xref: /freebsd/sys/powerpc/booke/pmap.c (revision 95d45410b5100e07f6f98450bcd841a8945d4726)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/param.h>
55 #include <sys/malloc.h>
56 #include <sys/ktr.h>
57 #include <sys/proc.h>
58 #include <sys/user.h>
59 #include <sys/queue.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/linker.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/rwlock.h>
67 #include <sys/sched.h>
68 #include <sys/smp.h>
69 #include <sys/vmmeter.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_pager.h>
80 #include <vm/uma.h>
81 
82 #include <machine/cpu.h>
83 #include <machine/pcb.h>
84 #include <machine/platform.h>
85 
86 #include <machine/tlb.h>
87 #include <machine/spr.h>
88 #include <machine/md_var.h>
89 #include <machine/mmuvar.h>
90 #include <machine/pmap.h>
91 #include <machine/pte.h>
92 
93 #include "mmu_if.h"
94 
95 #ifdef  DEBUG
96 #define debugf(fmt, args...) printf(fmt, ##args)
97 #else
98 #define debugf(fmt, args...)
99 #endif
100 
101 #define TODO			panic("%s: not implemented", __func__);
102 
103 extern int dumpsys_minidump;
104 
105 extern unsigned char _etext[];
106 extern unsigned char _end[];
107 
108 extern uint32_t *bootinfo;
109 
110 #ifdef SMP
111 extern uint32_t bp_ntlb1s;
112 #endif
113 
114 vm_paddr_t kernload;
115 vm_offset_t kernstart;
116 vm_size_t kernsize;
117 
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
121 
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
127 
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
131 
132 static struct mtx tlbivax_mutex;
133 
134 /*
135  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136  * by idle thred only, no lock required.
137  */
138 static vm_offset_t zero_page_idle_va;
139 
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
144 
145 /**************************************************************************/
146 /* PMAP */
147 /**************************************************************************/
148 
149 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, boolean_t);
151 
152 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
154 
155 /*
156  * If user pmap is processed with mmu_booke_remove and the resident count
157  * drops to 0, there are no more pages to remove, so we need not continue.
158  */
159 #define PMAP_REMOVE_DONE(pmap) \
160 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
161 
162 extern void tid_flush(tlbtid_t);
163 
164 /**************************************************************************/
165 /* TLB and TID handling */
166 /**************************************************************************/
167 
168 /* Translation ID busy table */
169 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
170 
171 /*
172  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
173  * core revisions and should be read from h/w registers during early config.
174  */
175 uint32_t tlb0_entries;
176 uint32_t tlb0_ways;
177 uint32_t tlb0_entries_per_way;
178 
179 #define TLB0_ENTRIES		(tlb0_entries)
180 #define TLB0_WAYS		(tlb0_ways)
181 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
182 
183 #define TLB1_ENTRIES 16
184 
185 /* In-ram copy of the TLB1 */
186 static tlb_entry_t tlb1[TLB1_ENTRIES];
187 
188 /* Next free entry in the TLB1 */
189 static unsigned int tlb1_idx;
190 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
191 
192 static tlbtid_t tid_alloc(struct pmap *);
193 
194 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
195 
196 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
197 static void tlb1_write_entry(unsigned int);
198 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
199 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
200 
201 static vm_size_t tsize2size(unsigned int);
202 static unsigned int size2tsize(vm_size_t);
203 static unsigned int ilog2(unsigned int);
204 
205 static void set_mas4_defaults(void);
206 
207 static inline void tlb0_flush_entry(vm_offset_t);
208 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
209 
210 /**************************************************************************/
211 /* Page table management */
212 /**************************************************************************/
213 
214 static struct rwlock_padalign pvh_global_lock;
215 
216 /* Data for the pv entry allocation mechanism */
217 static uma_zone_t pvzone;
218 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
219 
220 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
221 
222 #ifndef PMAP_SHPGPERPROC
223 #define PMAP_SHPGPERPROC	200
224 #endif
225 
226 static void ptbl_init(void);
227 static struct ptbl_buf *ptbl_buf_alloc(void);
228 static void ptbl_buf_free(struct ptbl_buf *);
229 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
230 
231 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
232 static void ptbl_free(mmu_t, pmap_t, unsigned int);
233 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
234 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
235 
236 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
237 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
238 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
239 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
240 
241 static pv_entry_t pv_alloc(void);
242 static void pv_free(pv_entry_t);
243 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
244 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
245 
246 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
247 #define PTBL_BUFS		(128 * 16)
248 
249 struct ptbl_buf {
250 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
251 	vm_offset_t kva;		/* va of mapping */
252 };
253 
254 /* ptbl free list and a lock used for access synchronization. */
255 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
256 static struct mtx ptbl_buf_freelist_lock;
257 
258 /* Base address of kva space allocated fot ptbl bufs. */
259 static vm_offset_t ptbl_buf_pool_vabase;
260 
261 /* Pointer to ptbl_buf structures. */
262 static struct ptbl_buf *ptbl_bufs;
263 
264 void pmap_bootstrap_ap(volatile uint32_t *);
265 
266 /*
267  * Kernel MMU interface
268  */
269 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
270 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
271 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
272     vm_size_t, vm_offset_t);
273 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
274 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
275     vm_offset_t, vm_page_t *, vm_offset_t, int);
276 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
277     vm_prot_t, boolean_t);
278 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
279     vm_page_t, vm_prot_t);
280 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t);
282 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
283 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
284     vm_prot_t);
285 static void		mmu_booke_init(mmu_t);
286 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
287 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
288 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
289 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
290 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
291     int);
292 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
293     vm_paddr_t *);
294 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
295     vm_object_t, vm_pindex_t, vm_size_t);
296 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
297 static void		mmu_booke_page_init(mmu_t, vm_page_t);
298 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
299 static void		mmu_booke_pinit(mmu_t, pmap_t);
300 static void		mmu_booke_pinit0(mmu_t, pmap_t);
301 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
302     vm_prot_t);
303 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
304 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
305 static void		mmu_booke_release(mmu_t, pmap_t);
306 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
308 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
309 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_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_paddr_t, vm_size_t);
317 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
318 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
319 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
320 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
321 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
322 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
323 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
324 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
325     vm_size_t);
326 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
327     vm_size_t, vm_size_t *);
328 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
329     vm_size_t, vm_offset_t);
330 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
331 
332 static mmu_method_t mmu_booke_methods[] = {
333 	/* pmap dispatcher interface */
334 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
335 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
336 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
337 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
338 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
339 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
340 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
341 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
342 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
343 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
344 	MMUMETHOD(mmu_init,		mmu_booke_init),
345 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
346 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
347 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
348 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
349 	MMUMETHOD(mmu_map,		mmu_booke_map),
350 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
351 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
352 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
353 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
354 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
355 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
356 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
357 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
358 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
359 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
360 	MMUMETHOD(mmu_release,		mmu_booke_release),
361 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
362 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
363 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
364 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
365 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
366 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
367 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
368 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
369 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
370 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
371 
372 	/* Internal interfaces */
373 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
374 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
375 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
376 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
377 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
378 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
379 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
380 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
381 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
382 
383 	/* dumpsys() support */
384 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
385 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
386 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
387 
388 	{ 0, 0 }
389 };
390 
391 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
392 
393 static __inline uint32_t
394 tlb_calc_wimg(vm_offset_t pa, vm_memattr_t ma)
395 {
396 	uint32_t attrib;
397 	int i;
398 
399 	if (ma != VM_MEMATTR_DEFAULT) {
400 		switch (ma) {
401 		case VM_MEMATTR_UNCACHEABLE:
402 			return (PTE_I | PTE_G);
403 		case VM_MEMATTR_WRITE_COMBINING:
404 		case VM_MEMATTR_WRITE_BACK:
405 		case VM_MEMATTR_PREFETCHABLE:
406 			return (PTE_I);
407 		case VM_MEMATTR_WRITE_THROUGH:
408 			return (PTE_W | PTE_M);
409 		}
410 	}
411 
412 	/*
413 	 * Assume the page is cache inhibited and access is guarded unless
414 	 * it's in our available memory array.
415 	 */
416 	attrib = _TLB_ENTRY_IO;
417 	for (i = 0; i < physmem_regions_sz; i++) {
418 		if ((pa >= physmem_regions[i].mr_start) &&
419 		    (pa < (physmem_regions[i].mr_start +
420 		     physmem_regions[i].mr_size))) {
421 			attrib = _TLB_ENTRY_MEM;
422 			break;
423 		}
424 	}
425 
426 	return (attrib);
427 }
428 
429 static inline void
430 tlb_miss_lock(void)
431 {
432 #ifdef SMP
433 	struct pcpu *pc;
434 
435 	if (!smp_started)
436 		return;
437 
438 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
439 		if (pc != pcpup) {
440 
441 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
442 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
443 
444 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
445 			    ("tlb_miss_lock: tried to lock self"));
446 
447 			tlb_lock(pc->pc_booke_tlb_lock);
448 
449 			CTR1(KTR_PMAP, "%s: locked", __func__);
450 		}
451 	}
452 #endif
453 }
454 
455 static inline void
456 tlb_miss_unlock(void)
457 {
458 #ifdef SMP
459 	struct pcpu *pc;
460 
461 	if (!smp_started)
462 		return;
463 
464 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
465 		if (pc != pcpup) {
466 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
467 			    __func__, pc->pc_cpuid);
468 
469 			tlb_unlock(pc->pc_booke_tlb_lock);
470 
471 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
472 		}
473 	}
474 #endif
475 }
476 
477 /* Return number of entries in TLB0. */
478 static __inline void
479 tlb0_get_tlbconf(void)
480 {
481 	uint32_t tlb0_cfg;
482 
483 	tlb0_cfg = mfspr(SPR_TLB0CFG);
484 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
485 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
486 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
487 }
488 
489 /* Initialize pool of kva ptbl buffers. */
490 static void
491 ptbl_init(void)
492 {
493 	int i;
494 
495 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
496 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
497 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
498 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
499 
500 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
501 	TAILQ_INIT(&ptbl_buf_freelist);
502 
503 	for (i = 0; i < PTBL_BUFS; i++) {
504 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
505 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
506 	}
507 }
508 
509 /* Get a ptbl_buf from the freelist. */
510 static struct ptbl_buf *
511 ptbl_buf_alloc(void)
512 {
513 	struct ptbl_buf *buf;
514 
515 	mtx_lock(&ptbl_buf_freelist_lock);
516 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
517 	if (buf != NULL)
518 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
519 	mtx_unlock(&ptbl_buf_freelist_lock);
520 
521 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
522 
523 	return (buf);
524 }
525 
526 /* Return ptbl buff to free pool. */
527 static void
528 ptbl_buf_free(struct ptbl_buf *buf)
529 {
530 
531 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
532 
533 	mtx_lock(&ptbl_buf_freelist_lock);
534 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
535 	mtx_unlock(&ptbl_buf_freelist_lock);
536 }
537 
538 /*
539  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
540  */
541 static void
542 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
543 {
544 	struct ptbl_buf *pbuf;
545 
546 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
547 
548 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
549 
550 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
551 		if (pbuf->kva == (vm_offset_t)ptbl) {
552 			/* Remove from pmap ptbl buf list. */
553 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
554 
555 			/* Free corresponding ptbl buf. */
556 			ptbl_buf_free(pbuf);
557 			break;
558 		}
559 }
560 
561 /* Allocate page table. */
562 static pte_t *
563 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
564 {
565 	vm_page_t mtbl[PTBL_PAGES];
566 	vm_page_t m;
567 	struct ptbl_buf *pbuf;
568 	unsigned int pidx;
569 	pte_t *ptbl;
570 	int i;
571 
572 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
573 	    (pmap == kernel_pmap), pdir_idx);
574 
575 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
576 	    ("ptbl_alloc: invalid pdir_idx"));
577 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
578 	    ("pte_alloc: valid ptbl entry exists!"));
579 
580 	pbuf = ptbl_buf_alloc();
581 	if (pbuf == NULL)
582 		panic("pte_alloc: couldn't alloc kernel virtual memory");
583 
584 	ptbl = (pte_t *)pbuf->kva;
585 
586 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
587 
588 	/* Allocate ptbl pages, this will sleep! */
589 	for (i = 0; i < PTBL_PAGES; i++) {
590 		pidx = (PTBL_PAGES * pdir_idx) + i;
591 		while ((m = vm_page_alloc(NULL, pidx,
592 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
593 
594 			PMAP_UNLOCK(pmap);
595 			rw_wunlock(&pvh_global_lock);
596 			VM_WAIT;
597 			rw_wlock(&pvh_global_lock);
598 			PMAP_LOCK(pmap);
599 		}
600 		mtbl[i] = m;
601 	}
602 
603 	/* Map allocated pages into kernel_pmap. */
604 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
605 
606 	/* Zero whole ptbl. */
607 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
608 
609 	/* Add pbuf to the pmap ptbl bufs list. */
610 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
611 
612 	return (ptbl);
613 }
614 
615 /* Free ptbl pages and invalidate pdir entry. */
616 static void
617 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
618 {
619 	pte_t *ptbl;
620 	vm_paddr_t pa;
621 	vm_offset_t va;
622 	vm_page_t m;
623 	int i;
624 
625 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
626 	    (pmap == kernel_pmap), pdir_idx);
627 
628 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
629 	    ("ptbl_free: invalid pdir_idx"));
630 
631 	ptbl = pmap->pm_pdir[pdir_idx];
632 
633 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
634 
635 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
636 
637 	/*
638 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
639 	 * don't attempt to look up the page tables we are releasing.
