xref: /freebsd/sys/powerpc/booke/pmap.c (revision 93e779a26c651610ac6e7986d67ecc9ed2cadcbf)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include "opt_kstack_pages.h"
55 
56 #include <sys/param.h>
57 #include <sys/conf.h>
58 #include <sys/malloc.h>
59 #include <sys/ktr.h>
60 #include <sys/proc.h>
61 #include <sys/user.h>
62 #include <sys/queue.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/kerneldump.h>
66 #include <sys/linker.h>
67 #include <sys/msgbuf.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/vmmeter.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_pager.h>
84 #include <vm/uma.h>
85 
86 #include <machine/cpu.h>
87 #include <machine/pcb.h>
88 #include <machine/platform.h>
89 
90 #include <machine/tlb.h>
91 #include <machine/spr.h>
92 #include <machine/md_var.h>
93 #include <machine/mmuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/pte.h>
96 
97 #include "mmu_if.h"
98 
99 #ifdef  DEBUG
100 #define debugf(fmt, args...) printf(fmt, ##args)
101 #else
102 #define debugf(fmt, args...)
103 #endif
104 
105 #define TODO			panic("%s: not implemented", __func__);
106 
107 extern unsigned char _etext[];
108 extern unsigned char _end[];
109 
110 extern uint32_t *bootinfo;
111 
112 #ifdef SMP
113 extern uint32_t bp_ntlb1s;
114 #endif
115 
116 vm_paddr_t kernload;
117 vm_offset_t kernstart;
118 vm_size_t kernsize;
119 
120 /* Message buffer and tables. */
121 static vm_offset_t data_start;
122 static vm_size_t data_end;
123 
124 /* Phys/avail memory regions. */
125 static struct mem_region *availmem_regions;
126 static int availmem_regions_sz;
127 static struct mem_region *physmem_regions;
128 static int physmem_regions_sz;
129 
130 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
131 static vm_offset_t zero_page_va;
132 static struct mtx zero_page_mutex;
133 
134 static struct mtx tlbivax_mutex;
135 
136 /*
137  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
138  * by idle thred only, no lock required.
139  */
140 static vm_offset_t zero_page_idle_va;
141 
142 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
143 static vm_offset_t copy_page_src_va;
144 static vm_offset_t copy_page_dst_va;
145 static struct mtx copy_page_mutex;
146 
147 /**************************************************************************/
148 /* PMAP */
149 /**************************************************************************/
150 
151 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
152     vm_prot_t, u_int flags, int8_t psind);
153 
154 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
155 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
156 
157 /*
158  * If user pmap is processed with mmu_booke_remove and the resident count
159  * drops to 0, there are no more pages to remove, so we need not continue.
160  */
161 #define PMAP_REMOVE_DONE(pmap) \
162 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
163 
164 extern void tid_flush(tlbtid_t tid, int tlb0_ways, int tlb0_entries_per_way);
165 extern int elf32_nxstack;
166 
167 /**************************************************************************/
168 /* TLB and TID handling */
169 /**************************************************************************/
170 
171 /* Translation ID busy table */
172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 
174 /*
175  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
176  * core revisions and should be read from h/w registers during early config.
177  */
178 uint32_t tlb0_entries;
179 uint32_t tlb0_ways;
180 uint32_t tlb0_entries_per_way;
181 uint32_t tlb1_entries;
182 
183 #define TLB0_ENTRIES		(tlb0_entries)
184 #define TLB0_WAYS		(tlb0_ways)
185 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
186 
187 #define TLB1_ENTRIES (tlb1_entries)
188 #define TLB1_MAXENTRIES	64
189 
190 /* In-ram copy of the TLB1 */
191 static tlb_entry_t tlb1[TLB1_MAXENTRIES];
192 
193 /* Next free entry in the TLB1 */
194 static unsigned int tlb1_idx;
195 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
196 
197 static tlbtid_t tid_alloc(struct pmap *);
198 
199 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
200 
201 static int tlb1_set_entry(vm_offset_t, vm_paddr_t, vm_size_t, uint32_t);
202 static void tlb1_write_entry(unsigned int);
203 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
204 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
205 
206 static vm_size_t tsize2size(unsigned int);
207 static unsigned int size2tsize(vm_size_t);
208 static unsigned int ilog2(unsigned int);
209 
210 static void set_mas4_defaults(void);
211 
212 static inline void tlb0_flush_entry(vm_offset_t);
213 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
214 
215 /**************************************************************************/
216 /* Page table management */
217 /**************************************************************************/
218 
219 static struct rwlock_padalign pvh_global_lock;
220 
221 /* Data for the pv entry allocation mechanism */
222 static uma_zone_t pvzone;
223 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
224 
225 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
226 
227 #ifndef PMAP_SHPGPERPROC
228 #define PMAP_SHPGPERPROC	200
229 #endif
230 
231 static void ptbl_init(void);
232 static struct ptbl_buf *ptbl_buf_alloc(void);
233 static void ptbl_buf_free(struct ptbl_buf *);
234 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
235 
236 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
237 static void ptbl_free(mmu_t, pmap_t, unsigned int);
238 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
239 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
240 
241 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
242 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
243 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
244 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
245 
246 static pv_entry_t pv_alloc(void);
247 static void pv_free(pv_entry_t);
248 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
249 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
250 
251 static void booke_pmap_init_qpages(void);
252 
253 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
254 #define PTBL_BUFS		(128 * 16)
255 
256 struct ptbl_buf {
257 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
258 	vm_offset_t kva;		/* va of mapping */
259 };
260 
261 /* ptbl free list and a lock used for access synchronization. */
262 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
263 static struct mtx ptbl_buf_freelist_lock;
264 
265 /* Base address of kva space allocated fot ptbl bufs. */
266 static vm_offset_t ptbl_buf_pool_vabase;
267 
268 /* Pointer to ptbl_buf structures. */
269 static struct ptbl_buf *ptbl_bufs;
270 
271 #ifdef SMP
272 void pmap_bootstrap_ap(volatile uint32_t *);
273 #endif
274 
275 /*
276  * Kernel MMU interface
277  */
278 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
279 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
280     vm_size_t, vm_offset_t);
281 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
282 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
283     vm_offset_t, vm_page_t *, vm_offset_t, int);
284 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
285     vm_prot_t, u_int flags, int8_t psind);
286 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
287     vm_page_t, vm_prot_t);
288 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
289     vm_prot_t);
290 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
291 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
292     vm_prot_t);
293 static void		mmu_booke_init(mmu_t);
294 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
295 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
296 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
297 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
298 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
299     int);
300 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
301     vm_paddr_t *);
302 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
303     vm_object_t, vm_pindex_t, vm_size_t);
304 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
305 static void		mmu_booke_page_init(mmu_t, vm_page_t);
306 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
307 static void		mmu_booke_pinit(mmu_t, pmap_t);
308 static void		mmu_booke_pinit0(mmu_t, pmap_t);
309 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
310     vm_prot_t);
311 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
312 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
313 static void		mmu_booke_release(mmu_t, pmap_t);
314 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
315 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
316 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
317 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
318 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
319 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
320 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
321 static void		mmu_booke_activate(mmu_t, struct thread *);
322 static void		mmu_booke_deactivate(mmu_t, struct thread *);
323 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
324 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
325 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
326 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
327 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
328 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
329 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
330 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
331 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
332 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
333     vm_size_t);
334 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
335     void **);
336 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
337     void *);
338 static void		mmu_booke_scan_init(mmu_t);
339 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
340 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
341 
342 static mmu_method_t mmu_booke_methods[] = {
343 	/* pmap dispatcher interface */
344 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
345 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
346 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
347 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
348 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
349 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
350 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
351 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
352 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
353 	MMUMETHOD(mmu_init,		mmu_booke_init),
354 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
355 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
356 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
357 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
358 	MMUMETHOD(mmu_map,		mmu_booke_map),
359 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
360 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
361 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
362 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
363 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
364 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
365 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
366 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
367 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
368 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
369 	MMUMETHOD(mmu_release,		mmu_booke_release),
370 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
371 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
372 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
373 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
374 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
375 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
376 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
377 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
378 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
379 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
380 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
381 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
382 
383 	/* Internal interfaces */
384 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
385 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
386 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
387 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
388 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
389 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
390 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
391 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
392 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
393 
394 	/* dumpsys() support */
395 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
396 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
397 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
398 
399 	{ 0, 0 }
400 };
401 
402 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
403 
404 static __inline uint32_t
405 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
406 {
407 	uint32_t attrib;
408 	int i;
409 
410 	if (ma != VM_MEMATTR_DEFAULT) {
411 		switch (ma) {
412 		case VM_MEMATTR_UNCACHEABLE:
413 			return (PTE_I | PTE_G);
414 		case VM_MEMATTR_WRITE_COMBINING:
415 		case VM_MEMATTR_WRITE_BACK:
416 		case VM_MEMATTR_PREFETCHABLE:
417 			return (PTE_I);
418 		case VM_MEMATTR_WRITE_THROUGH:
419 			return (PTE_W | PTE_M);
420 		}
421 	}
422 
423 	/*
424 	 * Assume the page is cache inhibited and access is guarded unless
425 	 * it's in our available memory array.
