xref: /freebsd/sys/powerpc/booke/pmap.c (revision ec0e626bafb335b30c499d06066997f54b10c092)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/param.h>
55 #include <sys/conf.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
66 #include <sys/lock.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
70 #include <sys/smp.h>
71 #include <sys/vmmeter.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
82 #include <vm/uma.h>
83 
84 #include <machine/cpu.h>
85 #include <machine/pcb.h>
86 #include <machine/platform.h>
87 
88 #include <machine/tlb.h>
89 #include <machine/spr.h>
90 #include <machine/md_var.h>
91 #include <machine/mmuvar.h>
92 #include <machine/pmap.h>
93 #include <machine/pte.h>
94 
95 #include "mmu_if.h"
96 
97 #ifdef  DEBUG
98 #define debugf(fmt, args...) printf(fmt, ##args)
99 #else
100 #define debugf(fmt, args...)
101 #endif
102 
103 #define TODO			panic("%s: not implemented", __func__);
104 
105 extern unsigned char _etext[];
106 extern unsigned char _end[];
107 
108 extern uint32_t *bootinfo;
109 
110 #ifdef SMP
111 extern uint32_t bp_ntlb1s;
112 #endif
113 
114 vm_paddr_t kernload;
115 vm_offset_t kernstart;
116 vm_size_t kernsize;
117 
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
121 
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
127 
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
131 
132 static struct mtx tlbivax_mutex;
133 
134 /*
135  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136  * by idle thred only, no lock required.
137  */
138 static vm_offset_t zero_page_idle_va;
139 
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
144 
145 /**************************************************************************/
146 /* PMAP */
147 /**************************************************************************/
148 
149 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, u_int flags, int8_t psind);
151 
152 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
154 
155 /*
156  * If user pmap is processed with mmu_booke_remove and the resident count
157  * drops to 0, there are no more pages to remove, so we need not continue.
158  */
159 #define PMAP_REMOVE_DONE(pmap) \
160 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
161 
162 extern void tid_flush(tlbtid_t tid, int tlb0_ways, int tlb0_entries_per_way);
163 extern int elf32_nxstack;
164 
165 /**************************************************************************/
166 /* TLB and TID handling */
167 /**************************************************************************/
168 
169 /* Translation ID busy table */
170 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
171 
172 /*
173  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
174  * core revisions and should be read from h/w registers during early config.
175  */
176 uint32_t tlb0_entries;
177 uint32_t tlb0_ways;
178 uint32_t tlb0_entries_per_way;
179 
180 #define TLB0_ENTRIES		(tlb0_entries)
181 #define TLB0_WAYS		(tlb0_ways)
182 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
183 
184 #define TLB1_ENTRIES 16
185 
186 /* In-ram copy of the TLB1 */
187 static tlb_entry_t tlb1[TLB1_ENTRIES];
188 
189 /* Next free entry in the TLB1 */
190 static unsigned int tlb1_idx;
191 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
192 
193 static tlbtid_t tid_alloc(struct pmap *);
194 
195 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
196 
197 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
198 static void tlb1_write_entry(unsigned int);
199 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
200 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
201 
202 static vm_size_t tsize2size(unsigned int);
203 static unsigned int size2tsize(vm_size_t);
204 static unsigned int ilog2(unsigned int);
205 
206 static void set_mas4_defaults(void);
207 
208 static inline void tlb0_flush_entry(vm_offset_t);
209 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
210 
211 /**************************************************************************/
212 /* Page table management */
213 /**************************************************************************/
214 
215 static struct rwlock_padalign pvh_global_lock;
216 
217 /* Data for the pv entry allocation mechanism */
218 static uma_zone_t pvzone;
219 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
220 
221 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
222 
223 #ifndef PMAP_SHPGPERPROC
224 #define PMAP_SHPGPERPROC	200
225 #endif
226 
227 static void ptbl_init(void);
228 static struct ptbl_buf *ptbl_buf_alloc(void);
229 static void ptbl_buf_free(struct ptbl_buf *);
230 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
231 
232 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
233 static void ptbl_free(mmu_t, pmap_t, unsigned int);
234 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
235 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
236 
237 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
238 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
239 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
240 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
241 
242 static pv_entry_t pv_alloc(void);
243 static void pv_free(pv_entry_t);
244 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
245 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
246 
247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
248 #define PTBL_BUFS		(128 * 16)
249 
250 struct ptbl_buf {
251 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
252 	vm_offset_t kva;		/* va of mapping */
253 };
254 
255 /* ptbl free list and a lock used for access synchronization. */
256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
257 static struct mtx ptbl_buf_freelist_lock;
258 
259 /* Base address of kva space allocated fot ptbl bufs. */
260 static vm_offset_t ptbl_buf_pool_vabase;
261 
262 /* Pointer to ptbl_buf structures. */
263 static struct ptbl_buf *ptbl_bufs;
264 
265 void pmap_bootstrap_ap(volatile uint32_t *);
266 
267 /*
268  * Kernel MMU interface
269  */
270 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
271 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
272     vm_size_t, vm_offset_t);
273 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
274 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
275     vm_offset_t, vm_page_t *, vm_offset_t, int);
276 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
277     vm_prot_t, u_int flags, int8_t psind);
278 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
279     vm_page_t, vm_prot_t);
280 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t);
282 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
283 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
284     vm_prot_t);
285 static void		mmu_booke_init(mmu_t);
286 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
287 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
288 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
289 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
290 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
291     int);
292 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
293     vm_paddr_t *);
294 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
295     vm_object_t, vm_pindex_t, vm_size_t);
296 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
297 static void		mmu_booke_page_init(mmu_t, vm_page_t);
298 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
299 static void		mmu_booke_pinit(mmu_t, pmap_t);
300 static void		mmu_booke_pinit0(mmu_t, pmap_t);
301 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
302     vm_prot_t);
303 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
304 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
305 static void		mmu_booke_release(mmu_t, pmap_t);
306 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
308 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
309 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
310 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
311 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
312 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
313 static void		mmu_booke_activate(mmu_t, struct thread *);
314 static void		mmu_booke_deactivate(mmu_t, struct thread *);
315 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
316 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
317 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
318 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
319 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
320 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
321 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
322 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
323 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
324 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
325     vm_size_t);
326 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
327     void **);
328 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
329     void *);
330 static void		mmu_booke_scan_init(mmu_t);
331 
332 static mmu_method_t mmu_booke_methods[] = {
333 	/* pmap dispatcher interface */
334 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
335 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
336 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
337 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
338 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
339 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
340 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
341 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
342 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
343 	MMUMETHOD(mmu_init,		mmu_booke_init),
344 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
345 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
346 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
347 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
348 	MMUMETHOD(mmu_map,		mmu_booke_map),
349 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
350 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
351 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
352 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
353 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
354 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
355 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
356 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
357 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
358 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
359 	MMUMETHOD(mmu_release,		mmu_booke_release),
360 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
361 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
362 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
363 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
364 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
365 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
366 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
367 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
368 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
369 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
370 
371 	/* Internal interfaces */
372 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
373 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
374 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
375 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
376 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
377 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
378 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
379 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
380 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
381 
382 	/* dumpsys() support */
383 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
384 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
385 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
386 
387 	{ 0, 0 }
388 };
389 
390 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
391 
392 static __inline uint32_t
393 tlb_calc_wimg(vm_offset_t pa, vm_memattr_t ma)
394 {
395 	uint32_t attrib;
396 	int i;
397 
398 	if (ma != VM_MEMATTR_DEFAULT) {
399 		switch (ma) {
400 		case VM_MEMATTR_UNCACHEABLE:
401 			return (PTE_I | PTE_G);
402 		case VM_MEMATTR_WRITE_COMBINING:
403 		case VM_MEMATTR_WRITE_BACK:
404 		case VM_MEMATTR_PREFETCHABLE:
405 			return (PTE_I);
406 		case VM_MEMATTR_WRITE_THROUGH:
407 			return (PTE_W | PTE_M);
408 		}
409 	}
410 
411 	/*
412 	 * Assume the page is cache inhibited and access is guarded unless
413 	 * it's in our available memory array.
414 	 */
415 	attrib = _TLB_ENTRY_IO;
416 	for (i = 0; i < physmem_regions_sz; i++) {
417 		if ((pa >= physmem_regions[i].mr_start) &&
418 		    (pa < (physmem_regions[i].mr_start +
419 		     physmem_regions[i].mr_size))) {
420 			attrib = _TLB_ENTRY_MEM;
421 			break;
422 		}
423 	}
424 
425 	return (attrib);
426 }
427 
428 static inline void
429 tlb_miss_lock(void)
430 {
431 #ifdef SMP
432 	struct pcpu *pc;
433 
434 	if (!smp_started)
435 		return;
436 
437 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
438 		if (pc != pcpup) {
439 
440 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
441 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
442 
443 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
444 			    ("tlb_miss_lock: tried to lock self"));
445 
446 			tlb_lock(pc->pc_booke_tlb_lock);
447 
448 			CTR1(KTR_PMAP, "%s: locked", __func__);
449 		}
450 	}
451 #endif
452 }
453 
454 static inline void
455 tlb_miss_unlock(void)
456 {
457 #ifdef SMP
458 	struct pcpu *pc;
459 
460 	if (!smp_started)
461 		return;
462 
463 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
464 		if (pc != pcpup) {
465 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
466 			    __func__, pc->pc_cpuid);
467 
468 			tlb_unlock(pc->pc_booke_tlb_lock);
469 
470 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
471 		}
472 	}
473 #endif
474 }
475 
476 /* Return number of entries in TLB0. */
477 static __inline void
478 tlb0_get_tlbconf(void)
479 {
480 	uint32_t tlb0_cfg;
481 
482 	tlb0_cfg = mfspr(SPR_TLB0CFG);
483 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
484 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
485 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
486 }
487 
488 /* Initialize pool of kva ptbl buffers. */
489 static void
490 ptbl_init(void)
491 {
492 	int i;
493 
494 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
495 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
496 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
497 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
498 
499 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
500 	TAILQ_INIT(&ptbl_buf_freelist);
501 
502 	for (i = 0; i < PTBL_BUFS; i++) {
503 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
504 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
505 	}
506 }
507 
508 /* Get a ptbl_buf from the freelist. */
509 static struct ptbl_buf *
510 ptbl_buf_alloc(void)
511 {
512 	struct ptbl_buf *buf;
513 
514 	mtx_lock(&ptbl_buf_freelist_lock);
515 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
516 	if (buf != NULL)
517 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
518 	mtx_unlock(&ptbl_buf_freelist_lock);
519 
520 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
521 
522 	return (buf);
523 }
524 
525 /* Return ptbl buff to free pool. */
526 static void
527 ptbl_buf_free(struct ptbl_buf *buf)
528 {
529 
530 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
531 
532 	mtx_lock(&ptbl_buf_freelist_lock);
533 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
534 	mtx_unlock(&ptbl_buf_freelist_lock);
535 }
536 
537 /*
538  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
539  */
540 static void
541 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
542 {
543 	struct ptbl_buf *pbuf;
544 
545 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
546 
547 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
548 
549 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
550 		if (pbuf->kva == (vm_offset_t)ptbl) {
551 			/* Remove from pmap ptbl buf list. */
552 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
553 
554 			/* Free corresponding ptbl buf. */
555 			ptbl_buf_free(pbuf);
556 			break;
557 		}
558 }
559 
560 /* Allocate page table. */
561 static pte_t *
562 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
563 {
564 	vm_page_t mtbl[PTBL_PAGES];
565 	vm_page_t m;
566 	struct ptbl_buf *pbuf;
567 	unsigned int pidx;
568 	pte_t *ptbl;
569 	int i, j;
570 
571 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
572 	    (pmap == kernel_pmap), pdir_idx);
573 
574 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
575 	    ("ptbl_alloc: invalid pdir_idx"));
576 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
577 	    ("pte_alloc: valid ptbl entry exists!"));
578 
579 	pbuf = ptbl_buf_alloc();
580 	if (pbuf == NULL)
581 		panic("pte_alloc: couldn't alloc kernel virtual memory");
582 
583 	ptbl = (pte_t *)pbuf->kva;
584 
585 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
586 
587 	/* Allocate ptbl pages, this will sleep! */
588 	for (i = 0; i < PTBL_PAGES; i++) {
589 		pidx = (PTBL_PAGES * pdir_idx) + i;
590 		while ((m = vm_page_alloc(NULL, pidx,
591 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
592 			PMAP_UNLOCK(pmap);
593 			rw_wunlock(&pvh_global_lock);
594 			if (nosleep) {
595 				ptbl_free_pmap_ptbl(pmap, ptbl);
596 				for (j = 0; j < i; j++)
597 					vm_page_free(mtbl[j]);
598 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
599 				return (NULL);
600 			}
601 			VM_WAIT;
602 			rw_wlock(&pvh_global_lock);
603 			PMAP_LOCK(pmap);
604 		}
605 		mtbl[i] = m;
606 	}
607 
608 	/* Map allocated pages into kernel_pmap. */
609 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
610 
611 	/* Zero whole ptbl. */
612 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
613 
614 	/* Add pbuf to the pmap ptbl bufs list. */
615 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
616 
617 	return (ptbl);
618 }
619 
620 /* Free ptbl pages and invalidate pdir entry. */
621 static void
622 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
623 {
624 	pte_t *ptbl;
625 	vm_paddr_t pa;
626 	vm_offset_t va;
627 	vm_page_t m;
628 	int i;
629 
630 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
631 	    (pmap == kernel_pmap), pdir_idx);
632 
633 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
634 	    ("ptbl_free: invalid pdir_idx"));
635 
636 	ptbl = pmap->pm_pdir[pdir_idx];
637 
638 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
639 
640 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
641 
642 	/*
643 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
644 	 * don't attempt to look up the page tables we are releasing.