640 	 */
641 	mtx_lock_spin(&tlbivax_mutex);
642 	tlb_miss_lock();
643 
644 	pmap->pm_pdir[pdir_idx] = NULL;
645 
646 	tlb_miss_unlock();
647 	mtx_unlock_spin(&tlbivax_mutex);
648 
649 	for (i = 0; i < PTBL_PAGES; i++) {
650 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
651 		pa = pte_vatopa(mmu, kernel_pmap, va);
652 		m = PHYS_TO_VM_PAGE(pa);
653 		vm_page_free_zero(m);
654 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
655 		mmu_booke_kremove(mmu, va);
656 	}
657 
658 	ptbl_free_pmap_ptbl(pmap, ptbl);
659 }
660 
661 /*
662  * Decrement ptbl pages hold count and attempt to free ptbl pages.
663  * Called when removing pte entry from ptbl.
664  *
665  * Return 1 if ptbl pages were freed.
666  */
667 static int
668 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
669 {
670 	pte_t *ptbl;
671 	vm_paddr_t pa;
672 	vm_page_t m;
673 	int i;
674 
675 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
676 	    (pmap == kernel_pmap), pdir_idx);
677 
678 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
679 	    ("ptbl_unhold: invalid pdir_idx"));
680 	KASSERT((pmap != kernel_pmap),
681 	    ("ptbl_unhold: unholding kernel ptbl!"));
682 
683 	ptbl = pmap->pm_pdir[pdir_idx];
684 
685 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
686 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
687 	    ("ptbl_unhold: non kva ptbl"));
688 
689 	/* decrement hold count */
690 	for (i = 0; i < PTBL_PAGES; i++) {
691 		pa = pte_vatopa(mmu, kernel_pmap,
692 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
693 		m = PHYS_TO_VM_PAGE(pa);
694 		m->wire_count--;
695 	}
696 
697 	/*
698 	 * Free ptbl pages if there are no pte etries in this ptbl.
699 	 * wire_count has the same value for all ptbl pages, so check the last
700 	 * page.
701 	 */
702 	if (m->wire_count == 0) {
703 		ptbl_free(mmu, pmap, pdir_idx);
704 
705 		//debugf("ptbl_unhold: e (freed ptbl)\n");
706 		return (1);
707 	}
708 
709 	return (0);
710 }
711 
712 /*
713  * Increment hold count for ptbl pages. This routine is used when a new pte
714  * entry is being inserted into the ptbl.
715  */
716 static void
717 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
718 {
719 	vm_paddr_t pa;
720 	pte_t *ptbl;
721 	vm_page_t m;
722 	int i;
723 
724 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
725 	    pdir_idx);
726 
727 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
728 	    ("ptbl_hold: invalid pdir_idx"));
729 	KASSERT((pmap != kernel_pmap),
730 	    ("ptbl_hold: holding kernel ptbl!"));
731 
732 	ptbl = pmap->pm_pdir[pdir_idx];
733 
734 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
735 
736 	for (i = 0; i < PTBL_PAGES; i++) {
737 		pa = pte_vatopa(mmu, kernel_pmap,
738 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
739 		m = PHYS_TO_VM_PAGE(pa);
740 		m->wire_count++;
741 	}
742 }
743 
744 /* Allocate pv_entry structure. */
745 pv_entry_t
746 pv_alloc(void)
747 {
748 	pv_entry_t pv;
749 
750 	pv_entry_count++;
751 	if (pv_entry_count > pv_entry_high_water)
752 		pagedaemon_wakeup();
753 	pv = uma_zalloc(pvzone, M_NOWAIT);
754 
755 	return (pv);
756 }
757 
758 /* Free pv_entry structure. */
759 static __inline void
760 pv_free(pv_entry_t pve)
761 {
762 
763 	pv_entry_count--;
764 	uma_zfree(pvzone, pve);
765 }
766 
767 
768 /* Allocate and initialize pv_entry structure. */
769 static void
770 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
771 {
772 	pv_entry_t pve;
773 
774 	//int su = (pmap == kernel_pmap);
775 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
776 	//	(u_int32_t)pmap, va, (u_int32_t)m);
777 
778 	pve = pv_alloc();
779 	if (pve == NULL)
780 		panic("pv_insert: no pv entries!");
781 
782 	pve->pv_pmap = pmap;
783 	pve->pv_va = va;
784 
785 	/* add to pv_list */
786 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
787 	rw_assert(&pvh_global_lock, RA_WLOCKED);
788 
789 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
790 
791 	//debugf("pv_insert: e\n");
792 }
793 
794 /* Destroy pv entry. */
795 static void
796 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
797 {
798 	pv_entry_t pve;
799 
800 	//int su = (pmap == kernel_pmap);
801 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
802 
803 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
804 	rw_assert(&pvh_global_lock, RA_WLOCKED);
805 
806 	/* find pv entry */
807 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
808 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
809 			/* remove from pv_list */
810 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
811 			if (TAILQ_EMPTY(&m->md.pv_list))
812 				vm_page_aflag_clear(m, PGA_WRITEABLE);
813 
814 			/* free pv entry struct */
815 			pv_free(pve);
816 			break;
817 		}
818 	}
819 
820 	//debugf("pv_remove: e\n");
821 }
822 
823 /*
824  * Clean pte entry, try to free page table page if requested.
825  *
826  * Return 1 if ptbl pages were freed, otherwise return 0.
827  */
828 static int
829 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
830 {
831 	unsigned int pdir_idx = PDIR_IDX(va);
832 	unsigned int ptbl_idx = PTBL_IDX(va);
833 	vm_page_t m;
834 	pte_t *ptbl;
835 	pte_t *pte;
836 
837 	//int su = (pmap == kernel_pmap);
838 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
839 	//		su, (u_int32_t)pmap, va, flags);
840 
841 	ptbl = pmap->pm_pdir[pdir_idx];
842 	KASSERT(ptbl, ("pte_remove: null ptbl"));
843 
844 	pte = &ptbl[ptbl_idx];
845 
846 	if (pte == NULL || !PTE_ISVALID(pte))
847 		return (0);
848 
849 	if (PTE_ISWIRED(pte))
850 		pmap->pm_stats.wired_count--;
851 
852 	/* Handle managed entry. */
853 	if (PTE_ISMANAGED(pte)) {
854 		/* Get vm_page_t for mapped pte. */
855 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
856 
857 		if (PTE_ISMODIFIED(pte))
858 			vm_page_dirty(m);
859 
860 		if (PTE_ISREFERENCED(pte))
861 			vm_page_aflag_set(m, PGA_REFERENCED);
862 
863 		pv_remove(pmap, va, m);
864 	}
865 
866 	mtx_lock_spin(&tlbivax_mutex);
867 	tlb_miss_lock();
868 
869 	tlb0_flush_entry(va);
870 	pte->flags = 0;
871 	pte->rpn = 0;
872 
873 	tlb_miss_unlock();
874 	mtx_unlock_spin(&tlbivax_mutex);
875 
876 	pmap->pm_stats.resident_count--;
877 
878 	if (flags & PTBL_UNHOLD) {
879 		//debugf("pte_remove: e (unhold)\n");
880 		return (ptbl_unhold(mmu, pmap, pdir_idx));
881 	}
882 
883 	//debugf("pte_remove: e\n");
884 	return (0);
885 }
886 
887 /*
888  * Insert PTE for a given page and virtual address.
889  */
890 static void
891 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
892 {
893 	unsigned int pdir_idx = PDIR_IDX(va);
894 	unsigned int ptbl_idx = PTBL_IDX(va);
895 	pte_t *ptbl, *pte;
896 
897 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
898 	    pmap == kernel_pmap, pmap, va);
899 
900 	/* Get the page table pointer. */
901 	ptbl = pmap->pm_pdir[pdir_idx];
902 
903 	if (ptbl == NULL) {
904 		/* Allocate page table pages. */
905 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
906 	} else {
907 		/*
908 		 * Check if there is valid mapping for requested
909 		 * va, if there is, remove it.
910 		 */
911 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
912 		if (PTE_ISVALID(pte)) {
913 			pte_remove(mmu, pmap, va, PTBL_HOLD);
914 		} else {
915 			/*
916 			 * pte is not used, increment hold count
917 			 * for ptbl pages.
918 			 */
919 			if (pmap != kernel_pmap)
920 				ptbl_hold(mmu, pmap, pdir_idx);
921 		}
922 	}
923 
924 	/*
925 	 * Insert pv_entry into pv_list for mapped page if part of managed
926 	 * memory.
927 	 */
928 	if ((m->oflags & VPO_UNMANAGED) == 0) {
929 		flags |= PTE_MANAGED;
930 
931 		/* Create and insert pv entry. */
932 		pv_insert(pmap, va, m);
933 	}
934 
935 	pmap->pm_stats.resident_count++;
936 
937 	mtx_lock_spin(&tlbivax_mutex);
938 	tlb_miss_lock();
939 
940 	tlb0_flush_entry(va);
941 	if (pmap->pm_pdir[pdir_idx] == NULL) {
942 		/*
943 		 * If we just allocated a new page table, hook it in
944 		 * the pdir.
945 		 */
946 		pmap->pm_pdir[pdir_idx] = ptbl;
947 	}
948 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
949 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
950 	pte->flags |= (PTE_VALID | flags);
951 
952 	tlb_miss_unlock();
953 	mtx_unlock_spin(&tlbivax_mutex);
954 }
955 
956 /* Return the pa for the given pmap/va. */
957 static vm_paddr_t
958 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
959 {
960 	vm_paddr_t pa = 0;
961 	pte_t *pte;
962 
963 	pte = pte_find(mmu, pmap, va);
964 	if ((pte != NULL) && PTE_ISVALID(pte))
965 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
966 	return (pa);
967 }
968 
969 /* Get a pointer to a PTE in a page table. */
970 static pte_t *
971 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
972 {
973 	unsigned int pdir_idx = PDIR_IDX(va);
974 	unsigned int ptbl_idx = PTBL_IDX(va);
975 
976 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
977 
978 	if (pmap->pm_pdir[pdir_idx])
979 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
980 
981 	return (NULL);
982 }
983 
984 /**************************************************************************/
985 /* PMAP related */
986 /**************************************************************************/
987 
988 /*
989  * This is called during booke_init, before the system is really initialized.