426 	 */
427 	attrib = _TLB_ENTRY_IO;
428 	for (i = 0; i < physmem_regions_sz; i++) {
429 		if ((pa >= physmem_regions[i].mr_start) &&
430 		    (pa < (physmem_regions[i].mr_start +
431 		     physmem_regions[i].mr_size))) {
432 			attrib = _TLB_ENTRY_MEM;
433 			break;
434 		}
435 	}
436 
437 	return (attrib);
438 }
439 
440 static inline void
441 tlb_miss_lock(void)
442 {
443 #ifdef SMP
444 	struct pcpu *pc;
445 
446 	if (!smp_started)
447 		return;
448 
449 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
450 		if (pc != pcpup) {
451 
452 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
453 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
454 
455 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
456 			    ("tlb_miss_lock: tried to lock self"));
457 
458 			tlb_lock(pc->pc_booke_tlb_lock);
459 
460 			CTR1(KTR_PMAP, "%s: locked", __func__);
461 		}
462 	}
463 #endif
464 }
465 
466 static inline void
467 tlb_miss_unlock(void)
468 {
469 #ifdef SMP
470 	struct pcpu *pc;
471 
472 	if (!smp_started)
473 		return;
474 
475 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
476 		if (pc != pcpup) {
477 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
478 			    __func__, pc->pc_cpuid);
479 
480 			tlb_unlock(pc->pc_booke_tlb_lock);
481 
482 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
483 		}
484 	}
485 #endif
486 }
487 
488 /* Return number of entries in TLB0. */
489 static __inline void
490 tlb0_get_tlbconf(void)
491 {
492 	uint32_t tlb0_cfg;
493 
494 	tlb0_cfg = mfspr(SPR_TLB0CFG);
495 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
496 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
497 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
498 }
499 
500 /* Return number of entries in TLB1. */
501 static __inline void
502 tlb1_get_tlbconf(void)
503 {
504 	uint32_t tlb1_cfg;
505 
506 	tlb1_cfg = mfspr(SPR_TLB1CFG);
507 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
508 }
509 
510 /* Initialize pool of kva ptbl buffers. */
511 static void
512 ptbl_init(void)
513 {
514 	int i;
515 
516 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
517 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
518 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
519 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
520 
521 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
522 	TAILQ_INIT(&ptbl_buf_freelist);
523 
524 	for (i = 0; i < PTBL_BUFS; i++) {
525 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
526 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
527 	}
528 }
529 
530 /* Get a ptbl_buf from the freelist. */
531 static struct ptbl_buf *
532 ptbl_buf_alloc(void)
533 {
534 	struct ptbl_buf *buf;
535 
536 	mtx_lock(&ptbl_buf_freelist_lock);
537 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
538 	if (buf != NULL)
539 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
540 	mtx_unlock(&ptbl_buf_freelist_lock);
541 
542 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
543 
544 	return (buf);
545 }
546 
547 /* Return ptbl buff to free pool. */
548 static void
549 ptbl_buf_free(struct ptbl_buf *buf)
550 {
551 
552 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
553 
554 	mtx_lock(&ptbl_buf_freelist_lock);
555 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
556 	mtx_unlock(&ptbl_buf_freelist_lock);
557 }
558 
559 /*
560  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
561  */
562 static void
563 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
564 {
565 	struct ptbl_buf *pbuf;
566 
567 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
568 
569 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
570 
571 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
572 		if (pbuf->kva == (vm_offset_t)ptbl) {
573 			/* Remove from pmap ptbl buf list. */
574 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
575 
576 			/* Free corresponding ptbl buf. */
577 			ptbl_buf_free(pbuf);
578 			break;
579 		}
580 }
581 
582 /* Allocate page table. */
583 static pte_t *
584 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
585 {
586 	vm_page_t mtbl[PTBL_PAGES];
587 	vm_page_t m;
588 	struct ptbl_buf *pbuf;
589 	unsigned int pidx;
590 	pte_t *ptbl;
591 	int i, j;
592 
593 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
594 	    (pmap == kernel_pmap), pdir_idx);
595 
596 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
597 	    ("ptbl_alloc: invalid pdir_idx"));
598 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
599 	    ("pte_alloc: valid ptbl entry exists!"));
600 
601 	pbuf = ptbl_buf_alloc();
602 	if (pbuf == NULL)
603 		panic("pte_alloc: couldn't alloc kernel virtual memory");
604 
605 	ptbl = (pte_t *)pbuf->kva;
606 
607 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
608 
609 	/* Allocate ptbl pages, this will sleep! */
610 	for (i = 0; i < PTBL_PAGES; i++) {
611 		pidx = (PTBL_PAGES * pdir_idx) + i;
612 		while ((m = vm_page_alloc(NULL, pidx,
613 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
614 			PMAP_UNLOCK(pmap);
615 			rw_wunlock(&pvh_global_lock);
616 			if (nosleep) {
617 				ptbl_free_pmap_ptbl(pmap, ptbl);
618 				for (j = 0; j < i; j++)
619 					vm_page_free(mtbl[j]);
620 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
621 				return (NULL);
622 			}
623 			VM_WAIT;
624 			rw_wlock(&pvh_global_lock);
625 			PMAP_LOCK(pmap);
626 		}
627 		mtbl[i] = m;
628 	}
629 
630 	/* Map allocated pages into kernel_pmap. */
631 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
632 
633 	/* Zero whole ptbl. */
634 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
635 
636 	/* Add pbuf to the pmap ptbl bufs list. */
637 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
638 
639 	return (ptbl);
640 }
641 
642 /* Free ptbl pages and invalidate pdir entry. */
643 static void
644 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
645 {
646 	pte_t *ptbl;
647 	vm_paddr_t pa;
648 	vm_offset_t va;
649 	vm_page_t m;
650 	int i;
651 
652 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
653 	    (pmap == kernel_pmap), pdir_idx);
654 
655 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
656 	    ("ptbl_free: invalid pdir_idx"));
657 
658 	ptbl = pmap->pm_pdir[pdir_idx];
659 
660 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
661 
662 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
663 
664 	/*
665 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
666 	 * don't attempt to look up the page tables we are releasing.
667 	 */
668 	mtx_lock_spin(&tlbivax_mutex);
669 	tlb_miss_lock();
670 
671 	pmap->pm_pdir[pdir_idx] = NULL;
672 
673 	tlb_miss_unlock();
674 	mtx_unlock_spin(&tlbivax_mutex);
675 
676 	for (i = 0; i < PTBL_PAGES; i++) {
677 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
678 		pa = pte_vatopa(mmu, kernel_pmap, va);
679 		m = PHYS_TO_VM_PAGE(pa);
680 		vm_page_free_zero(m);
681 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
682 		mmu_booke_kremove(mmu, va);
683 	}
684 
685 	ptbl_free_pmap_ptbl(pmap, ptbl);
686 }
687 
688 /*
689  * Decrement ptbl pages hold count and attempt to free ptbl pages.
690  * Called when removing pte entry from ptbl.
691  *
692  * Return 1 if ptbl pages were freed.
693  */
694 static int
695 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
696 {
697 	pte_t *ptbl;
698 	vm_paddr_t pa;
699 	vm_page_t m;
700 	int i;
701 
702 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
703 	    (pmap == kernel_pmap), pdir_idx);
704 
705 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
706 	    ("ptbl_unhold: invalid pdir_idx"));
707 	KASSERT((pmap != kernel_pmap),
708 	    ("ptbl_unhold: unholding kernel ptbl!"));
709 
710 	ptbl = pmap->pm_pdir[pdir_idx];
711 
712 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
713 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
714 	    ("ptbl_unhold: non kva ptbl"));
715 
716 	/* decrement hold count */
717 	for (i = 0; i < PTBL_PAGES; i++) {
718 		pa = pte_vatopa(mmu, kernel_pmap,
719 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
720 		m = PHYS_TO_VM_PAGE(pa);
721 		m->wire_count--;
722 	}
723 
724 	/*
725 	 * Free ptbl pages if there are no pte etries in this ptbl.
726 	 * wire_count has the same value for all ptbl pages, so check the last
727 	 * page.
728 	 */
729 	if (m->wire_count == 0) {
730 		ptbl_free(mmu, pmap, pdir_idx);
731 
732 		//debugf("ptbl_unhold: e (freed ptbl)\n");
733 		return (1);
734 	}
735 
736 	return (0);
737 }
738 
739 /*
740  * Increment hold count for ptbl pages. This routine is used when a new pte
741  * entry is being inserted into the ptbl.
742  */
743 static void
744 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
745 {
746 	vm_paddr_t pa;
747 	pte_t *ptbl;
748 	vm_page_t m;
749 	int i;
750 
751 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
752 	    pdir_idx);
753 
754 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
755 	    ("ptbl_hold: invalid pdir_idx"));
756 	KASSERT((pmap != kernel_pmap),
757 	    ("ptbl_hold: holding kernel ptbl!"));
758 
759 	ptbl = pmap->pm_pdir[pdir_idx];
760 
761 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
762 
763 	for (i = 0; i < PTBL_PAGES; i++) {
764 		pa = pte_vatopa(mmu, kernel_pmap,
765 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
766 		m = PHYS_TO_VM_PAGE(pa);
767 		m->wire_count++;
768 	}
769 }
770 
771 /* Allocate pv_entry structure. */
772 pv_entry_t
773 pv_alloc(void)
774 {
775 	pv_entry_t pv;
776 
777 	pv_entry_count++;
778 	if (pv_entry_count > pv_entry_high_water)
779 		pagedaemon_wakeup();
780 	pv = uma_zalloc(pvzone, M_NOWAIT);
781 
782 	return (pv);
783 }
784 
785 /* Free pv_entry structure. */
786 static __inline void
787 pv_free(pv_entry_t pve)
788 {
789 
790 	pv_entry_count--;
791 	uma_zfree(pvzone, pve);
792 }
793 
794 
795 /* Allocate and initialize pv_entry structure. */
796 static void
797 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
798 {
799 	pv_entry_t pve;
800 
801 	//int su = (pmap == kernel_pmap);
802 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
803 	//	(u_int32_t)pmap, va, (u_int32_t)m);
804 
805 	pve = pv_alloc();
806 	if (pve == NULL)
807 		panic("pv_insert: no pv entries!");
808 
809 	pve->pv_pmap = pmap;
810 	pve->pv_va = va;
811 
812 	/* add to pv_list */
813 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
814 	rw_assert(&pvh_global_lock, RA_WLOCKED);
815 
816 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
817 
818 	//debugf("pv_insert: e\n");
819 }
820 
821 /* Destroy pv entry. */
822 static void
823 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
824 {
825 	pv_entry_t pve;
826 
827 	//int su = (pmap == kernel_pmap);
828 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
829 
830 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
831 	rw_assert(&pvh_global_lock, RA_WLOCKED);
832 
833 	/* find pv entry */
834 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
835 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
836 			/* remove from pv_list */
837 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
838 			if (TAILQ_EMPTY(&m->md.pv_list))
839 				vm_page_aflag_clear(m, PGA_WRITEABLE);
840 
841 			/* free pv entry struct */
842 			pv_free(pve);
843 			break;
844 		}
845 	}
846 
847 	//debugf("pv_remove: e\n");
848 }
849 
850 /*
851  * Clean pte entry, try to free page table page if requested.
852  *
853  * Return 1 if ptbl pages were freed, otherwise return 0.
854  */
855 static int
856 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
857 {
858 	unsigned int pdir_idx = PDIR_IDX(va);
859 	unsigned int ptbl_idx = PTBL_IDX(va);
860 	vm_page_t m;
861 	pte_t *ptbl;
862 	pte_t *pte;
863 
864 	//int su = (pmap == kernel_pmap);
865 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
866 	//		su, (u_int32_t)pmap, va, flags);
867 
868 	ptbl = pmap->pm_pdir[pdir_idx];
869 	KASSERT(ptbl, ("pte_remove: null ptbl"));
870 
871 	pte = &ptbl[ptbl_idx];
872 
873 	if (pte == NULL || !PTE_ISVALID(pte))
874 		return (0);
875 
876 	if (PTE_ISWIRED(pte))
877 		pmap->pm_stats.wired_count--;
878 
879 	/* Handle managed entry. */
880 	if (PTE_ISMANAGED(pte)) {
881 		/* Get vm_page_t for mapped pte. */
882 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
883 
884 		if (PTE_ISMODIFIED(pte))
885 			vm_page_dirty(m);
886 
887 		if (PTE_ISREFERENCED(pte))
888 			vm_page_aflag_set(m, PGA_REFERENCED);
889 
890 		pv_remove(pmap, va, m);
891 	}
892 
893 	mtx_lock_spin(&tlbivax_mutex);
894 	tlb_miss_lock();
895 
896 	tlb0_flush_entry(va);
897 	pte->flags = 0;
898 	pte->rpn = 0;
899 
900 	tlb_miss_unlock();
901 	mtx_unlock_spin(&tlbivax_mutex);
902 
903 	pmap->pm_stats.resident_count--;
904 
905 	if (flags & PTBL_UNHOLD) {
906 		//debugf("pte_remove: e (unhold)\n");
907 		return (ptbl_unhold(mmu, pmap, pdir_idx));
908 	}
909 
910 	//debugf("pte_remove: e\n");
911 	return (0);
912 }
913 
914 /*
915  * Insert PTE for a given page and virtual address.
916  */
917 static int
918 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
919     boolean_t nosleep)
920 {
921 	unsigned int pdir_idx = PDIR_IDX(va);
922 	unsigned int ptbl_idx = PTBL_IDX(va);
923 	pte_t *ptbl, *pte;
924 
925 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
926 	    pmap == kernel_pmap, pmap, va);
927 
928 	/* Get the page table pointer. */
929 	ptbl = pmap->pm_pdir[pdir_idx];
930 
931 	if (ptbl == NULL) {
932 		/* Allocate page table pages. */
933 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
934 		if (ptbl == NULL) {
935 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
936 			return (ENOMEM);
937 		}
938 	} else {
939 		/*
940 		 * Check if there is valid mapping for requested
941 		 * va, if there is, remove it.
942 		 */
943 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
944 		if (PTE_ISVALID(pte)) {
945 			pte_remove(mmu, pmap, va, PTBL_HOLD);
946 		} else {
947 			/*
948 			 * pte is not used, increment hold count
949 			 * for ptbl pages.
950 			 */
951 			if (pmap != kernel_pmap)
952 				ptbl_hold(mmu, pmap, pdir_idx);
953 		}
954 	}
955 
956 	/*
957 	 * Insert pv_entry into pv_list for mapped page if part of managed
958 	 * memory.
959 	 */
960 	if ((m->oflags & VPO_UNMANAGED) == 0) {
961 		flags |= PTE_MANAGED;
962 
963 		/* Create and insert pv entry. */
964 		pv_insert(pmap, va, m);
965 	}
966 
967 	pmap->pm_stats.resident_count++;
968 
969 	mtx_lock_spin(&tlbivax_mutex);
970 	tlb_miss_lock();
971 
972 	tlb0_flush_entry(va);
973 	if (pmap->pm_pdir[pdir_idx] == NULL) {
974 		/*
975 		 * If we just allocated a new page table, hook it in
976 		 * the pdir.
