = 0; i < PTBL_PAGES; i++) {
700 		pa = pte_vatopa(mmu, kernel_pmap,
701 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
702 		m = PHYS_TO_VM_PAGE(pa);
703 		m->wire_count++;
704 	}
705 }
706 
707 /* Allocate pv_entry structure. */
708 pv_entry_t
709 pv_alloc(void)
710 {
711 	pv_entry_t pv;
712 
713 	pv_entry_count++;
714 	if (pv_entry_count > pv_entry_high_water)
715 		pagedaemon_wakeup();
716 	pv = uma_zalloc(pvzone, M_NOWAIT);
717 
718 	return (pv);
719 }
720 
721 /* Free pv_entry structure. */
722 static __inline void
723 pv_free(pv_entry_t pve)
724 {
725 
726 	pv_entry_count--;
727 	uma_zfree(pvzone, pve);
728 }
729 
730 
731 /* Allocate and initialize pv_entry structure. */
732 static void
733 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
734 {
735 	pv_entry_t pve;
736 
737 	//int su = (pmap == kernel_pmap);
738 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
739 	//	(u_int32_t)pmap, va, (u_int32_t)m);
740 
741 	pve = pv_alloc();
742 	if (pve == NULL)
743 		panic("pv_insert: no pv entries!");
744 
745 	pve->pv_pmap = pmap;
746 	pve->pv_va = va;
747 
748 	/* add to pv_list */
749 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
750 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
751 
752 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
753 
754 	//debugf("pv_insert: e\n");
755 }
756 
757 /* Destroy pv entry. */
758 static void
759 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
760 {
761 	pv_entry_t pve;
762 
763 	//int su = (pmap == kernel_pmap);
764 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
765 
766 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
767 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
768 
769 	/* find pv entry */
770 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
771 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
772 			/* remove from pv_list */
773 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
774 			if (TAILQ_EMPTY(&m->md.pv_list))
775 				vm_page_flag_clear(m, PG_WRITEABLE);
776 
777 			/* free pv entry struct */
778 			pv_free(pve);
779 			break;
780 		}
781 	}
782 
783 	//debugf("pv_remove: e\n");
784 }
785 
786 /*
787  * Clean pte entry, try to free page table page if requested.
788  *
789  * Return 1 if ptbl pages were freed, otherwise return 0.
790  */
791 static int
792 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
793 {
794 	unsigned int pdir_idx = PDIR_IDX(va);
795 	unsigned int ptbl_idx = PTBL_IDX(va);
796 	vm_page_t m;
797 	pte_t *ptbl;
798 	pte_t *pte;
799 
800 	//int su = (pmap == kernel_pmap);
801 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
802 	//		su, (u_int32_t)pmap, va, flags);
803 
804 	ptbl = pmap->pm_pdir[pdir_idx];
805 	KASSERT(ptbl, ("pte_remove: null ptbl"));
806 
807 	pte = &ptbl[ptbl_idx];
808 
809 	if (pte == NULL || !PTE_ISVALID(pte))
810 		return (0);
811 
812 	if (PTE_ISWIRED(pte))
813 		pmap->pm_stats.wired_count--;
814 
815 	/* Handle managed entry. */
816 	if (PTE_ISMANAGED(pte)) {
817 		/* Get vm_page_t for mapped pte. */
818 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
819 
820 		if (PTE_ISMODIFIED(pte))
821 			vm_page_dirty(m);
822 
823 		if (PTE_ISREFERENCED(pte))
824 			vm_page_flag_set(m, PG_REFERENCED);
825 
826 		pv_remove(pmap, va, m);
827 	}
828 
829 	mtx_lock_spin(&tlbivax_mutex);
830 	tlb_miss_lock();
831 
832 	tlb0_flush_entry(va);
833 	pte->flags = 0;
834 	pte->rpn = 0;
835 
836 	tlb_miss_unlock();
837 	mtx_unlock_spin(&tlbivax_mutex);
838 
839 	pmap->pm_stats.resident_count--;
840 
841 	if (flags & PTBL_UNHOLD) {
842 		//debugf("pte_remove: e (unhold)\n");
843 		return (ptbl_unhold(mmu, pmap, pdir_idx));
844 	}
845 
846 	//debugf("pte_remove: e\n");
847 	return (0);
848 }
849 
850 /*
851  * Insert PTE for a given page and virtual address.
852  */
853 static void
854 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
855 {
856 	unsigned int pdir_idx = PDIR_IDX(va);
857 	unsigned int ptbl_idx = PTBL_IDX(va);
858 	pte_t *ptbl, *pte;
859 
860 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
861 	    pmap == kernel_pmap, pmap, va);
862 
863 	/* Get the page table pointer. */
864 	ptbl = pmap->pm_pdir[pdir_idx];
865 
866 	if (ptbl == NULL) {
867 		/* Allocate page table pages. */
868 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
869 	} else {
870 		/*
871 		 * Check if there is valid mapping for requested
872 		 * va, if there is, remove it.
873 		 */
874 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
875 		if (PTE_ISVALID(pte)) {
876 			pte_remove(mmu, pmap, va, PTBL_HOLD);
877 		} else {
878 			/*
879 			 * pte is not used, increment hold count
880 			 * for ptbl pages.
881 			 */
882 			if (pmap != kernel_pmap)
883 				ptbl_hold(mmu, pmap, pdir_idx);
884 		}
885 	}
886 
887 	/*
888 	 * Insert pv_entry into pv_list for mapped page if part of managed
889 	 * memory.
890 	 */
891         if ((m->flags & PG_FICTITIOUS) == 0) {
892 		if ((m->flags & PG_UNMANAGED) == 0) {
893 			flags |= PTE_MANAGED;
894 
895 			/* Create and insert pv entry. */
896 			pv_insert(pmap, va, m);
897 		}
898 	}
899 
900 	pmap->pm_stats.resident_count++;
901 
902 	mtx_lock_spin(&tlbivax_mutex);
903 	tlb_miss_lock();
904 
905 	tlb0_flush_entry(va);
906 	if (pmap->pm_pdir[pdir_idx] == NULL) {
907 		/*
908 		 * If we just allocated a new page table, hook it in
909 		 * the pdir.
910 		 */
911 		pmap->pm_pdir[pdir_idx] = ptbl;
912 	}
913 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
914 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
915 	pte->flags |= (PTE_VALID | flags);
916 
917 	tlb_miss_unlock();
918 	mtx_unlock_spin(&tlbivax_mutex);
919 }
920 
921 /* Return the pa for the given pmap/va. */
922 static vm_paddr_t
923 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
924 {
925 	vm_paddr_t pa = 0;
926 	pte_t *pte;
927 
928 	pte = pte_find(mmu, pmap, va);
929 	if ((pte != NULL) && PTE_ISVALID(pte))
930 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
931 	return (pa);
932 }
933 
934 /* Get a pointer to a PTE in a page table. */
935 static pte_t *
936 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
937 {
938 	unsigned int pdir_idx = PDIR_IDX(va);
939 	unsigned int ptbl_idx = PTBL_IDX(va);
940 
941 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
942 
943 	if (pmap->pm_pdir[pdir_idx])
944 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
945 
946 	return (NULL);
947 }
948 
949 /**************************************************************************/
950 /* PMAP related */
951 /**************************************************************************/
952 
953 /*
954  * This is called during e500_init, before the system is really initialized.
955  */
956 static void
957 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
958 {
959 	vm_offset_t phys_kernelend;
960 	struct mem_region *mp, *mp1;
961 	int cnt, i, j;
962 	u_int s, e, sz;
963 	u_int phys_avail_count;
964 	vm_size_t physsz, hwphyssz, kstack0_sz;
965 	vm_offset_t kernel_pdir, kstack0, va;
966 	vm_paddr_t kstack0_phys;
967 	void *dpcpu;
968 	pte_t *pte;
969 
970 	debugf("mmu_booke_bootstrap: entered\n");
971 
972 	/* Initialize invalidation mutex */
973 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
974 
975 	/* Read TLB0 size and associativity. */
976 	tlb0_get_tlbconf();
977 
978 	/* Align kernel start and end address (kernel image). */
979 	kernstart = trunc_page(start);
980 	data_start = round_page(kernelend);
981 	kernsize = data_start - kernstart;
982 
983 	data_end = data_start;
984 
985 	/* Allocate space for the message buffer. */
986 	msgbufp = (struct msgbuf *)data_end;
987 	data_end += MSGBUF_SIZE;
988 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
989 	    data_end);
990 
991 	data_end = round_page(data_end);
992 
993 	/* Allocate the dynamic per-cpu area. */
994 	dpcpu = (void *)data_end;
995 	data_end += DPCPU_SIZE;
996 	dpcpu_init(dpcpu, 0);
997 
998 	/* Allocate space for ptbl_bufs. */
999 	ptbl_bufs = (struct ptbl_buf *)data_end;
1000 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1001 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1002 	    data_end);
1003 
1004 	data_end = round_page(data_end);
1005 
1006 	/* Allocate PTE tables for kernel KVA. */
1007 	kernel_pdir = data_end;
1008 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1009 	    PDIR_SIZE - 1) / PDIR_SIZE;
1010 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1011 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1012 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1013 
1014 	debugf(" data_end: 0x%08x\n", data_end);
1015 	if (data_end - kernstart > 0x1000000) {
1016 		data_end = (data_end + 0x3fffff) & ~0x3fffff;
1017 		tlb1_mapin_region(kernstart + 0x1000000,
1018 		    kernload + 0x1000000, data_end - kernstart - 0x1000000);
1019 	} else
1020 		data_end = (data_end + 0xffffff) & ~0xffffff;
1021 
1022 	debugf(" updated data_end: 0x%08x\n", data_end);
1023 
1024 	kernsize += data_end - data_start;
1025 
1026 	/*
1027 	 * Clear the structures - note we can only do it safely after the
1028 	 * possible additional TLB1 translations are in place (above) so that
1029 	 * all range up to the currently calculated 'data_end' is covered.
