xref: /freebsd/sys/powerpc/booke/pmap.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include "opt_kstack_pages.h"
55 
56 #include <sys/param.h>
57 #include <sys/conf.h>
58 #include <sys/malloc.h>
59 #include <sys/ktr.h>
60 #include <sys/proc.h>
61 #include <sys/user.h>
62 #include <sys/queue.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/kerneldump.h>
66 #include <sys/linker.h>
67 #include <sys/msgbuf.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/vmmeter.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_pager.h>
84 #include <vm/uma.h>
85 
86 #include <machine/cpu.h>
87 #include <machine/pcb.h>
88 #include <machine/platform.h>
89 
90 #include <machine/tlb.h>
91 #include <machine/spr.h>
92 #include <machine/md_var.h>
93 #include <machine/mmuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/pte.h>
96 
97 #include "mmu_if.h"
98 
99 #ifdef  DEBUG
100 #define debugf(fmt, args...) printf(fmt, ##args)
101 #else
102 #define debugf(fmt, args...)
103 #endif
104 
105 #define TODO			panic("%s: not implemented", __func__);
106 
107 extern unsigned char _etext[];
108 extern unsigned char _end[];
109 
110 extern uint32_t *bootinfo;
111 
112 #ifdef SMP
113 extern uint32_t bp_ntlb1s;
114 #endif
115 
116 vm_paddr_t kernload;
117 vm_offset_t kernstart;
118 vm_size_t kernsize;
119 
120 /* Message buffer and tables. */
121 static vm_offset_t data_start;
122 static vm_size_t data_end;
123 
124 /* Phys/avail memory regions. */
125 static struct mem_region *availmem_regions;
126 static int availmem_regions_sz;
127 static struct mem_region *physmem_regions;
128 static int physmem_regions_sz;
129 
130 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
131 static vm_offset_t zero_page_va;
132 static struct mtx zero_page_mutex;
133 
134 static struct mtx tlbivax_mutex;
135 
136 /*
137  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
138  * by idle thred only, no lock required.
139  */
140 static vm_offset_t zero_page_idle_va;
141 
142 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
143 static vm_offset_t copy_page_src_va;
144 static vm_offset_t copy_page_dst_va;
145 static struct mtx copy_page_mutex;
146 
147 /**************************************************************************/
148 /* PMAP */
149 /**************************************************************************/
150 
151 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
152     vm_prot_t, u_int flags, int8_t psind);
153 
154 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
155 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
156 
157 /*
158  * If user pmap is processed with mmu_booke_remove and the resident count
159  * drops to 0, there are no more pages to remove, so we need not continue.
160  */
161 #define PMAP_REMOVE_DONE(pmap) \
162 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
163 
164 extern void tid_flush(tlbtid_t tid, int tlb0_ways, int tlb0_entries_per_way);
165 extern int elf32_nxstack;
166 
167 /**************************************************************************/
168 /* TLB and TID handling */
169 /**************************************************************************/
170 
171 /* Translation ID busy table */
172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 
174 /*
175  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
176  * core revisions and should be read from h/w registers during early config.
177  */
178 uint32_t tlb0_entries;
179 uint32_t tlb0_ways;
180 uint32_t tlb0_entries_per_way;
181 
182 #define TLB0_ENTRIES		(tlb0_entries)
183 #define TLB0_WAYS		(tlb0_ways)
184 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
185 
186 #define TLB1_ENTRIES 16
187 
188 /* In-ram copy of the TLB1 */
189 static tlb_entry_t tlb1[TLB1_ENTRIES];
190 
191 /* Next free entry in the TLB1 */
192 static unsigned int tlb1_idx;
193 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
194 
195 static tlbtid_t tid_alloc(struct pmap *);
196 
197 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
198 
199 static int tlb1_set_entry(vm_offset_t, vm_paddr_t, vm_size_t, uint32_t);
200 static void tlb1_write_entry(unsigned int);
201 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
202 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
203 
204 static vm_size_t tsize2size(unsigned int);
205 static unsigned int size2tsize(vm_size_t);
206 static unsigned int ilog2(unsigned int);
207 
208 static void set_mas4_defaults(void);
209 
210 static inline void tlb0_flush_entry(vm_offset_t);
211 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
212 
213 /**************************************************************************/
214 /* Page table management */
215 /**************************************************************************/
216 
217 static struct rwlock_padalign pvh_global_lock;
218 
219 /* Data for the pv entry allocation mechanism */
220 static uma_zone_t pvzone;
221 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
222 
223 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
224 
225 #ifndef PMAP_SHPGPERPROC
226 #define PMAP_SHPGPERPROC	200
227 #endif
228 
229 static void ptbl_init(void);
230 static struct ptbl_buf *ptbl_buf_alloc(void);
231 static void ptbl_buf_free(struct ptbl_buf *);
232 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
233 
234 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
235 static void ptbl_free(mmu_t, pmap_t, unsigned int);
236 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
237 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
238 
239 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
240 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
241 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
242 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
243 
244 static pv_entry_t pv_alloc(void);
245 static void pv_free(pv_entry_t);
246 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
247 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
248 
249 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
250 #define PTBL_BUFS		(128 * 16)
251 
252 struct ptbl_buf {
253 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
254 	vm_offset_t kva;		/* va of mapping */
255 };
256 
257 /* ptbl free list and a lock used for access synchronization. */
258 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
259 static struct mtx ptbl_buf_freelist_lock;
260 
261 /* Base address of kva space allocated fot ptbl bufs. */
262 static vm_offset_t ptbl_buf_pool_vabase;
263 
264 /* Pointer to ptbl_buf structures. */
265 static struct ptbl_buf *ptbl_bufs;
266 
267 #ifdef SMP
268 void pmap_bootstrap_ap(volatile uint32_t *);
269 #endif
270 
271 /*
272  * Kernel MMU interface
273  */
274 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
275 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
276     vm_size_t, vm_offset_t);
277 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
278 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
279     vm_offset_t, vm_page_t *, vm_offset_t, int);
280 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t, u_int flags, int8_t psind);
282 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
283     vm_page_t, vm_prot_t);
284 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
285     vm_prot_t);
286 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
287 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
288     vm_prot_t);
289 static void		mmu_booke_init(mmu_t);
290 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
291 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
292 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
293 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
294 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
295     int);
296 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
297     vm_paddr_t *);
298 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
299     vm_object_t, vm_pindex_t, vm_size_t);
300 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
301 static void		mmu_booke_page_init(mmu_t, vm_page_t);
302 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
303 static void		mmu_booke_pinit(mmu_t, pmap_t);
304 static void		mmu_booke_pinit0(mmu_t, pmap_t);
305 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
306     vm_prot_t);
307 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
308 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
309 static void		mmu_booke_release(mmu_t, pmap_t);
310 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
311 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
312 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
313 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
314 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
315 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
316 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
317 static void		mmu_booke_activate(mmu_t, struct thread *);
318 static void		mmu_booke_deactivate(mmu_t, struct thread *);
319 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
320 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
321 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
322 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
323 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
324 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
325 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
326 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
327 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
328 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
329     vm_size_t);
330 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
331     void **);
332 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
333     void *);
334 static void		mmu_booke_scan_init(mmu_t);
335 
336 static mmu_method_t mmu_booke_methods[] = {
337 	/* pmap dispatcher interface */
338 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
339 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
340 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
341 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
342 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
343 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
344 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
345 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
346 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
347 	MMUMETHOD(mmu_init,		mmu_booke_init),
348 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
349 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
350 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
351 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
352 	MMUMETHOD(mmu_map,		mmu_booke_map),
353 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
354 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
355 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
356 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
357 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
358 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
359 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
360 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
361 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
362 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
363 	MMUMETHOD(mmu_release,		mmu_booke_release),
364 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
365 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
366 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
367 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
368 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
369 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
370 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
371 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
372 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
373 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
374 
375 	/* Internal interfaces */
376 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
377 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
378 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
379 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
380 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
381 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
382 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
383 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
384 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
385 
386 	/* dumpsys() support */
387 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
388 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
389 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
390 
391 	{ 0, 0 }
392 };
393 
394 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
395 
396 static __inline uint32_t
397 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
398 {
399 	uint32_t attrib;
400 	int i;
401 
402 	if (ma != VM_MEMATTR_DEFAULT) {
403 		switch (ma) {
404 		case VM_MEMATTR_UNCACHEABLE:
405 			return (PTE_I | PTE_G);
406 		case VM_MEMATTR_WRITE_COMBINING:
407 		case VM_MEMATTR_WRITE_BACK:
408 		case VM_MEMATTR_PREFETCHABLE:
409 			return (PTE_I);
410 		case VM_MEMATTR_WRITE_THROUGH:
411 			return (PTE_W | PTE_M);
412 		}
413 	}
414 
415 	/*
416 	 * Assume the page is cache inhibited and access is guarded unless
417 	 * it's in our available memory array.