990  */
991 static void
992 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
993 {
994 	vm_offset_t phys_kernelend;
995 	struct mem_region *mp, *mp1;
996 	int cnt, i, j;
997 	u_int s, e, sz;
998 	u_int phys_avail_count;
999 	vm_size_t physsz, hwphyssz, kstack0_sz;
1000 	vm_offset_t kernel_pdir, kstack0, va;
1001 	vm_paddr_t kstack0_phys;
1002 	void *dpcpu;
1003 	pte_t *pte;
1004 
1005 	debugf("mmu_booke_bootstrap: entered\n");
1006 
1007 	/* Initialize invalidation mutex */
1008 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1009 
1010 	/* Read TLB0 size and associativity. */
1011 	tlb0_get_tlbconf();
1012 
1013 	/*
1014 	 * Align kernel start and end address (kernel image).
1015 	 * Note that kernel end does not necessarily relate to kernsize.
1016 	 * kernsize is the size of the kernel that is actually mapped.
1017 	 * Also note that "start - 1" is deliberate. With SMP, the
1018 	 * entry point is exactly a page from the actual load address.
1019 	 * As such, trunc_page() has no effect and we're off by a page.
1020 	 * Since we always have the ELF header between the load address
1021 	 * and the entry point, we can safely subtract 1 to compensate.
1022 	 */
1023 	kernstart = trunc_page(start - 1);
1024 	data_start = round_page(kernelend);
1025 	data_end = data_start;
1026 
1027 	/*
1028 	 * Addresses of preloaded modules (like file systems) use
1029 	 * physical addresses. Make sure we relocate those into
1030 	 * virtual addresses.
1031 	 */
1032 	preload_addr_relocate = kernstart - kernload;
1033 
1034 	/* Allocate the dynamic per-cpu area. */
1035 	dpcpu = (void *)data_end;
1036 	data_end += DPCPU_SIZE;
1037 
1038 	/* Allocate space for the message buffer. */
1039 	msgbufp = (struct msgbuf *)data_end;
1040 	data_end += msgbufsize;
1041 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1042 	    data_end);
1043 
1044 	data_end = round_page(data_end);
1045 
1046 	/* Allocate space for ptbl_bufs. */
1047 	ptbl_bufs = (struct ptbl_buf *)data_end;
1048 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1049 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1050 	    data_end);
1051 
1052 	data_end = round_page(data_end);
1053 
1054 	/* Allocate PTE tables for kernel KVA. */
1055 	kernel_pdir = data_end;
1056 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1057 	    PDIR_SIZE - 1) / PDIR_SIZE;
1058 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1059 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1060 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1061 
1062 	debugf(" data_end: 0x%08x\n", data_end);
1063 	if (data_end - kernstart > kernsize) {
1064 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1065 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1066 	}
1067 	data_end = kernstart + kernsize;
1068 	debugf(" updated data_end: 0x%08x\n", data_end);
1069 
1070 	/*
1071 	 * Clear the structures - note we can only do it safely after the
1072 	 * possible additional TLB1 translations are in place (above) so that
1073 	 * all range up to the currently calculated 'data_end' is covered.
1074 	 */
1075 	dpcpu_init(dpcpu, 0);
1076 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1077 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1078 
1079 	/*******************************************************/
1080 	/* Set the start and end of kva. */
1081 	/*******************************************************/
1082 	virtual_avail = round_page(data_end);
1083 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1084 
1085 	/* Allocate KVA space for page zero/copy operations. */
1086 	zero_page_va = virtual_avail;
1087 	virtual_avail += PAGE_SIZE;
1088 	zero_page_idle_va = virtual_avail;
1089 	virtual_avail += PAGE_SIZE;
1090 	copy_page_src_va = virtual_avail;
1091 	virtual_avail += PAGE_SIZE;
1092 	copy_page_dst_va = virtual_avail;
1093 	virtual_avail += PAGE_SIZE;
1094 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1095 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1096 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1097 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1098 
1099 	/* Initialize page zero/copy mutexes. */
1100 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1101 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1102 
1103 	/* Allocate KVA space for ptbl bufs. */
1104 	ptbl_buf_pool_vabase = virtual_avail;
1105 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1106 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1107 	    ptbl_buf_pool_vabase, virtual_avail);
1108 
1109 	/* Calculate corresponding physical addresses for the kernel region. */
1110 	phys_kernelend = kernload + kernsize;
1111 	debugf("kernel image and allocated data:\n");
1112 	debugf(" kernload    = 0x%08x\n", kernload);
1113 	debugf(" kernstart   = 0x%08x\n", kernstart);
1114 	debugf(" kernsize    = 0x%08x\n", kernsize);
1115 
1116 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1117 		panic("mmu_booke_bootstrap: phys_avail too small");
1118 
1119 	/*
1120 	 * Remove kernel physical address range from avail regions list. Page
1121 	 * align all regions.  Non-page aligned memory isn't very interesting
1122 	 * to us.  Also, sort the entries for ascending addresses.
1123 	 */
1124 
1125 	/* Retrieve phys/avail mem regions */
1126 	mem_regions(&physmem_regions, &physmem_regions_sz,
1127 	    &availmem_regions, &availmem_regions_sz);
1128 	sz = 0;
1129 	cnt = availmem_regions_sz;
1130 	debugf("processing avail regions:\n");
1131 	for (mp = availmem_regions; mp->mr_size; mp++) {
1132 		s = mp->mr_start;
1133 		e = mp->mr_start + mp->mr_size;
1134 		debugf(" %08x-%08x -> ", s, e);
1135 		/* Check whether this region holds all of the kernel. */
1136 		if (s < kernload && e > phys_kernelend) {
1137 			availmem_regions[cnt].mr_start = phys_kernelend;
1138 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1139 			e = kernload;
1140 		}
1141 		/* Look whether this regions starts within the kernel. */
1142 		if (s >= kernload && s < phys_kernelend) {
1143 			if (e <= phys_kernelend)
1144 				goto empty;
1145 			s = phys_kernelend;
1146 		}
1147 		/* Now look whether this region ends within the kernel. */
1148 		if (e > kernload && e <= phys_kernelend) {
1149 			if (s >= kernload)
1150 				goto empty;
1151 			e = kernload;
1152 		}
1153 		/* Now page align the start and size of the region. */
1154 		s = round_page(s);
1155 		e = trunc_page(e);
1156 		if (e < s)
1157 			e = s;
1158 		sz = e - s;
1159 		debugf("%08x-%08x = %x\n", s, e, sz);
1160 
1161 		/* Check whether some memory is left here. */
1162 		if (sz == 0) {
1163 		empty:
1164 			memmove(mp, mp + 1,
1165 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1166 			cnt--;
1167 			mp--;
1168 			continue;
1169 		}
1170 
1171 		/* Do an insertion sort. */
1172 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1173 			if (s < mp1->mr_start)
1174 				break;
1175 		if (mp1 < mp) {
1176 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1177 			mp1->mr_start = s;
1178 			mp1->mr_size = sz;
1179 		} else {
1180 			mp->mr_start = s;
1181 			mp->mr_size = sz;
1182 		}
1183 	}
1184 	availmem_regions_sz = cnt;
1185 
1186 	/*******************************************************/
1187 	/* Steal physical memory for kernel stack from the end */
1188 	/* of the first avail region                           */
1189 	/*******************************************************/
1190 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1191 	kstack0_phys = availmem_regions[0].mr_start +
1192 	    availmem_regions[0].mr_size;
1193 	kstack0_phys -= kstack0_sz;
1194 	availmem_regions[0].mr_size -= kstack0_sz;
1195 
1196 	/*******************************************************/
1197 	/* Fill in phys_avail table, based on availmem_regions */
1198 	/*******************************************************/
1199 	phys_avail_count = 0;
1200 	physsz = 0;
1201 	hwphyssz = 0;
1202 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1203 
1204 	debugf("fill in phys_avail:\n");
1205 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1206 
1207 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1208 		    availmem_regions[i].mr_start,
1209 		    availmem_regions[i].mr_start +
1210 		        availmem_regions[i].mr_size,
1211 		    availmem_regions[i].mr_size);
1212 
1213 		if (hwphyssz != 0 &&
1214 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1215 			debugf(" hw.physmem adjust\n");
1216 			if (physsz < hwphyssz) {
1217 				phys_avail[j] = availmem_regions[i].mr_start;
1218 				phys_avail[j + 1] =
1219 				    availmem_regions[i].mr_start +
1220 				    hwphyssz - physsz;
1221 				physsz = hwphyssz;
1222 				phys_avail_count++;
1223 			}
1224 			break;
1225 		}
1226 
1227 		phys_avail[j] = availmem_regions[i].mr_start;
1228 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1229 		    availmem_regions[i].mr_size;
1230 		phys_avail_count++;
1231 		physsz += availmem_regions[i].mr_size;
1232 	}
1233 	physmem = btoc(physsz);
1234 
1235 	/* Calculate the last available physical address. */
1236 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1237 		;
1238 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1239 
1240 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1241 	debugf("phys_avail_count = %d\n", phys_avail_count);
1242 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1243 	    physmem);
1244 
1245 	/*******************************************************/
1246 	/* Initialize (statically allocated) kernel pmap. */
1247 	/*******************************************************/
1248 	PMAP_LOCK_INIT(kernel_pmap);
1249 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1250 
1251 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1252 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1253 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1254 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1255 
1256 	/* Initialize kernel pdir */
1257 	for (i = 0; i < kernel_ptbls; i++)
1258 		kernel_pmap->pm_pdir[kptbl_min + i] =
1259 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1260 
1261 	for (i = 0; i < MAXCPU; i++) {
1262 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1263 
1264 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1265 		tidbusy[i][0] = kernel_pmap;
1266 	}
1267 
1268 	/*
1269 	 * Fill in PTEs covering kernel code and data. They are not required
1270 	 * for address translation, as this area is covered by static TLB1
1271 	 * entries, but for pte_vatopa() to work correctly with kernel area
1272 	 * addresses.
1273 	 */
1274 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1275 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1276 		pte->rpn = kernload + (va - kernstart);
1277 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1278 		    PTE_VALID;
1279 	}
1280 	/* Mark kernel_pmap active on all CPUs */
1281 	CPU_FILL(&kernel_pmap->pm_active);
1282 
1283  	/*
1284 	 * Initialize the global pv list lock.
1285 	 */
1286 	rw_init(&pvh_global_lock, "pmap pv global");
1287 
1288 	/*******************************************************/
1289 	/* Final setup */
1290 	/*******************************************************/
1291 
1292 	/* Enter kstack0 into kernel map, provide guard page */
1293 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1294 	thread0.td_kstack = kstack0;
1295 	thread0.td_kstack_pages = KSTACK_PAGES;
1296 
1297 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1298 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1299 	    kstack0_phys, kstack0_phys + kstack0_sz);
1300 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1301 
1302 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1303 	for (i = 0; i < KSTACK_PAGES; i++) {
1304 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1305 		kstack0 += PAGE_SIZE;
1306 		kstack0_phys += PAGE_SIZE;
1307 	}
1308 
1309 	debugf("virtual_avail = %08x\n", virtual_avail);
1310 	debugf("virtual_end   = %08x\n", virtual_end);
1311 
1312 	debugf("mmu_booke_bootstrap: exit\n");
1313 }
1314 
1315 void
1316 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1317 {
1318 	int i;
1319 
1320 	/*
1321 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1322 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1323 	 * these values directly to (re)program AP's TLB1 hardware.