645 	 */
646 	mtx_lock_spin(&tlbivax_mutex);
647 	tlb_miss_lock();
648 
649 	pmap->pm_pdir[pdir_idx] = NULL;
650 
651 	tlb_miss_unlock();
652 	mtx_unlock_spin(&tlbivax_mutex);
653 
654 	for (i = 0; i < PTBL_PAGES; i++) {
655 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
656 		pa = pte_vatopa(mmu, kernel_pmap, va);
657 		m = PHYS_TO_VM_PAGE(pa);
658 		vm_page_free_zero(m);
659 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
660 		mmu_booke_kremove(mmu, va);
661 	}
662 
663 	ptbl_free_pmap_ptbl(pmap, ptbl);
664 }
665 
666 /*
667  * Decrement ptbl pages hold count and attempt to free ptbl pages.
668  * Called when removing pte entry from ptbl.
669  *
670  * Return 1 if ptbl pages were freed.
671  */
672 static int
673 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
674 {
675 	pte_t *ptbl;
676 	vm_paddr_t pa;
677 	vm_page_t m;
678 	int i;
679 
680 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
681 	    (pmap == kernel_pmap), pdir_idx);
682 
683 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
684 	    ("ptbl_unhold: invalid pdir_idx"));
685 	KASSERT((pmap != kernel_pmap),
686 	    ("ptbl_unhold: unholding kernel ptbl!"));
687 
688 	ptbl = pmap->pm_pdir[pdir_idx];
689 
690 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
691 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
692 	    ("ptbl_unhold: non kva ptbl"));
693 
694 	/* decrement hold count */
695 	for (i = 0; i < PTBL_PAGES; i++) {
696 		pa = pte_vatopa(mmu, kernel_pmap,
697 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
698 		m = PHYS_TO_VM_PAGE(pa);
699 		m->wire_count--;
700 	}
701 
702 	/*
703 	 * Free ptbl pages if there are no pte etries in this ptbl.
704 	 * wire_count has the same value for all ptbl pages, so check the last
705 	 * page.
706 	 */
707 	if (m->wire_count == 0) {
708 		ptbl_free(mmu, pmap, pdir_idx);
709 
710 		//debugf("ptbl_unhold: e (freed ptbl)\n");
711 		return (1);
712 	}
713 
714 	return (0);
715 }
716 
717 /*
718  * Increment hold count for ptbl pages. This routine is used when a new pte
719  * entry is being inserted into the ptbl.
720  */
721 static void
722 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
723 {
724 	vm_paddr_t pa;
725 	pte_t *ptbl;
726 	vm_page_t m;
727 	int i;
728 
729 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
730 	    pdir_idx);
731 
732 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
733 	    ("ptbl_hold: invalid pdir_idx"));
734 	KASSERT((pmap != kernel_pmap),
735 	    ("ptbl_hold: holding kernel ptbl!"));
736 
737 	ptbl = pmap->pm_pdir[pdir_idx];
738 
739 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
740 
741 	for (i = 0; i < PTBL_PAGES; i++) {
742 		pa = pte_vatopa(mmu, kernel_pmap,
743 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
744 		m = PHYS_TO_VM_PAGE(pa);
745 		m->wire_count++;
746 	}
747 }
748 
749 /* Allocate pv_entry structure. */
750 pv_entry_t
751 pv_alloc(void)
752 {
753 	pv_entry_t pv;
754 
755 	pv_entry_count++;
756 	if (pv_entry_count > pv_entry_high_water)
757 		pagedaemon_wakeup();
758 	pv = uma_zalloc(pvzone, M_NOWAIT);
759 
760 	return (pv);
761 }
762 
763 /* Free pv_entry structure. */
764 static __inline void
765 pv_free(pv_entry_t pve)
766 {
767 
768 	pv_entry_count--;
769 	uma_zfree(pvzone, pve);
770 }
771 
772 
773 /* Allocate and initialize pv_entry structure. */
774 static void
775 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
776 {
777 	pv_entry_t pve;
778 
779 	//int su = (pmap == kernel_pmap);
780 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
781 	//	(u_int32_t)pmap, va, (u_int32_t)m);
782 
783 	pve = pv_alloc();
784 	if (pve == NULL)
785 		panic("pv_insert: no pv entries!");
786 
787 	pve->pv_pmap = pmap;
788 	pve->pv_va = va;
789 
790 	/* add to pv_list */
791 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
792 	rw_assert(&pvh_global_lock, RA_WLOCKED);
793 
794 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
795 
796 	//debugf("pv_insert: e\n");
797 }
798 
799 /* Destroy pv entry. */
800 static void
801 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
802 {
803 	pv_entry_t pve;
804 
805 	//int su = (pmap == kernel_pmap);
806 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
807 
808 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
809 	rw_assert(&pvh_global_lock, RA_WLOCKED);
810 
811 	/* find pv entry */
812 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
813 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
814 			/* remove from pv_list */
815 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
816 			if (TAILQ_EMPTY(&m->md.pv_list))
817 				vm_page_aflag_clear(m, PGA_WRITEABLE);
818 
819 			/* free pv entry struct */
820 			pv_free(pve);
821 			break;
822 		}
823 	}
824 
825 	//debugf("pv_remove: e\n");
826 }
827 
828 /*
829  * Clean pte entry, try to free page table page if requested.
830  *
831  * Return 1 if ptbl pages were freed, otherwise return 0.
832  */
833 static int
834 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
835 {
836 	unsigned int pdir_idx = PDIR_IDX(va);
837 	unsigned int ptbl_idx = PTBL_IDX(va);
838 	vm_page_t m;
839 	pte_t *ptbl;
840 	pte_t *pte;
841 
842 	//int su = (pmap == kernel_pmap);
843 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
844 	//		su, (u_int32_t)pmap, va, flags);
845 
846 	ptbl = pmap->pm_pdir[pdir_idx];
847 	KASSERT(ptbl, ("pte_remove: null ptbl"));
848 
849 	pte = &ptbl[ptbl_idx];
850 
851 	if (pte == NULL || !PTE_ISVALID(pte))
852 		return (0);
853 
854 	if (PTE_ISWIRED(pte))
855 		pmap->pm_stats.wired_count--;
856 
857 	/* Handle managed entry. */
858 	if (PTE_ISMANAGED(pte)) {
859 		/* Get vm_page_t for mapped pte. */
860 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
861 
862 		if (PTE_ISMODIFIED(pte))
863 			vm_page_dirty(m);
864 
865 		if (PTE_ISREFERENCED(pte))
866 			vm_page_aflag_set(m, PGA_REFERENCED);
867 
868 		pv_remove(pmap, va, m);
869 	}
870 
871 	mtx_lock_spin(&tlbivax_mutex);
872 	tlb_miss_lock();
873 
874 	tlb0_flush_entry(va);
875 	pte->flags = 0;
876 	pte->rpn = 0;
877 
878 	tlb_miss_unlock();
879 	mtx_unlock_spin(&tlbivax_mutex);
880 
881 	pmap->pm_stats.resident_count--;
882 
883 	if (flags & PTBL_UNHOLD) {
884 		//debugf("pte_remove: e (unhold)\n");
885 		return (ptbl_unhold(mmu, pmap, pdir_idx));
886 	}
887 
888 	//debugf("pte_remove: e\n");
889 	return (0);
890 }
891 
892 /*
893  * Insert PTE for a given page and virtual address.
894  */
895 static int
896 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
897     boolean_t nosleep)
898 {
899 	unsigned int pdir_idx = PDIR_IDX(va);
900 	unsigned int ptbl_idx = PTBL_IDX(va);
901 	pte_t *ptbl, *pte;
902 
903 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
904 	    pmap == kernel_pmap, pmap, va);
905 
906 	/* Get the page table pointer. */
907 	ptbl = pmap->pm_pdir[pdir_idx];
908 
909 	if (ptbl == NULL) {
910 		/* Allocate page table pages. */
911 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
912 		if (ptbl == NULL) {
913 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
914 			return (ENOMEM);
915 		}
916 	} else {
917 		/*
918 		 * Check if there is valid mapping for requested
919 		 * va, if there is, remove it.
920 		 */
921 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
922 		if (PTE_ISVALID(pte)) {
923 			pte_remove(mmu, pmap, va, PTBL_HOLD);
924 		} else {
925 			/*
926 			 * pte is not used, increment hold count
927 			 * for ptbl pages.
928 			 */
929 			if (pmap != kernel_pmap)
930 				ptbl_hold(mmu, pmap, pdir_idx);
931 		}
932 	}
933 
934 	/*
935 	 * Insert pv_entry into pv_list for mapped page if part of managed
936 	 * memory.
937 	 */
938 	if ((m->oflags & VPO_UNMANAGED) == 0) {
939 		flags |= PTE_MANAGED;
940 
941 		/* Create and insert pv entry. */
942 		pv_insert(pmap, va, m);
943 	}
944 
945 	pmap->pm_stats.resident_count++;
946 
947 	mtx_lock_spin(&tlbivax_mutex);
948 	tlb_miss_lock();
949 
950 	tlb0_flush_entry(va);
951 	if (pmap->pm_pdir[pdir_idx] == NULL) {
952 		/*
953 		 * If we just allocated a new page table, hook it in
954 		 * the pdir.