1030 	 */
1031 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1032 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1033 
1034 	/*******************************************************/
1035 	/* Set the start and end of kva. */
1036 	/*******************************************************/
1037 	virtual_avail = round_page(data_end);
1038 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1039 
1040 	/* Allocate KVA space for page zero/copy operations. */
1041 	zero_page_va = virtual_avail;
1042 	virtual_avail += PAGE_SIZE;
1043 	zero_page_idle_va = virtual_avail;
1044 	virtual_avail += PAGE_SIZE;
1045 	copy_page_src_va = virtual_avail;
1046 	virtual_avail += PAGE_SIZE;
1047 	copy_page_dst_va = virtual_avail;
1048 	virtual_avail += PAGE_SIZE;
1049 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1050 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1051 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1052 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1053 
1054 	/* Initialize page zero/copy mutexes. */
1055 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1056 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1057 
1058 	/* Allocate KVA space for ptbl bufs. */
1059 	ptbl_buf_pool_vabase = virtual_avail;
1060 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1061 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1062 	    ptbl_buf_pool_vabase, virtual_avail);
1063 
1064 	/* Calculate corresponding physical addresses for the kernel region. */
1065 	phys_kernelend = kernload + kernsize;
1066 	debugf("kernel image and allocated data:\n");
1067 	debugf(" kernload    = 0x%08x\n", kernload);
1068 	debugf(" kernstart   = 0x%08x\n", kernstart);
1069 	debugf(" kernsize    = 0x%08x\n", kernsize);
1070 
1071 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1072 		panic("mmu_booke_bootstrap: phys_avail too small");
1073 
1074 	/*
1075 	 * Remove kernel physical address range from avail regions list. Page
1076 	 * align all regions.  Non-page aligned memory isn't very interesting
1077 	 * to us.  Also, sort the entries for ascending addresses.
1078 	 */
1079 
1080 	/* Retrieve phys/avail mem regions */
1081 	mem_regions(&physmem_regions, &physmem_regions_sz,
1082 	    &availmem_regions, &availmem_regions_sz);
1083 	sz = 0;
1084 	cnt = availmem_regions_sz;
1085 	debugf("processing avail regions:\n");
1086 	for (mp = availmem_regions; mp->mr_size; mp++) {
1087 		s = mp->mr_start;
1088 		e = mp->mr_start + mp->mr_size;
1089 		debugf(" %08x-%08x -> ", s, e);
1090 		/* Check whether this region holds all of the kernel. */
1091 		if (s < kernload && e > phys_kernelend) {
1092 			availmem_regions[cnt].mr_start = phys_kernelend;
1093 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1094 			e = kernload;
1095 		}
1096 		/* Look whether this regions starts within the kernel. */
1097 		if (s >= kernload && s < phys_kernelend) {
1098 			if (e <= phys_kernelend)
1099 				goto empty;
1100 			s = phys_kernelend;
1101 		}
1102 		/* Now look whether this region ends within the kernel. */
1103 		if (e > kernload && e <= phys_kernelend) {
1104 			if (s >= kernload)
1105 				goto empty;
1106 			e = kernload;
1107 		}
1108 		/* Now page align the start and size of the region. */
1109 		s = round_page(s);
1110 		e = trunc_page(e);
1111 		if (e < s)
1112 			e = s;
1113 		sz = e - s;
1114 		debugf("%08x-%08x = %x\n", s, e, sz);
1115 
1116 		/* Check whether some memory is left here. */
1117 		if (sz == 0) {
1118 		empty:
1119 			memmove(mp, mp + 1,
1120 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1121 			cnt--;
1122 			mp--;
1123 			continue;
1124 		}
1125 
1126 		/* Do an insertion sort. */
1127 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1128 			if (s < mp1->mr_start)
1129 				break;
1130 		if (mp1 < mp) {
1131 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1132 			mp1->mr_start = s;
1133 			mp1->mr_size = sz;
1134 		} else {
1135 			mp->mr_start = s;
1136 			mp->mr_size = sz;
1137 		}
1138 	}
1139 	availmem_regions_sz = cnt;
1140 
1141 	/*******************************************************/
1142 	/* Steal physical memory for kernel stack from the end */
1143 	/* of the first avail region                           */
1144 	/*******************************************************/
1145 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1146 	kstack0_phys = availmem_regions[0].mr_start +
1147 	    availmem_regions[0].mr_size;
1148 	kstack0_phys -= kstack0_sz;
1149 	availmem_regions[0].mr_size -= kstack0_sz;
1150 
1151 	/*******************************************************/
1152 	/* Fill in phys_avail table, based on availmem_regions */
1153 	/*******************************************************/
1154 	phys_avail_count = 0;
1155 	physsz = 0;
1156 	hwphyssz = 0;
1157 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1158 
1159 	debugf("fill in phys_avail:\n");
1160 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1161 
1162 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1163 		    availmem_regions[i].mr_start,
1164 		    availmem_regions[i].mr_start +
1165 		        availmem_regions[i].mr_size,
1166 		    availmem_regions[i].mr_size);
1167 
1168 		if (hwphyssz != 0 &&
1169 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1170 			debugf(" hw.physmem adjust\n");
1171 			if (physsz < hwphyssz) {
1172 				phys_avail[j] = availmem_regions[i].mr_start;
1173 				phys_avail[j + 1] =
1174 				    availmem_regions[i].mr_start +
1175 				    hwphyssz - physsz;
1176 				physsz = hwphyssz;
1177 				phys_avail_count++;
1178 			}
1179 			break;
1180 		}
1181 
1182 		phys_avail[j] = availmem_regions[i].mr_start;
1183 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1184 		    availmem_regions[i].mr_size;
1185 		phys_avail_count++;
1186 		physsz += availmem_regions[i].mr_size;
1187 	}
1188 	physmem = btoc(physsz);
1189 
1190 	/* Calculate the last available physical address. */
1191 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1192 		;
1193 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1194 
1195 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1196 	debugf("phys_avail_count = %d\n", phys_avail_count);
1197 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1198 	    physmem);
1199 
1200 	/*******************************************************/
1201 	/* Initialize (statically allocated) kernel pmap. */
1202 	/*******************************************************/
1203 	PMAP_LOCK_INIT(kernel_pmap);
1204 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1205 
1206 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1207 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1208 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1209 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1210 
1211 	/* Initialize kernel pdir */
1212 	for (i = 0; i < kernel_ptbls; i++)
1213 		kernel_pmap->pm_pdir[kptbl_min + i] =
1214 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1215 
1216 	for (i = 0; i < MAXCPU; i++) {
1217 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1218 
1219 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1220 		tidbusy[i][0] = kernel_pmap;
1221 	}
1222 
1223 	/*
1224 	 * Fill in PTEs covering kernel code and data. They are not required
1225 	 * for address translation, as this area is covered by static TLB1
1226 	 * entries, but for pte_vatopa() to work correctly with kernel area
1227 	 * addresses.
1228 	 */
1229 	for (va = KERNBASE; va < data_end; va += PAGE_SIZE) {
1230 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1231 		pte->rpn = kernload + (va - KERNBASE);
1232 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1233 		    PTE_VALID;
1234 	}
1235 	/* Mark kernel_pmap active on all CPUs */
1236 	kernel_pmap->pm_active = ~0;
1237 
1238 	/*******************************************************/
1239 	/* Final setup */
1240 	/*******************************************************/
1241 
1242 	/* Enter kstack0 into kernel map, provide guard page */
1243 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1244 	thread0.td_kstack = kstack0;
1245 	thread0.td_kstack_pages = KSTACK_PAGES;
1246 
1247 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1248 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1249 	    kstack0_phys, kstack0_phys + kstack0_sz);
1250 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1251 
1252 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1253 	for (i = 0; i < KSTACK_PAGES; i++) {
1254 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1255 		kstack0 += PAGE_SIZE;
1256 		kstack0_phys += PAGE_SIZE;
1257 	}
1258 
1259 	debugf("virtual_avail = %08x\n", virtual_avail);
1260 	debugf("virtual_end   = %08x\n", virtual_end);
1261 
1262 	debugf("mmu_booke_bootstrap: exit\n");
1263 }
1264 
1265 void
1266 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1267 {
1268 	int i;
1269 
1270 	/*
1271 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1272 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1273 	 * these values directly to (re)program AP's TLB1 hardware.