418 	 */
419 	attrib = _TLB_ENTRY_IO;
420 	for (i = 0; i < physmem_regions_sz; i++) {
421 		if ((pa >= physmem_regions[i].mr_start) &&
422 		    (pa < (physmem_regions[i].mr_start +
423 		     physmem_regions[i].mr_size))) {
424 			attrib = _TLB_ENTRY_MEM;
425 			break;
426 		}
427 	}
428 
429 	return (attrib);
430 }
431 
432 static inline void
433 tlb_miss_lock(void)
434 {
435 #ifdef SMP
436 	struct pcpu *pc;
437 
438 	if (!smp_started)
439 		return;
440 
441 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
442 		if (pc != pcpup) {
443 
444 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
445 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
446 
447 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
448 			    ("tlb_miss_lock: tried to lock self"));
449 
450 			tlb_lock(pc->pc_booke_tlb_lock);
451 
452 			CTR1(KTR_PMAP, "%s: locked", __func__);
453 		}
454 	}
455 #endif
456 }
457 
458 static inline void
459 tlb_miss_unlock(void)
460 {
461 #ifdef SMP
462 	struct pcpu *pc;
463 
464 	if (!smp_started)
465 		return;
466 
467 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
468 		if (pc != pcpup) {
469 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
470 			    __func__, pc->pc_cpuid);
471 
472 			tlb_unlock(pc->pc_booke_tlb_lock);
473 
474 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
475 		}
476 	}
477 #endif
478 }
479 
480 /* Return number of entries in TLB0. */
481 static __inline void
482 tlb0_get_tlbconf(void)
483 {
484 	uint32_t tlb0_cfg;
485 
486 	tlb0_cfg = mfspr(SPR_TLB0CFG);
487 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
488 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
489 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
490 }
491 
492 /* Initialize pool of kva ptbl buffers. */
493 static void
494 ptbl_init(void)
495 {
496 	int i;
497 
498 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
499 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
500 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
501 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
502 
503 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
504 	TAILQ_INIT(&ptbl_buf_freelist);
505 
506 	for (i = 0; i < PTBL_BUFS; i++) {
507 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
508 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
509 	}
510 }
511 
512 /* Get a ptbl_buf from the freelist. */
513 static struct ptbl_buf *
514 ptbl_buf_alloc(void)
515 {
516 	struct ptbl_buf *buf;
517 
518 	mtx_lock(&ptbl_buf_freelist_lock);
519 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
520 	if (buf != NULL)
521 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
522 	mtx_unlock(&ptbl_buf_freelist_lock);
523 
524 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
525 
526 	return (buf);
527 }
528 
529 /* Return ptbl buff to free pool. */
530 static void
531 ptbl_buf_free(struct ptbl_buf *buf)
532 {
533 
534 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
535 
536 	mtx_lock(&ptbl_buf_freelist_lock);
537 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
538 	mtx_unlock(&ptbl_buf_freelist_lock);
539 }
540 
541 /*
542  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
543  */
544 static void
545 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
546 {
547 	struct ptbl_buf *pbuf;
548 
549 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
550 
551 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
552 
553 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
554 		if (pbuf->kva == (vm_offset_t)ptbl) {
555 			/* Remove from pmap ptbl buf list. */
556 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
557 
558 			/* Free corresponding ptbl buf. */
559 			ptbl_buf_free(pbuf);
560 			break;
561 		}
562 }
563 
564 /* Allocate page table. */
565 static pte_t *
566 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
567 {
568 	vm_page_t mtbl[PTBL_PAGES];
569 	vm_page_t m;
570 	struct ptbl_buf *pbuf;
571 	unsigned int pidx;
572 	pte_t *ptbl;
573 	int i, j;
574 
575 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
576 	    (pmap == kernel_pmap), pdir_idx);
577 
578 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
579 	    ("ptbl_alloc: invalid pdir_idx"));
580 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
581 	    ("pte_alloc: valid ptbl entry exists!"));
582 
583 	pbuf = ptbl_buf_alloc();
584 	if (pbuf == NULL)
585 		panic("pte_alloc: couldn't alloc kernel virtual memory");
586 
587 	ptbl = (pte_t *)pbuf->kva;
588 
589 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
590 
591 	/* Allocate ptbl pages, this will sleep! */
592 	for (i = 0; i < PTBL_PAGES; i++) {
593 		pidx = (PTBL_PAGES * pdir_idx) + i;
594 		while ((m = vm_page_alloc(NULL, pidx,
595 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
596 			PMAP_UNLOCK(pmap);
597 			rw_wunlock(&pvh_global_lock);
598 			if (nosleep) {
599 				ptbl_free_pmap_ptbl(pmap, ptbl);
600 				for (j = 0; j < i; j++)
601 					vm_page_free(mtbl[j]);
602 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
603 				return (NULL);
604 			}
605 			VM_WAIT;
606 			rw_wlock(&pvh_global_lock);
607 			PMAP_LOCK(pmap);
608 		}
609 		mtbl[i] = m;
610 	}
611 
612 	/* Map allocated pages into kernel_pmap. */
613 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
614 
615 	/* Zero whole ptbl. */
616 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
617 
618 	/* Add pbuf to the pmap ptbl bufs list. */
619 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
620 
621 	return (ptbl);
622 }
623 
624 /* Free ptbl pages and invalidate pdir entry. */
625 static void
626 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
627 {
628 	pte_t *ptbl;
629 	vm_paddr_t pa;
630 	vm_offset_t va;
631 	vm_page_t m;
632 	int i;
633 
634 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
635 	    (pmap == kernel_pmap), pdir_idx);
636 
637 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
638 	    ("ptbl_free: invalid pdir_idx"));
639 
640 	ptbl = pmap->pm_pdir[pdir_idx];
641 
642 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
643 
644 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
645 
646 	/*
647 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
648 	 * don't attempt to look up the page tables we are releasing.
649 	 */
650 	mtx_lock_spin(&tlbivax_mutex);
651 	tlb_miss_lock();
652 
653 	pmap->pm_pdir[pdir_idx] = NULL;
654 
655 	tlb_miss_unlock();
656 	mtx_unlock_spin(&tlbivax_mutex);
657 
658 	for (i = 0; i < PTBL_PAGES; i++) {
659 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
660 		pa = pte_vatopa(mmu, kernel_pmap, va);
661 		m = PHYS_TO_VM_PAGE(pa);
662 		vm_page_free_zero(m);
663 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
664 		mmu_booke_kremove(mmu, va);
665 	}
666 
667 	ptbl_free_pmap_ptbl(pmap, ptbl);
668 }
669 
670 /*
671  * Decrement ptbl pages hold count and attempt to free ptbl pages.
672  * Called when removing pte entry from ptbl.
673  *
674  * Return 1 if ptbl pages were freed.
675  */
676 static int
677 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
678 {
679 	pte_t *ptbl;
680 	vm_paddr_t pa;
681 	vm_page_t m;
682 	int i;
683 
684 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
685 	    (pmap == kernel_pmap), pdir_idx);
686 
687 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
688 	    ("ptbl_unhold: invalid pdir_idx"));
689 	KASSERT((pmap != kernel_pmap),
690 	    ("ptbl_unhold: unholding kernel ptbl!"));
691 
692 	ptbl = pmap->pm_pdir[pdir_idx];
693 
694 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
695 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
696 	    ("ptbl_unhold: non kva ptbl"));
697 
698 	/* decrement hold count */
699 	for (i = 0; i < PTBL_PAGES; i++) {
700 		pa = pte_vatopa(mmu, kernel_pmap,
701 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
702 		m = PHYS_TO_VM_PAGE(pa);
703 		m->wire_count--;
704 	}
705 
706 	/*
707 	 * Free ptbl pages if there are no pte etries in this ptbl.