955 		 */
956 		pmap->pm_pdir[pdir_idx] = ptbl;
957 	}
958 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
959 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
960 	pte->flags |= (PTE_VALID | flags);
961 
962 	tlb_miss_unlock();
963 	mtx_unlock_spin(&tlbivax_mutex);
964 	return (0);
965 }
966 
967 /* Return the pa for the given pmap/va. */
968 static vm_paddr_t
969 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
970 {
971 	vm_paddr_t pa = 0;
972 	pte_t *pte;
973 
974 	pte = pte_find(mmu, pmap, va);
975 	if ((pte != NULL) && PTE_ISVALID(pte))
976 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
977 	return (pa);
978 }
979 
980 /* Get a pointer to a PTE in a page table. */
981 static pte_t *
982 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
983 {
984 	unsigned int pdir_idx = PDIR_IDX(va);
985 	unsigned int ptbl_idx = PTBL_IDX(va);
986 
987 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
988 
989 	if (pmap->pm_pdir[pdir_idx])
990 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
991 
992 	return (NULL);
993 }
994 
995 /**************************************************************************/
996 /* PMAP related */
997 /**************************************************************************/
998 
999 /*
1000  * This is called during booke_init, before the system is really initialized.
1001  */
1002 static void
1003 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1004 {
1005 	vm_offset_t phys_kernelend;
1006 	struct mem_region *mp, *mp1;
1007 	int cnt, i, j;
1008 	u_int s, e, sz;
1009 	u_int phys_avail_count;
1010 	vm_size_t physsz, hwphyssz, kstack0_sz;
1011 	vm_offset_t kernel_pdir, kstack0, va;
1012 	vm_paddr_t kstack0_phys;
1013 	void *dpcpu;
1014 	pte_t *pte;
1015 
1016 	debugf("mmu_booke_bootstrap: entered\n");
1017 
1018 	/* Set interesting system properties */
1019 	hw_direct_map = 0;
1020 	elf32_nxstack = 1;
1021 
1022 	/* Initialize invalidation mutex */
1023 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1024 
1025 	/* Read TLB0 size and associativity. */
1026 	tlb0_get_tlbconf();
1027 
1028 	/*
1029 	 * Align kernel start and end address (kernel image).
1030 	 * Note that kernel end does not necessarily relate to kernsize.
1031 	 * kernsize is the size of the kernel that is actually mapped.
1032 	 * Also note that "start - 1" is deliberate. With SMP, the
1033 	 * entry point is exactly a page from the actual load address.
1034 	 * As such, trunc_page() has no effect and we're off by a page.
1035 	 * Since we always have the ELF header between the load address
1036 	 * and the entry point, we can safely subtract 1 to compensate.
1037 	 */
1038 	kernstart = trunc_page(start - 1);
1039 	data_start = round_page(kernelend);
1040 	data_end = data_start;
1041 
1042 	/*
1043 	 * Addresses of preloaded modules (like file systems) use
1044 	 * physical addresses. Make sure we relocate those into
1045 	 * virtual addresses.
1046 	 */
1047 	preload_addr_relocate = kernstart - kernload;
1048 
1049 	/* Allocate the dynamic per-cpu area. */
1050 	dpcpu = (void *)data_end;
1051 	data_end += DPCPU_SIZE;
1052 
1053 	/* Allocate space for the message buffer. */
1054 	msgbufp = (struct msgbuf *)data_end;
1055 	data_end += msgbufsize;
1056 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1057 	    data_end);
1058 
1059 	data_end = round_page(data_end);
1060 
1061 	/* Allocate space for ptbl_bufs. */
1062 	ptbl_bufs = (struct ptbl_buf *)data_end;
1063 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1064 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1065 	    data_end);
1066 
1067 	data_end = round_page(data_end);
1068 
1069 	/* Allocate PTE tables for kernel KVA. */
1070 	kernel_pdir = data_end;
1071 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1072 	    PDIR_SIZE - 1) / PDIR_SIZE;
1073 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1074 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1075 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1076 
1077 	debugf(" data_end: 0x%08x\n", data_end);
1078 	if (data_end - kernstart > kernsize) {
1079 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1080 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1081 	}
1082 	data_end = kernstart + kernsize;
1083 	debugf(" updated data_end: 0x%08x\n", data_end);
1084 
1085 	/*
1086 	 * Clear the structures - note we can only do it safely after the
1087 	 * possible additional TLB1 translations are in place (above) so that
1088 	 * all range up to the currently calculated 'data_end' is covered.
1089 	 */
1090 	dpcpu_init(dpcpu, 0);
1091 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1092 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1093 
1094 	/*******************************************************/
1095 	/* Set the start and end of kva. */
1096 	/*******************************************************/
1097 	virtual_avail = round_page(data_end);
1098 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1099 
1100 	/* Allocate KVA space for page zero/copy operations. */
1101 	zero_page_va = virtual_avail;
1102 	virtual_avail += PAGE_SIZE;
1103 	zero_page_idle_va = virtual_avail;
1104 	virtual_avail += PAGE_SIZE;
1105 	copy_page_src_va = virtual_avail;
1106 	virtual_avail += PAGE_SIZE;
1107 	copy_page_dst_va = virtual_avail;
1108 	virtual_avail += PAGE_SIZE;
1109 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1110 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1111 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1112 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1113 
1114 	/* Initialize page zero/copy mutexes. */
1115 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1116 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1117 
1118 	/* Allocate KVA space for ptbl bufs. */
1119 	ptbl_buf_pool_vabase = virtual_avail;
1120 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1121 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1122 	    ptbl_buf_pool_vabase, virtual_avail);
1123 
1124 	/* Calculate corresponding physical addresses for the kernel region. */
1125 	phys_kernelend = kernload + kernsize;
1126 	debugf("kernel image and allocated data:\n");
1127 	debugf(" kernload    = 0x%08x\n", kernload);
1128 	debugf(" kernstart   = 0x%08x\n", kernstart);
1129 	debugf(" kernsize    = 0x%08x\n", kernsize);
1130 
1131 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1132 		panic("mmu_booke_bootstrap: phys_avail too small");
1133 
1134 	/*
1135 	 * Remove kernel physical address range from avail regions list. Page
1136 	 * align all regions.  Non-page aligned memory isn't very interesting
1137 	 * to us.  Also, sort the entries for ascending addresses.
1138 	 */
1139 
1140 	/* Retrieve phys/avail mem regions */
1141 	mem_regions(&physmem_regions, &physmem_regions_sz,
1142 	    &availmem_regions, &availmem_regions_sz);
1143 	sz = 0;
1144 	cnt = availmem_regions_sz;
1145 	debugf("processing avail regions:\n");
1146 	for (mp = availmem_regions; mp->mr_size; mp++) {
1147 		s = mp->mr_start;
1148 		e = mp->mr_start + mp->mr_size;
1149 		debugf(" %08x-%08x -> ", s, e);
1150 		/* Check whether this region holds all of the kernel. */
1151 		if (s < kernload && e > phys_kernelend) {
1152 			availmem_regions[cnt].mr_start = phys_kernelend;
1153 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1154 			e = kernload;
1155 		}
1156 		/* Look whether this regions starts within the kernel. */
1157 		if (s >= kernload && s < phys_kernelend) {
1158 			if (e <= phys_kernelend)
1159 				goto empty;
1160 			s = phys_kernelend;
1161 		}
1162 		/* Now look whether this region ends within the kernel. */
1163 		if (e > kernload && e <= phys_kernelend) {
1164 			if (s >= kernload)
1165 				goto empty;
1166 			e = kernload;
1167 		}
1168 		/* Now page align the start and size of the region. */
1169 		s = round_page(s);
1170 		e = trunc_page(e);
1171 		if (e < s)
1172 			e = s;
1173 		sz = e - s;
1174 		debugf("%08x-%08x = %x\n", s, e, sz);
1175 
1176 		/* Check whether some memory is left here. */
1177 		if (sz == 0) {
1178 		empty:
1179 			memmove(mp, mp + 1,
1180 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1181 			cnt--;
1182 			mp--;
1183 			continue;
1184 		}
1185 
1186 		/* Do an insertion sort. */
1187 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1188 			if (s < mp1->mr_start)
1189 				break;
1190 		if (mp1 < mp) {
1191 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1192 			mp1->mr_start = s;
1193 			mp1->mr_size = sz;
1194 		} else {
1195 			mp->mr_start = s;
1196 			mp->mr_size = sz;
1197 		}
1198 	}
1199 	availmem_regions_sz = cnt;
1200 
1201 	/*******************************************************/
1202 	/* Steal physical memory for kernel stack from the end */
1203 	/* of the first avail region                           */
1204 	/*******************************************************/
1205 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1206 	kstack0_phys = availmem_regions[0].mr_start +
1207 	    availmem_regions[0].mr_size;
1208 	kstack0_phys -= kstack0_sz;
1209 	availmem_regions[0].mr_size -= kstack0_sz;
1210 
1211 	/*******************************************************/
1212 	/* Fill in phys_avail table, based on availmem_regions */
1213 	/*******************************************************/
1214 	phys_avail_count = 0;
1215 	physsz = 0;
1216 	hwphyssz = 0;
1217 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1218 
1219 	debugf("fill in phys_avail:\n");
1220 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1221 
1222 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1223 		    availmem_regions[i].mr_start,
1224 		    availmem_regions[i].mr_start +
1225 		        availmem_regions[i].mr_size,
1226 		    availmem_regions[i].mr_size);
1227 
1228 		if (hwphyssz != 0 &&
1229 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1230 			debugf(" hw.physmem adjust\n");
1231 			if (physsz < hwphyssz) {
1232 				phys_avail[j] = availmem_regions[i].mr_start;
1233 				phys_avail[j + 1] =
1234 				    availmem_regions[i].mr_start +
1235 				    hwphyssz - physsz;
1236 				physsz = hwphyssz;
1237 				phys_avail_count++;
1238 			}
1239 			break;
1240 		}
1241 
1242 		phys_avail[j] = availmem_regions[i].mr_start;
1243 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1244 		    availmem_regions[i].mr_size;
1245 		phys_avail_count++;
1246 		physsz += availmem_regions[i].mr_size;
1247 	}
1248 	physmem = btoc(physsz);
1249 
1250 	/* Calculate the last available physical address. */
1251 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1252 		;
1253 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1254 
1255 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1256 	debugf("phys_avail_count = %d\n", phys_avail_count);
1257 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1258 	    physmem);
1259 
1260 	/*******************************************************/
1261 	/* Initialize (statically allocated) kernel pmap. */
1262 	/*******************************************************/
1263 	PMAP_LOCK_INIT(kernel_pmap);
1264 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1265 
1266 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1267 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1268 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1269 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1270 
1271 	/* Initialize kernel pdir */
1272 	for (i = 0; i < kernel_ptbls; i++)
1273 		kernel_pmap->pm_pdir[kptbl_min + i] =
1274 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1275 
1276 	for (i = 0; i < MAXCPU; i++) {
1277 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1278 
1279 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1280 		tidbusy[i][0] = kernel_pmap;
1281 	}
1282 
1283 	/*
1284 	 * Fill in PTEs covering kernel code and data. They are not required
1285 	 * for address translation, as this area is covered by static TLB1
1286 	 * entries, but for pte_vatopa() to work correctly with kernel area
1287 	 * addresses.
1288 	 */
1289 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1290 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1291 		pte->rpn = kernload + (va - kernstart);
1292 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1293 		    PTE_VALID;
1294 	}
1295 	/* Mark kernel_pmap active on all CPUs */
1296 	CPU_FILL(&kernel_pmap->pm_active);
1297 
1298  	/*
1299 	 * Initialize the global pv list lock.