977 		 */
978 		pmap->pm_pdir[pdir_idx] = ptbl;
979 	}
980 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
981 	pte->rpn = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
982 	pte->flags |= (PTE_VALID | flags);
983 
984 	tlb_miss_unlock();
985 	mtx_unlock_spin(&tlbivax_mutex);
986 	return (0);
987 }
988 
989 /* Return the pa for the given pmap/va. */
990 static vm_paddr_t
991 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
992 {
993 	vm_paddr_t pa = 0;
994 	pte_t *pte;
995 
996 	pte = pte_find(mmu, pmap, va);
997 	if ((pte != NULL) && PTE_ISVALID(pte))
998 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
999 	return (pa);
1000 }
1001 
1002 /* Get a pointer to a PTE in a page table. */
1003 static pte_t *
1004 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1005 {
1006 	unsigned int pdir_idx = PDIR_IDX(va);
1007 	unsigned int ptbl_idx = PTBL_IDX(va);
1008 
1009 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1010 
1011 	if (pmap->pm_pdir[pdir_idx])
1012 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1013 
1014 	return (NULL);
1015 }
1016 
1017 /**************************************************************************/
1018 /* PMAP related */
1019 /**************************************************************************/
1020 
1021 /*
1022  * This is called during booke_init, before the system is really initialized.
1023  */
1024 static void
1025 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1026 {
1027 	vm_offset_t phys_kernelend;
1028 	struct mem_region *mp, *mp1;
1029 	int cnt, i, j;
1030 	u_int s, e, sz;
1031 	u_int phys_avail_count;
1032 	vm_size_t physsz, hwphyssz, kstack0_sz;
1033 	vm_offset_t kernel_pdir, kstack0, va;
1034 	vm_paddr_t kstack0_phys;
1035 	void *dpcpu;
1036 	pte_t *pte;
1037 
1038 	debugf("mmu_booke_bootstrap: entered\n");
1039 
1040 	/* Set interesting system properties */
1041 	hw_direct_map = 0;
1042 	elf32_nxstack = 1;
1043 
1044 	/* Initialize invalidation mutex */
1045 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1046 
1047 	/* Read TLB0 size and associativity. */
1048 	tlb0_get_tlbconf();
1049 
1050 	/*
1051 	 * Align kernel start and end address (kernel image).
1052 	 * Note that kernel end does not necessarily relate to kernsize.
1053 	 * kernsize is the size of the kernel that is actually mapped.
1054 	 */
1055 	kernstart = trunc_page(start);
1056 	data_start = round_page(kernelend);
1057 	data_end = data_start;
1058 
1059 	/*
1060 	 * Addresses of preloaded modules (like file systems) use
1061 	 * physical addresses. Make sure we relocate those into
1062 	 * virtual addresses.
1063 	 */
1064 	preload_addr_relocate = kernstart - kernload;
1065 
1066 	/* Allocate the dynamic per-cpu area. */
1067 	dpcpu = (void *)data_end;
1068 	data_end += DPCPU_SIZE;
1069 
1070 	/* Allocate space for the message buffer. */
1071 	msgbufp = (struct msgbuf *)data_end;
1072 	data_end += msgbufsize;
1073 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1074 	    data_end);
1075 
1076 	data_end = round_page(data_end);
1077 
1078 	/* Allocate space for ptbl_bufs. */
1079 	ptbl_bufs = (struct ptbl_buf *)data_end;
1080 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1081 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1082 	    data_end);
1083 
1084 	data_end = round_page(data_end);
1085 
1086 	/* Allocate PTE tables for kernel KVA. */
1087 	kernel_pdir = data_end;
1088 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1089 	    PDIR_SIZE - 1) / PDIR_SIZE;
1090 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1091 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1092 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1093 
1094 	debugf(" data_end: 0x%08x\n", data_end);
1095 	if (data_end - kernstart > kernsize) {
1096 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1097 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1098 	}
1099 	data_end = kernstart + kernsize;
1100 	debugf(" updated data_end: 0x%08x\n", data_end);
1101 
1102 	/*
1103 	 * Clear the structures - note we can only do it safely after the
1104 	 * possible additional TLB1 translations are in place (above) so that
1105 	 * all range up to the currently calculated 'data_end' is covered.
1106 	 */
1107 	dpcpu_init(dpcpu, 0);
1108 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1109 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1110 
1111 	/*******************************************************/
1112 	/* Set the start and end of kva. */
1113 	/*******************************************************/
1114 	virtual_avail = round_page(data_end);
1115 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1116 
1117 	/* Allocate KVA space for page zero/copy operations. */
1118 	zero_page_va = virtual_avail;
1119 	virtual_avail += PAGE_SIZE;
1120 	zero_page_idle_va = virtual_avail;
1121 	virtual_avail += PAGE_SIZE;
1122 	copy_page_src_va = virtual_avail;
1123 	virtual_avail += PAGE_SIZE;
1124 	copy_page_dst_va = virtual_avail;
1125 	virtual_avail += PAGE_SIZE;
1126 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1127 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1128 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1129 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1130 
1131 	/* Initialize page zero/copy mutexes. */
1132 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1133 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1134 
1135 	/* Allocate KVA space for ptbl bufs. */
1136 	ptbl_buf_pool_vabase = virtual_avail;
1137 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1138 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1139 	    ptbl_buf_pool_vabase, virtual_avail);
1140 
1141 	/* Calculate corresponding physical addresses for the kernel region. */
1142 	phys_kernelend = kernload + kernsize;
1143 	debugf("kernel image and allocated data:\n");
1144 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1145 	debugf(" kernstart   = 0x%08x\n", kernstart);
1146 	debugf(" kernsize    = 0x%08x\n", kernsize);
1147 
1148 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1149 		panic("mmu_booke_bootstrap: phys_avail too small");
1150 
1151 	/*
1152 	 * Remove kernel physical address range from avail regions list. Page
1153 	 * align all regions.  Non-page aligned memory isn't very interesting
1154 	 * to us.  Also, sort the entries for ascending addresses.
1155 	 */
1156 
1157 	/* Retrieve phys/avail mem regions */
1158 	mem_regions(&physmem_regions, &physmem_regions_sz,
1159 	    &availmem_regions, &availmem_regions_sz);
1160 	sz = 0;
1161 	cnt = availmem_regions_sz;
1162 	debugf("processing avail regions:\n");
1163 	for (mp = availmem_regions; mp->mr_size; mp++) {
1164 		s = mp->mr_start;
1165 		e = mp->mr_start + mp->mr_size;
1166 		debugf(" %08x-%08x -> ", s, e);
1167 		/* Check whether this region holds all of the kernel. */
1168 		if (s < kernload && e > phys_kernelend) {
1169 			availmem_regions[cnt].mr_start = phys_kernelend;
1170 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1171 			e = kernload;
1172 		}
1173 		/* Look whether this regions starts within the kernel. */
1174 		if (s >= kernload && s < phys_kernelend) {
1175 			if (e <= phys_kernelend)
1176 				goto empty;
1177 			s = phys_kernelend;
1178 		}
1179 		/* Now look whether this region ends within the kernel. */
1180 		if (e > kernload && e <= phys_kernelend) {
1181 			if (s >= kernload)
1182 				goto empty;
1183 			e = kernload;
1184 		}
1185 		/* Now page align the start and size of the region. */
1186 		s = round_page(s);
1187 		e = trunc_page(e);
1188 		if (e < s)
1189 			e = s;
1190 		sz = e - s;
1191 		debugf("%08x-%08x = %x\n", s, e, sz);
1192 
1193 		/* Check whether some memory is left here. */
1194 		if (sz == 0) {
1195 		empty:
1196 			memmove(mp, mp + 1,
1197 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1198 			cnt--;
1199 			mp--;
1200 			continue;
1201 		}
1202 
1203 		/* Do an insertion sort. */
1204 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1205 			if (s < mp1->mr_start)
1206 				break;
1207 		if (mp1 < mp) {
1208 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1209 			mp1->mr_start = s;
1210 			mp1->mr_size = sz;
1211 		} else {
1212 			mp->mr_start = s;
1213 			mp->mr_size = sz;
1214 		}
1215 	}
1216 	availmem_regions_sz = cnt;
1217 
1218 	/*******************************************************/
1219 	/* Steal physical memory for kernel stack from the end */
1220 	/* of the first avail region                           */
1221 	/*******************************************************/
1222 	kstack0_sz = kstack_pages * PAGE_SIZE;
1223 	kstack0_phys = availmem_regions[0].mr_start +
1224 	    availmem_regions[0].mr_size;
1225 	kstack0_phys -= kstack0_sz;
1226 	availmem_regions[0].mr_size -= kstack0_sz;
1227 
1228 	/*******************************************************/
1229 	/* Fill in phys_avail table, based on availmem_regions */
1230 	/*******************************************************/
1231 	phys_avail_count = 0;
1232 	physsz = 0;
1233 	hwphyssz = 0;
1234 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1235 
1236 	debugf("fill in phys_avail:\n");
1237 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1238 
1239 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1240 		    availmem_regions[i].mr_start,
1241 		    availmem_regions[i].mr_start +
1242 		        availmem_regions[i].mr_size,
1243 		    availmem_regions[i].mr_size);
1244 
1245 		if (hwphyssz != 0 &&
1246 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1247 			debugf(" hw.physmem adjust\n");
1248 			if (physsz < hwphyssz) {
1249 				phys_avail[j] = availmem_regions[i].mr_start;
1250 				phys_avail[j + 1] =
1251 				    availmem_regions[i].mr_start +
1252 				    hwphyssz - physsz;
1253 				physsz = hwphyssz;
1254 				phys_avail_count++;
1255 			}
1256 			break;
1257 		}
1258 
1259 		phys_avail[j] = availmem_regions[i].mr_start;
1260 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1261 		    availmem_regions[i].mr_size;
1262 		phys_avail_count++;
1263 		physsz += availmem_regions[i].mr_size;
1264 	}
1265 	physmem = btoc(physsz);
1266 
1267 	/* Calculate the last available physical address. */
1268 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1269 		;
1270 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1271 
1272 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1273 	debugf("phys_avail_count = %d\n", phys_avail_count);
1274 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1275 	    physmem);
1276 
1277 	/*******************************************************/
1278 	/* Initialize (statically allocated) kernel pmap. */
1279 	/*******************************************************/
1280 	PMAP_LOCK_INIT(kernel_pmap);
1281 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1282 
1283 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1284 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1285 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1286 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1287 
1288 	/* Initialize kernel pdir */
1289 	for (i = 0; i < kernel_ptbls; i++)
1290 		kernel_pmap->pm_pdir[kptbl_min + i] =
1291 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1292 
1293 	for (i = 0; i < MAXCPU; i++) {
1294 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1295 
1296 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1297 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1298 	}
1299 
1300 	/*
1301 	 * Fill in PTEs covering kernel code and data. They are not required
1302 	 * for address translation, as this area is covered by static TLB1
1303 	 * entries, but for pte_vatopa() to work correctly with kernel area
1304 	 * addresses.
1305 	 */
1306 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1307 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1308 		pte->rpn = kernload + (va - kernstart);
1309 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1310 		    PTE_VALID;
1311 	}
1312 	/* Mark kernel_pmap active on all CPUs */
1313 	CPU_FILL(&kernel_pmap->pm_active);
1314 
1315  	/*
1316 	 * Initialize the global pv list lock.
1317 	 */
1318 	rw_init(&pvh_global_lock, "pmap pv global");
1319 
1320 	/*******************************************************/
1321 	/* Final setup */
1322 	/*******************************************************/
1323 
1324 	/* Enter kstack0 into kernel map, provide guard page */
1325 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1326 	thread0.td_kstack = kstack0;
1327 	thread0.td_kstack_pages = kstack_pages;
1328 
1329 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1330 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1331 	    kstack0_phys, kstack0_phys + kstack0_sz);
1332 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1333 
1334 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1335 	for (i = 0; i < kstack_pages; i++) {
1336 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1337 		kstack0 += PAGE_SIZE;
1338 		kstack0_phys += PAGE_SIZE;
1339 	}
1340 
1341 	pmap_bootstrapped = 1;
1342 
1343 	debugf("virtual_avail = %08x\n", virtual_avail);
1344 	debugf("virtual_end   = %08x\n", virtual_end);
1345 
1346 	debugf("mmu_booke_bootstrap: exit\n");
1347 }
1348 
1349 #ifdef SMP
1350 void
1351 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1352 {
1353 	int i;
1354 
1355 	/*
1356 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1357 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1358 	 * these values directly to (re)program AP's TLB1 hardware.