708 	 * wire_count has the same value for all ptbl pages, so check the last
709 	 * page.
710 	 */
711 	if (m->wire_count == 0) {
712 		ptbl_free(mmu, pmap, pdir_idx);
713 
714 		//debugf("ptbl_unhold: e (freed ptbl)\n");
715 		return (1);
716 	}
717 
718 	return (0);
719 }
720 
721 /*
722  * Increment hold count for ptbl pages. This routine is used when a new pte
723  * entry is being inserted into the ptbl.
724  */
725 static void
726 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
727 {
728 	vm_paddr_t pa;
729 	pte_t *ptbl;
730 	vm_page_t m;
731 	int i;
732 
733 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
734 	    pdir_idx);
735 
736 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
737 	    ("ptbl_hold: invalid pdir_idx"));
738 	KASSERT((pmap != kernel_pmap),
739 	    ("ptbl_hold: holding kernel ptbl!"));
740 
741 	ptbl = pmap->pm_pdir[pdir_idx];
742 
743 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
744 
745 	for (i = 0; i < PTBL_PAGES; i++) {
746 		pa = pte_vatopa(mmu, kernel_pmap,
747 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
748 		m = PHYS_TO_VM_PAGE(pa);
749 		m->wire_count++;
750 	}
751 }
752 
753 /* Allocate pv_entry structure. */
754 pv_entry_t
755 pv_alloc(void)
756 {
757 	pv_entry_t pv;
758 
759 	pv_entry_count++;
760 	if (pv_entry_count > pv_entry_high_water)
761 		pagedaemon_wakeup();
762 	pv = uma_zalloc(pvzone, M_NOWAIT);
763 
764 	return (pv);
765 }
766 
767 /* Free pv_entry structure. */
768 static __inline void
769 pv_free(pv_entry_t pve)
770 {
771 
772 	pv_entry_count--;
773 	uma_zfree(pvzone, pve);
774 }
775 
776 
777 /* Allocate and initialize pv_entry structure. */
778 static void
779 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
780 {
781 	pv_entry_t pve;
782 
783 	//int su = (pmap == kernel_pmap);
784 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
785 	//	(u_int32_t)pmap, va, (u_int32_t)m);
786 
787 	pve = pv_alloc();
788 	if (pve == NULL)
789 		panic("pv_insert: no pv entries!");
790 
791 	pve->pv_pmap = pmap;
792 	pve->pv_va = va;
793 
794 	/* add to pv_list */
795 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
796 	rw_assert(&pvh_global_lock, RA_WLOCKED);
797 
798 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
799 
800 	//debugf("pv_insert: e\n");
801 }
802 
803 /* Destroy pv entry. */
804 static void
805 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
806 {
807 	pv_entry_t pve;
808 
809 	//int su = (pmap == kernel_pmap);
810 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
811 
812 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
813 	rw_assert(&pvh_global_lock, RA_WLOCKED);
814 
815 	/* find pv entry */
816 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
817 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
818 			/* remove from pv_list */
819 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
820 			if (TAILQ_EMPTY(&m->md.pv_list))
821 				vm_page_aflag_clear(m, PGA_WRITEABLE);
822 
823 			/* free pv entry struct */
824 			pv_free(pve);
825 			break;
826 		}
827 	}
828 
829 	//debugf("pv_remove: e\n");
830 }
831 
832 /*
833  * Clean pte entry, try to free page table page if requested.
834  *
835  * Return 1 if ptbl pages were freed, otherwise return 0.
836  */
837 static int
838 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
839 {
840 	unsigned int pdir_idx = PDIR_IDX(va);
841 	unsigned int ptbl_idx = PTBL_IDX(va);
842 	vm_page_t m;
843 	pte_t *ptbl;
844 	pte_t *pte;
845 
846 	//int su = (pmap == kernel_pmap);
847 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
848 	//		su, (u_int32_t)pmap, va, flags);
849 
850 	ptbl = pmap->pm_pdir[pdir_idx];
851 	KASSERT(ptbl, ("pte_remove: null ptbl"));
852 
853 	pte = &ptbl[ptbl_idx];
854 
855 	if (pte == NULL || !PTE_ISVALID(pte))
856 		return (0);
857 
858 	if (PTE_ISWIRED(pte))
859 		pmap->pm_stats.wired_count--;
860 
861 	/* Handle managed entry. */
862 	if (PTE_ISMANAGED(pte)) {
863 		/* Get vm_page_t for mapped pte. */
864 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
865 
866 		if (PTE_ISMODIFIED(pte))
867 			vm_page_dirty(m);
868 
869 		if (PTE_ISREFERENCED(pte))
870 			vm_page_aflag_set(m, PGA_REFERENCED);
871 
872 		pv_remove(pmap, va, m);
873 	}
874 
875 	mtx_lock_spin(&tlbivax_mutex);
876 	tlb_miss_lock();
877 
878 	tlb0_flush_entry(va);
879 	pte->flags = 0;
880 	pte->rpn = 0;
881 
882 	tlb_miss_unlock();
883 	mtx_unlock_spin(&tlbivax_mutex);
884 
885 	pmap->pm_stats.resident_count--;
886 
887 	if (flags & PTBL_UNHOLD) {
888 		//debugf("pte_remove: e (unhold)\n");
889 		return (ptbl_unhold(mmu, pmap, pdir_idx));
890 	}
891 
892 	//debugf("pte_remove: e\n");
893 	return (0);
894 }
895 
896 /*
897  * Insert PTE for a given page and virtual address.
898  */
899 static int
900 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
901     boolean_t nosleep)
902 {
903 	unsigned int pdir_idx = PDIR_IDX(va);
904 	unsigned int ptbl_idx = PTBL_IDX(va);
905 	pte_t *ptbl, *pte;
906 
907 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
908 	    pmap == kernel_pmap, pmap, va);
909 
910 	/* Get the page table pointer. */
911 	ptbl = pmap->pm_pdir[pdir_idx];
912 
913 	if (ptbl == NULL) {
914 		/* Allocate page table pages. */
915 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
916 		if (ptbl == NULL) {
917 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
918 			return (ENOMEM);
919 		}
920 	} else {
921 		/*
922 		 * Check if there is valid mapping for requested
923 		 * va, if there is, remove it.
924 		 */
925 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
926 		if (PTE_ISVALID(pte)) {
927 			pte_remove(mmu, pmap, va, PTBL_HOLD);
928 		} else {
929 			/*
930 			 * pte is not used, increment hold count
931 			 * for ptbl pages.
932 			 */
933 			if (pmap != kernel_pmap)
934 				ptbl_hold(mmu, pmap, pdir_idx);
935 		}
936 	}
937 
938 	/*
939 	 * Insert pv_entry into pv_list for mapped page if part of managed
940 	 * memory.
941 	 */
942 	if ((m->oflags & VPO_UNMANAGED) == 0) {
943 		flags |= PTE_MANAGED;
944 
945 		/* Create and insert pv entry. */
946 		pv_insert(pmap, va, m);
947 	}
948 
949 	pmap->pm_stats.resident_count++;
950 
951 	mtx_lock_spin(&tlbivax_mutex);
952 	tlb_miss_lock();
953 
954 	tlb0_flush_entry(va);
955 	if (pmap->pm_pdir[pdir_idx] == NULL) {
956 		/*
957 		 * If we just allocated a new page table, hook it in
958 		 * the pdir.