1300 	 */
1301 	rw_init(&pvh_global_lock, "pmap pv global");
1302 
1303 	/*******************************************************/
1304 	/* Final setup */
1305 	/*******************************************************/
1306 
1307 	/* Enter kstack0 into kernel map, provide guard page */
1308 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1309 	thread0.td_kstack = kstack0;
1310 	thread0.td_kstack_pages = KSTACK_PAGES;
1311 
1312 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1313 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1314 	    kstack0_phys, kstack0_phys + kstack0_sz);
1315 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1316 
1317 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1318 	for (i = 0; i < KSTACK_PAGES; i++) {
1319 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1320 		kstack0 += PAGE_SIZE;
1321 		kstack0_phys += PAGE_SIZE;
1322 	}
1323 
1324 	pmap_bootstrapped = 1;
1325 
1326 	debugf("virtual_avail = %08x\n", virtual_avail);
1327 	debugf("virtual_end   = %08x\n", virtual_end);
1328 
1329 	debugf("mmu_booke_bootstrap: exit\n");
1330 }
1331 
1332 void
1333 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1334 {
1335 	int i;
1336 
1337 	/*
1338 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1339 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1340 	 * these values directly to (re)program AP's TLB1 hardware.
1341 	 */
1342 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1343 		/* Skip invalid entries */
1344 		if (!(tlb1[i].mas1 & MAS1_VALID))
1345 			continue;
1346 
1347 		tlb1_write_entry(i);
1348 	}
1349 
1350 	set_mas4_defaults();
1351 }
1352 
1353 /*
1354  * Get the physical page address for the given pmap/virtual address.
1355  */
1356 static vm_paddr_t
1357 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1358 {
1359 	vm_paddr_t pa;
1360 
1361 	PMAP_LOCK(pmap);
1362 	pa = pte_vatopa(mmu, pmap, va);
1363 	PMAP_UNLOCK(pmap);
1364 
1365 	return (pa);
1366 }
1367 
1368 /*
1369  * Extract the physical page address associated with the given
1370  * kernel virtual address.
1371  */
1372 static vm_paddr_t
1373 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1374 {
1375 	int i;
1376 
1377 	/* Check TLB1 mappings */
1378 	for (i = 0; i < tlb1_idx; i++) {
1379 		if (!(tlb1[i].mas1 & MAS1_VALID))
1380 			continue;
1381 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1382 			return (tlb1[i].phys + (va - tlb1[i].virt));
1383 	}
1384 
1385 	return (pte_vatopa(mmu, kernel_pmap, va));
1386 }
1387 
1388 /*
1389  * Initialize the pmap module.
1390  * Called by vm_init, to initialize any structures that the pmap
1391  * system needs to map virtual memory.
1392  */
1393 static void
1394 mmu_booke_init(mmu_t mmu)
1395 {
1396 	int shpgperproc = PMAP_SHPGPERPROC;
1397 
1398 	/*
1399 	 * Initialize the address space (zone) for the pv entries.  Set a
1400 	 * high water mark so that the system can recover from excessive
1401 	 * numbers of pv entries.
1402 	 */
1403 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1404 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1405 
1406 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1407 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1408 
1409 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1410 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1411 
1412 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1413 
1414 	/* Pre-fill pvzone with initial number of pv entries. */
1415 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1416 
1417 	/* Initialize ptbl allocation. */
1418 	ptbl_init();
1419 }
1420 
1421 /*
1422  * Map a list of wired pages into kernel virtual address space.  This is
1423  * intended for temporary mappings which do not need page modification or
1424  * references recorded.  Existing mappings in the region are overwritten.
1425  */
1426 static void
1427 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1428 {
1429 	vm_offset_t va;
1430 
1431 	va = sva;
1432 	while (count-- > 0) {
1433 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1434 		va += PAGE_SIZE;
1435 		m++;
1436 	}
1437 }
1438 
1439 /*
1440  * Remove page mappings from kernel virtual address space.  Intended for
1441  * temporary mappings entered by mmu_booke_qenter.
1442  */
1443 static void
1444 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1445 {
1446 	vm_offset_t va;
1447 
1448 	va = sva;
1449 	while (count-- > 0) {
1450 		mmu_booke_kremove(mmu, va);
1451 		va += PAGE_SIZE;
1452 	}
1453 }
1454 
1455 /*
1456  * Map a wired page into kernel virtual address space.
1457  */
1458 static void
1459 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1460 {
1461 
1462 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1463 }
1464 
1465 static void
1466 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1467 {
1468 	unsigned int pdir_idx = PDIR_IDX(va);
1469 	unsigned int ptbl_idx = PTBL_IDX(va);
1470 	uint32_t flags;
1471 	pte_t *pte;
1472 
1473 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1474 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1475 
1476 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1477 	flags |= tlb_calc_wimg(pa, ma);
1478 
1479 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1480 
1481 	mtx_lock_spin(&tlbivax_mutex);
1482 	tlb_miss_lock();
1483 
1484 	if (PTE_ISVALID(pte)) {
1485 
1486 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1487 
1488 		/* Flush entry from TLB0 */
1489 		tlb0_flush_entry(va);
1490 	}
1491 
1492 	pte->rpn = pa & ~PTE_PA_MASK;
1493 	pte->flags = flags;
1494 
1495 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1496 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1497 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1498 
1499 	/* Flush the real memory from the instruction cache. */
1500 	if ((flags & (PTE_I | PTE_G)) == 0) {
1501 		__syncicache((void *)va, PAGE_SIZE);
1502 	}
1503 
1504 	tlb_miss_unlock();
1505 	mtx_unlock_spin(&tlbivax_mutex);
1506 }
1507 
1508 /*
1509  * Remove a page from kernel page table.
1510  */
1511 static void
1512 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1513 {
1514 	unsigned int pdir_idx = PDIR_IDX(va);
1515 	unsigned int ptbl_idx = PTBL_IDX(va);
1516 	pte_t *pte;
1517 
1518 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1519 
1520 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1521 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1522 	    ("mmu_booke_kremove: invalid va"));
1523 
1524 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1525 
1526 	if (!PTE_ISVALID(pte)) {
1527 
1528 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1529 
1530 		return;
1531 	}
1532 
1533 	mtx_lock_spin(&tlbivax_mutex);
1534 	tlb_miss_lock();
1535 
1536 	/* Invalidate entry in TLB0, update PTE. */
1537 	tlb0_flush_entry(va);
1538 	pte->flags = 0;
1539 	pte->rpn = 0;
1540 
1541 	tlb_miss_unlock();
1542 	mtx_unlock_spin(&tlbivax_mutex);
1543 }
1544 
1545 /*
1546  * Initialize pmap associated with process 0.
1547  */
1548 static void
1549 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1550 {
1551 
1552 	PMAP_LOCK_INIT(pmap);
1553 	mmu_booke_pinit(mmu, pmap);
1554 	PCPU_SET(curpmap, pmap);
1555 }
1556 
1557 /*
1558  * Initialize a preallocated and zeroed pmap structure,
1559  * such as one in a vmspace structure.
1560  */
1561 static void
1562 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1563 {
1564 	int i;
1565 
1566 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1567 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1568 
1569 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1570 
1571 	for (i = 0; i < MAXCPU; i++)
1572 		pmap->pm_tid[i] = TID_NONE;
1573 	CPU_ZERO(&kernel_pmap->pm_active);
1574 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1575 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1576 	TAILQ_INIT(&pmap->pm_ptbl_list);
1577 }
1578 
1579 /*
1580  * Release any resources held by the given physical map.
1581  * Called when a pmap initialized by mmu_booke_pinit is being released.
1582  * Should only be called if the map contains no valid mappings.
1583  */
1584 static void
1585 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1586 {
1587 
1588 	KASSERT(pmap->pm_stats.resident_count == 0,
1589 	    ("pmap_release: pmap resident count %ld != 0",
1590 	    pmap->pm_stats.resident_count));
1591 }
1592 
1593 /*
1594  * Insert the given physical page at the specified virtual address in the
1595  * target physical map with the protection requested. If specified the page
1596  * will be wired down.
1597  */
1598 static int
1599 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1600     vm_prot_t prot, u_int flags, int8_t psind)
1601 {
1602 	int error;
1603 
1604 	rw_wlock(&pvh_global_lock);
1605 	PMAP_LOCK(pmap);
1606 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1607 	rw_wunlock(&pvh_global_lock);
1608 	PMAP_UNLOCK(pmap);
1609 	return (error);
1610 }
1611 
1612 static int
1613 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1614     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1615 {
1616 	pte_t *pte;
1617 	vm_paddr_t pa;
1618 	uint32_t flags;
1619 	int error, su, sync;
1620 
1621 	pa = VM_PAGE_TO_PHYS(m);
1622 	su = (pmap == kernel_pmap);
1623 	sync = 0;
1624 
1625 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1626 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1627 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1628 	//		(u_int32_t)m, va, pa, prot, flags);
1629 
1630 	if (su) {
1631 		KASSERT(((va >= virtual_avail) &&
1632 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1633 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1634 	} else {
1635 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1636 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1637 	}
1638 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1639 		VM_OBJECT_ASSERT_LOCKED(m->object);
1640 
1641 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1642 
1643 	/*
1644 	 * If there is an existing mapping, and the physical address has not
1645 	 * changed, must be protection or wiring change.
1646 	 */
1647 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1648 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1649 
1650 		/*
1651 		 * Before actually updating pte->flags we calculate and
1652 		 * prepare its new value in a helper var.
1653 		 */
1654 		flags = pte->flags;
1655 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1656 
1657 		/* Wiring change, just update stats. */
1658 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1659 			if (!PTE_ISWIRED(pte)) {
1660 				flags |= PTE_WIRED;
1661 				pmap->pm_stats.wired_count++;
1662 			}
1663 		} else {
1664 			if (PTE_ISWIRED(pte)) {
1665 				flags &= ~PTE_WIRED;
1666 				pmap->pm_stats.wired_count--;
1667 			}
1668 		}
1669 
1670 		if (prot & VM_PROT_WRITE) {
1671 			/* Add write permissions. */
1672 			flags |= PTE_SW;
1673 			if (!su)
1674 				flags |= PTE_UW;
1675 
1676 			if ((flags & PTE_MANAGED) != 0)
1677 				vm_page_aflag_set(m, PGA_WRITEABLE);
1678 		} else {
1679 			/* Handle modified pages, sense modify status. */
1680 
1681 			/*
1682 			 * The PTE_MODIFIED flag could be set by underlying
1683 			 * TLB misses since we last read it (above), possibly
1684 			 * other CPUs could update it so we check in the PTE
1685 			 * directly rather than rely on that saved local flags
1686 			 * copy.
1687 			 */
1688 			if (PTE_ISMODIFIED(pte))
1689 				vm_page_dirty(m);
1690 		}
1691 
1692 		if (prot & VM_PROT_EXECUTE) {
1693 			flags |= PTE_SX;
1694 			if (!su)
1695 				flags |= PTE_UX;
1696 
1697 			/*
1698 			 * Check existing flags for execute permissions: if we
1699 			 * are turning execute permissions on, icache should
1700 			 * be flushed.
1701 			 */
1702 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1703 				sync++;
1704 		}
1705 
1706 		flags &= ~PTE_REFERENCED;
1707 
1708 		/*
1709 		 * The new flags value is all calculated -- only now actually
1710 		 * update the PTE.
1711 		 */
1712 		mtx_lock_spin(&tlbivax_mutex);
1713 		tlb_miss_lock();
1714 
1715 		tlb0_flush_entry(va);
1716 		pte->flags = flags;
1717 
1718 		tlb_miss_unlock();
1719 		mtx_unlock_spin(&tlbivax_mutex);
1720 
1721 	} else {
1722 		/*
1723 		 * If there is an existing mapping, but it's for a different
1724 		 * physical address, pte_enter() will delete the old mapping.