1274 	 */
1275 	for (i = 0; i < tlb1_idx; i ++) {
1276 		/* Skip invalid entries */
1277 		if (!(tlb1[i].mas1 & MAS1_VALID))
1278 			continue;
1279 
1280 		tlb1_write_entry(i);
1281 	}
1282 
1283 	set_mas4_defaults();
1284 }
1285 
1286 /*
1287  * Get the physical page address for the given pmap/virtual address.
1288  */
1289 static vm_paddr_t
1290 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1291 {
1292 	vm_paddr_t pa;
1293 
1294 	PMAP_LOCK(pmap);
1295 	pa = pte_vatopa(mmu, pmap, va);
1296 	PMAP_UNLOCK(pmap);
1297 
1298 	return (pa);
1299 }
1300 
1301 /*
1302  * Extract the physical page address associated with the given
1303  * kernel virtual address.
1304  */
1305 static vm_paddr_t
1306 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1307 {
1308 
1309 	return (pte_vatopa(mmu, kernel_pmap, va));
1310 }
1311 
1312 /*
1313  * Initialize the pmap module.
1314  * Called by vm_init, to initialize any structures that the pmap
1315  * system needs to map virtual memory.
1316  */
1317 static void
1318 mmu_booke_init(mmu_t mmu)
1319 {
1320 	int shpgperproc = PMAP_SHPGPERPROC;
1321 
1322 	/*
1323 	 * Initialize the address space (zone) for the pv entries.  Set a
1324 	 * high water mark so that the system can recover from excessive
1325 	 * numbers of pv entries.
1326 	 */
1327 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1328 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1329 
1330 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1331 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1332 
1333 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1334 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1335 
1336 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1337 
1338 	/* Pre-fill pvzone with initial number of pv entries. */
1339 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1340 
1341 	/* Initialize ptbl allocation. */
1342 	ptbl_init();
1343 }
1344 
1345 /*
1346  * Map a list of wired pages into kernel virtual address space.  This is
1347  * intended for temporary mappings which do not need page modification or
1348  * references recorded.  Existing mappings in the region are overwritten.
1349  */
1350 static void
1351 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1352 {
1353 	vm_offset_t va;
1354 
1355 	va = sva;
1356 	while (count-- > 0) {
1357 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1358 		va += PAGE_SIZE;
1359 		m++;
1360 	}
1361 }
1362 
1363 /*
1364  * Remove page mappings from kernel virtual address space.  Intended for
1365  * temporary mappings entered by mmu_booke_qenter.
1366  */
1367 static void
1368 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1369 {
1370 	vm_offset_t va;
1371 
1372 	va = sva;
1373 	while (count-- > 0) {
1374 		mmu_booke_kremove(mmu, va);
1375 		va += PAGE_SIZE;
1376 	}
1377 }
1378 
1379 /*
1380  * Map a wired page into kernel virtual address space.
1381  */
1382 static void
1383 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1384 {
1385 	unsigned int pdir_idx = PDIR_IDX(va);
1386 	unsigned int ptbl_idx = PTBL_IDX(va);
1387 	uint32_t flags;
1388 	pte_t *pte;
1389 
1390 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1391 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1392 
1393 	flags = 0;
1394 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1395 	flags |= PTE_M;
1396 
1397 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1398 
1399 	mtx_lock_spin(&tlbivax_mutex);
1400 	tlb_miss_lock();
1401 
1402 	if (PTE_ISVALID(pte)) {
1403 
1404 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1405 
1406 		/* Flush entry from TLB0 */
1407 		tlb0_flush_entry(va);
1408 	}
1409 
1410 	pte->rpn = pa & ~PTE_PA_MASK;
1411 	pte->flags = flags;
1412 
1413 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1414 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1415 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1416 
1417 	/* Flush the real memory from the instruction cache. */
1418 	if ((flags & (PTE_I | PTE_G)) == 0) {
1419 		__syncicache((void *)va, PAGE_SIZE);
1420 	}
1421 
1422 	tlb_miss_unlock();
1423 	mtx_unlock_spin(&tlbivax_mutex);
1424 }
1425 
1426 /*
1427  * Remove a page from kernel page table.
1428  */
1429 static void
1430 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1431 {
1432 	unsigned int pdir_idx = PDIR_IDX(va);
1433 	unsigned int ptbl_idx = PTBL_IDX(va);
1434 	pte_t *pte;
1435 
1436 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1437 
1438 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1439 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1440 	    ("mmu_booke_kremove: invalid va"));
1441 
1442 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1443 
1444 	if (!PTE_ISVALID(pte)) {
1445 
1446 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1447 
1448 		return;
1449 	}
1450 
1451 	mtx_lock_spin(&tlbivax_mutex);
1452 	tlb_miss_lock();
1453 
1454 	/* Invalidate entry in TLB0, update PTE. */
1455 	tlb0_flush_entry(va);
1456 	pte->flags = 0;
1457 	pte->rpn = 0;
1458 
1459 	tlb_miss_unlock();
1460 	mtx_unlock_spin(&tlbivax_mutex);
1461 }
1462 
1463 /*
1464  * Initialize pmap associated with process 0.
1465  */
1466 static void
1467 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1468 {
1469 
1470 	mmu_booke_pinit(mmu, pmap);
1471 	PCPU_SET(curpmap, pmap);
1472 }
1473 
1474 /*
1475  * Initialize a preallocated and zeroed pmap structure,
1476  * such as one in a vmspace structure.
1477  */
1478 static void
1479 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1480 {
1481 	int i;
1482 
1483 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1484 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1485 
1486 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1487 
1488 	PMAP_LOCK_INIT(pmap);
1489 	for (i = 0; i < MAXCPU; i++)
1490 		pmap->pm_tid[i] = TID_NONE;
1491 	pmap->pm_active = 0;
1492 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1493 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1494 	TAILQ_INIT(&pmap->pm_ptbl_list);
1495 }
1496 
1497 /*
1498  * Release any resources held by the given physical map.
1499  * Called when a pmap initialized by mmu_booke_pinit is being released.
1500  * Should only be called if the map contains no valid mappings.
1501  */
1502 static void
1503 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1504 {
1505 
1506 	printf("mmu_booke_release: s\n");
1507 
1508 	KASSERT(pmap->pm_stats.resident_count == 0,
1509 	    ("pmap_release: pmap resident count %ld != 0",
1510 	    pmap->pm_stats.resident_count));
1511 
1512 	PMAP_LOCK_DESTROY(pmap);
1513 }
1514 
1515 /*
1516  * Insert the given physical page at the specified virtual address in the
1517  * target physical map with the protection requested. If specified the page
1518  * will be wired down.
1519  */
1520 static void
1521 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1522     vm_prot_t prot, boolean_t wired)
1523 {
1524 
1525 	vm_page_lock_queues();
1526 	PMAP_LOCK(pmap);
1527 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1528 	vm_page_unlock_queues();
1529 	PMAP_UNLOCK(pmap);
1530 }
1531 
1532 static void
1533 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1534     vm_prot_t prot, boolean_t wired)
1535 {
1536 	pte_t *pte;
1537 	vm_paddr_t pa;
1538 	uint32_t flags;
1539 	int su, sync;
1540 
1541 	pa = VM_PAGE_TO_PHYS(m);
1542 	su = (pmap == kernel_pmap);
1543 	sync = 0;
1544 
1545 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1546 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1547 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1548 	//		(u_int32_t)m, va, pa, prot, wired);
1549 
1550 	if (su) {
1551 		KASSERT(((va >= virtual_avail) &&
1552 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1553 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1554 	} else {
1555 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1556 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1557 	}
1558 
1559 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1560 
1561 	/*
1562 	 * If there is an existing mapping, and the physical address has not
1563 	 * changed, must be protection or wiring change.
1564 	 */
1565 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1566 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1567 
1568 		/*
1569 		 * Before actually updating pte->flags we calculate and
1570 		 * prepare its new value in a helper var.