1359 	 */
1360 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1361 		/* Skip invalid entries */
1362 		if (!(tlb1[i].mas1 & MAS1_VALID))
1363 			continue;
1364 
1365 		tlb1_write_entry(i);
1366 	}
1367 
1368 	set_mas4_defaults();
1369 }
1370 #endif
1371 
1372 static void
1373 booke_pmap_init_qpages(void)
1374 {
1375 	struct pcpu *pc;
1376 	int i;
1377 
1378 	CPU_FOREACH(i) {
1379 		pc = pcpu_find(i);
1380 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1381 		if (pc->pc_qmap_addr == 0)
1382 			panic("pmap_init_qpages: unable to allocate KVA");
1383 	}
1384 }
1385 
1386 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1387 
1388 /*
1389  * Get the physical page address for the given pmap/virtual address.
1390  */
1391 static vm_paddr_t
1392 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1393 {
1394 	vm_paddr_t pa;
1395 
1396 	PMAP_LOCK(pmap);
1397 	pa = pte_vatopa(mmu, pmap, va);
1398 	PMAP_UNLOCK(pmap);
1399 
1400 	return (pa);
1401 }
1402 
1403 /*
1404  * Extract the physical page address associated with the given
1405  * kernel virtual address.
1406  */
1407 static vm_paddr_t
1408 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1409 {
1410 	int i;
1411 
1412 	/* Check TLB1 mappings */
1413 	for (i = 0; i < tlb1_idx; i++) {
1414 		if (!(tlb1[i].mas1 & MAS1_VALID))
1415 			continue;
1416 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1417 			return (tlb1[i].phys + (va - tlb1[i].virt));
1418 	}
1419 
1420 	return (pte_vatopa(mmu, kernel_pmap, va));
1421 }
1422 
1423 /*
1424  * Initialize the pmap module.
1425  * Called by vm_init, to initialize any structures that the pmap
1426  * system needs to map virtual memory.
1427  */
1428 static void
1429 mmu_booke_init(mmu_t mmu)
1430 {
1431 	int shpgperproc = PMAP_SHPGPERPROC;
1432 
1433 	/*
1434 	 * Initialize the address space (zone) for the pv entries.  Set a
1435 	 * high water mark so that the system can recover from excessive
1436 	 * numbers of pv entries.
1437 	 */
1438 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1439 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1440 
1441 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1442 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1443 
1444 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1445 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1446 
1447 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1448 
1449 	/* Pre-fill pvzone with initial number of pv entries. */
1450 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1451 
1452 	/* Initialize ptbl allocation. */
1453 	ptbl_init();
1454 }
1455 
1456 /*
1457  * Map a list of wired pages into kernel virtual address space.  This is
1458  * intended for temporary mappings which do not need page modification or
1459  * references recorded.  Existing mappings in the region are overwritten.
1460  */
1461 static void
1462 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1463 {
1464 	vm_offset_t va;
1465 
1466 	va = sva;
1467 	while (count-- > 0) {
1468 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1469 		va += PAGE_SIZE;
1470 		m++;
1471 	}
1472 }
1473 
1474 /*
1475  * Remove page mappings from kernel virtual address space.  Intended for
1476  * temporary mappings entered by mmu_booke_qenter.
1477  */
1478 static void
1479 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1480 {
1481 	vm_offset_t va;
1482 
1483 	va = sva;
1484 	while (count-- > 0) {
1485 		mmu_booke_kremove(mmu, va);
1486 		va += PAGE_SIZE;
1487 	}
1488 }
1489 
1490 /*
1491  * Map a wired page into kernel virtual address space.
1492  */
1493 static void
1494 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1495 {
1496 
1497 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1498 }
1499 
1500 static void
1501 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1502 {
1503 	unsigned int pdir_idx = PDIR_IDX(va);
1504 	unsigned int ptbl_idx = PTBL_IDX(va);
1505 	uint32_t flags;
1506 	pte_t *pte;
1507 
1508 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1509 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1510 
1511 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1512 	flags |= tlb_calc_wimg(pa, ma);
1513 
1514 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1515 
1516 	mtx_lock_spin(&tlbivax_mutex);
1517 	tlb_miss_lock();
1518 
1519 	if (PTE_ISVALID(pte)) {
1520 
1521 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1522 
1523 		/* Flush entry from TLB0 */
1524 		tlb0_flush_entry(va);
1525 	}
1526 
1527 	pte->rpn = PTE_RPN_FROM_PA(pa);
1528 	pte->flags = flags;
1529 
1530 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1531 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1532 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1533 
1534 	/* Flush the real memory from the instruction cache. */
1535 	if ((flags & (PTE_I | PTE_G)) == 0) {
1536 		__syncicache((void *)va, PAGE_SIZE);
1537 	}
1538 
1539 	tlb_miss_unlock();
1540 	mtx_unlock_spin(&tlbivax_mutex);
1541 }
1542 
1543 /*
1544  * Remove a page from kernel page table.
1545  */
1546 static void
1547 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1548 {
1549 	unsigned int pdir_idx = PDIR_IDX(va);
1550 	unsigned int ptbl_idx = PTBL_IDX(va);
1551 	pte_t *pte;
1552 
1553 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1554 
1555 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1556 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1557 	    ("mmu_booke_kremove: invalid va"));
1558 
1559 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1560 
1561 	if (!PTE_ISVALID(pte)) {
1562 
1563 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1564 
1565 		return;
1566 	}
1567 
1568 	mtx_lock_spin(&tlbivax_mutex);
1569 	tlb_miss_lock();
1570 
1571 	/* Invalidate entry in TLB0, update PTE. */
1572 	tlb0_flush_entry(va);
1573 	pte->flags = 0;
1574 	pte->rpn = 0;
1575 
1576 	tlb_miss_unlock();
1577 	mtx_unlock_spin(&tlbivax_mutex);
1578 }
1579 
1580 /*
1581  * Initialize pmap associated with process 0.
1582  */
1583 static void
1584 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1585 {
1586 
1587 	PMAP_LOCK_INIT(pmap);
1588 	mmu_booke_pinit(mmu, pmap);
1589 	PCPU_SET(curpmap, pmap);
1590 }
1591 
1592 /*
1593  * Initialize a preallocated and zeroed pmap structure,
1594  * such as one in a vmspace structure.
1595  */
1596 static void
1597 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1598 {
1599 	int i;
1600 
1601 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1602 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1603 
1604 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1605 
1606 	for (i = 0; i < MAXCPU; i++)
1607 		pmap->pm_tid[i] = TID_NONE;
1608 	CPU_ZERO(&kernel_pmap->pm_active);
1609 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1610 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1611 	TAILQ_INIT(&pmap->pm_ptbl_list);
1612 }
1613 
1614 /*
1615  * Release any resources held by the given physical map.
1616  * Called when a pmap initialized by mmu_booke_pinit is being released.
1617  * Should only be called if the map contains no valid mappings.
1618  */
1619 static void
1620 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1621 {
1622 
1623 	KASSERT(pmap->pm_stats.resident_count == 0,
1624 	    ("pmap_release: pmap resident count %ld != 0",
1625 	    pmap->pm_stats.resident_count));
1626 }
1627 
1628 /*
1629  * Insert the given physical page at the specified virtual address in the
1630  * target physical map with the protection requested. If specified the page
1631  * will be wired down.
1632  */
1633 static int
1634 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1635     vm_prot_t prot, u_int flags, int8_t psind)
1636 {
1637 	int error;
1638 
1639 	rw_wlock(&pvh_global_lock);
1640 	PMAP_LOCK(pmap);
1641 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1642 	rw_wunlock(&pvh_global_lock);
1643 	PMAP_UNLOCK(pmap);
1644 	return (error);
1645 }
1646 
1647 static int
1648 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1649     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1650 {
1651 	pte_t *pte;
1652 	vm_paddr_t pa;
1653 	uint32_t flags;
1654 	int error, su, sync;
1655 
1656 	pa = VM_PAGE_TO_PHYS(m);
1657 	su = (pmap == kernel_pmap);
1658 	sync = 0;
1659 
1660 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1661 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1662 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1663 	//		(u_int32_t)m, va, pa, prot, flags);
1664 
1665 	if (su) {
1666 		KASSERT(((va >= virtual_avail) &&
1667 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1668 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1669 	} else {
1670 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1671 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1672 	}
1673 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1674 		VM_OBJECT_ASSERT_LOCKED(m->object);
1675 
1676 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1677 
1678 	/*
1679 	 * If there is an existing mapping, and the physical address has not
1680 	 * changed, must be protection or wiring change.
1681 	 */
1682 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1683 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1684 
1685 		/*
1686 		 * Before actually updating pte->flags we calculate and
1687 		 * prepare its new value in a helper var.
1688 		 */
1689 		flags = pte->flags;
1690 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1691 
1692 		/* Wiring change, just update stats. */
1693 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1694 			if (!PTE_ISWIRED(pte)) {
1695 				flags |= PTE_WIRED;
1696 				pmap->pm_stats.wired_count++;
1697 			}
1698 		} else {
1699 			if (PTE_ISWIRED(pte)) {
1700 				flags &= ~PTE_WIRED;
1701 				pmap->pm_stats.wired_count--;
1702 			}
1703 		}
1704 
1705 		if (prot & VM_PROT_WRITE) {
1706 			/* Add write permissions. */
1707 			flags |= PTE_SW;
1708 			if (!su)
1709 				flags |= PTE_UW;
1710 
1711 			if ((flags & PTE_MANAGED) != 0)
1712 				vm_page_aflag_set(m, PGA_WRITEABLE);
1713 		} else {
1714 			/* Handle modified pages, sense modify status. */
1715 
1716 			/*
1717 			 * The PTE_MODIFIED flag could be set by underlying
1718 			 * TLB misses since we last read it (above), possibly
1719 			 * other CPUs could update it so we check in the PTE
1720 			 * directly rather than rely on that saved local flags
1721 			 * copy.
1722 			 */
1723 			if (PTE_ISMODIFIED(pte))
1724 				vm_page_dirty(m);
1725 		}
1726 
1727 		if (prot & VM_PROT_EXECUTE) {
1728 			flags |= PTE_SX;
1729 			if (!su)
1730 				flags |= PTE_UX;
1731 
1732 			/*
1733 			 * Check existing flags for execute permissions: if we
1734 			 * are turning execute permissions on, icache should
1735 			 * be flushed.
1736 			 */
1737 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1738 				sync++;
1739 		}
1740 
1741 		flags &= ~PTE_REFERENCED;
1742 
1743 		/*
1744 		 * The new flags value is all calculated -- only now actually
1745 		 * update the PTE.
1746 		 */
1747 		mtx_lock_spin(&tlbivax_mutex);
1748 		tlb_miss_lock();
1749 
1750 		tlb0_flush_entry(va);
1751 		pte->flags = flags;
1752 
1753 		tlb_miss_unlock();
1754 		mtx_unlock_spin(&tlbivax_mutex);
1755 
1756 	} else {
1757 		/*
1758 		 * If there is an existing mapping, but it's for a different
1759 		 * physical address, pte_enter() will delete the old mapping.