1324 	 */
1325 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1326 		/* Skip invalid entries */
1327 		if (!(tlb1[i].mas1 & MAS1_VALID))
1328 			continue;
1329 
1330 		tlb1_write_entry(i);
1331 	}
1332 
1333 	set_mas4_defaults();
1334 }
1335 
1336 /*
1337  * Get the physical page address for the given pmap/virtual address.
1338  */
1339 static vm_paddr_t
1340 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1341 {
1342 	vm_paddr_t pa;
1343 
1344 	PMAP_LOCK(pmap);
1345 	pa = pte_vatopa(mmu, pmap, va);
1346 	PMAP_UNLOCK(pmap);
1347 
1348 	return (pa);
1349 }
1350 
1351 /*
1352  * Extract the physical page address associated with the given
1353  * kernel virtual address.
1354  */
1355 static vm_paddr_t
1356 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1357 {
1358 	int i;
1359 
1360 	/* Check TLB1 mappings */
1361 	for (i = 0; i < tlb1_idx; i++) {
1362 		if (!(tlb1[i].mas1 & MAS1_VALID))
1363 			continue;
1364 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1365 			return (tlb1[i].phys + (va - tlb1[i].virt));
1366 	}
1367 
1368 	return (pte_vatopa(mmu, kernel_pmap, va));
1369 }
1370 
1371 /*
1372  * Initialize the pmap module.
1373  * Called by vm_init, to initialize any structures that the pmap
1374  * system needs to map virtual memory.
1375  */
1376 static void
1377 mmu_booke_init(mmu_t mmu)
1378 {
1379 	int shpgperproc = PMAP_SHPGPERPROC;
1380 
1381 	/*
1382 	 * Initialize the address space (zone) for the pv entries.  Set a
1383 	 * high water mark so that the system can recover from excessive
1384 	 * numbers of pv entries.
1385 	 */
1386 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1387 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1388 
1389 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1390 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1391 
1392 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1393 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1394 
1395 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1396 
1397 	/* Pre-fill pvzone with initial number of pv entries. */
1398 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1399 
1400 	/* Initialize ptbl allocation. */
1401 	ptbl_init();
1402 }
1403 
1404 /*
1405  * Map a list of wired pages into kernel virtual address space.  This is
1406  * intended for temporary mappings which do not need page modification or
1407  * references recorded.  Existing mappings in the region are overwritten.
1408  */
1409 static void
1410 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1411 {
1412 	vm_offset_t va;
1413 
1414 	va = sva;
1415 	while (count-- > 0) {
1416 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1417 		va += PAGE_SIZE;
1418 		m++;
1419 	}
1420 }
1421 
1422 /*
1423  * Remove page mappings from kernel virtual address space.  Intended for
1424  * temporary mappings entered by mmu_booke_qenter.
1425  */
1426 static void
1427 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1428 {
1429 	vm_offset_t va;
1430 
1431 	va = sva;
1432 	while (count-- > 0) {
1433 		mmu_booke_kremove(mmu, va);
1434 		va += PAGE_SIZE;
1435 	}
1436 }
1437 
1438 /*
1439  * Map a wired page into kernel virtual address space.
1440  */
1441 static void
1442 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1443 {
1444 
1445 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1446 }
1447 
1448 static void
1449 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1450 {
1451 	unsigned int pdir_idx = PDIR_IDX(va);
1452 	unsigned int ptbl_idx = PTBL_IDX(va);
1453 	uint32_t flags;
1454 	pte_t *pte;
1455 
1456 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1457 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1458 
1459 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1460 	flags |= tlb_calc_wimg(pa, ma);
1461 
1462 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1463 
1464 	mtx_lock_spin(&tlbivax_mutex);
1465 	tlb_miss_lock();
1466 
1467 	if (PTE_ISVALID(pte)) {
1468 
1469 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1470 
1471 		/* Flush entry from TLB0 */
1472 		tlb0_flush_entry(va);
1473 	}
1474 
1475 	pte->rpn = pa & ~PTE_PA_MASK;
1476 	pte->flags = flags;
1477 
1478 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1479 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1480 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1481 
1482 	/* Flush the real memory from the instruction cache. */
1483 	if ((flags & (PTE_I | PTE_G)) == 0) {
1484 		__syncicache((void *)va, PAGE_SIZE);
1485 	}
1486 
1487 	tlb_miss_unlock();
1488 	mtx_unlock_spin(&tlbivax_mutex);
1489 }
1490 
1491 /*
1492  * Remove a page from kernel page table.
1493  */
1494 static void
1495 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1496 {
1497 	unsigned int pdir_idx = PDIR_IDX(va);
1498 	unsigned int ptbl_idx = PTBL_IDX(va);
1499 	pte_t *pte;
1500 
1501 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1502 
1503 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1504 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1505 	    ("mmu_booke_kremove: invalid va"));
1506 
1507 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1508 
1509 	if (!PTE_ISVALID(pte)) {
1510 
1511 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1512 
1513 		return;
1514 	}
1515 
1516 	mtx_lock_spin(&tlbivax_mutex);
1517 	tlb_miss_lock();
1518 
1519 	/* Invalidate entry in TLB0, update PTE. */
1520 	tlb0_flush_entry(va);
1521 	pte->flags = 0;
1522 	pte->rpn = 0;
1523 
1524 	tlb_miss_unlock();
1525 	mtx_unlock_spin(&tlbivax_mutex);
1526 }
1527 
1528 /*
1529  * Initialize pmap associated with process 0.
1530  */
1531 static void
1532 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1533 {
1534 
1535 	PMAP_LOCK_INIT(pmap);
1536 	mmu_booke_pinit(mmu, pmap);
1537 	PCPU_SET(curpmap, pmap);
1538 }
1539 
1540 /*
1541  * Initialize a preallocated and zeroed pmap structure,
1542  * such as one in a vmspace structure.
1543  */
1544 static void
1545 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1546 {
1547 	int i;
1548 
1549 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1550 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1551 
1552 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1553 
1554 	for (i = 0; i < MAXCPU; i++)
1555 		pmap->pm_tid[i] = TID_NONE;
1556 	CPU_ZERO(&kernel_pmap->pm_active);
1557 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1558 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1559 	TAILQ_INIT(&pmap->pm_ptbl_list);
1560 }
1561 
1562 /*
1563  * Release any resources held by the given physical map.
1564  * Called when a pmap initialized by mmu_booke_pinit is being released.
1565  * Should only be called if the map contains no valid mappings.
1566  */
1567 static void
1568 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1569 {
1570 
1571 	KASSERT(pmap->pm_stats.resident_count == 0,
1572 	    ("pmap_release: pmap resident count %ld != 0",
1573 	    pmap->pm_stats.resident_count));
1574 }
1575 
1576 /*
1577  * Insert the given physical page at the specified virtual address in the
1578  * target physical map with the protection requested. If specified the page
1579  * will be wired down.
1580  */
1581 static void
1582 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1583     vm_prot_t prot, boolean_t wired)
1584 {
1585 
1586 	rw_wlock(&pvh_global_lock);
1587 	PMAP_LOCK(pmap);
1588 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1589 	rw_wunlock(&pvh_global_lock);
1590 	PMAP_UNLOCK(pmap);
1591 }
1592 
1593 static void
1594 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1595     vm_prot_t prot, boolean_t wired)
1596 {
1597 	pte_t *pte;
1598 	vm_paddr_t pa;
1599 	uint32_t flags;
1600 	int su, sync;
1601 
1602 	pa = VM_PAGE_TO_PHYS(m);
1603 	su = (pmap == kernel_pmap);
1604 	sync = 0;
1605 
1606 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1607 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1608 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1609 	//		(u_int32_t)m, va, pa, prot, wired);
1610 
1611 	if (su) {
1612 		KASSERT(((va >= virtual_avail) &&
1613 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1614 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1615 	} else {
1616 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1617 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1618 	}
1619 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1620 		VM_OBJECT_ASSERT_LOCKED(m->object);
1621 
1622 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1623 
1624 	/*
1625 	 * If there is an existing mapping, and the physical address has not
1626 	 * changed, must be protection or wiring change.
1627 	 */
1628 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1629 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1630 
1631 		/*
1632 		 * Before actually updating pte->flags we calculate and
1633 		 * prepare its new value in a helper var.
1634 		 */
1635 		flags = pte->flags;
1636 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1637 
1638 		/* Wiring change, just update stats. */
1639 		if (wired) {
1640 			if (!PTE_ISWIRED(pte)) {
1641 				flags |= PTE_WIRED;
1642 				pmap->pm_stats.wired_count++;
1643 			}
1644 		} else {
1645 			if (PTE_ISWIRED(pte)) {
1646 				flags &= ~PTE_WIRED;
1647 				pmap->pm_stats.wired_count--;
1648 			}
1649 		}
1650 
1651 		if (prot & VM_PROT_WRITE) {
1652 			/* Add write permissions. */
1653 			flags |= PTE_SW;
1654 			if (!su)
1655 				flags |= PTE_UW;
1656 
1657 			if ((flags & PTE_MANAGED) != 0)
1658 				vm_page_aflag_set(m, PGA_WRITEABLE);
1659 		} else {
1660 			/* Handle modified pages, sense modify status. */
1661 
1662 			/*
1663 			 * The PTE_MODIFIED flag could be set by underlying
1664 			 * TLB misses since we last read it (above), possibly
1665 			 * other CPUs could update it so we check in the PTE
1666 			 * directly rather than rely on that saved local flags
1667 			 * copy.
1668 			 */
1669 			if (PTE_ISMODIFIED(pte))
1670 				vm_page_dirty(m);
1671 		}
1672 
1673 		if (prot & VM_PROT_EXECUTE) {
1674 			flags |= PTE_SX;
1675 			if (!su)
1676 				flags |= PTE_UX;
1677 
1678 			/*
1679 			 * Check existing flags for execute permissions: if we
1680 			 * are turning execute permissions on, icache should
1681 			 * be flushed.
1682 			 */
1683 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1684 				sync++;
1685 		}
1686 
1687 		flags &= ~PTE_REFERENCED;
1688 
1689 		/*
1690 		 * The new flags value is all calculated -- only now actually
1691 		 * update the PTE.
1692 		 */
1693 		mtx_lock_spin(&tlbivax_mutex);
1694 		tlb_miss_lock();
1695 
1696 		tlb0_flush_entry(va);
1697 		pte->flags = flags;
1698 
1699 		tlb_miss_unlock();
1700 		mtx_unlock_spin(&tlbivax_mutex);
1701 
1702 	} else {
1703 		/*
1704 		 * If there is an existing mapping, but it's for a different
1705 		 * physical address, pte_enter() will delete the old mapping.