1571 		 */
1572 		flags = pte->flags;
1573 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1574 
1575 		/* Wiring change, just update stats. */
1576 		if (wired) {
1577 			if (!PTE_ISWIRED(pte)) {
1578 				flags |= PTE_WIRED;
1579 				pmap->pm_stats.wired_count++;
1580 			}
1581 		} else {
1582 			if (PTE_ISWIRED(pte)) {
1583 				flags &= ~PTE_WIRED;
1584 				pmap->pm_stats.wired_count--;
1585 			}
1586 		}
1587 
1588 		if (prot & VM_PROT_WRITE) {
1589 			/* Add write permissions. */
1590 			flags |= PTE_SW;
1591 			if (!su)
1592 				flags |= PTE_UW;
1593 
1594 			vm_page_flag_set(m, PG_WRITEABLE);
1595 		} else {
1596 			/* Handle modified pages, sense modify status. */
1597 
1598 			/*
1599 			 * The PTE_MODIFIED flag could be set by underlying
1600 			 * TLB misses since we last read it (above), possibly
1601 			 * other CPUs could update it so we check in the PTE
1602 			 * directly rather than rely on that saved local flags
1603 			 * copy.
1604 			 */
1605 			if (PTE_ISMODIFIED(pte))
1606 				vm_page_dirty(m);
1607 		}
1608 
1609 		if (prot & VM_PROT_EXECUTE) {
1610 			flags |= PTE_SX;
1611 			if (!su)
1612 				flags |= PTE_UX;
1613 
1614 			/*
1615 			 * Check existing flags for execute permissions: if we
1616 			 * are turning execute permissions on, icache should
1617 			 * be flushed.
1618 			 */
1619 			if ((flags & (PTE_UX | PTE_SX)) == 0)
1620 				sync++;
1621 		}
1622 
1623 		flags &= ~PTE_REFERENCED;
1624 
1625 		/*
1626 		 * The new flags value is all calculated -- only now actually
1627 		 * update the PTE.
1628 		 */
1629 		mtx_lock_spin(&tlbivax_mutex);
1630 		tlb_miss_lock();
1631 
1632 		tlb0_flush_entry(va);
1633 		pte->flags = flags;
1634 
1635 		tlb_miss_unlock();
1636 		mtx_unlock_spin(&tlbivax_mutex);
1637 
1638 	} else {
1639 		/*
1640 		 * If there is an existing mapping, but it's for a different
1641 		 * physical address, pte_enter() will delete the old mapping.
1642 		 */
1643 		//if ((pte != NULL) && PTE_ISVALID(pte))
1644 		//	debugf("mmu_booke_enter_locked: replace\n");
1645 		//else
1646 		//	debugf("mmu_booke_enter_locked: new\n");
1647 
1648 		/* Now set up the flags and install the new mapping. */
1649 		flags = (PTE_SR | PTE_VALID);
1650 		flags |= PTE_M;
1651 
1652 		if (!su)
1653 			flags |= PTE_UR;
1654 
1655 		if (prot & VM_PROT_WRITE) {
1656 			flags |= PTE_SW;
1657 			if (!su)
1658 				flags |= PTE_UW;
1659 
1660 			vm_page_flag_set(m, PG_WRITEABLE);
1661 		}
1662 
1663 		if (prot & VM_PROT_EXECUTE) {
1664 			flags |= PTE_SX;
1665 			if (!su)
1666 				flags |= PTE_UX;
1667 		}
1668 
1669 		/* If its wired update stats. */
1670 		if (wired) {
1671 			pmap->pm_stats.wired_count++;
1672 			flags |= PTE_WIRED;
1673 		}
1674 
1675 		pte_enter(mmu, pmap, m, va, flags);
1676 
1677 		/* Flush the real memory from the instruction cache. */
1678 		if (prot & VM_PROT_EXECUTE)
1679 			sync++;
1680 	}
1681 
1682 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1683 		__syncicache((void *)va, PAGE_SIZE);
1684 		sync = 0;
1685 	}
1686 }
1687 
1688 /*
1689  * Maps a sequence of resident pages belonging to the same object.
1690  * The sequence begins with the given page m_start.  This page is
1691  * mapped at the given virtual address start.  Each subsequent page is
1692  * mapped at a virtual address that is offset from start by the same
1693  * amount as the page is offset from m_start within the object.  The
1694  * last page in the sequence is the page with the largest offset from
1695  * m_start that can be mapped at a virtual address less than the given
1696  * virtual address end.  Not every virtual page between start and end
1697  * is mapped; only those for which a resident page exists with the
1698  * corresponding offset from m_start are mapped.
1699  */
1700 static void
1701 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1702     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1703 {
1704 	vm_page_t m;
1705 	vm_pindex_t diff, psize;
1706 
1707 	psize = atop(end - start);
1708 	m = m_start;
1709 	PMAP_LOCK(pmap);
1710 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1711 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1712 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1713 		m = TAILQ_NEXT(m, listq);
1714 	}
1715 	PMAP_UNLOCK(pmap);
1716 }
1717 
1718 static void
1719 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1720     vm_prot_t prot)
1721 {
1722 
1723 	PMAP_LOCK(pmap);
1724 	mmu_booke_enter_locked(mmu, pmap, va, m,
1725 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1726 	PMAP_UNLOCK(pmap);
1727 }
1728 
1729 /*
1730  * Remove the given range of addresses from the specified map.
1731  *
1732  * It is assumed that the start and end are properly rounded to the page size.
1733  */
1734 static void
1735 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1736 {
1737 	pte_t *pte;
1738 	uint8_t hold_flag;
1739 
1740 	int su = (pmap == kernel_pmap);
1741 
1742 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1743 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1744 
1745 	if (su) {
1746 		KASSERT(((va >= virtual_avail) &&
1747 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1748 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1749 	} else {
1750 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1751 		    ("mmu_booke_remove: user pmap, non user va"));
1752 	}
1753 
1754 	if (PMAP_REMOVE_DONE(pmap)) {
1755 		//debugf("mmu_booke_remove: e (empty)\n");
1756 		return;
1757 	}
1758 
1759 	hold_flag = PTBL_HOLD_FLAG(pmap);
1760 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1761 
1762 	vm_page_lock_queues();
1763 	PMAP_LOCK(pmap);
1764 	for (; va < endva; va += PAGE_SIZE) {
1765 		pte = pte_find(mmu, pmap, va);
1766 		if ((pte != NULL) && PTE_ISVALID(pte))
1767 			pte_remove(mmu, pmap, va, hold_flag);
1768 	}
1769 	PMAP_UNLOCK(pmap);
1770 	vm_page_unlock_queues();
1771 
1772 	//debugf("mmu_booke_remove: e\n");
1773 }
1774 
1775 /*
1776  * Remove physical page from all pmaps in which it resides.
1777  */
1778 static void
1779 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1780 {
1781 	pv_entry_t pv, pvn;
1782 	uint8_t hold_flag;
1783 
1784 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1785 
1786 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1787 		pvn = TAILQ_NEXT(pv, pv_link);
1788 
1789 		PMAP_LOCK(pv->pv_pmap);
1790 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1791 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1792 		PMAP_UNLOCK(pv->pv_pmap);
1793 	}
1794 	vm_page_flag_clear(m, PG_WRITEABLE);
1795 }
1796 
1797 /*
1798  * Map a range of physical addresses into kernel virtual address space.
1799  */
1800 static vm_offset_t
1801 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1802     vm_offset_t pa_end, int prot)
1803 {
1804 	vm_offset_t sva = *virt;
1805 	vm_offset_t va = sva;
1806 
1807 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1808 	//		sva, pa_start, pa_end);
1809 
1810 	while (pa_start < pa_end) {
1811 		mmu_booke_kenter(mmu, va, pa_start);
1812 		va += PAGE_SIZE;
1813 		pa_start += PAGE_SIZE;
1814 	}
1815 	*virt = va;
1816 
1817 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1818 	return (sva);
1819 }
1820 
1821 /*
1822  * The pmap must be activated before it's address space can be accessed in any
1823  * way.
1824  */
1825 static void
1826 mmu_booke_activate(mmu_t mmu, struct thread *td)
1827 {
1828 	pmap_t pmap;
1829 
1830 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1831 
1832 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1833 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1834 
1835 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1836 
1837 	mtx_lock_spin(&sched_lock);
1838 
1839 	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
1840 	PCPU_SET(curpmap, pmap);
1841 
1842 	if (pmap->pm_tid[PCPU_GET(cpuid)] == TID_NONE)
1843 		tid_alloc(pmap);
1844 
1845 	/* Load PID0 register with pmap tid value. */
1846 	mtspr(SPR_PID0, pmap->pm_tid[PCPU_GET(cpuid)]);
1847 	__asm __volatile("isync");
1848 
1849 	mtx_unlock_spin(&sched_lock);
1850 
1851 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1852 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1853 }
1854 
1855 /*
1856  * Deactivate the specified process's address space.
1857  */
1858 static void
1859 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1860 {
1861 	pmap_t pmap;
1862 
1863 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1864 
1865 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1866 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1867 
1868 	atomic_clear_int(&pmap->pm_active, PCPU_GET(cpumask));
1869 	PCPU_SET(curpmap, NULL);
1870 }
1871 
1872 /*
1873  * Copy the range specified by src_addr/len
1874  * from the source map to the range dst_addr/len
1875  * in the destination map.