1760 		 */
1761 		//if ((pte != NULL) && PTE_ISVALID(pte))
1762 		//	debugf("mmu_booke_enter_locked: replace\n");
1763 		//else
1764 		//	debugf("mmu_booke_enter_locked: new\n");
1765 
1766 		/* Now set up the flags and install the new mapping. */
1767 		flags = (PTE_SR | PTE_VALID);
1768 		flags |= PTE_M;
1769 
1770 		if (!su)
1771 			flags |= PTE_UR;
1772 
1773 		if (prot & VM_PROT_WRITE) {
1774 			flags |= PTE_SW;
1775 			if (!su)
1776 				flags |= PTE_UW;
1777 
1778 			if ((m->oflags & VPO_UNMANAGED) == 0)
1779 				vm_page_aflag_set(m, PGA_WRITEABLE);
1780 		}
1781 
1782 		if (prot & VM_PROT_EXECUTE) {
1783 			flags |= PTE_SX;
1784 			if (!su)
1785 				flags |= PTE_UX;
1786 		}
1787 
1788 		/* If its wired update stats. */
1789 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1790 			flags |= PTE_WIRED;
1791 
1792 		error = pte_enter(mmu, pmap, m, va, flags,
1793 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1794 		if (error != 0)
1795 			return (KERN_RESOURCE_SHORTAGE);
1796 
1797 		if ((flags & PMAP_ENTER_WIRED) != 0)
1798 			pmap->pm_stats.wired_count++;
1799 
1800 		/* Flush the real memory from the instruction cache. */
1801 		if (prot & VM_PROT_EXECUTE)
1802 			sync++;
1803 	}
1804 
1805 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1806 		__syncicache((void *)va, PAGE_SIZE);
1807 		sync = 0;
1808 	}
1809 
1810 	return (KERN_SUCCESS);
1811 }
1812 
1813 /*
1814  * Maps a sequence of resident pages belonging to the same object.
1815  * The sequence begins with the given page m_start.  This page is
1816  * mapped at the given virtual address start.  Each subsequent page is
1817  * mapped at a virtual address that is offset from start by the same
1818  * amount as the page is offset from m_start within the object.  The
1819  * last page in the sequence is the page with the largest offset from
1820  * m_start that can be mapped at a virtual address less than the given
1821  * virtual address end.  Not every virtual page between start and end
1822  * is mapped; only those for which a resident page exists with the
1823  * corresponding offset from m_start are mapped.
1824  */
1825 static void
1826 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1827     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1828 {
1829 	vm_page_t m;
1830 	vm_pindex_t diff, psize;
1831 
1832 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1833 
1834 	psize = atop(end - start);
1835 	m = m_start;
1836 	rw_wlock(&pvh_global_lock);
1837 	PMAP_LOCK(pmap);
1838 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1839 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1840 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1841 		    PMAP_ENTER_NOSLEEP, 0);
1842 		m = TAILQ_NEXT(m, listq);
1843 	}
1844 	rw_wunlock(&pvh_global_lock);
1845 	PMAP_UNLOCK(pmap);
1846 }
1847 
1848 static void
1849 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1850     vm_prot_t prot)
1851 {
1852 
1853 	rw_wlock(&pvh_global_lock);
1854 	PMAP_LOCK(pmap);
1855 	mmu_booke_enter_locked(mmu, pmap, va, m,
1856 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1857 	    0);
1858 	rw_wunlock(&pvh_global_lock);
1859 	PMAP_UNLOCK(pmap);
1860 }
1861 
1862 /*
1863  * Remove the given range of addresses from the specified map.
1864  *
1865  * It is assumed that the start and end are properly rounded to the page size.
1866  */
1867 static void
1868 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1869 {
1870 	pte_t *pte;
1871 	uint8_t hold_flag;
1872 
1873 	int su = (pmap == kernel_pmap);
1874 
1875 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1876 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1877 
1878 	if (su) {
1879 		KASSERT(((va >= virtual_avail) &&
1880 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1881 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1882 	} else {
1883 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1884 		    ("mmu_booke_remove: user pmap, non user va"));
1885 	}
1886 
1887 	if (PMAP_REMOVE_DONE(pmap)) {
1888 		//debugf("mmu_booke_remove: e (empty)\n");
1889 		return;
1890 	}
1891 
1892 	hold_flag = PTBL_HOLD_FLAG(pmap);
1893 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1894 
1895 	rw_wlock(&pvh_global_lock);
1896 	PMAP_LOCK(pmap);
1897 	for (; va < endva; va += PAGE_SIZE) {
1898 		pte = pte_find(mmu, pmap, va);
1899 		if ((pte != NULL) && PTE_ISVALID(pte))
1900 			pte_remove(mmu, pmap, va, hold_flag);
1901 	}
1902 	PMAP_UNLOCK(pmap);
1903 	rw_wunlock(&pvh_global_lock);
1904 
1905 	//debugf("mmu_booke_remove: e\n");
1906 }
1907 
1908 /*
1909  * Remove physical page from all pmaps in which it resides.
1910  */
1911 static void
1912 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1913 {
1914 	pv_entry_t pv, pvn;
1915 	uint8_t hold_flag;
1916 
1917 	rw_wlock(&pvh_global_lock);
1918 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1919 		pvn = TAILQ_NEXT(pv, pv_link);
1920 
1921 		PMAP_LOCK(pv->pv_pmap);
1922 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1923 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1924 		PMAP_UNLOCK(pv->pv_pmap);
1925 	}
1926 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1927 	rw_wunlock(&pvh_global_lock);
1928 }
1929 
1930 /*
1931  * Map a range of physical addresses into kernel virtual address space.
1932  */
1933 static vm_offset_t
1934 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1935     vm_paddr_t pa_end, int prot)
1936 {
1937 	vm_offset_t sva = *virt;
1938 	vm_offset_t va = sva;
1939 
1940 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1941 	//		sva, pa_start, pa_end);
1942 
1943 	while (pa_start < pa_end) {
1944 		mmu_booke_kenter(mmu, va, pa_start);
1945 		va += PAGE_SIZE;
1946 		pa_start += PAGE_SIZE;
1947 	}
1948 	*virt = va;
1949 
1950 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1951 	return (sva);
1952 }
1953 
1954 /*
1955  * The pmap must be activated before it's address space can be accessed in any
1956  * way.
1957  */
1958 static void
1959 mmu_booke_activate(mmu_t mmu, struct thread *td)
1960 {
1961 	pmap_t pmap;
1962 	u_int cpuid;
1963 
1964 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1965 
1966 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1967 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1968 
1969 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1970 
1971 	sched_pin();
1972 
1973 	cpuid = PCPU_GET(cpuid);
1974 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1975 	PCPU_SET(curpmap, pmap);
1976 
1977 	if (pmap->pm_tid[cpuid] == TID_NONE)
1978 		tid_alloc(pmap);
1979 
1980 	/* Load PID0 register with pmap tid value. */
1981 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1982 	__asm __volatile("isync");
1983 
1984 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1985 
1986 	sched_unpin();
1987 
1988 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1989 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1990 }
1991 
1992 /*
1993  * Deactivate the specified process's address space.
1994  */
1995 static void
1996 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1997 {
1998 	pmap_t pmap;
1999 
2000 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2001 
2002 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2003 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2004 
2005 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2006 
2007 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2008 	PCPU_SET(curpmap, NULL);
2009 }
2010 
2011 /*
2012  * Copy the range specified by src_addr/len
2013  * from the source map to the range dst_addr/len
2014  * in the destination map.
2015  *
2016  * This routine is only advisory and need not do anything.
2017  */
2018 static void
2019 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2020     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2021 {
2022 
2023 }
2024 
2025 /*
2026  * Set the physical protection on the specified range of this map as requested.
2027  */
2028 static void
2029 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2030     vm_prot_t prot)
2031 {
2032 	vm_offset_t va;
2033 	vm_page_t m;
2034 	pte_t *pte;
2035 
2036 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2037 		mmu_booke_remove(mmu, pmap, sva, eva);
2038 		return;
2039 	}
2040 
2041 	if (prot & VM_PROT_WRITE)
2042 		return;
2043 
2044 	PMAP_LOCK(pmap);
2045 	for (va = sva; va < eva; va += PAGE_SIZE) {
2046 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2047 			if (PTE_ISVALID(pte)) {
2048 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2049 
2050 				mtx_lock_spin(&tlbivax_mutex);
2051 				tlb_miss_lock();
2052 
2053 				/* Handle modified pages. */
2054 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2055 					vm_page_dirty(m);
2056 
2057 				tlb0_flush_entry(va);
2058 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2059 
2060 				tlb_miss_unlock();
2061 				mtx_unlock_spin(&tlbivax_mutex);
2062 			}
2063 		}
2064 	}
2065 	PMAP_UNLOCK(pmap);
2066 }
2067 
2068 /*
2069  * Clear the write and modified bits in each of the given page's mappings.
2070  */
2071 static void
2072 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2073 {
2074 	pv_entry_t pv;
2075 	pte_t *pte;
2076 
2077 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2078 	    ("mmu_booke_remove_write: page %p is not managed", m));
2079 
2080 	/*
2081 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2082 	 * set by another thread while the object is locked.  Thus,
2083 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2084 	 */
2085 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2086 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2087 		return;
2088 	rw_wlock(&pvh_global_lock);
2089 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2090 		PMAP_LOCK(pv->pv_pmap);
2091 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2092 			if (PTE_ISVALID(pte)) {
2093 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2094 
2095 				mtx_lock_spin(&tlbivax_mutex);
2096 				tlb_miss_lock();
2097 
2098 				/* Handle modified pages. */
2099 				if (PTE_ISMODIFIED(pte))
2100 					vm_page_dirty(m);
2101 
2102 				/* Flush mapping from TLB0. */
2103 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2104 
2105 				tlb_miss_unlock();
2106 				mtx_unlock_spin(&tlbivax_mutex);
2107 			}
2108 		}
2109 		PMAP_UNLOCK(pv->pv_pmap);
2110 	}
2111 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2112 	rw_wunlock(&pvh_global_lock);
2113 }
2114 
2115 static void
2116 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2117 {
2118 	pte_t *pte;
2119 	pmap_t pmap;
2120 	vm_page_t m;
2121 	vm_offset_t addr;
2122 	vm_paddr_t pa = 0;
2123 	int active, valid;
2124 
2125 	va = trunc_page(va);
2126 	sz = round_page(sz);
2127 
2128 	rw_wlock(&pvh_global_lock);
2129 	pmap = PCPU_GET(curpmap);
2130 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2131 	while (sz > 0) {
2132 		PMAP_LOCK(pm);
2133 		pte = pte_find(mmu, pm, va);
2134 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2135 		if (valid)
2136 			pa = PTE_PA(pte);
2137 		PMAP_UNLOCK(pm);
2138 		if (valid) {
2139 			if (!active) {
2140 				/* Create a mapping in the active pmap. */
2141 				addr = 0;
2142 				m = PHYS_TO_VM_PAGE(pa);
2143 				PMAP_LOCK(pmap);
2144 				pte_enter(mmu, pmap, m, addr,
2145 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2146 				__syncicache((void *)addr, PAGE_SIZE);
2147 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2148 				PMAP_UNLOCK(pmap);
2149 			} else
2150 				__syncicache((void *)va, PAGE_SIZE);
2151 		}
2152 		va += PAGE_SIZE;
2153 		sz -= PAGE_SIZE;
2154 	}
2155 	rw_wunlock(&pvh_global_lock);
2156 }
2157 
2158 /*
2159  * Atomically extract and hold the physical page with the given
2160  * pmap and virtual address pair if that mapping permits the given
2161  * protection.
2162  */
2163 static vm_page_t
2164 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2165     vm_prot_t prot)
2166 {
2167 	pte_t *pte;
2168 	vm_page_t m;
2169 	uint32_t pte_wbit;
2170 	vm_paddr_t pa;
2171 
2172 	m = NULL;
2173 	pa = 0;
2174 	PMAP_LOCK(pmap);
2175 retry:
2176 	pte = pte_find(mmu, pmap, va);
2177 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2178 		if (pmap == kernel_pmap)
2179 			pte_wbit = PTE_SW;
2180 		else
2181 			pte_wbit = PTE_UW;
2182 
2183 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2184 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2185 				goto retry;
2186 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2187 			vm_page_hold(m);
2188 		}
2189 	}
2190 
2191 	PA_UNLOCK_COND(pa);
2192 	PMAP_UNLOCK(pmap);
2193 	return (m);
2194 }
2195 
2196 /*
2197  * Initialize a vm_page's machine-dependent fields.