1725 		 */
1726 		//if ((pte != NULL) && PTE_ISVALID(pte))
1727 		//	debugf("mmu_booke_enter_locked: replace\n");
1728 		//else
1729 		//	debugf("mmu_booke_enter_locked: new\n");
1730 
1731 		/* Now set up the flags and install the new mapping. */
1732 		flags = (PTE_SR | PTE_VALID);
1733 		flags |= PTE_M;
1734 
1735 		if (!su)
1736 			flags |= PTE_UR;
1737 
1738 		if (prot & VM_PROT_WRITE) {
1739 			flags |= PTE_SW;
1740 			if (!su)
1741 				flags |= PTE_UW;
1742 
1743 			if ((m->oflags & VPO_UNMANAGED) == 0)
1744 				vm_page_aflag_set(m, PGA_WRITEABLE);
1745 		}
1746 
1747 		if (prot & VM_PROT_EXECUTE) {
1748 			flags |= PTE_SX;
1749 			if (!su)
1750 				flags |= PTE_UX;
1751 		}
1752 
1753 		/* If its wired update stats. */
1754 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1755 			flags |= PTE_WIRED;
1756 
1757 		error = pte_enter(mmu, pmap, m, va, flags,
1758 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1759 		if (error != 0)
1760 			return (KERN_RESOURCE_SHORTAGE);
1761 
1762 		if ((flags & PMAP_ENTER_WIRED) != 0)
1763 			pmap->pm_stats.wired_count++;
1764 
1765 		/* Flush the real memory from the instruction cache. */
1766 		if (prot & VM_PROT_EXECUTE)
1767 			sync++;
1768 	}
1769 
1770 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1771 		__syncicache((void *)va, PAGE_SIZE);
1772 		sync = 0;
1773 	}
1774 
1775 	return (KERN_SUCCESS);
1776 }
1777 
1778 /*
1779  * Maps a sequence of resident pages belonging to the same object.
1780  * The sequence begins with the given page m_start.  This page is
1781  * mapped at the given virtual address start.  Each subsequent page is
1782  * mapped at a virtual address that is offset from start by the same
1783  * amount as the page is offset from m_start within the object.  The
1784  * last page in the sequence is the page with the largest offset from
1785  * m_start that can be mapped at a virtual address less than the given
1786  * virtual address end.  Not every virtual page between start and end
1787  * is mapped; only those for which a resident page exists with the
1788  * corresponding offset from m_start are mapped.
1789  */
1790 static void
1791 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1792     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1793 {
1794 	vm_page_t m;
1795 	vm_pindex_t diff, psize;
1796 
1797 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1798 
1799 	psize = atop(end - start);
1800 	m = m_start;
1801 	rw_wlock(&pvh_global_lock);
1802 	PMAP_LOCK(pmap);
1803 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1804 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1805 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1806 		    PMAP_ENTER_NOSLEEP, 0);
1807 		m = TAILQ_NEXT(m, listq);
1808 	}
1809 	rw_wunlock(&pvh_global_lock);
1810 	PMAP_UNLOCK(pmap);
1811 }
1812 
1813 static void
1814 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1815     vm_prot_t prot)
1816 {
1817 
1818 	rw_wlock(&pvh_global_lock);
1819 	PMAP_LOCK(pmap);
1820 	mmu_booke_enter_locked(mmu, pmap, va, m,
1821 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1822 	    0);
1823 	rw_wunlock(&pvh_global_lock);
1824 	PMAP_UNLOCK(pmap);
1825 }
1826 
1827 /*
1828  * Remove the given range of addresses from the specified map.
1829  *
1830  * It is assumed that the start and end are properly rounded to the page size.
1831  */
1832 static void
1833 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1834 {
1835 	pte_t *pte;
1836 	uint8_t hold_flag;
1837 
1838 	int su = (pmap == kernel_pmap);
1839 
1840 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1841 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1842 
1843 	if (su) {
1844 		KASSERT(((va >= virtual_avail) &&
1845 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1846 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1847 	} else {
1848 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1849 		    ("mmu_booke_remove: user pmap, non user va"));
1850 	}
1851 
1852 	if (PMAP_REMOVE_DONE(pmap)) {
1853 		//debugf("mmu_booke_remove: e (empty)\n");
1854 		return;
1855 	}
1856 
1857 	hold_flag = PTBL_HOLD_FLAG(pmap);
1858 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1859 
1860 	rw_wlock(&pvh_global_lock);
1861 	PMAP_LOCK(pmap);
1862 	for (; va < endva; va += PAGE_SIZE) {
1863 		pte = pte_find(mmu, pmap, va);
1864 		if ((pte != NULL) && PTE_ISVALID(pte))
1865 			pte_remove(mmu, pmap, va, hold_flag);
1866 	}
1867 	PMAP_UNLOCK(pmap);
1868 	rw_wunlock(&pvh_global_lock);
1869 
1870 	//debugf("mmu_booke_remove: e\n");
1871 }
1872 
1873 /*
1874  * Remove physical page from all pmaps in which it resides.
1875  */
1876 static void
1877 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1878 {
1879 	pv_entry_t pv, pvn;
1880 	uint8_t hold_flag;
1881 
1882 	rw_wlock(&pvh_global_lock);
1883 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1884 		pvn = TAILQ_NEXT(pv, pv_link);
1885 
1886 		PMAP_LOCK(pv->pv_pmap);
1887 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1888 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1889 		PMAP_UNLOCK(pv->pv_pmap);
1890 	}
1891 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1892 	rw_wunlock(&pvh_global_lock);
1893 }
1894 
1895 /*
1896  * Map a range of physical addresses into kernel virtual address space.
1897  */
1898 static vm_offset_t
1899 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1900     vm_paddr_t pa_end, int prot)
1901 {
1902 	vm_offset_t sva = *virt;
1903 	vm_offset_t va = sva;
1904 
1905 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1906 	//		sva, pa_start, pa_end);
1907 
1908 	while (pa_start < pa_end) {
1909 		mmu_booke_kenter(mmu, va, pa_start);
1910 		va += PAGE_SIZE;
1911 		pa_start += PAGE_SIZE;
1912 	}
1913 	*virt = va;
1914 
1915 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1916 	return (sva);
1917 }
1918 
1919 /*
1920  * The pmap must be activated before it's address space can be accessed in any
1921  * way.
1922  */
1923 static void
1924 mmu_booke_activate(mmu_t mmu, struct thread *td)
1925 {
1926 	pmap_t pmap;
1927 	u_int cpuid;
1928 
1929 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1930 
1931 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1932 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1933 
1934 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1935 
1936 	sched_pin();
1937 
1938 	cpuid = PCPU_GET(cpuid);
1939 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1940 	PCPU_SET(curpmap, pmap);
1941 
1942 	if (pmap->pm_tid[cpuid] == TID_NONE)
1943 		tid_alloc(pmap);
1944 
1945 	/* Load PID0 register with pmap tid value. */
1946 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1947 	__asm __volatile("isync");
1948 
1949 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1950 
1951 	sched_unpin();
1952 
1953 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1954 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1955 }
1956 
1957 /*
1958  * Deactivate the specified process's address space.
1959  */
1960 static void
1961 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1962 {
1963 	pmap_t pmap;
1964 
1965 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1966 
1967 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1968 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1969 
1970 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1971 
1972 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1973 	PCPU_SET(curpmap, NULL);
1974 }
1975 
1976 /*
1977  * Copy the range specified by src_addr/len
1978  * from the source map to the range dst_addr/len
1979  * in the destination map.
1980  *
1981  * This routine is only advisory and need not do anything.
1982  */
1983 static void
1984 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1985     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1986 {
1987 
1988 }
1989 
1990 /*
1991  * Set the physical protection on the specified range of this map as requested.
1992  */
1993 static void
1994 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1995     vm_prot_t prot)
1996 {
1997 	vm_offset_t va;
1998 	vm_page_t m;
1999 	pte_t *pte;
2000 
2001 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2002 		mmu_booke_remove(mmu, pmap, sva, eva);
2003 		return;
2004 	}
2005 
2006 	if (prot & VM_PROT_WRITE)
2007 		return;
2008 
2009 	PMAP_LOCK(pmap);
2010 	for (va = sva; va < eva; va += PAGE_SIZE) {
2011 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2012 			if (PTE_ISVALID(pte)) {
2013 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2014 
2015 				mtx_lock_spin(&tlbivax_mutex);
2016 				tlb_miss_lock();
2017 
2018 				/* Handle modified pages. */
2019 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2020 					vm_page_dirty(m);
2021 
2022 				tlb0_flush_entry(va);
2023 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2024 
2025 				tlb_miss_unlock();
2026 				mtx_unlock_spin(&tlbivax_mutex);
2027 			}
2028 		}
2029 	}
2030 	PMAP_UNLOCK(pmap);
2031 }
2032 
2033 /*
2034  * Clear the write and modified bits in each of the given page's mappings.
2035  */
2036 static void
2037 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2038 {
2039 	pv_entry_t pv;
2040 	pte_t *pte;
2041 
2042 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2043 	    ("mmu_booke_remove_write: page %p is not managed", m));
2044 
2045 	/*
2046 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2047 	 * set by another thread while the object is locked.  Thus,
2048 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2049 	 */
2050 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2051 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2052 		return;
2053 	rw_wlock(&pvh_global_lock);
2054 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2055 		PMAP_LOCK(pv->pv_pmap);
2056 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2057 			if (PTE_ISVALID(pte)) {
2058 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2059 
2060 				mtx_lock_spin(&tlbivax_mutex);
2061 				tlb_miss_lock();
2062 
2063 				/* Handle modified pages. */
2064 				if (PTE_ISMODIFIED(pte))
2065 					vm_page_dirty(m);
2066 
2067 				/* Flush mapping from TLB0. */
2068 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2069 
2070 				tlb_miss_unlock();
2071 				mtx_unlock_spin(&tlbivax_mutex);
2072 			}
2073 		}
2074 		PMAP_UNLOCK(pv->pv_pmap);
2075 	}
2076 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2077 	rw_wunlock(&pvh_global_lock);
2078 }
2079 
2080 static void
2081 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2082 {
2083 	pte_t *pte;
2084 	pmap_t pmap;
2085 	vm_page_t m;
2086 	vm_offset_t addr;
2087 	vm_paddr_t pa = 0;
2088 	int active, valid;
2089 
2090 	va = trunc_page(va);
2091 	sz = round_page(sz);
2092 
2093 	rw_wlock(&pvh_global_lock);
2094 	pmap = PCPU_GET(curpmap);
2095 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2096 	while (sz > 0) {
2097 		PMAP_LOCK(pm);
2098 		pte = pte_find(mmu, pm, va);
2099 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2100 		if (valid)
2101 			pa = PTE_PA(pte);
2102 		PMAP_UNLOCK(pm);
2103 		if (valid) {
2104 			if (!active) {
2105 				/* Create a mapping in the active pmap. */
2106 				addr = 0;
2107 				m = PHYS_TO_VM_PAGE(pa);
2108 				PMAP_LOCK(pmap);
2109 				pte_enter(mmu, pmap, m, addr,
2110 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2111 				__syncicache((void *)addr, PAGE_SIZE);
2112 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2113 				PMAP_UNLOCK(pmap);
2114 			} else
2115 				__syncicache((void *)va, PAGE_SIZE);
2116 		}
2117 		va += PAGE_SIZE;
2118 		sz -= PAGE_SIZE;
2119 	}
2120 	rw_wunlock(&pvh_global_lock);
2121 }
2122 
2123 /*
2124  * Atomically extract and hold the physical page with the given
2125  * pmap and virtual address pair if that mapping permits the given
2126  * protection.