1876  *
1877  * This routine is only advisory and need not do anything.
1878  */
1879 static void
1880 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1881     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1882 {
1883 
1884 }
1885 
1886 /*
1887  * Set the physical protection on the specified range of this map as requested.
1888  */
1889 static void
1890 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1891     vm_prot_t prot)
1892 {
1893 	vm_offset_t va;
1894 	vm_page_t m;
1895 	pte_t *pte;
1896 
1897 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1898 		mmu_booke_remove(mmu, pmap, sva, eva);
1899 		return;
1900 	}
1901 
1902 	if (prot & VM_PROT_WRITE)
1903 		return;
1904 
1905 	vm_page_lock_queues();
1906 	PMAP_LOCK(pmap);
1907 	for (va = sva; va < eva; va += PAGE_SIZE) {
1908 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1909 			if (PTE_ISVALID(pte)) {
1910 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1911 
1912 				mtx_lock_spin(&tlbivax_mutex);
1913 				tlb_miss_lock();
1914 
1915 				/* Handle modified pages. */
1916 				if (PTE_ISMODIFIED(pte))
1917 					vm_page_dirty(m);
1918 
1919 				/* Referenced pages. */
1920 				if (PTE_ISREFERENCED(pte))
1921 					vm_page_flag_set(m, PG_REFERENCED);
1922 
1923 				tlb0_flush_entry(va);
1924 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1925 				    PTE_REFERENCED);
1926 
1927 				tlb_miss_unlock();
1928 				mtx_unlock_spin(&tlbivax_mutex);
1929 			}
1930 		}
1931 	}
1932 	PMAP_UNLOCK(pmap);
1933 	vm_page_unlock_queues();
1934 }
1935 
1936 /*
1937  * Clear the write and modified bits in each of the given page's mappings.
1938  */
1939 static void
1940 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1941 {
1942 	pv_entry_t pv;
1943 	pte_t *pte;
1944 
1945 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1946 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1947 	    (m->flags & PG_WRITEABLE) == 0)
1948 		return;
1949 
1950 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1951 		PMAP_LOCK(pv->pv_pmap);
1952 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1953 			if (PTE_ISVALID(pte)) {
1954 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1955 
1956 				mtx_lock_spin(&tlbivax_mutex);
1957 				tlb_miss_lock();
1958 
1959 				/* Handle modified pages. */
1960 				if (PTE_ISMODIFIED(pte))
1961 					vm_page_dirty(m);
1962 
1963 				/* Referenced pages. */
1964 				if (PTE_ISREFERENCED(pte))
1965 					vm_page_flag_set(m, PG_REFERENCED);
1966 
1967 				/* Flush mapping from TLB0. */
1968 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED |
1969 				    PTE_REFERENCED);
1970 
1971 				tlb_miss_unlock();
1972 				mtx_unlock_spin(&tlbivax_mutex);
1973 			}
1974 		}
1975 		PMAP_UNLOCK(pv->pv_pmap);
1976 	}
1977 	vm_page_flag_clear(m, PG_WRITEABLE);
1978 }
1979 
1980 static void
1981 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1982 {
1983 	pte_t *pte;
1984 	pmap_t pmap;
1985 	vm_page_t m;
1986 	vm_offset_t addr;
1987 	vm_paddr_t pa;
1988 	int active, valid;
1989 
1990 	va = trunc_page(va);
1991 	sz = round_page(sz);
1992 
1993 	vm_page_lock_queues();
1994 	pmap = PCPU_GET(curpmap);
1995 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
1996 	while (sz > 0) {
1997 		PMAP_LOCK(pm);
1998 		pte = pte_find(mmu, pm, va);
1999 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2000 		if (valid)
2001 			pa = PTE_PA(pte);
2002 		PMAP_UNLOCK(pm);
2003 		if (valid) {
2004 			if (!active) {
2005 				/* Create a mapping in the active pmap. */
2006 				addr = 0;
2007 				m = PHYS_TO_VM_PAGE(pa);
2008 				PMAP_LOCK(pmap);
2009 				pte_enter(mmu, pmap, m, addr,
2010 				    PTE_SR | PTE_VALID | PTE_UR);
2011 				__syncicache((void *)addr, PAGE_SIZE);
2012 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2013 				PMAP_UNLOCK(pmap);
2014 			} else
2015 				__syncicache((void *)va, PAGE_SIZE);
2016 		}
2017 		va += PAGE_SIZE;
2018 		sz -= PAGE_SIZE;
2019 	}
2020 	vm_page_unlock_queues();
2021 }
2022 
2023 /*
2024  * Atomically extract and hold the physical page with the given
2025  * pmap and virtual address pair if that mapping permits the given
2026  * protection.
2027  */
2028 static vm_page_t
2029 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2030     vm_prot_t prot)
2031 {
2032 	pte_t *pte;
2033 	vm_page_t m;
2034 	uint32_t pte_wbit;
2035 
2036 	m = NULL;
2037 	vm_page_lock_queues();
2038 	PMAP_LOCK(pmap);
2039 
2040 	pte = pte_find(mmu, pmap, va);
2041 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2042 		if (pmap == kernel_pmap)
2043 			pte_wbit = PTE_SW;
2044 		else
2045 			pte_wbit = PTE_UW;
2046 
2047 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2048 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2049 			vm_page_hold(m);
2050 		}
2051 	}
2052 
2053 	vm_page_unlock_queues();
2054 	PMAP_UNLOCK(pmap);
2055 	return (m);
2056 }
2057 
2058 /*
2059  * Initialize a vm_page's machine-dependent fields.
2060  */
2061 static void
2062 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2063 {
2064 
2065 	TAILQ_INIT(&m->md.pv_list);
2066 }
2067 
2068 /*
2069  * mmu_booke_zero_page_area zeros the specified hardware page by
2070  * mapping it into virtual memory and using bzero to clear
2071  * its contents.
2072  *
2073  * off and size must reside within a single page.
2074  */
2075 static void
2076 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2077 {
2078 	vm_offset_t va;
2079 
2080 	/* XXX KASSERT off and size are within a single page? */
2081 
2082 	mtx_lock(&zero_page_mutex);
2083 	va = zero_page_va;
2084 
2085 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2086 	bzero((caddr_t)va + off, size);
2087 	mmu_booke_kremove(mmu, va);
2088 
2089 	mtx_unlock(&zero_page_mutex);
2090 }
2091 
2092 /*
2093  * mmu_booke_zero_page zeros the specified hardware page.
2094  */
2095 static void
2096 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2097 {
2098 
2099 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2100 }
2101 
2102 /*
2103  * mmu_booke_copy_page copies the specified (machine independent) page by
2104  * mapping the page into virtual memory and using memcopy to copy the page,
2105  * one machine dependent page at a time.
2106  */
2107 static void
2108 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2109 {
2110 	vm_offset_t sva, dva;
2111 
2112 	sva = copy_page_src_va;
2113 	dva = copy_page_dst_va;
2114 
2115 	mtx_lock(&copy_page_mutex);
2116 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2117 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2118 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2119 	mmu_booke_kremove(mmu, dva);
2120 	mmu_booke_kremove(mmu, sva);
2121 	mtx_unlock(&copy_page_mutex);
2122 }
2123 
2124 /*
2125  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2126  * into virtual memory and using bzero to clear its contents. This is intended
2127  * to be called from the vm_pagezero process only and outside of Giant. No
2128  * lock is required.
2129  */
2130 static void
2131 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2132 {
2133 	vm_offset_t va;
2134 
2135 	va = zero_page_idle_va;
2136 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2137 	bzero((caddr_t)va, PAGE_SIZE);
2138 	mmu_booke_kremove(mmu, va);
2139 }
2140 
2141 /*
2142  * Return whether or not the specified physical page was modified
2143  * in any of physical maps.
2144  */
2145 static boolean_t
2146 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2147 {
2148 	pte_t *pte;
2149 	pv_entry_t pv;
2150 
2151 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2152 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2153 		return (FALSE);
2154 
2155 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2156 		PMAP_LOCK(pv->pv_pmap);
2157 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2158 			if (!PTE_ISVALID(pte))
2159 				goto make_sure_to_unlock;
2160 
2161 			if (PTE_ISMODIFIED(pte)) {
2162 				PMAP_UNLOCK(pv->pv_pmap);
2163 				return (TRUE);
2164 			}
2165 		}
2166 make_sure_to_unlock:
2167 		PMAP_UNLOCK(pv->pv_pmap);
2168 	}
2169 	return (FALSE);
2170 }
2171 
2172 /*
2173  * Return whether or not the specified virtual address is eligible
2174  * for prefault.
2175  */
2176 static boolean_t
2177 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2178 {
2179 
2180 	return (FALSE);
2181 }
2182 
2183 /*
2184  * Clear the modify bits on the specified physical page.