2198  */
2199 static void
2200 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2201 {
2202 
2203 	TAILQ_INIT(&m->md.pv_list);
2204 }
2205 
2206 /*
2207  * mmu_booke_zero_page_area zeros the specified hardware page by
2208  * mapping it into virtual memory and using bzero to clear
2209  * its contents.
2210  *
2211  * off and size must reside within a single page.
2212  */
2213 static void
2214 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2215 {
2216 	vm_offset_t va;
2217 
2218 	/* XXX KASSERT off and size are within a single page? */
2219 
2220 	mtx_lock(&zero_page_mutex);
2221 	va = zero_page_va;
2222 
2223 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2224 	bzero((caddr_t)va + off, size);
2225 	mmu_booke_kremove(mmu, va);
2226 
2227 	mtx_unlock(&zero_page_mutex);
2228 }
2229 
2230 /*
2231  * mmu_booke_zero_page zeros the specified hardware page.
2232  */
2233 static void
2234 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2235 {
2236 
2237 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2238 }
2239 
2240 /*
2241  * mmu_booke_copy_page copies the specified (machine independent) page by
2242  * mapping the page into virtual memory and using memcopy to copy the page,
2243  * one machine dependent page at a time.
2244  */
2245 static void
2246 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2247 {
2248 	vm_offset_t sva, dva;
2249 
2250 	sva = copy_page_src_va;
2251 	dva = copy_page_dst_va;
2252 
2253 	mtx_lock(&copy_page_mutex);
2254 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2255 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2256 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2257 	mmu_booke_kremove(mmu, dva);
2258 	mmu_booke_kremove(mmu, sva);
2259 	mtx_unlock(&copy_page_mutex);
2260 }
2261 
2262 static inline void
2263 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2264     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2265 {
2266 	void *a_cp, *b_cp;
2267 	vm_offset_t a_pg_offset, b_pg_offset;
2268 	int cnt;
2269 
2270 	mtx_lock(&copy_page_mutex);
2271 	while (xfersize > 0) {
2272 		a_pg_offset = a_offset & PAGE_MASK;
2273 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2274 		mmu_booke_kenter(mmu, copy_page_src_va,
2275 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2276 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2277 		b_pg_offset = b_offset & PAGE_MASK;
2278 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2279 		mmu_booke_kenter(mmu, copy_page_dst_va,
2280 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2281 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2282 		bcopy(a_cp, b_cp, cnt);
2283 		mmu_booke_kremove(mmu, copy_page_dst_va);
2284 		mmu_booke_kremove(mmu, copy_page_src_va);
2285 		a_offset += cnt;
2286 		b_offset += cnt;
2287 		xfersize -= cnt;
2288 	}
2289 	mtx_unlock(&copy_page_mutex);
2290 }
2291 
2292 /*
2293  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2294  * into virtual memory and using bzero to clear its contents. This is intended
2295  * to be called from the vm_pagezero process only and outside of Giant. No
2296  * lock is required.
2297  */
2298 static void
2299 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2300 {
2301 	vm_offset_t va;
2302 
2303 	va = zero_page_idle_va;
2304 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2305 	bzero((caddr_t)va, PAGE_SIZE);
2306 	mmu_booke_kremove(mmu, va);
2307 }
2308 
2309 static vm_offset_t
2310 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2311 {
2312 	vm_paddr_t paddr;
2313 	vm_offset_t qaddr;
2314 	uint32_t flags;
2315 	pte_t *pte;
2316 
2317 	paddr = VM_PAGE_TO_PHYS(m);
2318 
2319 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2320 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m));
2321 
2322 	critical_enter();
2323 	qaddr = PCPU_GET(qmap_addr);
2324 
2325 	pte = &(kernel_pmap->pm_pdir[PDIR_IDX(qaddr)][PTBL_IDX(qaddr)]);
2326 
2327 	KASSERT(pte->flags == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2328 
2329 	/*
2330 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2331  	 * not be present in other TLBs.  Is there a better instruction
2332 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2333 	 */
2334 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2335 	__asm __volatile("isync; msync");
2336 
2337 	pte->rpn = paddr & ~PTE_PA_MASK;
2338 	pte->flags = flags;
2339 
2340 	/* Flush the real memory from the instruction cache. */
2341 	if ((flags & (PTE_I | PTE_G)) == 0)
2342 		__syncicache((void *)qaddr, PAGE_SIZE);
2343 
2344 	return (qaddr);
2345 }
2346 
2347 static void
2348 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2349 {
2350 	pte_t *pte;
2351 
2352 	pte = &(kernel_pmap->pm_pdir[PDIR_IDX(addr)][PTBL_IDX(addr)]);
2353 
2354 	KASSERT(PCPU_GET(qmap_addr) == addr,
2355 	    ("mmu_booke_quick_remove_page: invalid address"));
2356 	KASSERT(pte->flags != 0,
2357 	    ("mmu_booke_quick_remove_page: PTE not in use"));
2358 
2359 	pte->flags = 0;
2360 	pte->rpn = 0;
2361 	critical_exit();
2362 }
2363 
2364 /*
2365  * Return whether or not the specified physical page was modified
2366  * in any of physical maps.
2367  */
2368 static boolean_t
2369 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2370 {
2371 	pte_t *pte;
2372 	pv_entry_t pv;
2373 	boolean_t rv;
2374 
2375 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2376 	    ("mmu_booke_is_modified: page %p is not managed", m));
2377 	rv = FALSE;
2378 
2379 	/*
2380 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2381 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2382 	 * is clear, no PTEs can be modified.
2383 	 */
2384 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2385 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2386 		return (rv);
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_ISMODIFIED(pte))
2393 				rv = TRUE;
2394 		}
2395 		PMAP_UNLOCK(pv->pv_pmap);
2396 		if (rv)
2397 			break;
2398 	}
2399 	rw_wunlock(&pvh_global_lock);
2400 	return (rv);
2401 }
2402 
2403 /*
2404  * Return whether or not the specified virtual address is eligible
2405  * for prefault.
2406  */
2407 static boolean_t
2408 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2409 {
2410 
2411 	return (FALSE);
2412 }
2413 
2414 /*
2415  * Return whether or not the specified physical page was referenced
2416  * in any physical maps.
2417  */
2418 static boolean_t
2419 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2420 {
2421 	pte_t *pte;
2422 	pv_entry_t pv;
2423 	boolean_t rv;
2424 
2425 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2426 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2427 	rv = FALSE;
2428 	rw_wlock(&pvh_global_lock);
2429 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2430 		PMAP_LOCK(pv->pv_pmap);
2431 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2432 		    PTE_ISVALID(pte)) {
2433 			if (PTE_ISREFERENCED(pte))
2434 				rv = TRUE;
2435 		}
2436 		PMAP_UNLOCK(pv->pv_pmap);
2437 		if (rv)
2438 			break;
2439 	}
2440 	rw_wunlock(&pvh_global_lock);
2441 	return (rv);
2442 }
2443 
2444 /*
2445  * Clear the modify bits on the specified physical page.
2446  */
2447 static void
2448 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2449 {
2450 	pte_t *pte;
2451 	pv_entry_t pv;
2452 
2453 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2454 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2455 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2456 	KASSERT(!vm_page_xbusied(m),
2457 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2458 
2459 	/*
2460 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2461 	 * If the object containing the page is locked and the page is not
2462 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2463 	 */
2464 	if ((m->aflags & PGA_WRITEABLE) == 0)
2465 		return;
2466 	rw_wlock(&pvh_global_lock);
2467 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2468 		PMAP_LOCK(pv->pv_pmap);
2469 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2470 		    PTE_ISVALID(pte)) {
2471 			mtx_lock_spin(&tlbivax_mutex);
2472 			tlb_miss_lock();
2473 
2474 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2475 				tlb0_flush_entry(pv->pv_va);
2476 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2477 				    PTE_REFERENCED);
2478 			}
2479 
2480 			tlb_miss_unlock();
2481 			mtx_unlock_spin(&tlbivax_mutex);
2482 		}
2483 		PMAP_UNLOCK(pv->pv_pmap);
2484 	}
2485 	rw_wunlock(&pvh_global_lock);
2486 }
2487 
2488 /*
2489  * Return a count of reference bits for a page, clearing those bits.
2490  * It is not necessary for every reference bit to be cleared, but it
2491  * is necessary that 0 only be returned when there are truly no
2492  * reference bits set.
2493  *
2494  * XXX: The exact number of bits to check and clear is a matter that
2495  * should be tested and standardized at some point in the future for
2496  * optimal aging of shared pages.
2497  */
2498 static int
2499 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2500 {
2501 	pte_t *pte;
2502 	pv_entry_t pv;
2503 	int count;
2504 
2505 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2506 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2507 	count = 0;
2508 	rw_wlock(&pvh_global_lock);
2509 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2510 		PMAP_LOCK(pv->pv_pmap);
2511 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2512 		    PTE_ISVALID(pte)) {
2513 			if (PTE_ISREFERENCED(pte)) {
2514 				mtx_lock_spin(&tlbivax_mutex);
2515 				tlb_miss_lock();
2516 
2517 				tlb0_flush_entry(pv->pv_va);
2518 				pte->flags &= ~PTE_REFERENCED;
2519 
2520 				tlb_miss_unlock();
2521 				mtx_unlock_spin(&tlbivax_mutex);
2522 
2523 				if (++count > 4) {
2524 					PMAP_UNLOCK(pv->pv_pmap);
2525 					break;
2526 				}
2527 			}
2528 		}
2529 		PMAP_UNLOCK(pv->pv_pmap);
2530 	}
2531 	rw_wunlock(&pvh_global_lock);
2532 	return (count);
2533 }
2534 
2535 /*
2536  * Clear the wired attribute from the mappings for the specified range of
2537  * addresses in the given pmap.  Every valid mapping within that range must
2538  * have the wired attribute set.  In contrast, invalid mappings cannot have
2539  * the wired attribute set, so they are ignored.
2540  *
2541  * The wired attribute of the page table entry is not a hardware feature, so
2542  * there is no need to invalidate any TLB entries.
2543  */
2544 static void
2545 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2546 {
2547 	vm_offset_t va;
2548 	pte_t *pte;
2549 
2550 	PMAP_LOCK(pmap);
2551 	for (va = sva; va < eva; va += PAGE_SIZE) {
2552 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2553 		    PTE_ISVALID(pte)) {
2554 			if (!PTE_ISWIRED(pte))
2555 				panic("mmu_booke_unwire: pte %p isn't wired",
2556 				    pte);
2557 			pte->flags &= ~PTE_WIRED;
2558 			pmap->pm_stats.wired_count--;
2559 		}
2560 	}
2561 	PMAP_UNLOCK(pmap);
2562 
2563 }
2564 
2565 /*
2566  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2567  * page.  This count may be changed upwards or downwards in the future; it is
2568  * only necessary that true be returned for a small subset of pmaps for proper
2569  * page aging.
2570  */
2571 static boolean_t
2572 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2573 {
2574 	pv_entry_t pv;
2575 	int loops;
2576 	boolean_t rv;
2577 
2578 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2579 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2580 	loops = 0;
2581 	rv = FALSE;
2582 	rw_wlock(&pvh_global_lock);
2583 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2584 		if (pv->pv_pmap == pmap) {
2585 			rv = TRUE;
2586 			break;
2587 		}
2588 		if (++loops >= 16)
2589 			break;
2590 	}
2591 	rw_wunlock(&pvh_global_lock);
2592 	return (rv);
2593 }
2594 
2595 /*
2596  * Return the number of managed mappings to the given physical page that are
2597  * wired.
2598  */
2599 static int
2600 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2601 {
2602 	pv_entry_t pv;
2603 	pte_t *pte;
2604 	int count = 0;
2605 
2606 	if ((m->oflags & VPO_UNMANAGED) != 0)
2607 		return (count);
2608 	rw_wlock(&pvh_global_lock);
2609 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2610 		PMAP_LOCK(pv->pv_pmap);
2611 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2612 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2613 				count++;
2614 		PMAP_UNLOCK(pv->pv_pmap);
2615 	}
2616 	rw_wunlock(&pvh_global_lock);
2617 	return (count);
2618 }
2619 
2620 static int
2621 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2622 {
2623 	int i;
2624 	vm_offset_t va;
2625 
2626 	/*
2627 	 * This currently does not work for entries that
2628 	 * overlap TLB1 entries.