959 		 */
960 		pmap->pm_pdir[pdir_idx] = ptbl;
961 	}
962 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
963 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
964 	pte->flags |= (PTE_VALID | flags);
965 
966 	tlb_miss_unlock();
967 	mtx_unlock_spin(&tlbivax_mutex);
968 	return (0);
969 }
970 
971 /* Return the pa for the given pmap/va. */
972 static vm_paddr_t
973 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
974 {
975 	vm_paddr_t pa = 0;
976 	pte_t *pte;
977 
978 	pte = pte_find(mmu, pmap, va);
979 	if ((pte != NULL) && PTE_ISVALID(pte))
980 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
981 	return (pa);
982 }
983 
984 /* Get a pointer to a PTE in a page table. */
985 static pte_t *
986 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
987 {
988 	unsigned int pdir_idx = PDIR_IDX(va);
989 	unsigned int ptbl_idx = PTBL_IDX(va);
990 
991 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
992 
993 	if (pmap->pm_pdir[pdir_idx])
994 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
995 
996 	return (NULL);
997 }
998 
999 /**************************************************************************/
1000 /* PMAP related */
1001 /**************************************************************************/
1002 
1003 /*
1004  * This is called during booke_init, before the system is really initialized.
1005  */
1006 static void
1007 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1008 {
1009 	vm_offset_t phys_kernelend;
1010 	struct mem_region *mp, *mp1;
1011 	int cnt, i, j;
1012 	u_int s, e, sz;
1013 	u_int phys_avail_count;
1014 	vm_size_t physsz, hwphyssz, kstack0_sz;
1015 	vm_offset_t kernel_pdir, kstack0, va;
1016 	vm_paddr_t kstack0_phys;
1017 	void *dpcpu;
1018 	pte_t *pte;
1019 
1020 	debugf("mmu_booke_bootstrap: entered\n");
1021 
1022 	/* Set interesting system properties */
1023 	hw_direct_map = 0;
1024 	elf32_nxstack = 1;
1025 
1026 	/* Initialize invalidation mutex */
1027 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1028 
1029 	/* Read TLB0 size and associativity. */
1030 	tlb0_get_tlbconf();
1031 
1032 	/*
1033 	 * Align kernel start and end address (kernel image).
1034 	 * Note that kernel end does not necessarily relate to kernsize.
1035 	 * kernsize is the size of the kernel that is actually mapped.
1036 	 */
1037 	kernstart = trunc_page(start);
1038 	data_start = round_page(kernelend);
1039 	data_end = data_start;
1040 
1041 	/*
1042 	 * Addresses of preloaded modules (like file systems) use
1043 	 * physical addresses. Make sure we relocate those into
1044 	 * virtual addresses.
1045 	 */
1046 	preload_addr_relocate = kernstart - kernload;
1047 
1048 	/* Allocate the dynamic per-cpu area. */
1049 	dpcpu = (void *)data_end;
1050 	data_end += DPCPU_SIZE;
1051 
1052 	/* Allocate space for the message buffer. */
1053 	msgbufp = (struct msgbuf *)data_end;
1054 	data_end += msgbufsize;
1055 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1056 	    data_end);
1057 
1058 	data_end = round_page(data_end);
1059 
1060 	/* Allocate space for ptbl_bufs. */
1061 	ptbl_bufs = (struct ptbl_buf *)data_end;
1062 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1063 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1064 	    data_end);
1065 
1066 	data_end = round_page(data_end);
1067 
1068 	/* Allocate PTE tables for kernel KVA. */
1069 	kernel_pdir = data_end;
1070 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1071 	    PDIR_SIZE - 1) / PDIR_SIZE;
1072 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1073 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1074 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1075 
1076 	debugf(" data_end: 0x%08x\n", data_end);
1077 	if (data_end - kernstart > kernsize) {
1078 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1079 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1080 	}
1081 	data_end = kernstart + kernsize;
1082 	debugf(" updated data_end: 0x%08x\n", data_end);
1083 
1084 	/*
1085 	 * Clear the structures - note we can only do it safely after the
1086 	 * possible additional TLB1 translations are in place (above) so that
1087 	 * all range up to the currently calculated 'data_end' is covered.
1088 	 */
1089 	dpcpu_init(dpcpu, 0);
1090 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1091 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1092 
1093 	/*******************************************************/
1094 	/* Set the start and end of kva. */
1095 	/*******************************************************/
1096 	virtual_avail = round_page(data_end);
1097 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1098 
1099 	/* Allocate KVA space for page zero/copy operations. */
1100 	zero_page_va = virtual_avail;
1101 	virtual_avail += PAGE_SIZE;
1102 	zero_page_idle_va = virtual_avail;
1103 	virtual_avail += PAGE_SIZE;
1104 	copy_page_src_va = virtual_avail;
1105 	virtual_avail += PAGE_SIZE;
1106 	copy_page_dst_va = virtual_avail;
1107 	virtual_avail += PAGE_SIZE;
1108 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1109 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1110 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1111 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1112 
1113 	/* Initialize page zero/copy mutexes. */
1114 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1115 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1116 
1117 	/* Allocate KVA space for ptbl bufs. */
1118 	ptbl_buf_pool_vabase = virtual_avail;
1119 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1120 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1121 	    ptbl_buf_pool_vabase, virtual_avail);
1122 
1123 	/* Calculate corresponding physical addresses for the kernel region. */
1124 	phys_kernelend = kernload + kernsize;
1125 	debugf("kernel image and allocated data:\n");
1126 	debugf(" kernload    = 0x%08x\n", kernload);
1127 	debugf(" kernstart   = 0x%08x\n", kernstart);
1128 	debugf(" kernsize    = 0x%08x\n", kernsize);
1129 
1130 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1131 		panic("mmu_booke_bootstrap: phys_avail too small");
1132 
1133 	/*
1134 	 * Remove kernel physical address range from avail regions list. Page
1135 	 * align all regions.  Non-page aligned memory isn't very interesting
1136 	 * to us.  Also, sort the entries for ascending addresses.