2127  */
2128 static vm_page_t
2129 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2130     vm_prot_t prot)
2131 {
2132 	pte_t *pte;
2133 	vm_page_t m;
2134 	uint32_t pte_wbit;
2135 	vm_paddr_t pa;
2136 
2137 	m = NULL;
2138 	pa = 0;
2139 	PMAP_LOCK(pmap);
2140 retry:
2141 	pte = pte_find(mmu, pmap, va);
2142 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2143 		if (pmap == kernel_pmap)
2144 			pte_wbit = PTE_SW;
2145 		else
2146 			pte_wbit = PTE_UW;
2147 
2148 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2149 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2150 				goto retry;
2151 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2152 			vm_page_hold(m);
2153 		}
2154 	}
2155 
2156 	PA_UNLOCK_COND(pa);
2157 	PMAP_UNLOCK(pmap);
2158 	return (m);
2159 }
2160 
2161 /*
2162  * Initialize a vm_page's machine-dependent fields.
2163  */
2164 static void
2165 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2166 {
2167 
2168 	TAILQ_INIT(&m->md.pv_list);
2169 }
2170 
2171 /*
2172  * mmu_booke_zero_page_area zeros the specified hardware page by
2173  * mapping it into virtual memory and using bzero to clear
2174  * its contents.
2175  *
2176  * off and size must reside within a single page.
2177  */
2178 static void
2179 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2180 {
2181 	vm_offset_t va;
2182 
2183 	/* XXX KASSERT off and size are within a single page? */
2184 
2185 	mtx_lock(&zero_page_mutex);
2186 	va = zero_page_va;
2187 
2188 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2189 	bzero((caddr_t)va + off, size);
2190 	mmu_booke_kremove(mmu, va);
2191 
2192 	mtx_unlock(&zero_page_mutex);
2193 }
2194 
2195 /*
2196  * mmu_booke_zero_page zeros the specified hardware page.
2197  */
2198 static void
2199 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2200 {
2201 
2202 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2203 }
2204 
2205 /*
2206  * mmu_booke_copy_page copies the specified (machine independent) page by
2207  * mapping the page into virtual memory and using memcopy to copy the page,
2208  * one machine dependent page at a time.
2209  */
2210 static void
2211 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2212 {
2213 	vm_offset_t sva, dva;
2214 
2215 	sva = copy_page_src_va;
2216 	dva = copy_page_dst_va;
2217 
2218 	mtx_lock(&copy_page_mutex);
2219 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2220 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2221 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2222 	mmu_booke_kremove(mmu, dva);
2223 	mmu_booke_kremove(mmu, sva);
2224 	mtx_unlock(&copy_page_mutex);
2225 }
2226 
2227 static inline void
2228 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2229     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2230 {
2231 	void *a_cp, *b_cp;
2232 	vm_offset_t a_pg_offset, b_pg_offset;
2233 	int cnt;
2234 
2235 	mtx_lock(&copy_page_mutex);
2236 	while (xfersize > 0) {
2237 		a_pg_offset = a_offset & PAGE_MASK;
2238 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2239 		mmu_booke_kenter(mmu, copy_page_src_va,
2240 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2241 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2242 		b_pg_offset = b_offset & PAGE_MASK;
2243 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2244 		mmu_booke_kenter(mmu, copy_page_dst_va,
2245 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2246 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2247 		bcopy(a_cp, b_cp, cnt);
2248 		mmu_booke_kremove(mmu, copy_page_dst_va);
2249 		mmu_booke_kremove(mmu, copy_page_src_va);
2250 		a_offset += cnt;
2251 		b_offset += cnt;
2252 		xfersize -= cnt;
2253 	}
2254 	mtx_unlock(&copy_page_mutex);
2255 }
2256 
2257 /*
2258  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2259  * into virtual memory and using bzero to clear its contents. This is intended
2260  * to be called from the vm_pagezero process only and outside of Giant. No
2261  * lock is required.
2262  */
2263 static void
2264 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2265 {
2266 	vm_offset_t va;
2267 
2268 	va = zero_page_idle_va;
2269 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2270 	bzero((caddr_t)va, PAGE_SIZE);
2271 	mmu_booke_kremove(mmu, va);
2272 }
2273 
2274 /*
2275  * Return whether or not the specified physical page was modified
2276  * in any of physical maps.
2277  */
2278 static boolean_t
2279 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2280 {
2281 	pte_t *pte;
2282 	pv_entry_t pv;
2283 	boolean_t rv;
2284 
2285 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2286 	    ("mmu_booke_is_modified: page %p is not managed", m));
2287 	rv = FALSE;
2288 
2289 	/*
2290 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2291 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2292 	 * is clear, no PTEs can be modified.
2293 	 */
2294 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2295 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2296 		return (rv);
2297 	rw_wlock(&pvh_global_lock);
2298 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2299 		PMAP_LOCK(pv->pv_pmap);
2300 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2301 		    PTE_ISVALID(pte)) {
2302 			if (PTE_ISMODIFIED(pte))
2303 				rv = TRUE;
2304 		}
2305 		PMAP_UNLOCK(pv->pv_pmap);
2306 		if (rv)
2307 			break;
2308 	}
2309 	rw_wunlock(&pvh_global_lock);
2310 	return (rv);
2311 }
2312 
2313 /*
2314  * Return whether or not the specified virtual address is eligible
2315  * for prefault.
2316  */
2317 static boolean_t
2318 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2319 {
2320 
2321 	return (FALSE);
2322 }
2323 
2324 /*
2325  * Return whether or not the specified physical page was referenced
2326  * in any physical maps.
2327  */
2328 static boolean_t
2329 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2330 {
2331 	pte_t *pte;
2332 	pv_entry_t pv;
2333 	boolean_t rv;
2334 
2335 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2336 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2337 	rv = FALSE;
2338 	rw_wlock(&pvh_global_lock);
2339 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2340 		PMAP_LOCK(pv->pv_pmap);
2341 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2342 		    PTE_ISVALID(pte)) {
2343 			if (PTE_ISREFERENCED(pte))
2344 				rv = TRUE;
2345 		}
2346 		PMAP_UNLOCK(pv->pv_pmap);
2347 		if (rv)
2348 			break;
2349 	}
2350 	rw_wunlock(&pvh_global_lock);
2351 	return (rv);
2352 }
2353 
2354 /*
2355  * Clear the modify bits on the specified physical page.
2356  */
2357 static void
2358 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2359 {
2360 	pte_t *pte;
2361 	pv_entry_t pv;
2362 
2363 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2364 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2365 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2366 	KASSERT(!vm_page_xbusied(m),
2367 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2368 
2369 	/*
2370 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2371 	 * If the object containing the page is locked and the page is not
2372 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2373 	 */
2374 	if ((m->aflags & PGA_WRITEABLE) == 0)
2375 		return;
2376 	rw_wlock(&pvh_global_lock);
2377 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2378 		PMAP_LOCK(pv->pv_pmap);
2379 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2380 		    PTE_ISVALID(pte)) {
2381 			mtx_lock_spin(&tlbivax_mutex);
2382 			tlb_miss_lock();
2383 
2384 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2385 				tlb0_flush_entry(pv->pv_va);
2386 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2387 				    PTE_REFERENCED);
2388 			}
2389 
2390 			tlb_miss_unlock();
2391 			mtx_unlock_spin(&tlbivax_mutex);
2392 		}
2393 		PMAP_UNLOCK(pv->pv_pmap);
2394 	}
2395 	rw_wunlock(&pvh_global_lock);
2396 }
2397 
2398 /*
2399  * Return a count of reference bits for a page, clearing those bits.
2400  * It is not necessary for every reference bit to be cleared, but it
2401  * is necessary that 0 only be returned when there are truly no
2402  * reference bits set.
2403  *
2404  * XXX: The exact number of bits to check and clear is a matter that
2405  * should be tested and standardized at some point in the future for
2406  * optimal aging of shared pages.
2407  */
2408 static int
2409 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2410 {
2411 	pte_t *pte;
2412 	pv_entry_t pv;
2413 	int count;
2414 
2415 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2416 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2417 	count = 0;
2418 	rw_wlock(&pvh_global_lock);
2419 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2420 		PMAP_LOCK(pv->pv_pmap);
2421 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2422 		    PTE_ISVALID(pte)) {
2423 			if (PTE_ISREFERENCED(pte)) {
2424 				mtx_lock_spin(&tlbivax_mutex);
2425 				tlb_miss_lock();
2426 
2427 				tlb0_flush_entry(pv->pv_va);
2428 				pte->flags &= ~PTE_REFERENCED;
2429 
2430 				tlb_miss_unlock();
2431 				mtx_unlock_spin(&tlbivax_mutex);
2432 
2433 				if (++count > 4) {
2434 					PMAP_UNLOCK(pv->pv_pmap);
2435 					break;
2436 				}
2437 			}
2438 		}
2439 		PMAP_UNLOCK(pv->pv_pmap);
2440 	}
2441 	rw_wunlock(&pvh_global_lock);
2442 	return (count);
2443 }
2444 
2445 /*
2446  * Clear the wired attribute from the mappings for the specified range of
2447  * addresses in the given pmap.  Every valid mapping within that range must
2448  * have the wired attribute set.  In contrast, invalid mappings cannot have
2449  * the wired attribute set, so they are ignored.
2450  *
2451  * The wired attribute of the page table entry is not a hardware feature, so
2452  * there is no need to invalidate any TLB entries.
2453  */
2454 static void
2455 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2456 {
2457 	vm_offset_t va;
2458 	pte_t *pte;
2459 
2460 	PMAP_LOCK(pmap);
2461 	for (va = sva; va < eva; va += PAGE_SIZE) {
2462 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2463 		    PTE_ISVALID(pte)) {
2464 			if (!PTE_ISWIRED(pte))
2465 				panic("mmu_booke_unwire: pte %p isn't wired",
2466 				    pte);
2467 			pte->flags &= ~PTE_WIRED;
2468 			pmap->pm_stats.wired_count--;
2469 		}
2470 	}
2471 	PMAP_UNLOCK(pmap);
2472 
2473 }
2474 
2475 /*
2476  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2477  * page.  This count may be changed upwards or downwards in the future; it is
2478  * only necessary that true be returned for a small subset of pmaps for proper
2479  * page aging.
2480  */
2481 static boolean_t
2482 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2483 {
2484 	pv_entry_t pv;
2485 	int loops;
2486 	boolean_t rv;
2487 
2488 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2489 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2490 	loops = 0;
2491 	rv = FALSE;
2492 	rw_wlock(&pvh_global_lock);
2493 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2494 		if (pv->pv_pmap == pmap) {
2495 			rv = TRUE;
2496 			break;
2497 		}
2498 		if (++loops >= 16)
2499 			break;
2500 	}
2501 	rw_wunlock(&pvh_global_lock);
2502 	return (rv);
2503 }
2504 
2505 /*
2506  * Return the number of managed mappings to the given physical page that are
2507  * wired.
2508  */
2509 static int
2510 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2511 {
2512 	pv_entry_t pv;
2513 	pte_t *pte;
2514 	int count = 0;
2515 
2516 	if ((m->oflags & VPO_UNMANAGED) != 0)
2517 		return (count);
2518 	rw_wlock(&pvh_global_lock);
2519 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2520 		PMAP_LOCK(pv->pv_pmap);
2521 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2522 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2523 				count++;
2524 		PMAP_UNLOCK(pv->pv_pmap);
2525 	}
2526 	rw_wunlock(&pvh_global_lock);
2527 	return (count);
2528 }
2529 
2530 static int
2531 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2532 {
2533 	int i;
2534 	vm_offset_t va;
2535 
2536 	/*
2537 	 * This currently does not work for entries that
2538 	 * overlap TLB1 entries.