2629 	 */
2630 	for (i = 0; i < tlb1_idx; i ++) {
2631 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2632 			return (0);
2633 	}
2634 
2635 	return (EFAULT);
2636 }
2637 
2638 void
2639 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2640 {
2641 	vm_paddr_t ppa;
2642 	vm_offset_t ofs;
2643 	vm_size_t gran;
2644 
2645 	/* Minidumps are based on virtual memory addresses. */
2646 	if (do_minidump) {
2647 		*va = (void *)(vm_offset_t)pa;
2648 		return;
2649 	}
2650 
2651 	/* Raw physical memory dumps don't have a virtual address. */
2652 	/* We always map a 256MB page at 256M. */
2653 	gran = 256 * 1024 * 1024;
2654 	ppa = pa & ~(gran - 1);
2655 	ofs = pa - ppa;
2656 	*va = (void *)gran;
2657 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2658 
2659 	if (sz > (gran - ofs))
2660 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2661 		    _TLB_ENTRY_IO);
2662 }
2663 
2664 void
2665 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2666 {
2667 	vm_paddr_t ppa;
2668 	vm_offset_t ofs;
2669 	vm_size_t gran;
2670 
2671 	/* Minidumps are based on virtual memory addresses. */
2672 	/* Nothing to do... */
2673 	if (do_minidump)
2674 		return;
2675 
2676 	/* Raw physical memory dumps don't have a virtual address. */
2677 	tlb1_idx--;
2678 	tlb1[tlb1_idx].mas1 = 0;
2679 	tlb1[tlb1_idx].mas2 = 0;
2680 	tlb1[tlb1_idx].mas3 = 0;
2681 	tlb1_write_entry(tlb1_idx);
2682 
2683 	gran = 256 * 1024 * 1024;
2684 	ppa = pa & ~(gran - 1);
2685 	ofs = pa - ppa;
2686 	if (sz > (gran - ofs)) {
2687 		tlb1_idx--;
2688 		tlb1[tlb1_idx].mas1 = 0;
2689 		tlb1[tlb1_idx].mas2 = 0;
2690 		tlb1[tlb1_idx].mas3 = 0;
2691 		tlb1_write_entry(tlb1_idx);
2692 	}
2693 }
2694 
2695 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2696 
2697 void
2698 mmu_booke_scan_init(mmu_t mmu)
2699 {
2700 	vm_offset_t va;
2701 	pte_t *pte;
2702 	int i;
2703 
2704 	if (!do_minidump) {
2705 		/* Initialize phys. segments for dumpsys(). */
2706 		memset(&dump_map, 0, sizeof(dump_map));
2707 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2708 		    &availmem_regions_sz);
2709 		for (i = 0; i < physmem_regions_sz; i++) {
2710 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2711 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2712 		}
2713 		return;
2714 	}
2715 
2716 	/* Virtual segments for minidumps: */
2717 	memset(&dump_map, 0, sizeof(dump_map));
2718 
2719 	/* 1st: kernel .data and .bss. */
2720 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2721 	dump_map[0].pa_size =
2722 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2723 
2724 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2725 	dump_map[1].pa_start = data_start;
2726 	dump_map[1].pa_size = data_end - data_start;
2727 
2728 	/* 3rd: kernel VM. */
2729 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2730 	/* Find start of next chunk (from va). */
2731 	while (va < virtual_end) {
2732 		/* Don't dump the buffer cache. */
2733 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2734 			va = kmi.buffer_eva;
2735 			continue;
2736 		}
2737 		pte = pte_find(mmu, kernel_pmap, va);
2738 		if (pte != NULL && PTE_ISVALID(pte))
2739 			break;
2740 		va += PAGE_SIZE;
2741 	}
2742 	if (va < virtual_end) {
2743 		dump_map[2].pa_start = va;
2744 		va += PAGE_SIZE;
2745 		/* Find last page in chunk. */
2746 		while (va < virtual_end) {
2747 			/* Don't run into the buffer cache. */
2748 			if (va == kmi.buffer_sva)
2749 				break;
2750 			pte = pte_find(mmu, kernel_pmap, va);
2751 			if (pte == NULL || !PTE_ISVALID(pte))
2752 				break;
2753 			va += PAGE_SIZE;
2754 		}
2755 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2756 	}
2757 }
2758 
2759 /*
2760  * Map a set of physical memory pages into the kernel virtual address space.
2761  * Return a pointer to where it is mapped. This routine is intended to be used
2762  * for mapping device memory, NOT real memory.
2763  */
2764 static void *
2765 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2766 {
2767 
2768 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2769 }
2770 
2771 static void *
2772 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2773 {
2774 	void *res;
2775 	uintptr_t va;
2776 	vm_size_t sz;
2777 	int i;
2778 
2779 	/*
2780 	 * Check if this is premapped in TLB1. Note: this should probably also
2781 	 * check whether a sequence of TLB1 entries exist that match the
2782 	 * requirement, but now only checks the easy case.
2783 	 */
2784 	if (ma == VM_MEMATTR_DEFAULT) {
2785 		for (i = 0; i < tlb1_idx; i++) {
2786 			if (!(tlb1[i].mas1 & MAS1_VALID))
2787 				continue;
2788 			if (pa >= tlb1[i].phys &&
2789 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2790 				return (void *)(tlb1[i].virt +
2791 				    (vm_offset_t)(pa - tlb1[i].phys));
2792 		}
2793 	}
2794 
2795 	size = roundup(size, PAGE_SIZE);
2796 
2797 	/*
2798 	 * We leave a hole for device direct mapping between the maximum user
2799 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2800 	 * devices are in there, just map them 1:1. If not, map them to the
2801 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2802 	 * addresses should be pulled from an allocator, but since we do not
2803 	 * ever free TLB1 entries, it is safe just to increment a counter.
2804 	 * Note that there isn't a lot of address space here (128 MB) and it
2805 	 * is not at all difficult to imagine running out, since that is a 4:1
2806 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2807 	 * mapped there.
2808 	 */
2809 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2810 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2811 		va = pa;
2812 	else
2813 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2814 	res = (void *)va;
2815 
2816 	do {
2817 		sz = 1 << (ilog2(size) & ~1);
2818 		if (va % sz != 0) {
2819 			do {
2820 				sz >>= 2;
2821 			} while (va % sz != 0);
2822 		}
2823 		if (bootverbose)
2824 			printf("Wiring VA=%x to PA=%llx (size=%x), "
2825 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2826 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2827 		size -= sz;
2828 		pa += sz;
2829 		va += sz;
2830 	} while (size > 0);
2831 
2832 	return (res);
2833 }
2834 
2835 /*
2836  * 'Unmap' a range mapped by mmu_booke_mapdev().
2837  */
2838 static void
2839 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2840 {
2841 #ifdef SUPPORTS_SHRINKING_TLB1
2842 	vm_offset_t base, offset;
2843 
2844 	/*
2845 	 * Unmap only if this is inside kernel virtual space.
2846 	 */
2847 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2848 		base = trunc_page(va);
2849 		offset = va & PAGE_MASK;
2850 		size = roundup(offset + size, PAGE_SIZE);
2851 		kva_free(base, size);
2852 	}
2853 #endif
2854 }
2855 
2856 /*
2857  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2858  * specified pmap. This eliminates the blast of soft faults on process startup
2859  * and immediately after an mmap.
2860  */
2861 static void
2862 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2863     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2864 {
2865 
2866 	VM_OBJECT_ASSERT_WLOCKED(object);
2867 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2868 	    ("mmu_booke_object_init_pt: non-device object"));
2869 }
2870 
2871 /*
2872  * Perform the pmap work for mincore.
2873  */
2874 static int
2875 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2876     vm_paddr_t *locked_pa)
2877 {
2878 
2879 	/* XXX: this should be implemented at some point */
2880 	return (0);
2881 }
2882 
2883 /**************************************************************************/
2884 /* TID handling */
2885 /**************************************************************************/
2886 
2887 /*
2888  * Allocate a TID. If necessary, steal one from someone else.
2889  * The new TID is flushed from the TLB before returning.
2890  */
2891 static tlbtid_t
2892 tid_alloc(pmap_t pmap)
2893 {
2894 	tlbtid_t tid;
2895 	int thiscpu;
2896 
2897 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2898 
2899 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2900 
2901 	thiscpu = PCPU_GET(cpuid);
2902 
2903 	tid = PCPU_GET(tid_next);
2904 	if (tid > TID_MAX)
2905 		tid = TID_MIN;
2906 	PCPU_SET(tid_next, tid + 1);
2907 
2908 	/* If we are stealing TID then clear the relevant pmap's field */
2909 	if (tidbusy[thiscpu][tid] != NULL) {
2910 
2911 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2912 
2913 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2914 
2915 		/* Flush all entries from TLB0 matching this TID. */
2916 		tid_flush(tid, tlb0_ways, tlb0_entries_per_way);
2917 	}
2918 
2919 	tidbusy[thiscpu][tid] = pmap;
2920 	pmap->pm_tid[thiscpu] = tid;
2921 	__asm __volatile("msync; isync");
2922 
2923 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2924 	    PCPU_GET(tid_next));
2925 
2926 	return (tid);
2927 }
2928 
2929 /**************************************************************************/
2930 /* TLB0 handling */
2931 /**************************************************************************/
2932 
2933 static void
2934 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2935     uint32_t mas7)
2936 {
2937 	int as;
2938 	char desc[3];
2939 	tlbtid_t tid;
2940 	vm_size_t size;
2941 	unsigned int tsize;
2942 
2943 	desc[2] = '\0';
2944 	if (mas1 & MAS1_VALID)
2945 		desc[0] = 'V';
2946 	else
2947 		desc[0] = ' ';
2948 
2949 	if (mas1 & MAS1_IPROT)
2950 		desc[1] = 'P';
2951 	else
2952 		desc[1] = ' ';
2953 
2954 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2955 	tid = MAS1_GETTID(mas1);
2956 
2957 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2958 	size = 0;
2959 	if (tsize)
2960 		size = tsize2size(tsize);
2961 
2962 	debugf("%3d: (%s) [AS=%d] "
2963 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2964 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2965 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2966 }
2967 
2968 /* Convert TLB0 va and way number to tlb0[] table index. */
2969 static inline unsigned int
2970 tlb0_tableidx(vm_offset_t va, unsigned int way)
2971 {
2972 	unsigned int idx;
2973 
2974 	idx = (way * TLB0_ENTRIES_PER_WAY);
2975 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2976 	return (idx);
2977 }
2978 
2979 /*
2980  * Invalidate TLB0 entry.
2981  */
2982 static inline void
2983 tlb0_flush_entry(vm_offset_t va)
2984 {
2985 
2986 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2987 
2988 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2989 
2990 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2991 	__asm __volatile("isync; msync");
2992 	__asm __volatile("tlbsync; msync");
2993 
2994 	CTR1(KTR_PMAP, "%s: e", __func__);
2995 }
2996 
2997 /* Print out contents of the MAS registers for each TLB0 entry */
2998 void
2999 tlb0_print_tlbentries(void)
3000 {
3001 	uint32_t mas0, mas1, mas2, mas3, mas7;
3002 	int entryidx, way, idx;
3003 
3004 	debugf("TLB0 entries:\n");
3005 	for (way = 0; way < TLB0_WAYS; way ++)
3006 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3007 
3008 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3009 			mtspr(SPR_MAS0, mas0);
3010 			__asm __volatile("isync");
3011 
3012 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3013 			mtspr(SPR_MAS2, mas2);
3014 
3015 			__asm __volatile("isync; tlbre");
3016 
3017 			mas1 = mfspr(SPR_MAS1);
3018 			mas2 = mfspr(SPR_MAS2);
3019 			mas3 = mfspr(SPR_MAS3);
3020 			mas7 = mfspr(SPR_MAS7);
3021 
3022 			idx = tlb0_tableidx(mas2, way);
3023 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3024 		}
3025 }
3026 
3027 /**************************************************************************/
3028 /* TLB1 handling */
3029 /**************************************************************************/
3030 
3031 /*
3032  * TLB1 mapping notes:
3033  *
3034  * TLB1[0]	Kernel text and data.