2185  */
2186 static void
2187 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2188 {
2189 	pte_t *pte;
2190 	pv_entry_t pv;
2191 
2192 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2193 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2194 		return;
2195 
2196 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2197 		PMAP_LOCK(pv->pv_pmap);
2198 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2199 			if (!PTE_ISVALID(pte))
2200 				goto make_sure_to_unlock;
2201 
2202 			mtx_lock_spin(&tlbivax_mutex);
2203 			tlb_miss_lock();
2204 
2205 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2206 				tlb0_flush_entry(pv->pv_va);
2207 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2208 				    PTE_REFERENCED);
2209 			}
2210 
2211 			tlb_miss_unlock();
2212 			mtx_unlock_spin(&tlbivax_mutex);
2213 		}
2214 make_sure_to_unlock:
2215 		PMAP_UNLOCK(pv->pv_pmap);
2216 	}
2217 }
2218 
2219 /*
2220  * Return a count of reference bits for a page, clearing those bits.
2221  * It is not necessary for every reference bit to be cleared, but it
2222  * is necessary that 0 only be returned when there are truly no
2223  * reference bits set.
2224  *
2225  * XXX: The exact number of bits to check and clear is a matter that
2226  * should be tested and standardized at some point in the future for
2227  * optimal aging of shared pages.
2228  */
2229 static int
2230 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2231 {
2232 	pte_t *pte;
2233 	pv_entry_t pv;
2234 	int count;
2235 
2236 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2237 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2238 		return (0);
2239 
2240 	count = 0;
2241 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2242 		PMAP_LOCK(pv->pv_pmap);
2243 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2244 			if (!PTE_ISVALID(pte))
2245 				goto make_sure_to_unlock;
2246 
2247 			if (PTE_ISREFERENCED(pte)) {
2248 				mtx_lock_spin(&tlbivax_mutex);
2249 				tlb_miss_lock();
2250 
2251 				tlb0_flush_entry(pv->pv_va);
2252 				pte->flags &= ~PTE_REFERENCED;
2253 
2254 				tlb_miss_unlock();
2255 				mtx_unlock_spin(&tlbivax_mutex);
2256 
2257 				if (++count > 4) {
2258 					PMAP_UNLOCK(pv->pv_pmap);
2259 					break;
2260 				}
2261 			}
2262 		}
2263 make_sure_to_unlock:
2264 		PMAP_UNLOCK(pv->pv_pmap);
2265 	}
2266 	return (count);
2267 }
2268 
2269 /*
2270  * Clear the reference bit on the specified physical page.
2271  */
2272 static void
2273 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2274 {
2275 	pte_t *pte;
2276 	pv_entry_t pv;
2277 
2278 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2279 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2280 		return;
2281 
2282 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2283 		PMAP_LOCK(pv->pv_pmap);
2284 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2285 			if (!PTE_ISVALID(pte))
2286 				goto make_sure_to_unlock;
2287 
2288 			if (PTE_ISREFERENCED(pte)) {
2289 				mtx_lock_spin(&tlbivax_mutex);
2290 				tlb_miss_lock();
2291 
2292 				tlb0_flush_entry(pv->pv_va);
2293 				pte->flags &= ~PTE_REFERENCED;
2294 
2295 				tlb_miss_unlock();
2296 				mtx_unlock_spin(&tlbivax_mutex);
2297 			}
2298 		}
2299 make_sure_to_unlock:
2300 		PMAP_UNLOCK(pv->pv_pmap);
2301 	}
2302 }
2303 
2304 /*
2305  * Change wiring attribute for a map/virtual-address pair.
2306  */
2307 static void
2308 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2309 {
2310 	pte_t *pte;;
2311 
2312 	PMAP_LOCK(pmap);
2313 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2314 		if (wired) {
2315 			if (!PTE_ISWIRED(pte)) {
2316 				pte->flags |= PTE_WIRED;
2317 				pmap->pm_stats.wired_count++;
2318 			}
2319 		} else {
2320 			if (PTE_ISWIRED(pte)) {
2321 				pte->flags &= ~PTE_WIRED;
2322 				pmap->pm_stats.wired_count--;
2323 			}
2324 		}
2325 	}
2326 	PMAP_UNLOCK(pmap);
2327 }
2328 
2329 /*
2330  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2331  * page.  This count may be changed upwards or downwards in the future; it is
2332  * only necessary that true be returned for a small subset of pmaps for proper
2333  * page aging.
2334  */
2335 static boolean_t
2336 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2337 {
2338 	pv_entry_t pv;
2339 	int loops;
2340 
2341 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2342 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
2343 		return (FALSE);
2344 
2345 	loops = 0;
2346 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2347 		if (pv->pv_pmap == pmap)
2348 			return (TRUE);
2349 
2350 		if (++loops >= 16)
2351 			break;
2352 	}
2353 	return (FALSE);
2354 }
2355 
2356 /*
2357  * Return the number of managed mappings to the given physical page that are
2358  * wired.
2359  */
2360 static int
2361 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2362 {
2363 	pv_entry_t pv;
2364 	pte_t *pte;
2365 	int count = 0;
2366 
2367 	if ((m->flags & PG_FICTITIOUS) != 0)
2368 		return (count);
2369 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
2370 
2371 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2372 		PMAP_LOCK(pv->pv_pmap);
2373 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2374 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2375 				count++;
2376 		PMAP_UNLOCK(pv->pv_pmap);
2377 	}
2378 
2379 	return (count);
2380 }
2381 
2382 static int
2383 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2384 {
2385 	int i;
2386 	vm_offset_t va;
2387 
2388 	/*
2389 	 * This currently does not work for entries that
2390 	 * overlap TLB1 entries.
2391 	 */
2392 	for (i = 0; i < tlb1_idx; i ++) {
2393 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2394 			return (0);
2395 	}
2396 
2397 	return (EFAULT);
2398 }
2399 
2400 vm_offset_t
2401 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2402     vm_size_t *sz)
2403 {
2404 	vm_paddr_t pa, ppa;
2405 	vm_offset_t va;
2406 	vm_size_t gran;
2407 
2408 	/* Raw physical memory dumps don't have a virtual address. */
2409 	if (md->md_vaddr == ~0UL) {
2410 		/* We always map a 256MB page at 256M. */
2411 		gran = 256 * 1024 * 1024;
2412 		pa = md->md_paddr + ofs;
2413 		ppa = pa & ~(gran - 1);
2414 		ofs = pa - ppa;
2415 		va = gran;
2416 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2417 		if (*sz > (gran - ofs))
2418 			*sz = gran - ofs;
2419 		return (va + ofs);
2420 	}
2421 
2422 	/* Minidumps are based on virtual memory addresses. */
2423 	va = md->md_vaddr + ofs;
2424 	if (va >= kernstart + kernsize) {
2425 		gran = PAGE_SIZE - (va & PAGE_MASK);
2426 		if (*sz > gran)
2427 			*sz = gran;
2428 	}
2429 	return (va);
2430 }
2431 
2432 void
2433 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2434     vm_offset_t va)
2435 {
2436 
2437 	/* Raw physical memory dumps don't have a virtual address. */
2438 	if (md->md_vaddr == ~0UL) {
2439 		tlb1_idx--;
2440 		tlb1[tlb1_idx].mas1 = 0;
2441 		tlb1[tlb1_idx].mas2 = 0;
2442 		tlb1[tlb1_idx].mas3 = 0;
2443 		tlb1_write_entry(tlb1_idx);
2444 		return;
2445 	}
2446 
2447 	/* Minidumps are based on virtual memory addresses. */
2448 	/* Nothing to do... */
2449 }
2450 
2451 struct pmap_md *
2452 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2453 {
2454 	static struct pmap_md md;
2455 	struct bi_mem_region *mr;
2456 	pte_t *pte;
2457 	vm_offset_t va;
2458 
2459 	if (dumpsys_minidump) {
2460 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2461 		if (prev == NULL) {
2462 			/* 1st: kernel .data and .bss. */
2463 			md.md_index = 1;
2464 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2465 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2466 			return (&md);
2467 		}
2468 		switch (prev->md_index) {
2469 		case 1:
2470 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2471 			md.md_index = 2;
2472 			md.md_vaddr = data_start;
2473 			md.md_size = data_end - data_start;
2474 			break;
2475 		case 2:
2476 			/* 3rd: kernel VM. */
2477 			va = prev->md_vaddr + prev->md_size;
2478 			/* Find start of next chunk (from va). */
2479 			while (va < virtual_end) {
2480 				/* Don't dump the buffer cache. */
2481 				if (va >= kmi.buffer_sva &&
2482 				    va < kmi.buffer_eva) {
2483 					va = kmi.buffer_eva;
2484 					continue;
2485 				}
2486 				pte = pte_find(mmu, kernel_pmap, va);
2487 				if (pte != NULL && PTE_ISVALID(pte))
2488 					break;
2489 				va += PAGE_SIZE;
2490 			}
2491 			if (va < virtual_end) {
2492 				md.md_vaddr = va;
2493 				va += PAGE_SIZE;
2494 				/* Find last page in chunk. */
2495 				while (va < virtual_end) {
2496 					/* Don't run into the buffer cache. */
2497 					if (va == kmi.buffer_sva)
2498 						break;
2499 					pte = pte_find(mmu, kernel_pmap, va);
2500 					if (pte == NULL || !PTE_ISVALID(pte))
2501 						break;
2502 					va += PAGE_SIZE;
2503 				}
2504 				md.md_size = va - md.md_vaddr;
2505 				break;
2506 			}
2507 			md.md_index = 3;
2508 			/* FALLTHROUGH */
2509 		default:
2510 			return (NULL);
2511 		}
2512 	} else { /* minidumps */
2513 		mr = bootinfo_mr();
2514 		if (prev == NULL) {
2515 			/* first physical chunk. */
2516 			md.md_paddr = mr->mem_base;
2517 			md.md_size = mr->mem_size;
2518 			md.md_vaddr = ~0UL;
2519 			md.md_index = 1;
2520 		} else if (md.md_index < bootinfo->bi_mem_reg_no) {
2521 			md.md_paddr = mr[md.md_index].mem_base;
2522 			md.md_size = mr[md.md_index].mem_size;
2523 			md.md_vaddr = ~0UL;
2524 			md.md_index++;
2525 		} else {
2526 			/* There's no next physical chunk. */
2527 			return (NULL);
2528 		}
2529 	}
2530 
2531 	return (&md);
2532 }
2533 
2534 /*
2535  * Map a set of physical memory pages into the kernel virtual address space.