1706 		 */
1707 		//if ((pte != NULL) && PTE_ISVALID(pte))
1708 		//	debugf("mmu_booke_enter_locked: replace\n");
1709 		//else
1710 		//	debugf("mmu_booke_enter_locked: new\n");
1711 
1712 		/* Now set up the flags and install the new mapping. */
1713 		flags = (PTE_SR | PTE_VALID);
1714 		flags |= PTE_M;
1715 
1716 		if (!su)
1717 			flags |= PTE_UR;
1718 
1719 		if (prot & VM_PROT_WRITE) {
1720 			flags |= PTE_SW;
1721 			if (!su)
1722 				flags |= PTE_UW;
1723 
1724 			if ((m->oflags & VPO_UNMANAGED) == 0)
1725 				vm_page_aflag_set(m, PGA_WRITEABLE);
1726 		}
1727 
1728 		if (prot & VM_PROT_EXECUTE) {
1729 			flags |= PTE_SX;
1730 			if (!su)
1731 				flags |= PTE_UX;
1732 		}
1733 
1734 		/* If its wired update stats. */
1735 		if (wired) {
1736 			pmap->pm_stats.wired_count++;
1737 			flags |= PTE_WIRED;
1738 		}
1739 
1740 		pte_enter(mmu, pmap, m, va, flags);
1741 
1742 		/* Flush the real memory from the instruction cache. */
1743 		if (prot & VM_PROT_EXECUTE)
1744 			sync++;
1745 	}
1746 
1747 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1748 		__syncicache((void *)va, PAGE_SIZE);
1749 		sync = 0;
1750 	}
1751 }
1752 
1753 /*
1754  * Maps a sequence of resident pages belonging to the same object.
1755  * The sequence begins with the given page m_start.  This page is
1756  * mapped at the given virtual address start.  Each subsequent page is
1757  * mapped at a virtual address that is offset from start by the same
1758  * amount as the page is offset from m_start within the object.  The
1759  * last page in the sequence is the page with the largest offset from
1760  * m_start that can be mapped at a virtual address less than the given
1761  * virtual address end.  Not every virtual page between start and end
1762  * is mapped; only those for which a resident page exists with the
1763  * corresponding offset from m_start are mapped.
1764  */
1765 static void
1766 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1767     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1768 {
1769 	vm_page_t m;
1770 	vm_pindex_t diff, psize;
1771 
1772 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1773 
1774 	psize = atop(end - start);
1775 	m = m_start;
1776 	rw_wlock(&pvh_global_lock);
1777 	PMAP_LOCK(pmap);
1778 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1779 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1780 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1781 		m = TAILQ_NEXT(m, listq);
1782 	}
1783 	rw_wunlock(&pvh_global_lock);
1784 	PMAP_UNLOCK(pmap);
1785 }
1786 
1787 static void
1788 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1789     vm_prot_t prot)
1790 {
1791 
1792 	rw_wlock(&pvh_global_lock);
1793 	PMAP_LOCK(pmap);
1794 	mmu_booke_enter_locked(mmu, pmap, va, m,
1795 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1796 	rw_wunlock(&pvh_global_lock);
1797 	PMAP_UNLOCK(pmap);
1798 }
1799 
1800 /*
1801  * Remove the given range of addresses from the specified map.
1802  *
1803  * It is assumed that the start and end are properly rounded to the page size.
1804  */
1805 static void
1806 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1807 {
1808 	pte_t *pte;
1809 	uint8_t hold_flag;
1810 
1811 	int su = (pmap == kernel_pmap);
1812 
1813 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1814 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1815 
1816 	if (su) {
1817 		KASSERT(((va >= virtual_avail) &&
1818 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1819 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1820 	} else {
1821 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1822 		    ("mmu_booke_remove: user pmap, non user va"));
1823 	}
1824 
1825 	if (PMAP_REMOVE_DONE(pmap)) {
1826 		//debugf("mmu_booke_remove: e (empty)\n");
1827 		return;
1828 	}
1829 
1830 	hold_flag = PTBL_HOLD_FLAG(pmap);
1831 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1832 
1833 	rw_wlock(&pvh_global_lock);
1834 	PMAP_LOCK(pmap);
1835 	for (; va < endva; va += PAGE_SIZE) {
1836 		pte = pte_find(mmu, pmap, va);
1837 		if ((pte != NULL) && PTE_ISVALID(pte))
1838 			pte_remove(mmu, pmap, va, hold_flag);
1839 	}
1840 	PMAP_UNLOCK(pmap);
1841 	rw_wunlock(&pvh_global_lock);
1842 
1843 	//debugf("mmu_booke_remove: e\n");
1844 }
1845 
1846 /*
1847  * Remove physical page from all pmaps in which it resides.
1848  */
1849 static void
1850 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1851 {
1852 	pv_entry_t pv, pvn;
1853 	uint8_t hold_flag;
1854 
1855 	rw_wlock(&pvh_global_lock);
1856 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1857 		pvn = TAILQ_NEXT(pv, pv_link);
1858 
1859 		PMAP_LOCK(pv->pv_pmap);
1860 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1861 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1862 		PMAP_UNLOCK(pv->pv_pmap);
1863 	}
1864 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1865 	rw_wunlock(&pvh_global_lock);
1866 }
1867 
1868 /*
1869  * Map a range of physical addresses into kernel virtual address space.
1870  */
1871 static vm_offset_t
1872 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1873     vm_paddr_t pa_end, int prot)
1874 {
1875 	vm_offset_t sva = *virt;
1876 	vm_offset_t va = sva;
1877 
1878 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1879 	//		sva, pa_start, pa_end);
1880 
1881 	while (pa_start < pa_end) {
1882 		mmu_booke_kenter(mmu, va, pa_start);
1883 		va += PAGE_SIZE;
1884 		pa_start += PAGE_SIZE;
1885 	}
1886 	*virt = va;
1887 
1888 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1889 	return (sva);
1890 }
1891 
1892 /*
1893  * The pmap must be activated before it's address space can be accessed in any
1894  * way.
1895  */
1896 static void
1897 mmu_booke_activate(mmu_t mmu, struct thread *td)
1898 {
1899 	pmap_t pmap;
1900 	u_int cpuid;
1901 
1902 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1903 
1904 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1905 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1906 
1907 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1908 
1909 	sched_pin();
1910 
1911 	cpuid = PCPU_GET(cpuid);
1912 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1913 	PCPU_SET(curpmap, pmap);
1914 
1915 	if (pmap->pm_tid[cpuid] == TID_NONE)
1916 		tid_alloc(pmap);
1917 
1918 	/* Load PID0 register with pmap tid value. */
1919 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1920 	__asm __volatile("isync");
1921 
1922 	sched_unpin();
1923 
1924 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1925 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1926 }
1927 
1928 /*
1929  * Deactivate the specified process's address space.
1930  */
1931 static void
1932 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1933 {
1934 	pmap_t pmap;
1935 
1936 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1937 
1938 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1939 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1940 
1941 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1942 	PCPU_SET(curpmap, NULL);
1943 }
1944 
1945 /*
1946  * Copy the range specified by src_addr/len
1947  * from the source map to the range dst_addr/len
1948  * in the destination map.
1949  *
1950  * This routine is only advisory and need not do anything.
1951  */
1952 static void
1953 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1954     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1955 {
1956 
1957 }
1958 
1959 /*
1960  * Set the physical protection on the specified range of this map as requested.
1961  */
1962 static void
1963 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1964     vm_prot_t prot)
1965 {
1966 	vm_offset_t va;
1967 	vm_page_t m;
1968 	pte_t *pte;
1969 
1970 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1971 		mmu_booke_remove(mmu, pmap, sva, eva);
1972 		return;
1973 	}
1974 
1975 	if (prot & VM_PROT_WRITE)
1976 		return;
1977 
1978 	PMAP_LOCK(pmap);
1979 	for (va = sva; va < eva; va += PAGE_SIZE) {
1980 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1981 			if (PTE_ISVALID(pte)) {
1982 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1983 
1984 				mtx_lock_spin(&tlbivax_mutex);
1985 				tlb_miss_lock();
1986 
1987 				/* Handle modified pages. */
1988 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1989 					vm_page_dirty(m);
1990 
1991 				tlb0_flush_entry(va);
1992 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1993 
1994 				tlb_miss_unlock();
1995 				mtx_unlock_spin(&tlbivax_mutex);
1996 			}
1997 		}
1998 	}
1999 	PMAP_UNLOCK(pmap);
2000 }
2001 
2002 /*
2003  * Clear the write and modified bits in each of the given page's mappings.
2004  */
2005 static void
2006 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2007 {
2008 	pv_entry_t pv;
2009 	pte_t *pte;
2010 
2011 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2012 	    ("mmu_booke_remove_write: page %p is not managed", m));
2013 
2014 	/*
2015 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2016 	 * set by another thread while the object is locked.  Thus,
2017 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2018 	 */
2019 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2020 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2021 		return;
2022 	rw_wlock(&pvh_global_lock);
2023 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2024 		PMAP_LOCK(pv->pv_pmap);
2025 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2026 			if (PTE_ISVALID(pte)) {
2027 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2028 
2029 				mtx_lock_spin(&tlbivax_mutex);
2030 				tlb_miss_lock();
2031 
2032 				/* Handle modified pages. */
2033 				if (PTE_ISMODIFIED(pte))
2034 					vm_page_dirty(m);
2035 
2036 				/* Flush mapping from TLB0. */
2037 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2038 
2039 				tlb_miss_unlock();
2040 				mtx_unlock_spin(&tlbivax_mutex);
2041 			}
2042 		}
2043 		PMAP_UNLOCK(pv->pv_pmap);
2044 	}
2045 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2046 	rw_wunlock(&pvh_global_lock);
2047 }
2048 
2049 static void
2050 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2051 {
2052 	pte_t *pte;
2053 	pmap_t pmap;
2054 	vm_page_t m;
2055 	vm_offset_t addr;
2056 	vm_paddr_t pa = 0;
2057 	int active, valid;
2058 
2059 	va = trunc_page(va);
2060 	sz = round_page(sz);
2061 
2062 	rw_wlock(&pvh_global_lock);
2063 	pmap = PCPU_GET(curpmap);
2064 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2065 	while (sz > 0) {
2066 		PMAP_LOCK(pm);
2067 		pte = pte_find(mmu, pm, va);
2068 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2069 		if (valid)
2070 			pa = PTE_PA(pte);
2071 		PMAP_UNLOCK(pm);
2072 		if (valid) {
2073 			if (!active) {
2074 				/* Create a mapping in the active pmap. */
2075 				addr = 0;
2076 				m = PHYS_TO_VM_PAGE(pa);
2077 				PMAP_LOCK(pmap);
2078 				pte_enter(mmu, pmap, m, addr,
2079 				    PTE_SR | PTE_VALID | PTE_UR);
2080 				__syncicache((void *)addr, PAGE_SIZE);
2081 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2082 				PMAP_UNLOCK(pmap);
2083 			} else
2084 				__syncicache((void *)va, PAGE_SIZE);
2085 		}
2086 		va += PAGE_SIZE;
2087 		sz -= PAGE_SIZE;
2088 	}
2089 	rw_wunlock(&pvh_global_lock);
2090 }
2091 
2092 /*
2093  * Atomically extract and hold the physical page with the given
2094  * pmap and virtual address pair if that mapping permits the given
2095  * protection.
2096  */
2097 static vm_page_t
2098 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2099     vm_prot_t prot)
2100 {
2101 	pte_t *pte;
2102 	vm_page_t m;
2103 	uint32_t pte_wbit;
2104 	vm_paddr_t pa;
2105 
2106 	m = NULL;
2107 	pa = 0;
2108 	PMAP_LOCK(pmap);
2109 retry:
2110 	pte = pte_find(mmu, pmap, va);
2111 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2112 		if (pmap == kernel_pmap)
2113 			pte_wbit = PTE_SW;
2114 		else
2115 			pte_wbit = PTE_UW;
2116 
2117 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2118 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2119 				goto retry;
2120 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2121 			vm_page_hold(m);
2122 		}
2123 	}
2124 
2125 	PA_UNLOCK_COND(pa);
2126 	PMAP_UNLOCK(pmap);
2127 	return (m);
2128 }
2129 
2130 /*
2131  * Initialize a vm_page's machine-dependent fields.