1137 	 */
1138 
1139 	/* Retrieve phys/avail mem regions */
1140 	mem_regions(&physmem_regions, &physmem_regions_sz,
1141 	    &availmem_regions, &availmem_regions_sz);
1142 	sz = 0;
1143 	cnt = availmem_regions_sz;
1144 	debugf("processing avail regions:\n");
1145 	for (mp = availmem_regions; mp->mr_size; mp++) {
1146 		s = mp->mr_start;
1147 		e = mp->mr_start + mp->mr_size;
1148 		debugf(" %08x-%08x -> ", s, e);
1149 		/* Check whether this region holds all of the kernel. */
1150 		if (s < kernload && e > phys_kernelend) {
1151 			availmem_regions[cnt].mr_start = phys_kernelend;
1152 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1153 			e = kernload;
1154 		}
1155 		/* Look whether this regions starts within the kernel. */
1156 		if (s >= kernload && s < phys_kernelend) {
1157 			if (e <= phys_kernelend)
1158 				goto empty;
1159 			s = phys_kernelend;
1160 		}
1161 		/* Now look whether this region ends within the kernel. */
1162 		if (e > kernload && e <= phys_kernelend) {
1163 			if (s >= kernload)
1164 				goto empty;
1165 			e = kernload;
1166 		}
1167 		/* Now page align the start and size of the region. */
1168 		s = round_page(s);
1169 		e = trunc_page(e);
1170 		if (e < s)
1171 			e = s;
1172 		sz = e - s;
1173 		debugf("%08x-%08x = %x\n", s, e, sz);
1174 
1175 		/* Check whether some memory is left here. */
1176 		if (sz == 0) {
1177 		empty:
1178 			memmove(mp, mp + 1,
1179 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1180 			cnt--;
1181 			mp--;
1182 			continue;
1183 		}
1184 
1185 		/* Do an insertion sort. */
1186 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1187 			if (s < mp1->mr_start)
1188 				break;
1189 		if (mp1 < mp) {
1190 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1191 			mp1->mr_start = s;
1192 			mp1->mr_size = sz;
1193 		} else {
1194 			mp->mr_start = s;
1195 			mp->mr_size = sz;
1196 		}
1197 	}
1198 	availmem_regions_sz = cnt;
1199 
1200 	/*******************************************************/
1201 	/* Steal physical memory for kernel stack from the end */
1202 	/* of the first avail region                           */
1203 	/*******************************************************/
1204 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1205 	kstack0_phys = availmem_regions[0].mr_start +
1206 	    availmem_regions[0].mr_size;
1207 	kstack0_phys -= kstack0_sz;
1208 	availmem_regions[0].mr_size -= kstack0_sz;
1209 
1210 	/*******************************************************/
1211 	/* Fill in phys_avail table, based on availmem_regions */
1212 	/*******************************************************/
1213 	phys_avail_count = 0;
1214 	physsz = 0;
1215 	hwphyssz = 0;
1216 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1217 
1218 	debugf("fill in phys_avail:\n");
1219 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1220 
1221 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1222 		    availmem_regions[i].mr_start,
1223 		    availmem_regions[i].mr_start +
1224 		        availmem_regions[i].mr_size,
1225 		    availmem_regions[i].mr_size);
1226 
1227 		if (hwphyssz != 0 &&
1228 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1229 			debugf(" hw.physmem adjust\n");
1230 			if (physsz < hwphyssz) {
1231 				phys_avail[j] = availmem_regions[i].mr_start;
1232 				phys_avail[j + 1] =
1233 				    availmem_regions[i].mr_start +
1234 				    hwphyssz - physsz;
1235 				physsz = hwphyssz;
1236 				phys_avail_count++;
1237 			}
1238 			break;
1239 		}
1240 
1241 		phys_avail[j] = availmem_regions[i].mr_start;
1242 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1243 		    availmem_regions[i].mr_size;
1244 		phys_avail_count++;
1245 		physsz += availmem_regions[i].mr_size;
1246 	}
1247 	physmem = btoc(physsz);
1248 
1249 	/* Calculate the last available physical address. */
1250 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1251 		;
1252 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1253 
1254 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1255 	debugf("phys_avail_count = %d\n", phys_avail_count);
1256 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1257 	    physmem);
1258 
1259 	/*******************************************************/
1260 	/* Initialize (statically allocated) kernel pmap. */
1261 	/*******************************************************/
1262 	PMAP_LOCK_INIT(kernel_pmap);
1263 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1264 
1265 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1266 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1267 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1268 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1269 
1270 	/* Initialize kernel pdir */
1271 	for (i = 0; i < kernel_ptbls; i++)
1272 		kernel_pmap->pm_pdir[kptbl_min + i] =
1273 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1274 
1275 	for (i = 0; i < MAXCPU; i++) {
1276 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1277 
1278 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1279 		tidbusy[i][0] = kernel_pmap;
1280 	}
1281 
1282 	/*
1283 	 * Fill in PTEs covering kernel code and data. They are not required
1284 	 * for address translation, as this area is covered by static TLB1
1285 	 * entries, but for pte_vatopa() to work correctly with kernel area
1286 	 * addresses.
1287 	 */
1288 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1289 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1290 		pte->rpn = kernload + (va - kernstart);
1291 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1292 		    PTE_VALID;
1293 	}
1294 	/* Mark kernel_pmap active on all CPUs */
1295 	CPU_FILL(&kernel_pmap->pm_active);
1296 
1297  	/*
1298 	 * Initialize the global pv list lock.
1299 	 */
1300 	rw_init(&pvh_global_lock, "pmap pv global");
1301 
1302 	/*******************************************************/
1303 	/* Final setup */
1304 	/*******************************************************/
1305 
1306 	/* Enter kstack0 into kernel map, provide guard page */
1307 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1308 	thread0.td_kstack = kstack0;
1309 	thread0.td_kstack_pages = KSTACK_PAGES;
1310 
1311 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1312 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1313 	    kstack0_phys, kstack0_phys + kstack0_sz);
1314 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1315 
1316 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1317 	for (i = 0; i < KSTACK_PAGES; i++) {
1318 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1319 		kstack0 += PAGE_SIZE;
1320 		kstack0_phys += PAGE_SIZE;
1321 	}
1322 
1323 	pmap_bootstrapped = 1;
1324 
1325 	debugf("virtual_avail = %08x\n", virtual_avail);
1326 	debugf("virtual_end   = %08x\n", virtual_end);
1327 
1328 	debugf("mmu_booke_bootstrap: exit\n");
1329 }
1330 
1331 #ifdef SMP
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 #endif
1353 
1354 /*
1355  * Get the physical page address for the given pmap/virtual address.
1356  */
1357 static vm_paddr_t
1358 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1359 {
1360 	vm_paddr_t pa;
1361 
1362 	PMAP_LOCK(pmap);
1363 	pa = pte_vatopa(mmu, pmap, va);
1364 	PMAP_UNLOCK(pmap);
1365 
1366 	return (pa);
1367 }
1368 
1369 /*
1370  * Extract the physical page address associated with the given
1371  * kernel virtual address.
1372  */
1373 static vm_paddr_t
1374 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1375 {
1376 	int i;
1377 
1378 	/* Check TLB1 mappings */
1379 	for (i = 0; i < tlb1_idx; i++) {
1380 		if (!(tlb1[i].mas1 & MAS1_VALID))
1381 			continue;
1382 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1383 			return (tlb1[i].phys + (va - tlb1[i].virt));
1384 	}
1385 
1386 	return (pte_vatopa(mmu, kernel_pmap, va));
1387 }
1388 
1389 /*
1390  * Initialize the pmap module.
1391  * Called by vm_init, to initialize any structures that the pmap
1392  * system needs to map virtual memory.
1393  */
1394 static void
1395 mmu_booke_init(mmu_t mmu)
1396 {
1397 	int shpgperproc = PMAP_SHPGPERPROC;
1398 
1399 	/*
1400 	 * Initialize the address space (zone) for the pv entries.  Set a
1401 	 * high water mark so that the system can recover from excessive
1402 	 * numbers of pv entries.
1403 	 */
1404 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1405 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1406 
1407 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1408 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1409 
1410 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1411 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1412 
1413 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1414 
1415 	/* Pre-fill pvzone with initial number of pv entries. */
1416 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1417 
1418 	/* Initialize ptbl allocation. */
1419 	ptbl_init();
1420 }
1421 
1422 /*
1423  * Map a list of wired pages into kernel virtual address space.  This is
1424  * intended for temporary mappings which do not need page modification or
1425  * references recorded.  Existing mappings in the region are overwritten.
1426  */
1427 static void
1428 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1429 {
1430 	vm_offset_t va;
1431 
1432 	va = sva;
1433 	while (count-- > 0) {
1434 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1435 		va += PAGE_SIZE;
1436 		m++;
1437 	}
1438 }
1439 
1440 /*
1441  * Remove page mappings from kernel virtual address space.  Intended for
1442  * temporary mappings entered by mmu_booke_qenter.
1443  */
1444 static void
1445 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1446 {
1447 	vm_offset_t va;
1448 
1449 	va = sva;
1450 	while (count-- > 0) {
1451 		mmu_booke_kremove(mmu, va);
1452 		va += PAGE_SIZE;
1453 	}
1454 }
1455 
1456 /*
1457  * Map a wired page into kernel virtual address space.
1458  */
1459 static void
1460 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1461 {
1462 
1463 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1464 }
1465 
1466 static void
1467 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1468 {
1469 	unsigned int pdir_idx = PDIR_IDX(va);
1470 	unsigned int ptbl_idx = PTBL_IDX(va);
1471 	uint32_t flags;
1472 	pte_t *pte;
1473 
1474 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1475 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1476 
1477 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1478 	flags |= tlb_calc_wimg(pa, ma);
1479 
1480 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1481 
1482 	mtx_lock_spin(&tlbivax_mutex);
1483 	tlb_miss_lock();
1484 
1485 	if (PTE_ISVALID(pte)) {
1486 
1487 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1488 
1489 		/* Flush entry from TLB0 */
1490 		tlb0_flush_entry(va);
1491 	}
1492 
1493 	pte->rpn = pa & ~PTE_PA_MASK;
1494 	pte->flags = flags;
1495 
1496 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1497 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1498 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1499 
1500 	/* Flush the real memory from the instruction cache. */
1501 	if ((flags & (PTE_I | PTE_G)) == 0) {
1502 		__syncicache((void *)va, PAGE_SIZE);
1503 	}
1504 
1505 	tlb_miss_unlock();
1506 	mtx_unlock_spin(&tlbivax_mutex);
1507 }
1508 
1509 /*
1510  * Remove a page from kernel page table.