3035  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3036  *		windows, other devices mappings.
3037  */
3038 
3039 /*
3040  * Write given entry to TLB1 hardware.
3041  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
3042  */
3043 static void
3044 tlb1_write_entry(unsigned int idx)
3045 {
3046 	uint32_t mas0;
3047 
3048 	//debugf("tlb1_write_entry: s\n");
3049 
3050 	/* Select entry */
3051 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3052 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
3053 
3054 	mtspr(SPR_MAS0, mas0);
3055 	__asm __volatile("isync");
3056 	mtspr(SPR_MAS1, tlb1[idx].mas1);
3057 	__asm __volatile("isync");
3058 	mtspr(SPR_MAS2, tlb1[idx].mas2);
3059 	__asm __volatile("isync");
3060 	mtspr(SPR_MAS3, tlb1[idx].mas3);
3061 	__asm __volatile("isync");
3062 	switch ((mfpvr() >> 16) & 0xFFFF) {
3063 	case FSL_E500mc:
3064 	case FSL_E5500:
3065 		mtspr(SPR_MAS8, 0);
3066 		__asm __volatile("isync");
3067 		/* FALLTHROUGH */
3068 	case FSL_E500v2:
3069 		mtspr(SPR_MAS7, tlb1[idx].mas7);
3070 		__asm __volatile("isync");
3071 		break;
3072 	default:
3073 		break;
3074 	}
3075 
3076 	__asm __volatile("tlbwe; isync; msync");
3077 
3078 	//debugf("tlb1_write_entry: e\n");
3079 }
3080 
3081 /*
3082  * Return the largest uint value log such that 2^log <= num.
3083  */
3084 static unsigned int
3085 ilog2(unsigned int num)
3086 {
3087 	int lz;
3088 
3089 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3090 	return (31 - lz);
3091 }
3092 
3093 /*
3094  * Convert TLB TSIZE value to mapped region size.
3095  */
3096 static vm_size_t
3097 tsize2size(unsigned int tsize)
3098 {
3099 
3100 	/*
3101 	 * size = 4^tsize KB
3102 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3103 	 */
3104 
3105 	return ((1 << (2 * tsize)) * 1024);
3106 }
3107 
3108 /*
3109  * Convert region size (must be power of 4) to TLB TSIZE value.
3110  */
3111 static unsigned int
3112 size2tsize(vm_size_t size)
3113 {
3114 
3115 	return (ilog2(size) / 2 - 5);
3116 }
3117 
3118 /*
3119  * Register permanent kernel mapping in TLB1.
3120  *
3121  * Entries are created starting from index 0 (current free entry is
3122  * kept in tlb1_idx) and are not supposed to be invalidated.
3123  */
3124 static int
3125 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3126     uint32_t flags)
3127 {
3128 	uint32_t ts, tid;
3129 	int tsize, index;
3130 
3131 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3132 	if (index >= TLB1_ENTRIES) {
3133 		printf("tlb1_set_entry: TLB1 full!\n");
3134 		return (-1);
3135 	}
3136 
3137 	/* Convert size to TSIZE */
3138 	tsize = size2tsize(size);
3139 
3140 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3141 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3142 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3143 
3144 	/*
3145 	 * Atomicity is preserved by the atomic increment above since nothing
3146 	 * is ever removed from tlb1.
3147 	 */
3148 
3149 	tlb1[index].phys = pa;
3150 	tlb1[index].virt = va;
3151 	tlb1[index].size = size;
3152 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3153 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3154 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3155 
3156 	/* Set supervisor RWX permission bits */
3157 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3158 	tlb1[index].mas7 = (pa >> 32) & MAS7_RPN;
3159 
3160 	tlb1_write_entry(index);
3161 
3162 	/*
3163 	 * XXX in general TLB1 updates should be propagated between CPUs,
3164 	 * since current design assumes to have the same TLB1 set-up on all
3165 	 * cores.
3166 	 */
3167 	return (0);
3168 }
3169 
3170 /*
3171  * Map in contiguous RAM region into the TLB1 using maximum of
3172  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3173  *
3174  * If necessary round up last entry size and return total size
3175  * used by all allocated entries.
3176  */
3177 vm_size_t
3178 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3179 {
3180 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3181 	vm_size_t mapped, pgsz, base, mask;
3182 	int idx, nents;
3183 
3184 	/* Round up to the next 1M */
3185 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3186 
3187 	mapped = 0;
3188 	idx = 0;
3189 	base = va;
3190 	pgsz = 64*1024*1024;
3191 	while (mapped < size) {
3192 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3193 			while (pgsz > (size - mapped))
3194 				pgsz >>= 2;
3195 			pgs[idx++] = pgsz;
3196 			mapped += pgsz;
3197 		}
3198 
3199 		/* We under-map. Correct for this. */
3200 		if (mapped < size) {
3201 			while (pgs[idx - 1] == pgsz) {
3202 				idx--;
3203 				mapped -= pgsz;
3204 			}
3205 			/* XXX We may increase beyond out starting point. */
3206 			pgsz <<= 2;
3207 			pgs[idx++] = pgsz;
3208 			mapped += pgsz;
3209 		}
3210 	}
3211 
3212 	nents = idx;
3213 	mask = pgs[0] - 1;
3214 	/* Align address to the boundary */
3215 	if (va & mask) {
3216 		va = (va + mask) & ~mask;
3217 		pa = (pa + mask) & ~mask;
3218 	}
3219 
3220 	for (idx = 0; idx < nents; idx++) {
3221 		pgsz = pgs[idx];
3222 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
3223 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3224 		pa += pgsz;
3225 		va += pgsz;
3226 	}
3227 
3228 	mapped = (va - base);
3229 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3230 	    mapped, mapped - size);
3231 	return (mapped);
3232 }
3233 
3234 /*
3235  * TLB1 initialization routine, to be called after the very first
3236  * assembler level setup done in locore.S.
3237  */
3238 void
3239 tlb1_init()
3240 {
3241 	uint32_t mas0, mas1, mas2, mas3, mas7;
3242 	uint32_t tsz;
3243 	int i;
3244 
3245 	tlb1_idx = 1;
3246 
3247 	tlb1_get_tlbconf();
3248 
3249 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3250 	mtspr(SPR_MAS0, mas0);
3251 	__asm __volatile("isync; tlbre");
3252 
3253 	mas1 = mfspr(SPR_MAS1);
3254 	mas2 = mfspr(SPR_MAS2);
3255 	mas3 = mfspr(SPR_MAS3);
3256 	mas7 = mfspr(SPR_MAS7);
3257 
3258 	tlb1[0].mas1 = mas1;
3259 	tlb1[0].mas2 = mfspr(SPR_MAS2);
3260 	tlb1[0].mas3 = mas3;
3261 	tlb1[0].mas7 = mas7;
3262 	tlb1[0].virt = mas2 & MAS2_EPN_MASK;
3263 	tlb1[0].phys =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
3264 	    (mas3 & MAS3_RPN);
3265 
3266 	kernload = tlb1[0].phys;
3267 
3268 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3269 	tlb1[0].size = (tsz > 0) ? tsize2size(tsz) : 0;
3270 	kernsize += tlb1[0].size;
3271 
3272 #ifdef SMP
3273 	bp_ntlb1s = tlb1_idx;
3274 #endif
3275 
3276 	/* Purge the remaining entries */
3277 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3278 		tlb1_write_entry(i);
3279 
3280 	/* Setup TLB miss defaults */
3281 	set_mas4_defaults();
3282 }
3283 
3284 vm_offset_t
3285 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3286 {
3287 	vm_paddr_t pa_base;
3288 	vm_offset_t va, sz;
3289 	int i;
3290 
3291 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3292 
3293 	for (i = 0; i < tlb1_idx; i++) {
3294 		if (!(tlb1[i].mas1 & MAS1_VALID))
3295 			continue;
3296 		if (pa >= tlb1[i].phys && (pa + size) <=
3297 		    (tlb1[i].phys + tlb1[i].size))
3298 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3299 	}
3300 
3301 	pa_base = rounddown(pa, PAGE_SIZE);
3302 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3303 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3304 	va = tlb1_map_base + (pa - pa_base);
3305 
3306 	do {
3307 		sz = 1 << (ilog2(size) & ~1);
3308 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3309 		size -= sz;
3310 		pa_base += sz;
3311 		tlb1_map_base += sz;
3312 	} while (size > 0);
3313 
3314 #ifdef SMP
3315 	bp_ntlb1s = tlb1_idx;
3316 #endif
3317 
3318 	return (va);
3319 }
3320 
3321 /*
3322  * Setup MAS4 defaults.
3323  * These values are loaded to MAS0-2 on a TLB miss.
3324  */
3325 static void
3326 set_mas4_defaults(void)
3327 {
3328 	uint32_t mas4;
3329 
3330 	/* Defaults: TLB0, PID0, TSIZED=4K */
3331 	mas4 = MAS4_TLBSELD0;
3332 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3333 #ifdef SMP
3334 	mas4 |= MAS4_MD;
3335 #endif
3336 	mtspr(SPR_MAS4, mas4);
3337 	__asm __volatile("isync");
3338 }
3339 
3340 /*
3341  * Print out contents of the MAS registers for each TLB1 entry
3342  */
3343 void
3344 tlb1_print_tlbentries(void)
3345 {
3346 	uint32_t mas0, mas1, mas2, mas3, mas7;
3347 	int i;
3348 
3349 	debugf("TLB1 entries:\n");
3350 	for (i = 0; i < TLB1_ENTRIES; i++) {
3351 
3352 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3353 		mtspr(SPR_MAS0, mas0);
3354 
3355 		__asm __volatile("isync; tlbre");
3356 
3357 		mas1 = mfspr(SPR_MAS1);
3358 		mas2 = mfspr(SPR_MAS2);
3359 		mas3 = mfspr(SPR_MAS3);
3360 		mas7 = mfspr(SPR_MAS7);
3361 
3362 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3363 	}
3364 }
3365 
3366 /*
3367  * Print out contents of the in-ram tlb1 table.
3368  */
3369 void
3370 tlb1_print_entries(void)
3371 {
3372 	int i;
3373 
3374 	debugf("tlb1[] table entries:\n");
3375 	for (i = 0; i < TLB1_ENTRIES; i++)
3376 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3,
3377 		    tlb1[i].mas7);
3378 }
3379 
3380 /*
3381  * Return 0 if the physical IO range is encompassed by one of the
3382  * the TLB1 entries, otherwise return related error code.
3383  */
3384 static int
3385 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3386 {
3387 	uint32_t prot;
3388 	vm_paddr_t pa_start;
3389 	vm_paddr_t pa_end;
3390 	unsigned int entry_tsize;
3391 	vm_size_t entry_size;
3392 
3393 	*va = (vm_offset_t)NULL;
3394 
3395 	/* Skip invalid entries */
3396 	if (!(tlb1[i].mas1 & MAS1_VALID))
3397 		return (EINVAL);
3398 
3399 	/*
3400 	 * The entry must be cache-inhibited, guarded, and r/w
3401 	 * so it can function as an i/o page
3402 	 */
3403 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3404 	if (prot != (MAS2_I | MAS2_G))
3405 		return (EPERM);
3406 
3407 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3408 	if (prot != (MAS3_SR | MAS3_SW))
3409 		return (EPERM);
3410 
3411 	/* The address should be within the entry range. */
3412 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3413 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3414 
3415 	entry_size = tsize2size(entry_tsize);
3416 	pa_start = (((vm_paddr_t)tlb1[i].mas7 & MAS7_RPN) << 32) |
3417 	    (tlb1[i].mas3 & MAS3_RPN);
3418 	pa_end = pa_start + entry_size;
3419 
3420 	if ((pa < pa_start) || ((pa + size) > pa_end))
3421 		return (ERANGE);
3422 
3423 	/* Return virtual address of this mapping. */
3424 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3425 	return (0);
3426 }
3427