2132  */
2133 static void
2134 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2135 {
2136 
2137 	TAILQ_INIT(&m->md.pv_list);
2138 }
2139 
2140 /*
2141  * mmu_booke_zero_page_area zeros the specified hardware page by
2142  * mapping it into virtual memory and using bzero to clear
2143  * its contents.
2144  *
2145  * off and size must reside within a single page.
2146  */
2147 static void
2148 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2149 {
2150 	vm_offset_t va;
2151 
2152 	/* XXX KASSERT off and size are within a single page? */
2153 
2154 	mtx_lock(&zero_page_mutex);
2155 	va = zero_page_va;
2156 
2157 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2158 	bzero((caddr_t)va + off, size);
2159 	mmu_booke_kremove(mmu, va);
2160 
2161 	mtx_unlock(&zero_page_mutex);
2162 }
2163 
2164 /*
2165  * mmu_booke_zero_page zeros the specified hardware page.
2166  */
2167 static void
2168 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2169 {
2170 
2171 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2172 }
2173 
2174 /*
2175  * mmu_booke_copy_page copies the specified (machine independent) page by
2176  * mapping the page into virtual memory and using memcopy to copy the page,
2177  * one machine dependent page at a time.
2178  */
2179 static void
2180 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2181 {
2182 	vm_offset_t sva, dva;
2183 
2184 	sva = copy_page_src_va;
2185 	dva = copy_page_dst_va;
2186 
2187 	mtx_lock(&copy_page_mutex);
2188 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2189 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2190 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2191 	mmu_booke_kremove(mmu, dva);
2192 	mmu_booke_kremove(mmu, sva);
2193 	mtx_unlock(&copy_page_mutex);
2194 }
2195 
2196 static inline void
2197 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2198     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2199 {
2200 	void *a_cp, *b_cp;
2201 	vm_offset_t a_pg_offset, b_pg_offset;
2202 	int cnt;
2203 
2204 	mtx_lock(&copy_page_mutex);
2205 	while (xfersize > 0) {
2206 		a_pg_offset = a_offset & PAGE_MASK;
2207 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2208 		mmu_booke_kenter(mmu, copy_page_src_va,
2209 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2210 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2211 		b_pg_offset = b_offset & PAGE_MASK;
2212 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2213 		mmu_booke_kenter(mmu, copy_page_dst_va,
2214 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2215 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2216 		bcopy(a_cp, b_cp, cnt);
2217 		mmu_booke_kremove(mmu, copy_page_dst_va);
2218 		mmu_booke_kremove(mmu, copy_page_src_va);
2219 		a_offset += cnt;
2220 		b_offset += cnt;
2221 		xfersize -= cnt;
2222 	}
2223 	mtx_unlock(&copy_page_mutex);
2224 }
2225 
2226 /*
2227  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2228  * into virtual memory and using bzero to clear its contents. This is intended
2229  * to be called from the vm_pagezero process only and outside of Giant. No
2230  * lock is required.
2231  */
2232 static void
2233 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2234 {
2235 	vm_offset_t va;
2236 
2237 	va = zero_page_idle_va;
2238 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2239 	bzero((caddr_t)va, PAGE_SIZE);
2240 	mmu_booke_kremove(mmu, va);
2241 }
2242 
2243 /*
2244  * Return whether or not the specified physical page was modified
2245  * in any of physical maps.
2246  */
2247 static boolean_t
2248 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2249 {
2250 	pte_t *pte;
2251 	pv_entry_t pv;
2252 	boolean_t rv;
2253 
2254 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2255 	    ("mmu_booke_is_modified: page %p is not managed", m));
2256 	rv = FALSE;
2257 
2258 	/*
2259 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2260 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2261 	 * is clear, no PTEs can be modified.
2262 	 */
2263 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2264 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2265 		return (rv);
2266 	rw_wlock(&pvh_global_lock);
2267 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2268 		PMAP_LOCK(pv->pv_pmap);
2269 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2270 		    PTE_ISVALID(pte)) {
2271 			if (PTE_ISMODIFIED(pte))
2272 				rv = TRUE;
2273 		}
2274 		PMAP_UNLOCK(pv->pv_pmap);
2275 		if (rv)
2276 			break;
2277 	}
2278 	rw_wunlock(&pvh_global_lock);
2279 	return (rv);
2280 }
2281 
2282 /*
2283  * Return whether or not the specified virtual address is eligible
2284  * for prefault.
2285  */
2286 static boolean_t
2287 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2288 {
2289 
2290 	return (FALSE);
2291 }
2292 
2293 /*
2294  * Return whether or not the specified physical page was referenced
2295  * in any physical maps.
2296  */
2297 static boolean_t
2298 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2299 {
2300 	pte_t *pte;
2301 	pv_entry_t pv;
2302 	boolean_t rv;
2303 
2304 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2305 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2306 	rv = FALSE;
2307 	rw_wlock(&pvh_global_lock);
2308 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2309 		PMAP_LOCK(pv->pv_pmap);
2310 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2311 		    PTE_ISVALID(pte)) {
2312 			if (PTE_ISREFERENCED(pte))
2313 				rv = TRUE;
2314 		}
2315 		PMAP_UNLOCK(pv->pv_pmap);
2316 		if (rv)
2317 			break;
2318 	}
2319 	rw_wunlock(&pvh_global_lock);
2320 	return (rv);
2321 }
2322 
2323 /*
2324  * Clear the modify bits on the specified physical page.
2325  */
2326 static void
2327 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2328 {
2329 	pte_t *pte;
2330 	pv_entry_t pv;
2331 
2332 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2333 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2334 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2335 	KASSERT(!vm_page_xbusied(m),
2336 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2337 
2338 	/*
2339 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2340 	 * If the object containing the page is locked and the page is not
2341 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2342 	 */
2343 	if ((m->aflags & PGA_WRITEABLE) == 0)
2344 		return;
2345 	rw_wlock(&pvh_global_lock);
2346 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2347 		PMAP_LOCK(pv->pv_pmap);
2348 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2349 		    PTE_ISVALID(pte)) {
2350 			mtx_lock_spin(&tlbivax_mutex);
2351 			tlb_miss_lock();
2352 
2353 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2354 				tlb0_flush_entry(pv->pv_va);
2355 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2356 				    PTE_REFERENCED);
2357 			}
2358 
2359 			tlb_miss_unlock();
2360 			mtx_unlock_spin(&tlbivax_mutex);
2361 		}
2362 		PMAP_UNLOCK(pv->pv_pmap);
2363 	}
2364 	rw_wunlock(&pvh_global_lock);
2365 }
2366 
2367 /*
2368  * Return a count of reference bits for a page, clearing those bits.
2369  * It is not necessary for every reference bit to be cleared, but it
2370  * is necessary that 0 only be returned when there are truly no
2371  * reference bits set.
2372  *
2373  * XXX: The exact number of bits to check and clear is a matter that
2374  * should be tested and standardized at some point in the future for
2375  * optimal aging of shared pages.
2376  */
2377 static int
2378 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2379 {
2380 	pte_t *pte;
2381 	pv_entry_t pv;
2382 	int count;
2383 
2384 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2385 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2386 	count = 0;
2387 	rw_wlock(&pvh_global_lock);
2388 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2389 		PMAP_LOCK(pv->pv_pmap);
2390 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2391 		    PTE_ISVALID(pte)) {
2392 			if (PTE_ISREFERENCED(pte)) {
2393 				mtx_lock_spin(&tlbivax_mutex);
2394 				tlb_miss_lock();
2395 
2396 				tlb0_flush_entry(pv->pv_va);
2397 				pte->flags &= ~PTE_REFERENCED;
2398 
2399 				tlb_miss_unlock();
2400 				mtx_unlock_spin(&tlbivax_mutex);
2401 
2402 				if (++count > 4) {
2403 					PMAP_UNLOCK(pv->pv_pmap);
2404 					break;
2405 				}
2406 			}
2407 		}
2408 		PMAP_UNLOCK(pv->pv_pmap);
2409 	}
2410 	rw_wunlock(&pvh_global_lock);
2411 	return (count);
2412 }
2413 
2414 /*
2415  * Change wiring attribute for a map/virtual-address pair.
2416  */
2417 static void
2418 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2419 {
2420 	pte_t *pte;
2421 
2422 	PMAP_LOCK(pmap);
2423 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2424 		if (wired) {
2425 			if (!PTE_ISWIRED(pte)) {
2426 				pte->flags |= PTE_WIRED;
2427 				pmap->pm_stats.wired_count++;
2428 			}
2429 		} else {
2430 			if (PTE_ISWIRED(pte)) {
2431 				pte->flags &= ~PTE_WIRED;
2432 				pmap->pm_stats.wired_count--;
2433 			}
2434 		}
2435 	}
2436 	PMAP_UNLOCK(pmap);
2437 }
2438 
2439 /*
2440  * Clear the wired attribute from the mappings for the specified range of
2441  * addresses in the given pmap.  Every valid mapping within that range must
2442  * have the wired attribute set.  In contrast, invalid mappings cannot have
2443  * the wired attribute set, so they are ignored.
2444  *
2445  * The wired attribute of the page table entry is not a hardware feature, so
2446  * there is no need to invalidate any TLB entries.
2447  */
2448 static void
2449 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2450 {
2451 	vm_offset_t va;
2452 	pte_t *pte;
2453 
2454 	PMAP_LOCK(pmap);
2455 	for (va = sva; va < eva; va += PAGE_SIZE) {
2456 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2457 		    PTE_ISVALID(pte)) {
2458 			if (!PTE_ISWIRED(pte))
2459 				panic("mmu_booke_unwire: pte %p isn't wired",
2460 				    pte);
2461 			pte->flags &= ~PTE_WIRED;
2462 			pmap->pm_stats.wired_count--;
2463 		}
2464 	}
2465 	PMAP_UNLOCK(pmap);
2466 
2467 }
2468 
2469 /*
2470  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2471  * page.  This count may be changed upwards or downwards in the future; it is
2472  * only necessary that true be returned for a small subset of pmaps for proper
2473  * page aging.
2474  */
2475 static boolean_t
2476 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2477 {
2478 	pv_entry_t pv;
2479 	int loops;
2480 	boolean_t rv;
2481 
2482 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2483 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2484 	loops = 0;
2485 	rv = FALSE;
2486 	rw_wlock(&pvh_global_lock);
2487 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2488 		if (pv->pv_pmap == pmap) {
2489 			rv = TRUE;
2490 			break;
2491 		}
2492 		if (++loops >= 16)
2493 			break;
2494 	}
2495 	rw_wunlock(&pvh_global_lock);
2496 	return (rv);
2497 }
2498 
2499 /*
2500  * Return the number of managed mappings to the given physical page that are
2501  * wired.
2502  */
2503 static int
2504 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2505 {
2506 	pv_entry_t pv;
2507 	pte_t *pte;
2508 	int count = 0;
2509 
2510 	if ((m->oflags & VPO_UNMANAGED) != 0)
2511 		return (count);
2512 	rw_wlock(&pvh_global_lock);
2513 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2514 		PMAP_LOCK(pv->pv_pmap);
2515 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2516 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2517 				count++;
2518 		PMAP_UNLOCK(pv->pv_pmap);
2519 	}
2520 	rw_wunlock(&pvh_global_lock);
2521 	return (count);
2522 }
2523 
2524 static int
2525 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2526 {
2527 	int i;
2528 	vm_offset_t va;
2529 
2530 	/*
2531 	 * This currently does not work for entries that
2532 	 * overlap TLB1 entries.