2536  * Return a pointer to where it is mapped. This routine is intended to be used
2537  * for mapping device memory, NOT real memory.
2538  */
2539 static void *
2540 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2541 {
2542 	void *res;
2543 	uintptr_t va;
2544 	vm_size_t sz;
2545 
2546 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2547 	res = (void *)va;
2548 
2549 	do {
2550 		sz = 1 << (ilog2(size) & ~1);
2551 		if (bootverbose)
2552 			printf("Wiring VA=%x to PA=%x (size=%x), "
2553 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2554 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2555 		size -= sz;
2556 		pa += sz;
2557 		va += sz;
2558 	} while (size > 0);
2559 
2560 	return (res);
2561 }
2562 
2563 /*
2564  * 'Unmap' a range mapped by mmu_booke_mapdev().
2565  */
2566 static void
2567 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2568 {
2569 	vm_offset_t base, offset;
2570 
2571 	/*
2572 	 * Unmap only if this is inside kernel virtual space.
2573 	 */
2574 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2575 		base = trunc_page(va);
2576 		offset = va & PAGE_MASK;
2577 		size = roundup(offset + size, PAGE_SIZE);
2578 		kmem_free(kernel_map, base, size);
2579 	}
2580 }
2581 
2582 /*
2583  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2584  * specified pmap. This eliminates the blast of soft faults on process startup
2585  * and immediately after an mmap.
2586  */
2587 static void
2588 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2589     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2590 {
2591 
2592 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2593 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2594 	    ("mmu_booke_object_init_pt: non-device object"));
2595 }
2596 
2597 /*
2598  * Perform the pmap work for mincore.
2599  */
2600 static int
2601 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2602 {
2603 
2604 	TODO;
2605 	return (0);
2606 }
2607 
2608 /**************************************************************************/
2609 /* TID handling */
2610 /**************************************************************************/
2611 
2612 /*
2613  * Allocate a TID. If necessary, steal one from someone else.
2614  * The new TID is flushed from the TLB before returning.
2615  */
2616 static tlbtid_t
2617 tid_alloc(pmap_t pmap)
2618 {
2619 	tlbtid_t tid;
2620 	int thiscpu;
2621 
2622 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2623 
2624 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2625 
2626 	thiscpu = PCPU_GET(cpuid);
2627 
2628 	tid = PCPU_GET(tid_next);
2629 	if (tid > TID_MAX)
2630 		tid = TID_MIN;
2631 	PCPU_SET(tid_next, tid + 1);
2632 
2633 	/* If we are stealing TID then clear the relevant pmap's field */
2634 	if (tidbusy[thiscpu][tid] != NULL) {
2635 
2636 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2637 
2638 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2639 
2640 		/* Flush all entries from TLB0 matching this TID. */
2641 		tid_flush(tid);
2642 	}
2643 
2644 	tidbusy[thiscpu][tid] = pmap;
2645 	pmap->pm_tid[thiscpu] = tid;
2646 	__asm __volatile("msync; isync");
2647 
2648 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2649 	    PCPU_GET(tid_next));
2650 
2651 	return (tid);
2652 }
2653 
2654 /**************************************************************************/
2655 /* TLB0 handling */
2656 /**************************************************************************/
2657 
2658 static void
2659 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2660     uint32_t mas7)
2661 {
2662 	int as;
2663 	char desc[3];
2664 	tlbtid_t tid;
2665 	vm_size_t size;
2666 	unsigned int tsize;
2667 
2668 	desc[2] = '\0';
2669 	if (mas1 & MAS1_VALID)
2670 		desc[0] = 'V';
2671 	else
2672 		desc[0] = ' ';
2673 
2674 	if (mas1 & MAS1_IPROT)
2675 		desc[1] = 'P';
2676 	else
2677 		desc[1] = ' ';
2678 
2679 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2680 	tid = MAS1_GETTID(mas1);
2681 
2682 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2683 	size = 0;
2684 	if (tsize)
2685 		size = tsize2size(tsize);
2686 
2687 	debugf("%3d: (%s) [AS=%d] "
2688 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2689 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2690 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2691 }
2692 
2693 /* Convert TLB0 va and way number to tlb0[] table index. */
2694 static inline unsigned int
2695 tlb0_tableidx(vm_offset_t va, unsigned int way)
2696 {
2697 	unsigned int idx;
2698 
2699 	idx = (way * TLB0_ENTRIES_PER_WAY);
2700 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2701 	return (idx);
2702 }
2703 
2704 /*
2705  * Invalidate TLB0 entry.
2706  */
2707 static inline void
2708 tlb0_flush_entry(vm_offset_t va)
2709 {
2710 
2711 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2712 
2713 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2714 
2715 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2716 	__asm __volatile("isync; msync");
2717 	__asm __volatile("tlbsync; msync");
2718 
2719 	CTR1(KTR_PMAP, "%s: e", __func__);
2720 }
2721 
2722 /* Print out contents of the MAS registers for each TLB0 entry */
2723 void
2724 tlb0_print_tlbentries(void)
2725 {
2726 	uint32_t mas0, mas1, mas2, mas3, mas7;
2727 	int entryidx, way, idx;
2728 
2729 	debugf("TLB0 entries:\n");
2730 	for (way = 0; way < TLB0_WAYS; way ++)
2731 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2732 
2733 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2734 			mtspr(SPR_MAS0, mas0);
2735 			__asm __volatile("isync");
2736 
2737 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2738 			mtspr(SPR_MAS2, mas2);
2739 
2740 			__asm __volatile("isync; tlbre");
2741 
2742 			mas1 = mfspr(SPR_MAS1);
2743 			mas2 = mfspr(SPR_MAS2);
2744 			mas3 = mfspr(SPR_MAS3);
2745 			mas7 = mfspr(SPR_MAS7);
2746 
2747 			idx = tlb0_tableidx(mas2, way);
2748 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2749 		}
2750 }
2751 
2752 /**************************************************************************/
2753 /* TLB1 handling */
2754 /**************************************************************************/
2755 
2756 /*
2757  * TLB1 mapping notes:
2758  *
2759  * TLB1[0]	CCSRBAR
2760  * TLB1[1]	Kernel text and data.
2761  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2762  *		windows, other devices mappings.
2763  */
2764 
2765 /*
2766  * Write given entry to TLB1 hardware.
2767  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2768  */
2769 static void
2770 tlb1_write_entry(unsigned int idx)
2771 {
2772 	uint32_t mas0, mas7;
2773 
2774 	//debugf("tlb1_write_entry: s\n");
2775 
2776 	/* Clear high order RPN bits */
2777 	mas7 = 0;
2778 
2779 	/* Select entry */
2780 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2781 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2782 
2783 	mtspr(SPR_MAS0, mas0);
2784 	__asm __volatile("isync");
2785 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2786 	__asm __volatile("isync");
2787 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2788 	__asm __volatile("isync");
2789 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2790 	__asm __volatile("isync");
2791 	mtspr(SPR_MAS7, mas7);
2792 	__asm __volatile("isync; tlbwe; isync; msync");
2793 
2794 	//debugf("tlb1_write_entry: e\n");;
2795 }
2796 
2797 /*
2798  * Return the largest uint value log such that 2^log <= num.