2539 	 */
2540 	for (i = 0; i < tlb1_idx; i ++) {
2541 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2542 			return (0);
2543 	}
2544 
2545 	return (EFAULT);
2546 }
2547 
2548 void
2549 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2550 {
2551 	vm_paddr_t ppa;
2552 	vm_offset_t ofs;
2553 	vm_size_t gran;
2554 
2555 	/* Minidumps are based on virtual memory addresses. */
2556 	if (do_minidump) {
2557 		*va = (void *)pa;
2558 		return;
2559 	}
2560 
2561 	/* Raw physical memory dumps don't have a virtual address. */
2562 	/* We always map a 256MB page at 256M. */
2563 	gran = 256 * 1024 * 1024;
2564 	ppa = pa & ~(gran - 1);
2565 	ofs = pa - ppa;
2566 	*va = (void *)gran;
2567 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2568 
2569 	if (sz > (gran - ofs))
2570 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2571 		    _TLB_ENTRY_IO);
2572 }
2573 
2574 void
2575 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2576 {
2577 	vm_paddr_t ppa;
2578 	vm_offset_t ofs;
2579 	vm_size_t gran;
2580 
2581 	/* Minidumps are based on virtual memory addresses. */
2582 	/* Nothing to do... */
2583 	if (do_minidump)
2584 		return;
2585 
2586 	/* Raw physical memory dumps don't have a virtual address. */
2587 	tlb1_idx--;
2588 	tlb1[tlb1_idx].mas1 = 0;
2589 	tlb1[tlb1_idx].mas2 = 0;
2590 	tlb1[tlb1_idx].mas3 = 0;
2591 	tlb1_write_entry(tlb1_idx);
2592 
2593 	gran = 256 * 1024 * 1024;
2594 	ppa = pa & ~(gran - 1);
2595 	ofs = pa - ppa;
2596 	if (sz > (gran - ofs)) {
2597 		tlb1_idx--;
2598 		tlb1[tlb1_idx].mas1 = 0;
2599 		tlb1[tlb1_idx].mas2 = 0;
2600 		tlb1[tlb1_idx].mas3 = 0;
2601 		tlb1_write_entry(tlb1_idx);
2602 	}
2603 }
2604 
2605 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2606 
2607 void
2608 mmu_booke_scan_init(mmu_t mmu)
2609 {
2610 	vm_offset_t va;
2611 	pte_t *pte;
2612 	int i;
2613 
2614 	if (!do_minidump) {
2615 		/* Initialize phys. segments for dumpsys(). */
2616 		memset(&dump_map, 0, sizeof(dump_map));
2617 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2618 		    &availmem_regions_sz);
2619 		for (i = 0; i < physmem_regions_sz; i++) {
2620 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2621 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2622 		}
2623 		return;
2624 	}
2625 
2626 	/* Virtual segments for minidumps: */
2627 	memset(&dump_map, 0, sizeof(dump_map));
2628 
2629 	/* 1st: kernel .data and .bss. */
2630 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2631 	dump_map[0].pa_size =
2632 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2633 
2634 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2635 	dump_map[1].pa_start = data_start;
2636 	dump_map[1].pa_size = data_end - data_start;
2637 
2638 	/* 3rd: kernel VM. */
2639 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2640 	/* Find start of next chunk (from va). */
2641 	while (va < virtual_end) {
2642 		/* Don't dump the buffer cache. */
2643 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2644 			va = kmi.buffer_eva;
2645 			continue;
2646 		}
2647 		pte = pte_find(mmu, kernel_pmap, va);
2648 		if (pte != NULL && PTE_ISVALID(pte))
2649 			break;
2650 		va += PAGE_SIZE;
2651 	}
2652 	if (va < virtual_end) {
2653 		dump_map[2].pa_start = va;
2654 		va += PAGE_SIZE;
2655 		/* Find last page in chunk. */
2656 		while (va < virtual_end) {
2657 			/* Don't run into the buffer cache. */
2658 			if (va == kmi.buffer_sva)
2659 				break;
2660 			pte = pte_find(mmu, kernel_pmap, va);
2661 			if (pte == NULL || !PTE_ISVALID(pte))
2662 				break;
2663 			va += PAGE_SIZE;
2664 		}
2665 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2666 	}
2667 }
2668 
2669 /*
2670  * Map a set of physical memory pages into the kernel virtual address space.
2671  * Return a pointer to where it is mapped. This routine is intended to be used
2672  * for mapping device memory, NOT real memory.
2673  */
2674 static void *
2675 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2676 {
2677 
2678 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2679 }
2680 
2681 static void *
2682 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2683 {
2684 	void *res;
2685 	uintptr_t va;
2686 	vm_size_t sz;
2687 	int i;
2688 
2689 	/*
2690 	 * Check if this is premapped in TLB1. Note: this should probably also
2691 	 * check whether a sequence of TLB1 entries exist that match the
2692 	 * requirement, but now only checks the easy case.
2693 	 */
2694 	if (ma == VM_MEMATTR_DEFAULT) {
2695 		for (i = 0; i < tlb1_idx; i++) {
2696 			if (!(tlb1[i].mas1 & MAS1_VALID))
2697 				continue;
2698 			if (pa >= tlb1[i].phys &&
2699 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2700 				return (void *)(tlb1[i].virt +
2701 				    (pa - tlb1[i].phys));
2702 		}
2703 	}
2704 
2705 	size = roundup(size, PAGE_SIZE);
2706 
2707 	/*
2708 	 * We leave a hole for device direct mapping between the maximum user
2709 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2710 	 * devices are in there, just map them 1:1. If not, map them to the
2711 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2712 	 * addresses should be pulled from an allocator, but since we do not
2713 	 * ever free TLB1 entries, it is safe just to increment a counter.
2714 	 * Note that there isn't a lot of address space here (128 MB) and it
2715 	 * is not at all difficult to imagine running out, since that is a 4:1
2716 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2717 	 * mapped there.
2718 	 */
2719 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2720 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2721 		va = pa;
2722 	else
2723 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2724 	res = (void *)va;
2725 
2726 	do {
2727 		sz = 1 << (ilog2(size) & ~1);
2728 		if (bootverbose)
2729 			printf("Wiring VA=%x to PA=%x (size=%x), "
2730 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2731 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2732 		size -= sz;
2733 		pa += sz;
2734 		va += sz;
2735 	} while (size > 0);
2736 
2737 	return (res);
2738 }
2739 
2740 /*
2741  * 'Unmap' a range mapped by mmu_booke_mapdev().
2742  */
2743 static void
2744 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2745 {
2746 #ifdef SUPPORTS_SHRINKING_TLB1
2747 	vm_offset_t base, offset;
2748 
2749 	/*
2750 	 * Unmap only if this is inside kernel virtual space.
2751 	 */
2752 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2753 		base = trunc_page(va);
2754 		offset = va & PAGE_MASK;
2755 		size = roundup(offset + size, PAGE_SIZE);
2756 		kva_free(base, size);
2757 	}
2758 #endif
2759 }
2760 
2761 /*
2762  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2763  * specified pmap. This eliminates the blast of soft faults on process startup
2764  * and immediately after an mmap.
2765  */
2766 static void
2767 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2768     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2769 {
2770 
2771 	VM_OBJECT_ASSERT_WLOCKED(object);
2772 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2773 	    ("mmu_booke_object_init_pt: non-device object"));
2774 }
2775 
2776 /*
2777  * Perform the pmap work for mincore.
2778  */
2779 static int
2780 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2781     vm_paddr_t *locked_pa)
2782 {
2783 
2784 	/* XXX: this should be implemented at some point */
2785 	return (0);
2786 }
2787 
2788 /**************************************************************************/
2789 /* TID handling */
2790 /**************************************************************************/
2791 
2792 /*
2793  * Allocate a TID. If necessary, steal one from someone else.
2794  * The new TID is flushed from the TLB before returning.
2795  */
2796 static tlbtid_t
2797 tid_alloc(pmap_t pmap)
2798 {
2799 	tlbtid_t tid;
2800 	int thiscpu;
2801 
2802 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2803 
2804 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2805 
2806 	thiscpu = PCPU_GET(cpuid);
2807 
2808 	tid = PCPU_GET(tid_next);
2809 	if (tid > TID_MAX)
2810 		tid = TID_MIN;
2811 	PCPU_SET(tid_next, tid + 1);
2812 
2813 	/* If we are stealing TID then clear the relevant pmap's field */
2814 	if (tidbusy[thiscpu][tid] != NULL) {
2815 
2816 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2817 
2818 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2819 
2820 		/* Flush all entries from TLB0 matching this TID. */
2821 		tid_flush(tid, tlb0_ways, tlb0_entries_per_way);
2822 	}
2823 
2824 	tidbusy[thiscpu][tid] = pmap;
2825 	pmap->pm_tid[thiscpu] = tid;
2826 	__asm __volatile("msync; isync");
2827 
2828 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2829 	    PCPU_GET(tid_next));
2830 
2831 	return (tid);
2832 }
2833 
2834 /**************************************************************************/
2835 /* TLB0 handling */
2836 /**************************************************************************/
2837 
2838 static void
2839 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2840     uint32_t mas7)
2841 {
2842 	int as;
2843 	char desc[3];
2844 	tlbtid_t tid;
2845 	vm_size_t size;
2846 	unsigned int tsize;
2847 
2848 	desc[2] = '\0';
2849 	if (mas1 & MAS1_VALID)
2850 		desc[0] = 'V';
2851 	else
2852 		desc[0] = ' ';
2853 
2854 	if (mas1 & MAS1_IPROT)
2855 		desc[1] = 'P';
2856 	else
2857 		desc[1] = ' ';
2858 
2859 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2860 	tid = MAS1_GETTID(mas1);
2861 
2862 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2863 	size = 0;
2864 	if (tsize)
2865 		size = tsize2size(tsize);
2866 
2867 	debugf("%3d: (%s) [AS=%d] "
2868 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2869 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2870 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2871 }
2872 
2873 /* Convert TLB0 va and way number to tlb0[] table index. */
2874 static inline unsigned int
2875 tlb0_tableidx(vm_offset_t va, unsigned int way)
2876 {
2877 	unsigned int idx;
2878 
2879 	idx = (way * TLB0_ENTRIES_PER_WAY);
2880 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2881 	return (idx);
2882 }
2883 
2884 /*
2885  * Invalidate TLB0 entry.
2886  */
2887 static inline void
2888 tlb0_flush_entry(vm_offset_t va)
2889 {
2890 
2891 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2892 
2893 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2894 
2895 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2896 	__asm __volatile("isync; msync");
2897 	__asm __volatile("tlbsync; msync");
2898 
2899 	CTR1(KTR_PMAP, "%s: e", __func__);
2900 }
2901 
2902 /* Print out contents of the MAS registers for each TLB0 entry */
2903 void
2904 tlb0_print_tlbentries(void)
2905 {
2906 	uint32_t mas0, mas1, mas2, mas3, mas7;
2907 	int entryidx, way, idx;
2908 
2909 	debugf("TLB0 entries:\n");
2910 	for (way = 0; way < TLB0_WAYS; way ++)
2911 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2912 
2913 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2914 			mtspr(SPR_MAS0, mas0);
2915 			__asm __volatile("isync");
2916 
2917 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2918 			mtspr(SPR_MAS2, mas2);
2919 
2920 			__asm __volatile("isync; tlbre");
2921 
2922 			mas1 = mfspr(SPR_MAS1);
2923 			mas2 = mfspr(SPR_MAS2);
2924 			mas3 = mfspr(SPR_MAS3);
2925 			mas7 = mfspr(SPR_MAS7);
2926 
2927 			idx = tlb0_tableidx(mas2, way);
2928 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2929 		}
2930 }
2931 
2932 /**************************************************************************/
2933 /* TLB1 handling */
2934 /**************************************************************************/
2935 
2936 /*
2937  * TLB1 mapping notes:
2938  *
2939  * TLB1[0]	Kernel text and data.