2533 	 */
2534 	for (i = 0; i < tlb1_idx; i ++) {
2535 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2536 			return (0);
2537 	}
2538 
2539 	return (EFAULT);
2540 }
2541 
2542 vm_offset_t
2543 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2544     vm_size_t *sz)
2545 {
2546 	vm_paddr_t pa, ppa;
2547 	vm_offset_t va;
2548 	vm_size_t gran;
2549 
2550 	/* Raw physical memory dumps don't have a virtual address. */
2551 	if (md->md_vaddr == ~0UL) {
2552 		/* We always map a 256MB page at 256M. */
2553 		gran = 256 * 1024 * 1024;
2554 		pa = md->md_paddr + ofs;
2555 		ppa = pa & ~(gran - 1);
2556 		ofs = pa - ppa;
2557 		va = gran;
2558 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2559 		if (*sz > (gran - ofs))
2560 			*sz = gran - ofs;
2561 		return (va + ofs);
2562 	}
2563 
2564 	/* Minidumps are based on virtual memory addresses. */
2565 	va = md->md_vaddr + ofs;
2566 	if (va >= kernstart + kernsize) {
2567 		gran = PAGE_SIZE - (va & PAGE_MASK);
2568 		if (*sz > gran)
2569 			*sz = gran;
2570 	}
2571 	return (va);
2572 }
2573 
2574 void
2575 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2576     vm_offset_t va)
2577 {
2578 
2579 	/* Raw physical memory dumps don't have a virtual address. */
2580 	if (md->md_vaddr == ~0UL) {
2581 		tlb1_idx--;
2582 		tlb1[tlb1_idx].mas1 = 0;
2583 		tlb1[tlb1_idx].mas2 = 0;
2584 		tlb1[tlb1_idx].mas3 = 0;
2585 		tlb1_write_entry(tlb1_idx);
2586 		return;
2587 	}
2588 
2589 	/* Minidumps are based on virtual memory addresses. */
2590 	/* Nothing to do... */
2591 }
2592 
2593 struct pmap_md *
2594 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2595 {
2596 	static struct pmap_md md;
2597 	pte_t *pte;
2598 	vm_offset_t va;
2599 
2600 	if (dumpsys_minidump) {
2601 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2602 		if (prev == NULL) {
2603 			/* 1st: kernel .data and .bss. */
2604 			md.md_index = 1;
2605 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2606 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2607 			return (&md);
2608 		}
2609 		switch (prev->md_index) {
2610 		case 1:
2611 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2612 			md.md_index = 2;
2613 			md.md_vaddr = data_start;
2614 			md.md_size = data_end - data_start;
2615 			break;
2616 		case 2:
2617 			/* 3rd: kernel VM. */
2618 			va = prev->md_vaddr + prev->md_size;
2619 			/* Find start of next chunk (from va). */
2620 			while (va < virtual_end) {
2621 				/* Don't dump the buffer cache. */
2622 				if (va >= kmi.buffer_sva &&
2623 				    va < kmi.buffer_eva) {
2624 					va = kmi.buffer_eva;
2625 					continue;
2626 				}
2627 				pte = pte_find(mmu, kernel_pmap, va);
2628 				if (pte != NULL && PTE_ISVALID(pte))
2629 					break;
2630 				va += PAGE_SIZE;
2631 			}
2632 			if (va < virtual_end) {
2633 				md.md_vaddr = va;
2634 				va += PAGE_SIZE;
2635 				/* Find last page in chunk. */
2636 				while (va < virtual_end) {
2637 					/* Don't run into the buffer cache. */
2638 					if (va == kmi.buffer_sva)
2639 						break;
2640 					pte = pte_find(mmu, kernel_pmap, va);
2641 					if (pte == NULL || !PTE_ISVALID(pte))
2642 						break;
2643 					va += PAGE_SIZE;
2644 				}
2645 				md.md_size = va - md.md_vaddr;
2646 				break;
2647 			}
2648 			md.md_index = 3;
2649 			/* FALLTHROUGH */
2650 		default:
2651 			return (NULL);
2652 		}
2653 	} else { /* minidumps */
2654 		mem_regions(&physmem_regions, &physmem_regions_sz,
2655 		    &availmem_regions, &availmem_regions_sz);
2656 
2657 		if (prev == NULL) {
2658 			/* first physical chunk. */
2659 			md.md_paddr = physmem_regions[0].mr_start;
2660 			md.md_size = physmem_regions[0].mr_size;
2661 			md.md_vaddr = ~0UL;
2662 			md.md_index = 1;
2663 		} else if (md.md_index < physmem_regions_sz) {
2664 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2665 			md.md_size = physmem_regions[md.md_index].mr_size;
2666 			md.md_vaddr = ~0UL;
2667 			md.md_index++;
2668 		} else {
2669 			/* There's no next physical chunk. */
2670 			return (NULL);
2671 		}
2672 	}
2673 
2674 	return (&md);
2675 }
2676 
2677 /*
2678  * Map a set of physical memory pages into the kernel virtual address space.
2679  * Return a pointer to where it is mapped. This routine is intended to be used
2680  * for mapping device memory, NOT real memory.
2681  */
2682 static void *
2683 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2684 {
2685 
2686 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2687 }
2688 
2689 static void *
2690 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2691 {
2692 	void *res;
2693 	uintptr_t va;
2694 	vm_size_t sz;
2695 	int i;
2696 
2697 	/*
2698 	 * Check if this is premapped in TLB1. Note: this should probably also
2699 	 * check whether a sequence of TLB1 entries exist that match the
2700 	 * requirement, but now only checks the easy case.
2701 	 */
2702 	if (ma == VM_MEMATTR_DEFAULT) {
2703 		for (i = 0; i < tlb1_idx; i++) {
2704 			if (!(tlb1[i].mas1 & MAS1_VALID))
2705 				continue;
2706 			if (pa >= tlb1[i].phys &&
2707 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2708 				return (void *)(tlb1[i].virt +
2709 				    (pa - tlb1[i].phys));
2710 		}
2711 	}
2712 
2713 	size = roundup(size, PAGE_SIZE);
2714 
2715 	/*
2716 	 * We leave a hole for device direct mapping between the maximum user
2717 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2718 	 * devices are in there, just map them 1:1. If not, map them to the
2719 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2720 	 * addresses should be pulled from an allocator, but since we do not
2721 	 * ever free TLB1 entries, it is safe just to increment a counter.
2722 	 * Note that there isn't a lot of address space here (128 MB) and it
2723 	 * is not at all difficult to imagine running out, since that is a 4:1
2724 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2725 	 * mapped there.
2726 	 */
2727 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2728 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2729 		va = pa;
2730 	else
2731 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2732 	res = (void *)va;
2733 
2734 	do {
2735 		sz = 1 << (ilog2(size) & ~1);
2736 		if (bootverbose)
2737 			printf("Wiring VA=%x to PA=%x (size=%x), "
2738 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2739 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2740 		size -= sz;
2741 		pa += sz;
2742 		va += sz;
2743 	} while (size > 0);
2744 
2745 	return (res);
2746 }
2747 
2748 /*
2749  * 'Unmap' a range mapped by mmu_booke_mapdev().
2750  */
2751 static void
2752 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2753 {
2754 #ifdef SUPPORTS_SHRINKING_TLB1
2755 	vm_offset_t base, offset;
2756 
2757 	/*
2758 	 * Unmap only if this is inside kernel virtual space.
2759 	 */
2760 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2761 		base = trunc_page(va);
2762 		offset = va & PAGE_MASK;
2763 		size = roundup(offset + size, PAGE_SIZE);
2764 		kva_free(base, size);
2765 	}
2766 #endif
2767 }
2768 
2769 /*
2770  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2771  * specified pmap. This eliminates the blast of soft faults on process startup
2772  * and immediately after an mmap.
2773  */
2774 static void
2775 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2776     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2777 {
2778 
2779 	VM_OBJECT_ASSERT_WLOCKED(object);
2780 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2781 	    ("mmu_booke_object_init_pt: non-device object"));
2782 }
2783 
2784 /*
2785  * Perform the pmap work for mincore.
2786  */
2787 static int
2788 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2789     vm_paddr_t *locked_pa)
2790 {
2791 
2792 	/* XXX: this should be implemented at some point */
2793 	return (0);
2794 }
2795 
2796 /**************************************************************************/
2797 /* TID handling */
2798 /**************************************************************************/
2799 
2800 /*
2801  * Allocate a TID. If necessary, steal one from someone else.
2802  * The new TID is flushed from the TLB before returning.
2803  */
2804 static tlbtid_t
2805 tid_alloc(pmap_t pmap)
2806 {
2807 	tlbtid_t tid;
2808 	int thiscpu;
2809 
2810 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2811 
2812 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2813 
2814 	thiscpu = PCPU_GET(cpuid);
2815 
2816 	tid = PCPU_GET(tid_next);
2817 	if (tid > TID_MAX)
2818 		tid = TID_MIN;
2819 	PCPU_SET(tid_next, tid + 1);
2820 
2821 	/* If we are stealing TID then clear the relevant pmap's field */
2822 	if (tidbusy[thiscpu][tid] != NULL) {
2823 
2824 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2825 
2826 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2827 
2828 		/* Flush all entries from TLB0 matching this TID. */
2829 		tid_flush(tid);
2830 	}
2831 
2832 	tidbusy[thiscpu][tid] = pmap;
2833 	pmap->pm_tid[thiscpu] = tid;
2834 	__asm __volatile("msync; isync");
2835 
2836 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2837 	    PCPU_GET(tid_next));
2838 
2839 	return (tid);
2840 }
2841 
2842 /**************************************************************************/
2843 /* TLB0 handling */
2844 /**************************************************************************/
2845 
2846 static void
2847 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2848     uint32_t mas7)
2849 {
2850 	int as;
2851 	char desc[3];
2852 	tlbtid_t tid;
2853 	vm_size_t size;
2854 	unsigned int tsize;
2855 
2856 	desc[2] = '\0';
2857 	if (mas1 & MAS1_VALID)
2858 		desc[0] = 'V';
2859 	else
2860 		desc[0] = ' ';
2861 
2862 	if (mas1 & MAS1_IPROT)
2863 		desc[1] = 'P';
2864 	else
2865 		desc[1] = ' ';
2866 
2867 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2868 	tid = MAS1_GETTID(mas1);
2869 
2870 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2871 	size = 0;
2872 	if (tsize)
2873 		size = tsize2size(tsize);
2874 
2875 	debugf("%3d: (%s) [AS=%d] "
2876 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2877 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2878 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2879 }
2880 
2881 /* Convert TLB0 va and way number to tlb0[] table index. */
2882 static inline unsigned int
2883 tlb0_tableidx(vm_offset_t va, unsigned int way)
2884 {
2885 	unsigned int idx;
2886 
2887 	idx = (way * TLB0_ENTRIES_PER_WAY);
2888 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2889 	return (idx);
2890 }
2891 
2892 /*
2893  * Invalidate TLB0 entry.
2894  */
2895 static inline void
2896 tlb0_flush_entry(vm_offset_t va)
2897 {
2898 
2899 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2900 
2901 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2902 
2903 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2904 	__asm __volatile("isync; msync");
2905 	__asm __volatile("tlbsync; msync");
2906 
2907 	CTR1(KTR_PMAP, "%s: e", __func__);
2908 }
2909 
2910 /* Print out contents of the MAS registers for each TLB0 entry */
2911 void
2912 tlb0_print_tlbentries(void)
2913 {
2914 	uint32_t mas0, mas1, mas2, mas3, mas7;
2915 	int entryidx, way, idx;
2916 
2917 	debugf("TLB0 entries:\n");
2918 	for (way = 0; way < TLB0_WAYS; way ++)
2919 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2920 
2921 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2922 			mtspr(SPR_MAS0, mas0);
2923 			__asm __volatile("isync");
2924 
2925 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2926 			mtspr(SPR_MAS2, mas2);
2927 
2928 			__asm __volatile("isync; tlbre");
2929 
2930 			mas1 = mfspr(SPR_MAS1);
2931 			mas2 = mfspr(SPR_MAS2);
2932 			mas3 = mfspr(SPR_MAS3);
2933 			mas7 = mfspr(SPR_MAS7);
2934 
2935 			idx = tlb0_tableidx(mas2, way);
2936 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2937 		}
2938 }
2939 
2940 /**************************************************************************/
2941 /* TLB1 handling */
2942 /**************************************************************************/
2943 
2944 /*
2945  * TLB1 mapping notes:
2946  *
2947  * TLB1[0]	Kernel text and data.