1511  */
1512 static void
1513 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1514 {
1515 	unsigned int pdir_idx = PDIR_IDX(va);
1516 	unsigned int ptbl_idx = PTBL_IDX(va);
1517 	pte_t *pte;
1518 
1519 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1520 
1521 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1522 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1523 	    ("mmu_booke_kremove: invalid va"));
1524 
1525 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1526 
1527 	if (!PTE_ISVALID(pte)) {
1528 
1529 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1530 
1531 		return;
1532 	}
1533 
1534 	mtx_lock_spin(&tlbivax_mutex);
1535 	tlb_miss_lock();
1536 
1537 	/* Invalidate entry in TLB0, update PTE. */
1538 	tlb0_flush_entry(va);
1539 	pte->flags = 0;
1540 	pte->rpn = 0;
1541 
1542 	tlb_miss_unlock();
1543 	mtx_unlock_spin(&tlbivax_mutex);
1544 }
1545 
1546 /*
1547  * Initialize pmap associated with process 0.
1548  */
1549 static void
1550 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1551 {
1552 
1553 	PMAP_LOCK_INIT(pmap);
1554 	mmu_booke_pinit(mmu, pmap);
1555 	PCPU_SET(curpmap, pmap);
1556 }
1557 
1558 /*
1559  * Initialize a preallocated and zeroed pmap structure,
1560  * such as one in a vmspace structure.
1561  */
1562 static void
1563 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1564 {
1565 	int i;
1566 
1567 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1568 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1569 
1570 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1571 
1572 	for (i = 0; i < MAXCPU; i++)
1573 		pmap->pm_tid[i] = TID_NONE;
1574 	CPU_ZERO(&kernel_pmap->pm_active);
1575 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1576 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1577 	TAILQ_INIT(&pmap->pm_ptbl_list);
1578 }
1579 
1580 /*
1581  * Release any resources held by the given physical map.
1582  * Called when a pmap initialized by mmu_booke_pinit is being released.
1583  * Should only be called if the map contains no valid mappings.
1584  */
1585 static void
1586 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1587 {
1588 
1589 	KASSERT(pmap->pm_stats.resident_count == 0,
1590 	    ("pmap_release: pmap resident count %ld != 0",
1591 	    pmap->pm_stats.resident_count));
1592 }
1593 
1594 /*
1595  * Insert the given physical page at the specified virtual address in the
1596  * target physical map with the protection requested. If specified the page
1597  * will be wired down.
1598  */
1599 static int
1600 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1601     vm_prot_t prot, u_int flags, int8_t psind)
1602 {
1603 	int error;
1604 
1605 	rw_wlock(&pvh_global_lock);
1606 	PMAP_LOCK(pmap);
1607 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1608 	rw_wunlock(&pvh_global_lock);
1609 	PMAP_UNLOCK(pmap);
1610 	return (error);
1611 }
1612 
1613 static int
1614 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1615     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1616 {
1617 	pte_t *pte;
1618 	vm_paddr_t pa;
1619 	uint32_t flags;
1620 	int error, su, sync;
1621 
1622 	pa = VM_PAGE_TO_PHYS(m);
1623 	su = (pmap == kernel_pmap);
1624 	sync = 0;
1625 
1626 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1627 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1628 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1629 	//		(u_int32_t)m, va, pa, prot, flags);
1630 
1631 	if (su) {
1632 		KASSERT(((va >= virtual_avail) &&
1633 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1634 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1635 	} else {
1636 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1637 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1638 	}
1639 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1640 		VM_OBJECT_ASSERT_LOCKED(m->object);
1641 
1642 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1643 
1644 	/*
1645 	 * If there is an existing mapping, and the physical address has not
1646 	 * changed, must be protection or wiring change.
1647 	 */
1648 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1649 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1650 
1651 		/*
1652 		 * Before actually updating pte->flags we calculate and
1653 		 * prepare its new value in a helper var.
1654 		 */
1655 		flags = pte->flags;
1656 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1657 
1658 		/* Wiring change, just update stats. */
1659 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1660 			if (!PTE_ISWIRED(pte)) {
1661 				flags |= PTE_WIRED;
1662 				pmap->pm_stats.wired_count++;
1663 			}
1664 		} else {
1665 			if (PTE_ISWIRED(pte)) {
1666 				flags &= ~PTE_WIRED;
1667 				pmap->pm_stats.wired_count--;
1668 			}
1669 		}
1670 
1671 		if (prot & VM_PROT_WRITE) {
1672 			/* Add write permissions. */
1673 			flags |= PTE_SW;
1674 			if (!su)
1675 				flags |= PTE_UW;
1676 
1677 			if ((flags & PTE_MANAGED) != 0)
1678 				vm_page_aflag_set(m, PGA_WRITEABLE);
1679 		} else {
1680 			/* Handle modified pages, sense modify status. */
1681 
1682 			/*
1683 			 * The PTE_MODIFIED flag could be set by underlying
1684 			 * TLB misses since we last read it (above), possibly
1685 			 * other CPUs could update it so we check in the PTE
1686 			 * directly rather than rely on that saved local flags
1687 			 * copy.
1688 			 */
1689 			if (PTE_ISMODIFIED(pte))
1690 				vm_page_dirty(m);
1691 		}
1692 
1693 		if (prot & VM_PROT_EXECUTE) {
1694 			flags |= PTE_SX;
1695 			if (!su)
1696 				flags |= PTE_UX;
1697 
1698 			/*
1699 			 * Check existing flags for execute permissions: if we
1700 			 * are turning execute permissions on, icache should
1701 			 * be flushed.
1702 			 */
1703 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1704 				sync++;
1705 		}
1706 
1707 		flags &= ~PTE_REFERENCED;
1708 
1709 		/*
1710 		 * The new flags value is all calculated -- only now actually
1711 		 * update the PTE.
1712 		 */
1713 		mtx_lock_spin(&tlbivax_mutex);
1714 		tlb_miss_lock();
1715 
1716 		tlb0_flush_entry(va);
1717 		pte->flags = flags;
1718 
1719 		tlb_miss_unlock();
1720 		mtx_unlock_spin(&tlbivax_mutex);
1721 
1722 	} else {
1723 		/*
1724 		 * If there is an existing mapping, but it's for a different
1725 		 * physical address, pte_enter() will delete the old mapping.
1726 		 */
1727 		//if ((pte != NULL) && PTE_ISVALID(pte))
1728 		//	debugf("mmu_booke_enter_locked: replace\n");
1729 		//else
1730 		//	debugf("mmu_booke_enter_locked: new\n");
1731 
1732 		/* Now set up the flags and install the new mapping. */
1733 		flags = (PTE_SR | PTE_VALID);
1734 		flags |= PTE_M;
1735 
1736 		if (!su)
1737 			flags |= PTE_UR;
1738 
1739 		if (prot & VM_PROT_WRITE) {
1740 			flags |= PTE_SW;
1741 			if (!su)
1742 				flags |= PTE_UW;
1743 
1744 			if ((m->oflags & VPO_UNMANAGED) == 0)
1745 				vm_page_aflag_set(m, PGA_WRITEABLE);
1746 		}
1747 
1748 		if (prot & VM_PROT_EXECUTE) {
1749 			flags |= PTE_SX;
1750 			if (!su)
1751 				flags |= PTE_UX;
1752 		}
1753 
1754 		/* If its wired update stats. */
1755 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1756 			flags |= PTE_WIRED;
1757 
1758 		error = pte_enter(mmu, pmap, m, va, flags,
1759 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1760 		if (error != 0)
1761 			return (KERN_RESOURCE_SHORTAGE);
1762 
1763 		if ((flags & PMAP_ENTER_WIRED) != 0)
1764 			pmap->pm_stats.wired_count++;
1765 
1766 		/* Flush the real memory from the instruction cache. */
1767 		if (prot & VM_PROT_EXECUTE)
1768 			sync++;
1769 	}
1770 
1771 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1772 		__syncicache((void *)va, PAGE_SIZE);
1773 		sync = 0;
1774 	}
1775 
1776 	return (KERN_SUCCESS);
1777 }
1778 
1779 /*
1780  * Maps a sequence of resident pages belonging to the same object.