2940  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2941  *		windows, other devices mappings.
2942  */
2943 
2944 /*
2945  * Write given entry to TLB1 hardware.
2946  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2947  */
2948 static void
2949 tlb1_write_entry(unsigned int idx)
2950 {
2951 	uint32_t mas0, mas7;
2952 
2953 	//debugf("tlb1_write_entry: s\n");
2954 
2955 	/* Clear high order RPN bits */
2956 	mas7 = 0;
2957 
2958 	/* Select entry */
2959 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2960 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2961 
2962 	mtspr(SPR_MAS0, mas0);
2963 	__asm __volatile("isync");
2964 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2965 	__asm __volatile("isync");
2966 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2967 	__asm __volatile("isync");
2968 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2969 	__asm __volatile("isync");
2970 	mtspr(SPR_MAS7, mas7);
2971 	__asm __volatile("isync; tlbwe; isync; msync");
2972 
2973 	//debugf("tlb1_write_entry: e\n");
2974 }
2975 
2976 /*
2977  * Return the largest uint value log such that 2^log <= num.
2978  */
2979 static unsigned int
2980 ilog2(unsigned int num)
2981 {
2982 	int lz;
2983 
2984 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2985 	return (31 - lz);
2986 }
2987 
2988 /*
2989  * Convert TLB TSIZE value to mapped region size.
2990  */
2991 static vm_size_t
2992 tsize2size(unsigned int tsize)
2993 {
2994 
2995 	/*
2996 	 * size = 4^tsize KB
2997 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2998 	 */
2999 
3000 	return ((1 << (2 * tsize)) * 1024);
3001 }
3002 
3003 /*
3004  * Convert region size (must be power of 4) to TLB TSIZE value.
3005  */
3006 static unsigned int
3007 size2tsize(vm_size_t size)
3008 {
3009 
3010 	return (ilog2(size) / 2 - 5);
3011 }
3012 
3013 /*
3014  * Register permanent kernel mapping in TLB1.
3015  *
3016  * Entries are created starting from index 0 (current free entry is
3017  * kept in tlb1_idx) and are not supposed to be invalidated.
3018  */
3019 static int
3020 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
3021     uint32_t flags)
3022 {
3023 	uint32_t ts, tid;
3024 	int tsize, index;
3025 
3026 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3027 	if (index >= TLB1_ENTRIES) {
3028 		printf("tlb1_set_entry: TLB1 full!\n");
3029 		return (-1);
3030 	}
3031 
3032 	/* Convert size to TSIZE */
3033 	tsize = size2tsize(size);
3034 
3035 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3036 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3037 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3038 
3039 	/*
3040 	 * Atomicity is preserved by the atomic increment above since nothing
3041 	 * is ever removed from tlb1.
3042 	 */
3043 
3044 	tlb1[index].phys = pa;
3045 	tlb1[index].virt = va;
3046 	tlb1[index].size = size;
3047 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3048 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3049 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3050 
3051 	/* Set supervisor RWX permission bits */
3052 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3053 
3054 	tlb1_write_entry(index);
3055 
3056 	/*
3057 	 * XXX in general TLB1 updates should be propagated between CPUs,
3058 	 * since current design assumes to have the same TLB1 set-up on all
3059 	 * cores.
3060 	 */
3061 	return (0);
3062 }
3063 
3064 /*
3065  * Map in contiguous RAM region into the TLB1 using maximum of
3066  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3067  *
3068  * If necessary round up last entry size and return total size
3069  * used by all allocated entries.
3070  */
3071 vm_size_t
3072 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3073 {
3074 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3075 	vm_size_t mapped, pgsz, base, mask;
3076 	int idx, nents;
3077 
3078 	/* Round up to the next 1M */
3079 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3080 
3081 	mapped = 0;
3082 	idx = 0;
3083 	base = va;
3084 	pgsz = 64*1024*1024;
3085 	while (mapped < size) {
3086 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3087 			while (pgsz > (size - mapped))
3088 				pgsz >>= 2;
3089 			pgs[idx++] = pgsz;
3090 			mapped += pgsz;
3091 		}
3092 
3093 		/* We under-map. Correct for this. */
3094 		if (mapped < size) {
3095 			while (pgs[idx - 1] == pgsz) {
3096 				idx--;
3097 				mapped -= pgsz;
3098 			}
3099 			/* XXX We may increase beyond out starting point. */
3100 			pgsz <<= 2;
3101 			pgs[idx++] = pgsz;
3102 			mapped += pgsz;
3103 		}
3104 	}
3105 
3106 	nents = idx;
3107 	mask = pgs[0] - 1;
3108 	/* Align address to the boundary */
3109 	if (va & mask) {
3110 		va = (va + mask) & ~mask;
3111 		pa = (pa + mask) & ~mask;
3112 	}
3113 
3114 	for (idx = 0; idx < nents; idx++) {
3115 		pgsz = pgs[idx];
3116 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3117 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3118 		pa += pgsz;
3119 		va += pgsz;
3120 	}
3121 
3122 	mapped = (va - base);
3123 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3124 	    mapped, mapped - size);
3125 	return (mapped);
3126 }
3127 
3128 /*
3129  * TLB1 initialization routine, to be called after the very first
3130  * assembler level setup done in locore.S.
3131  */
3132 void
3133 tlb1_init()
3134 {
3135 	uint32_t mas0, mas1, mas2, mas3;
3136 	uint32_t tsz;
3137 	u_int i;
3138 
3139 	if (bootinfo != NULL && bootinfo[0] != 1) {
3140 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3141 	} else
3142 		tlb1_idx = 1;
3143 
3144 	/* The first entry/entries are used to map the kernel. */
3145 	for (i = 0; i < tlb1_idx; i++) {
3146 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3147 		mtspr(SPR_MAS0, mas0);
3148 		__asm __volatile("isync; tlbre");
3149 
3150 		mas1 = mfspr(SPR_MAS1);
3151 		if ((mas1 & MAS1_VALID) == 0)
3152 			continue;
3153 
3154 		mas2 = mfspr(SPR_MAS2);
3155 		mas3 = mfspr(SPR_MAS3);
3156 
3157 		tlb1[i].mas1 = mas1;
3158 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3159 		tlb1[i].mas3 = mas3;
3160 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3161 		tlb1[i].phys = mas3 & MAS3_RPN;
3162 
3163 		if (i == 0)
3164 			kernload = mas3 & MAS3_RPN;
3165 
3166 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3167 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3168 		kernsize += tlb1[i].size;
3169 	}
3170 
3171 #ifdef SMP
3172 	bp_ntlb1s = tlb1_idx;
3173 #endif
3174 
3175 	/* Purge the remaining entries */
3176 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3177 		tlb1_write_entry(i);
3178 
3179 	/* Setup TLB miss defaults */
3180 	set_mas4_defaults();
3181 }
3182 
3183 vm_offset_t
3184 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3185 {
3186 	vm_paddr_t pa_base;
3187 	vm_offset_t va, sz;
3188 	int i;
3189 
3190 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3191 
3192 	for (i = 0; i < tlb1_idx; i++) {
3193 		if (!(tlb1[i].mas1 & MAS1_VALID))
3194 			continue;
3195 		if (pa >= tlb1[i].phys && (pa + size) <=
3196 		    (tlb1[i].phys + tlb1[i].size))
3197 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3198 	}
3199 
3200 	pa_base = trunc_page(pa);
3201 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3202 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3203 	va = tlb1_map_base + (pa - pa_base);
3204 
3205 	do {
3206 		sz = 1 << (ilog2(size) & ~1);
3207 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3208 		size -= sz;
3209 		pa_base += sz;
3210 		tlb1_map_base += sz;
3211 	} while (size > 0);
3212 
3213 #ifdef SMP
3214 	bp_ntlb1s = tlb1_idx;
3215 #endif
3216 
3217 	return (va);
3218 }
3219 
3220 /*
3221  * Setup MAS4 defaults.
3222  * These values are loaded to MAS0-2 on a TLB miss.
3223  */
3224 static void
3225 set_mas4_defaults(void)
3226 {
3227 	uint32_t mas4;
3228 
3229 	/* Defaults: TLB0, PID0, TSIZED=4K */
3230 	mas4 = MAS4_TLBSELD0;
3231 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3232 #ifdef SMP
3233 	mas4 |= MAS4_MD;
3234 #endif
3235 	mtspr(SPR_MAS4, mas4);
3236 	__asm __volatile("isync");
3237 }
3238 
3239 /*
3240  * Print out contents of the MAS registers for each TLB1 entry
3241  */
3242 void
3243 tlb1_print_tlbentries(void)
3244 {
3245 	uint32_t mas0, mas1, mas2, mas3, mas7;
3246 	int i;
3247 
3248 	debugf("TLB1 entries:\n");
3249 	for (i = 0; i < TLB1_ENTRIES; i++) {
3250 
3251 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3252 		mtspr(SPR_MAS0, mas0);
3253 
3254 		__asm __volatile("isync; tlbre");
3255 
3256 		mas1 = mfspr(SPR_MAS1);
3257 		mas2 = mfspr(SPR_MAS2);
3258 		mas3 = mfspr(SPR_MAS3);
3259 		mas7 = mfspr(SPR_MAS7);
3260 
3261 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3262 	}
3263 }
3264 
3265 /*
3266  * Print out contents of the in-ram tlb1 table.
3267  */
3268 void
3269 tlb1_print_entries(void)
3270 {
3271 	int i;
3272 
3273 	debugf("tlb1[] table entries:\n");
3274 	for (i = 0; i < TLB1_ENTRIES; i++)
3275 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3276 }
3277 
3278 /*
3279  * Return 0 if the physical IO range is encompassed by one of the
3280  * the TLB1 entries, otherwise return related error code.
3281  */
3282 static int
3283 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3284 {
3285 	uint32_t prot;
3286 	vm_paddr_t pa_start;
3287 	vm_paddr_t pa_end;
3288 	unsigned int entry_tsize;
3289 	vm_size_t entry_size;
3290 
3291 	*va = (vm_offset_t)NULL;
3292 
3293 	/* Skip invalid entries */
3294 	if (!(tlb1[i].mas1 & MAS1_VALID))
3295 		return (EINVAL);
3296 
3297 	/*
3298 	 * The entry must be cache-inhibited, guarded, and r/w
3299 	 * so it can function as an i/o page
3300 	 */
3301 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3302 	if (prot != (MAS2_I | MAS2_G))
3303 		return (EPERM);
3304 
3305 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3306 	if (prot != (MAS3_SR | MAS3_SW))
3307 		return (EPERM);
3308 
3309 	/* The address should be within the entry range. */
3310 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3311 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3312 
3313 	entry_size = tsize2size(entry_tsize);
3314 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3315 	pa_end = pa_start + entry_size - 1;
3316 
3317 	if ((pa < pa_start) || ((pa + size) > pa_end))
3318 		return (ERANGE);
3319 
3320 	/* Return virtual address of this mapping. */
3321 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3322 	return (0);
3323 }
3324