2948  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2949  *		windows, other devices mappings.
2950  */
2951 
2952 /*
2953  * Write given entry to TLB1 hardware.
2954  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2955  */
2956 static void
2957 tlb1_write_entry(unsigned int idx)
2958 {
2959 	uint32_t mas0, mas7;
2960 
2961 	//debugf("tlb1_write_entry: s\n");
2962 
2963 	/* Clear high order RPN bits */
2964 	mas7 = 0;
2965 
2966 	/* Select entry */
2967 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2968 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2969 
2970 	mtspr(SPR_MAS0, mas0);
2971 	__asm __volatile("isync");
2972 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2973 	__asm __volatile("isync");
2974 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2975 	__asm __volatile("isync");
2976 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2977 	__asm __volatile("isync");
2978 	mtspr(SPR_MAS7, mas7);
2979 	__asm __volatile("isync; tlbwe; isync; msync");
2980 
2981 	//debugf("tlb1_write_entry: e\n");
2982 }
2983 
2984 /*
2985  * Return the largest uint value log such that 2^log <= num.
2986  */
2987 static unsigned int
2988 ilog2(unsigned int num)
2989 {
2990 	int lz;
2991 
2992 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2993 	return (31 - lz);
2994 }
2995 
2996 /*
2997  * Convert TLB TSIZE value to mapped region size.
2998  */
2999 static vm_size_t
3000 tsize2size(unsigned int tsize)
3001 {
3002 
3003 	/*
3004 	 * size = 4^tsize KB
3005 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3006 	 */
3007 
3008 	return ((1 << (2 * tsize)) * 1024);
3009 }
3010 
3011 /*
3012  * Convert region size (must be power of 4) to TLB TSIZE value.
3013  */
3014 static unsigned int
3015 size2tsize(vm_size_t size)
3016 {
3017 
3018 	return (ilog2(size) / 2 - 5);
3019 }
3020 
3021 /*
3022  * Register permanent kernel mapping in TLB1.
3023  *
3024  * Entries are created starting from index 0 (current free entry is
3025  * kept in tlb1_idx) and are not supposed to be invalidated.
3026  */
3027 static int
3028 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
3029     uint32_t flags)
3030 {
3031 	uint32_t ts, tid;
3032 	int tsize, index;
3033 
3034 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3035 	if (index >= TLB1_ENTRIES) {
3036 		printf("tlb1_set_entry: TLB1 full!\n");
3037 		return (-1);
3038 	}
3039 
3040 	/* Convert size to TSIZE */
3041 	tsize = size2tsize(size);
3042 
3043 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3044 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3045 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3046 
3047 	/*
3048 	 * Atomicity is preserved by the atomic increment above since nothing
3049 	 * is ever removed from tlb1.
3050 	 */
3051 
3052 	tlb1[index].phys = pa;
3053 	tlb1[index].virt = va;
3054 	tlb1[index].size = size;
3055 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3056 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3057 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3058 
3059 	/* Set supervisor RWX permission bits */
3060 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3061 
3062 	tlb1_write_entry(index);
3063 
3064 	/*
3065 	 * XXX in general TLB1 updates should be propagated between CPUs,
3066 	 * since current design assumes to have the same TLB1 set-up on all
3067 	 * cores.
3068 	 */
3069 	return (0);
3070 }
3071 
3072 /*
3073  * Map in contiguous RAM region into the TLB1 using maximum of
3074  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3075  *
3076  * If necessary round up last entry size and return total size
3077  * used by all allocated entries.
3078  */
3079 vm_size_t
3080 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3081 {
3082 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3083 	vm_size_t mapped, pgsz, base, mask;
3084 	int idx, nents;
3085 
3086 	/* Round up to the next 1M */
3087 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3088 
3089 	mapped = 0;
3090 	idx = 0;
3091 	base = va;
3092 	pgsz = 64*1024*1024;
3093 	while (mapped < size) {
3094 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3095 			while (pgsz > (size - mapped))
3096 				pgsz >>= 2;
3097 			pgs[idx++] = pgsz;
3098 			mapped += pgsz;
3099 		}
3100 
3101 		/* We under-map. Correct for this. */
3102 		if (mapped < size) {
3103 			while (pgs[idx - 1] == pgsz) {
3104 				idx--;
3105 				mapped -= pgsz;
3106 			}
3107 			/* XXX We may increase beyond out starting point. */
3108 			pgsz <<= 2;
3109 			pgs[idx++] = pgsz;
3110 			mapped += pgsz;
3111 		}
3112 	}
3113 
3114 	nents = idx;
3115 	mask = pgs[0] - 1;
3116 	/* Align address to the boundary */
3117 	if (va & mask) {
3118 		va = (va + mask) & ~mask;
3119 		pa = (pa + mask) & ~mask;
3120 	}
3121 
3122 	for (idx = 0; idx < nents; idx++) {
3123 		pgsz = pgs[idx];
3124 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3125 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3126 		pa += pgsz;
3127 		va += pgsz;
3128 	}
3129 
3130 	mapped = (va - base);
3131 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3132 	    mapped, mapped - size);
3133 	return (mapped);
3134 }
3135 
3136 /*
3137  * TLB1 initialization routine, to be called after the very first
3138  * assembler level setup done in locore.S.
3139  */
3140 void
3141 tlb1_init()
3142 {
3143 	uint32_t mas0, mas1, mas2, mas3;
3144 	uint32_t tsz;
3145 	u_int i;
3146 
3147 	if (bootinfo != NULL && bootinfo[0] != 1) {
3148 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3149 	} else
3150 		tlb1_idx = 1;
3151 
3152 	/* The first entry/entries are used to map the kernel. */
3153 	for (i = 0; i < tlb1_idx; i++) {
3154 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3155 		mtspr(SPR_MAS0, mas0);
3156 		__asm __volatile("isync; tlbre");
3157 
3158 		mas1 = mfspr(SPR_MAS1);
3159 		if ((mas1 & MAS1_VALID) == 0)
3160 			continue;
3161 
3162 		mas2 = mfspr(SPR_MAS2);
3163 		mas3 = mfspr(SPR_MAS3);
3164 
3165 		tlb1[i].mas1 = mas1;
3166 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3167 		tlb1[i].mas3 = mas3;
3168 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3169 		tlb1[i].phys = mas3 & MAS3_RPN;
3170 
3171 		if (i == 0)
3172 			kernload = mas3 & MAS3_RPN;
3173 
3174 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3175 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3176 		kernsize += tlb1[i].size;
3177 	}
3178 
3179 #ifdef SMP
3180 	bp_ntlb1s = tlb1_idx;
3181 #endif
3182 
3183 	/* Purge the remaining entries */
3184 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3185 		tlb1_write_entry(i);
3186 
3187 	/* Setup TLB miss defaults */
3188 	set_mas4_defaults();
3189 }
3190 
3191 vm_offset_t
3192 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3193 {
3194 	vm_paddr_t pa_base;
3195 	vm_offset_t va, sz;
3196 	int i;
3197 
3198 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3199 
3200 	for (i = 0; i < tlb1_idx; i++) {
3201 		if (!(tlb1[i].mas1 & MAS1_VALID))
3202 			continue;
3203 		if (pa >= tlb1[i].phys && (pa + size) <=
3204 		    (tlb1[i].phys + tlb1[i].size))
3205 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3206 	}
3207 
3208 	pa_base = trunc_page(pa);
3209 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3210 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3211 	va = tlb1_map_base + (pa - pa_base);
3212 
3213 	do {
3214 		sz = 1 << (ilog2(size) & ~1);
3215 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3216 		size -= sz;
3217 		pa_base += sz;
3218 		tlb1_map_base += sz;
3219 	} while (size > 0);
3220 
3221 #ifdef SMP
3222 	bp_ntlb1s = tlb1_idx;
3223 #endif
3224 
3225 	return (va);
3226 }
3227 
3228 /*
3229  * Setup MAS4 defaults.
3230  * These values are loaded to MAS0-2 on a TLB miss.
3231  */
3232 static void
3233 set_mas4_defaults(void)
3234 {
3235 	uint32_t mas4;
3236 
3237 	/* Defaults: TLB0, PID0, TSIZED=4K */
3238 	mas4 = MAS4_TLBSELD0;
3239 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3240 #ifdef SMP
3241 	mas4 |= MAS4_MD;
3242 #endif
3243 	mtspr(SPR_MAS4, mas4);
3244 	__asm __volatile("isync");
3245 }
3246 
3247 /*
3248  * Print out contents of the MAS registers for each TLB1 entry
3249  */
3250 void
3251 tlb1_print_tlbentries(void)
3252 {
3253 	uint32_t mas0, mas1, mas2, mas3, mas7;
3254 	int i;
3255 
3256 	debugf("TLB1 entries:\n");
3257 	for (i = 0; i < TLB1_ENTRIES; i++) {
3258 
3259 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3260 		mtspr(SPR_MAS0, mas0);
3261 
3262 		__asm __volatile("isync; tlbre");
3263 
3264 		mas1 = mfspr(SPR_MAS1);
3265 		mas2 = mfspr(SPR_MAS2);
3266 		mas3 = mfspr(SPR_MAS3);
3267 		mas7 = mfspr(SPR_MAS7);
3268 
3269 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3270 	}
3271 }
3272 
3273 /*
3274  * Print out contents of the in-ram tlb1 table.
3275  */
3276 void
3277 tlb1_print_entries(void)
3278 {
3279 	int i;
3280 
3281 	debugf("tlb1[] table entries:\n");
3282 	for (i = 0; i < TLB1_ENTRIES; i++)
3283 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3284 }
3285 
3286 /*
3287  * Return 0 if the physical IO range is encompassed by one of the
3288  * the TLB1 entries, otherwise return related error code.
3289  */
3290 static int
3291 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3292 {
3293 	uint32_t prot;
3294 	vm_paddr_t pa_start;
3295 	vm_paddr_t pa_end;
3296 	unsigned int entry_tsize;
3297 	vm_size_t entry_size;
3298 
3299 	*va = (vm_offset_t)NULL;
3300 
3301 	/* Skip invalid entries */
3302 	if (!(tlb1[i].mas1 & MAS1_VALID))
3303 		return (EINVAL);
3304 
3305 	/*
3306 	 * The entry must be cache-inhibited, guarded, and r/w
3307 	 * so it can function as an i/o page
3308 	 */
3309 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3310 	if (prot != (MAS2_I | MAS2_G))
3311 		return (EPERM);
3312 
3313 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3314 	if (prot != (MAS3_SR | MAS3_SW))
3315 		return (EPERM);
3316 
3317 	/* The address should be within the entry range. */
3318 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3319 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3320 
3321 	entry_size = tsize2size(entry_tsize);
3322 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3323 	pa_end = pa_start + entry_size - 1;
3324 
3325 	if ((pa < pa_start) || ((pa + size) > pa_end))
3326 		return (ERANGE);
3327 
3328 	/* Return virtual address of this mapping. */
3329 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3330 	return (0);
3331 }
3332