1781  * The sequence begins with the given page m_start.  This page is
1782  * mapped at the given virtual address start.  Each subsequent page is
1783  * mapped at a virtual address that is offset from start by the same
1784  * amount as the page is offset from m_start within the object.  The
1785  * last page in the sequence is the page with the largest offset from
1786  * m_start that can be mapped at a virtual address less than the given
1787  * virtual address end.  Not every virtual page between start and end
1788  * is mapped; only those for which a resident page exists with the
1789  * corresponding offset from m_start are mapped.
1790  */
1791 static void
1792 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1793     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1794 {
1795 	vm_page_t m;
1796 	vm_pindex_t diff, psize;
1797 
1798 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1799 
1800 	psize = atop(end - start);
1801 	m = m_start;
1802 	rw_wlock(&pvh_global_lock);
1803 	PMAP_LOCK(pmap);
1804 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1805 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1806 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1807 		    PMAP_ENTER_NOSLEEP, 0);
1808 		m = TAILQ_NEXT(m, listq);
1809 	}
1810 	rw_wunlock(&pvh_global_lock);
1811 	PMAP_UNLOCK(pmap);
1812 }
1813 
1814 static void
1815 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1816     vm_prot_t prot)
1817 {
1818 
1819 	rw_wlock(&pvh_global_lock);
1820 	PMAP_LOCK(pmap);
1821 	mmu_booke_enter_locked(mmu, pmap, va, m,
1822 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1823 	    0);
1824 	rw_wunlock(&pvh_global_lock);
1825 	PMAP_UNLOCK(pmap);
1826 }
1827 
1828 /*
1829  * Remove the given range of addresses from the specified map.
1830  *
1831  * It is assumed that the start and end are properly rounded to the page size.
1832  */
1833 static void
1834 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1835 {
1836 	pte_t *pte;
1837 	uint8_t hold_flag;
1838 
1839 	int su = (pmap == kernel_pmap);
1840 
1841 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1842 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1843 
1844 	if (su) {
1845 		KASSERT(((va >= virtual_avail) &&
1846 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1847 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1848 	} else {
1849 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1850 		    ("mmu_booke_remove: user pmap, non user va"));
1851 	}
1852 
1853 	if (PMAP_REMOVE_DONE(pmap)) {
1854 		//debugf("mmu_booke_remove: e (empty)\n");
1855 		return;
1856 	}
1857 
1858 	hold_flag = PTBL_HOLD_FLAG(pmap);
1859 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1860 
1861 	rw_wlock(&pvh_global_lock);
1862 	PMAP_LOCK(pmap);
1863 	for (; va < endva; va += PAGE_SIZE) {
1864 		pte = pte_find(mmu, pmap, va);
1865 		if ((pte != NULL) && PTE_ISVALID(pte))
1866 			pte_remove(mmu, pmap, va, hold_flag);
1867 	}
1868 	PMAP_UNLOCK(pmap);
1869 	rw_wunlock(&pvh_global_lock);
1870 
1871 	//debugf("mmu_booke_remove: e\n");
1872 }
1873 
1874 /*
1875  * Remove physical page from all pmaps in which it resides.
1876  */
1877 static void
1878 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1879 {
1880 	pv_entry_t pv, pvn;
1881 	uint8_t hold_flag;
1882 
1883 	rw_wlock(&pvh_global_lock);
1884 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1885 		pvn = TAILQ_NEXT(pv, pv_link);
1886 
1887 		PMAP_LOCK(pv->pv_pmap);
1888 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1889 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1890 		PMAP_UNLOCK(pv->pv_pmap);
1891 	}
1892 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1893 	rw_wunlock(&pvh_global_lock);
1894 }
1895 
1896 /*
1897  * Map a range of physical addresses into kernel virtual address space.
1898  */
1899 static vm_offset_t
1900 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1901     vm_paddr_t pa_end, int prot)
1902 {
1903 	vm_offset_t sva = *virt;
1904 	vm_offset_t va = sva;
1905 
1906 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1907 	//		sva, pa_start, pa_end);
1908 
1909 	while (pa_start < pa_end) {
1910 		mmu_booke_kenter(mmu, va, pa_start);
1911 		va += PAGE_SIZE;
1912 		pa_start += PAGE_SIZE;
1913 	}
1914 	*virt = va;
1915 
1916 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1917 	return (sva);
1918 }
1919 
1920 /*
1921  * The pmap must be activated before it's address space can be accessed in any
1922  * way.
1923  */
1924 static void
1925 mmu_booke_activate(mmu_t mmu, struct thread *td)
1926 {
1927 	pmap_t pmap;
1928 	u_int cpuid;
1929 
1930 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1931 
1932 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1933 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1934 
1935 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1936 
1937 	sched_pin();
1938 
1939 	cpuid = PCPU_GET(cpuid);
1940 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1941 	PCPU_SET(curpmap, pmap);
1942 
1943 	if (pmap->pm_tid[cpuid] == TID_NONE)
1944 		tid_alloc(pmap);
1945 
1946 	/* Load PID0 register with pmap tid value. */
1947 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1948 	__asm __volatile("isync");
1949 
1950 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1951 
1952 	sched_unpin();
1953 
1954 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1955 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1956 }
1957 
1958 /*
1959  * Deactivate the specified process's address space.
1960  */
1961 static void
1962 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1963 {
1964 	pmap_t pmap;
1965 
1966 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1967 
1968 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1969 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1970 
1971 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1972 
1973 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1974 	PCPU_SET(curpmap, NULL);
1975 }
1976 
1977 /*
1978  * Copy the range specified by src_addr/len
1979  * from the source map to the range dst_addr/len
1980  * in the destination map.
1981  *
1982  * This routine is only advisory and need not do anything.
1983  */
1984 static void
1985 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1986     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1987 {
1988 
1989 }
1990 
1991 /*
1992  * Set the physical protection on the specified range of this map as requested.
1993  */
1994 static void
1995 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1996     vm_prot_t prot)
1997 {
1998 	vm_offset_t va;
1999 	vm_page_t m;
2000 	pte_t *pte;
2001 
2002 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2003 		mmu_booke_remove(mmu, pmap, sva, eva);
2004 		return;
2005 	}
2006 
2007 	if (prot & VM_PROT_WRITE)
2008 		return;
2009 
2010 	PMAP_LOCK(pmap);
2011 	for (va = sva; va < eva; va += PAGE_SIZE) {
2012 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2013 			if (PTE_ISVALID(pte)) {
2014 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2015 
2016 				mtx_lock_spin(&tlbivax_mutex);
2017 				tlb_miss_lock();
2018 
2019 				/* Handle modified pages. */
2020 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2021 					vm_page_dirty(m);
2022 
2023 				tlb0_flush_entry(va);
2024 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2025 
2026 				tlb_miss_unlock();
2027 				mtx_unlock_spin(&tlbivax_mutex);
2028 			}
2029 		}
2030 	}
2031 	PMAP_UNLOCK(pmap);
2032 }
2033 
2034 /*
2035  * Clear the write and modified bits in each of the given page's mappings.