2799  */
2800 static unsigned int
2801 ilog2(unsigned int num)
2802 {
2803 	int lz;
2804 
2805 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2806 	return (31 - lz);
2807 }
2808 
2809 /*
2810  * Convert TLB TSIZE value to mapped region size.
2811  */
2812 static vm_size_t
2813 tsize2size(unsigned int tsize)
2814 {
2815 
2816 	/*
2817 	 * size = 4^tsize KB
2818 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2819 	 */
2820 
2821 	return ((1 << (2 * tsize)) * 1024);
2822 }
2823 
2824 /*
2825  * Convert region size (must be power of 4) to TLB TSIZE value.
2826  */
2827 static unsigned int
2828 size2tsize(vm_size_t size)
2829 {
2830 
2831 	return (ilog2(size) / 2 - 5);
2832 }
2833 
2834 /*
2835  * Register permanent kernel mapping in TLB1.
2836  *
2837  * Entries are created starting from index 0 (current free entry is
2838  * kept in tlb1_idx) and are not supposed to be invalidated.
2839  */
2840 static int
2841 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2842     uint32_t flags)
2843 {
2844 	uint32_t ts, tid;
2845 	int tsize;
2846 
2847 	if (tlb1_idx >= TLB1_ENTRIES) {
2848 		printf("tlb1_set_entry: TLB1 full!\n");
2849 		return (-1);
2850 	}
2851 
2852 	/* Convert size to TSIZE */
2853 	tsize = size2tsize(size);
2854 
2855 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2856 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2857 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2858 
2859 	/* XXX LOCK tlb1[] */
2860 
2861 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2862 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2863 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2864 
2865 	/* Set supervisor RWX permission bits */
2866 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2867 
2868 	tlb1_write_entry(tlb1_idx++);
2869 
2870 	/* XXX UNLOCK tlb1[] */
2871 
2872 	/*
2873 	 * XXX in general TLB1 updates should be propagated between CPUs,
2874 	 * since current design assumes to have the same TLB1 set-up on all
2875 	 * cores.
2876 	 */
2877 	return (0);
2878 }
2879 
2880 static int
2881 tlb1_entry_size_cmp(const void *a, const void *b)
2882 {
2883 	const vm_size_t *sza;
2884 	const vm_size_t *szb;
2885 
2886 	sza = a;
2887 	szb = b;
2888 	if (*sza > *szb)
2889 		return (-1);
2890 	else if (*sza < *szb)
2891 		return (1);
2892 	else
2893 		return (0);
2894 }
2895 
2896 /*
2897  * Map in contiguous RAM region into the TLB1 using maximum of
2898  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2899  *
2900  * If necessary round up last entry size and return total size
2901  * used by all allocated entries.
2902  */
2903 vm_size_t
2904 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2905 {
2906 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2907 	vm_size_t mapped_size, sz, esz;
2908 	unsigned int log;
2909 	int i;
2910 
2911 	CTR4(KTR_PMAP, "%s: region size = 0x%08x va = 0x%08x pa = 0x%08x",
2912 	    __func__, size, va, pa);
2913 
2914 	mapped_size = 0;
2915 	sz = size;
2916 	memset(entry_size, 0, sizeof(entry_size));
2917 
2918 	/* Calculate entry sizes. */
2919 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2920 
2921 		/* Largest region that is power of 4 and fits within size */
2922 		log = ilog2(sz) / 2;
2923 		esz = 1 << (2 * log);
2924 
2925 		/* If this is last entry cover remaining size. */
2926 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2927 			while (esz < sz)
2928 				esz = esz << 2;
2929 		}
2930 
2931 		entry_size[i] = esz;
2932 		mapped_size += esz;
2933 		if (esz < sz)
2934 			sz -= esz;
2935 		else
2936 			sz = 0;
2937 	}
2938 
2939 	/* Sort entry sizes, required to get proper entry address alignment. */
2940 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2941 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2942 
2943 	/* Load TLB1 entries. */
2944 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
2945 		esz = entry_size[i];
2946 		if (!esz)
2947 			break;
2948 
2949 		CTR5(KTR_PMAP, "%s: entry %d: sz  = 0x%08x (va = 0x%08x "
2950 		    "pa = 0x%08x)", __func__, tlb1_idx, esz, va, pa);
2951 
2952 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
2953 
2954 		va += esz;
2955 		pa += esz;
2956 	}
2957 
2958 	CTR3(KTR_PMAP, "%s: mapped size 0x%08x (wasted space 0x%08x)",
2959 	    __func__, mapped_size, mapped_size - size);
2960 
2961 	return (mapped_size);
2962 }
2963 
2964 /*
2965  * TLB1 initialization routine, to be called after the very first
2966  * assembler level setup done in locore.S.
2967  */
2968 void
2969 tlb1_init(vm_offset_t ccsrbar)
2970 {
2971 	uint32_t mas0;
2972 
2973 	/* TLB1[1] is used to map the kernel. Save that entry. */
2974 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(1);
2975 	mtspr(SPR_MAS0, mas0);
2976 	__asm __volatile("isync; tlbre");
2977 
2978 	tlb1[1].mas1 = mfspr(SPR_MAS1);
2979 	tlb1[1].mas2 = mfspr(SPR_MAS2);
2980 	tlb1[1].mas3 = mfspr(SPR_MAS3);
2981 
2982 	/* Map in CCSRBAR in TLB1[0] */
2983 	tlb1_idx = 0;
2984 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
2985 	/*
2986 	 * Set the next available TLB1 entry index. Note TLB[1] is reserved
2987 	 * for initial mapping of kernel text+data, which was set early in
2988 	 * locore, we need to skip this [busy] entry.
2989 	 */
2990 	tlb1_idx = 2;
2991 
2992 	/* Setup TLB miss defaults */
2993 	set_mas4_defaults();
2994 }
2995 
2996 /*
2997  * Setup MAS4 defaults.
2998  * These values are loaded to MAS0-2 on a TLB miss.
2999  */
3000 static void
3001 set_mas4_defaults(void)
3002 {
3003 	uint32_t mas4;
3004 
3005 	/* Defaults: TLB0, PID0, TSIZED=4K */
3006 	mas4 = MAS4_TLBSELD0;
3007 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3008 #ifdef SMP
3009 	mas4 |= MAS4_MD;
3010 #endif
3011 	mtspr(SPR_MAS4, mas4);
3012 	__asm __volatile("isync");
3013 }
3014 
3015 /*
3016  * Print out contents of the MAS registers for each TLB1 entry
3017  */
3018 void
3019 tlb1_print_tlbentries(void)
3020 {
3021 	uint32_t mas0, mas1, mas2, mas3, mas7;
3022 	int i;
3023 
3024 	debugf("TLB1 entries:\n");
3025 	for (i = 0; i < TLB1_ENTRIES; i++) {
3026 
3027 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3028 		mtspr(SPR_MAS0, mas0);
3029 
3030 		__asm __volatile("isync; tlbre");
3031 
3032 		mas1 = mfspr(SPR_MAS1);
3033 		mas2 = mfspr(SPR_MAS2);
3034 		mas3 = mfspr(SPR_MAS3);
3035 		mas7 = mfspr(SPR_MAS7);
3036 
3037 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3038 	}
3039 }
3040 
3041 /*
3042  * Print out contents of the in-ram tlb1 table.
3043  */
3044 void
3045 tlb1_print_entries(void)
3046 {
3047 	int i;
3048 
3049 	debugf("tlb1[] table entries:\n");
3050 	for (i = 0; i < TLB1_ENTRIES; i++)
3051 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3052 }
3053 
3054 /*
3055  * Return 0 if the physical IO range is encompassed by one of the
3056  * the TLB1 entries, otherwise return related error code.
3057  */
3058 static int
3059 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3060 {
3061 	uint32_t prot;
3062 	vm_paddr_t pa_start;
3063 	vm_paddr_t pa_end;
3064 	unsigned int entry_tsize;
3065 	vm_size_t entry_size;
3066 
3067 	*va = (vm_offset_t)NULL;
3068 
3069 	/* Skip invalid entries */
3070 	if (!(tlb1[i].mas1 & MAS1_VALID))
3071 		return (EINVAL);
3072 
3073 	/*
3074 	 * The entry must be cache-inhibited, guarded, and r/w
3075 	 * so it can function as an i/o page
3076 	 */
3077 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3078 	if (prot != (MAS2_I | MAS2_G))
3079 		return (EPERM);
3080 
3081 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3082 	if (prot != (MAS3_SR | MAS3_SW))
3083 		return (EPERM);
3084 
3085 	/* The address should be within the entry range. */
3086 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3087 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3088 
3089 	entry_size = tsize2size(entry_tsize);
3090 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3091 	pa_end = pa_start + entry_size - 1;
3092 
3093 	if ((pa < pa_start) || ((pa + size) > pa_end))
3094 		return (ERANGE);
3095 
3096 	/* Return virtual address of this mapping. */
3097 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3098 	return (0);
3099 }
3100