2036  */
2037 static void
2038 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2039 {
2040 	pv_entry_t pv;
2041 	pte_t *pte;
2042 
2043 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2044 	    ("mmu_booke_remove_write: page %p is not managed", m));
2045 
2046 	/*
2047 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2048 	 * set by another thread while the object is locked.  Thus,
2049 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2050 	 */
2051 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2052 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2053 		return;
2054 	rw_wlock(&pvh_global_lock);
2055 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2056 		PMAP_LOCK(pv->pv_pmap);
2057 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2058 			if (PTE_ISVALID(pte)) {
2059 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2060 
2061 				mtx_lock_spin(&tlbivax_mutex);
2062 				tlb_miss_lock();
2063 
2064 				/* Handle modified pages. */
2065 				if (PTE_ISMODIFIED(pte))
2066 					vm_page_dirty(m);
2067 
2068 				/* Flush mapping from TLB0. */
2069 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2070 
2071 				tlb_miss_unlock();
2072 				mtx_unlock_spin(&tlbivax_mutex);
2073 			}
2074 		}
2075 		PMAP_UNLOCK(pv->pv_pmap);
2076 	}
2077 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2078 	rw_wunlock(&pvh_global_lock);
2079 }
2080 
2081 static void
2082 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2083 {
2084 	pte_t *pte;
2085 	pmap_t pmap;
2086 	vm_page_t m;
2087 	vm_offset_t addr;
2088 	vm_paddr_t pa = 0;
2089 	int active, valid;
2090 
2091 	va = trunc_page(va);
2092 	sz = round_page(sz);
2093 
2094 	rw_wlock(&pvh_global_lock);
2095 	pmap = PCPU_GET(curpmap);
2096 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2097 	while (sz > 0) {
2098 		PMAP_LOCK(pm);
2099 		pte = pte_find(mmu, pm, va);
2100 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2101 		if (valid)
2102 			pa = PTE_PA(pte);
2103 		PMAP_UNLOCK(pm);
2104 		if (valid) {
2105 			if (!active) {
2106 				/* Create a mapping in the active pmap. */
2107 				addr = 0;
2108 				m = PHYS_TO_VM_PAGE(pa);
2109 				PMAP_LOCK(pmap);
2110 				pte_enter(mmu, pmap, m, addr,
2111 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2112 				__syncicache((void *)addr, PAGE_SIZE);
2113 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2114 				PMAP_UNLOCK(pmap);
2115 			} else
2116 				__syncicache((void *)va, PAGE_SIZE);
2117 		}
2118 		va += PAGE_SIZE;
2119 		sz -= PAGE_SIZE;
2120 	}
2121 	rw_wunlock(&pvh_global_lock);
2122 }
2123 
2124 /*
2125  * Atomically extract and hold the physical page with the given
2126  * pmap and virtual address pair if that mapping permits the given
2127  * protection.
2128  */
2129 static vm_page_t
2130 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2131     vm_prot_t prot)
2132 {
2133 	pte_t *pte;
2134 	vm_page_t m;
2135 	uint32_t pte_wbit;
2136 	vm_paddr_t pa;
2137 
2138 	m = NULL;
2139 	pa = 0;
2140 	PMAP_LOCK(pmap);
2141 retry:
2142 	pte = pte_find(mmu, pmap, va);
2143 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2144 		if (pmap == kernel_pmap)
2145 			pte_wbit = PTE_SW;
2146 		else
2147 			pte_wbit = PTE_UW;
2148 
2149 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2150 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2151 				goto retry;
2152 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2153 			vm_page_hold(m);
2154 		}
2155 	}
2156 
2157 	PA_UNLOCK_COND(pa);
2158 	PMAP_UNLOCK(pmap);
2159 	return (m);
2160 }
2161 
2162 /*
2163  * Initialize a vm_page's machine-dependent fields.
2164  */
2165 static void
2166 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2167 {
2168 
2169 	TAILQ_INIT(&m->md.pv_list);
2170 }
2171 
2172 /*
2173  * mmu_booke_zero_page_area zeros the specified hardware page by
2174  * mapping it into virtual memory and using bzero to clear
2175  * its contents.
2176  *
2177  * off and size must reside within a single page.
2178  */
2179 static void
2180 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2181 {
2182 	vm_offset_t va;
2183 
2184 	/* XXX KASSERT off and size are within a single page? */
2185 
2186 	mtx_lock(&zero_page_mutex);
2187 	va = zero_page_va;
2188 
2189 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2190 	bzero((caddr_t)va + off, size);
2191 	mmu_booke_kremove(mmu, va);
2192 
2193 	mtx_unlock(&zero_page_mutex);
2194 }
2195 
2196 /*
2197  * mmu_booke_zero_page zeros the specified hardware page.
2198  */
2199 static void
2200 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2201 {
2202 
2203 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2204 }
2205 
2206 /*
2207  * mmu_booke_copy_page copies the specified (machine independent) page by
2208  * mapping the page into virtual memory and using memcopy to copy the page,
2209  * one machine dependent page at a time.
2210  */
2211 static void
2212 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2213 {
2214 	vm_offset_t sva, dva;
2215 
2216 	sva = copy_page_src_va;
2217 	dva = copy_page_dst_va;
2218 
2219 	mtx_lock(&copy_page_mutex);
2220 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2221 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2222 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2223 	mmu_booke_kremove(mmu, dva);
2224 	mmu_booke_kremove(mmu, sva);
2225 	mtx_unlock(&copy_page_mutex);
2226 }
2227 
2228 static inline void
2229 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2230     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2231 {
2232 	void *a_cp, *b_cp;
2233 	vm_offset_t a_pg_offset, b_pg_offset;
2234 	int cnt;
2235 
2236 	mtx_lock(&copy_page_mutex);
2237 	while (xfersize > 0) {
2238 		a_pg_offset = a_offset & PAGE_MASK;
2239 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2240 		mmu_booke_kenter(mmu, copy_page_src_va,
2241 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2242 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2243 		b_pg_offset = b_offset & PAGE_MASK;
2244 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2245 		mmu_booke_kenter(mmu, copy_page_dst_va,
2246 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2247 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2248 		bcopy(a_cp, b_cp, cnt);
2249 		mmu_booke_kremove(mmu, copy_page_dst_va);
2250 		mmu_booke_kremove(mmu, copy_page_src_va);
2251 		a_offset += cnt;
2252 		b_offset += cnt;
2253 		xfersize -= cnt;
2254 	}
2255 	mtx_unlock(&copy_page_mutex);
2256 }
2257 
2258 /*
2259  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2260  * into virtual memory and using bzero to clear its contents. This is intended
2261  * to be called from the vm_pagezero process only and outside of Giant. No
2262  * lock is required.
2263  */
2264 static void
2265 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2266 {
2267 	vm_offset_t va;
2268 
2269 	va = zero_page_idle_va;
2270 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2271 	bzero((caddr_t)va, PAGE_SIZE);
2272 	mmu_booke_kremove(mmu, va);
2273 }
2274 
2275 /*
2276  * Return whether or not the specified physical page was modified
2277  * in any of physical maps.
2278  */
2279 static boolean_t
2280 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2281 {
2282 	pte_t *pte;
2283 	pv_entry_t pv;
2284 	boolean_t rv;
2285 
2286 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2287 	    ("mmu_booke_is_modified: page %p is not managed", m));
2288 	rv = FALSE;
2289 
2290 	/*
2291 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2292 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2293 	 * is clear, no PTEs can be modified.
2294 	 */
2295 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2296 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2297 		return (rv);
2298 	rw_wlock(&pvh_global_lock);
2299 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2300 		PMAP_LOCK(pv->pv_pmap);
2301 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2302 		    PTE_ISVALID(pte)) {
2303 			if (PTE_ISMODIFIED(pte))
2304 				rv = TRUE;
2305 		}
2306 		PMAP_UNLOCK(pv->pv_pmap);
2307 		if (rv)
2308 			break;
2309 	}
2310 	rw_wunlock(&pvh_global_lock);
2311 	return (rv);
2312 }
2313 
2314 /*
2315  * Return whether or not the specified virtual address is eligible
2316  * for prefault.
2317  */
2318 static boolean_t
2319 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2320 {
2321 
2322 	return (FALSE);
2323 }
2324 
2325 /*
2326  * Return whether or not the specified physical page was referenced
2327  * in any physical maps.
2328  */
2329 static boolean_t
2330 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2331 {
2332 	pte_t *pte;
2333 	pv_entry_t pv;
2334 	boolean_t rv;
2335 
2336 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2337 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2338 	rv = FALSE;
2339 	rw_wlock(&pvh_global_lock);
2340 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2341 		PMAP_LOCK(pv->pv_pmap);
2342 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2343 		    PTE_ISVALID(pte)) {
2344 			if (PTE_ISREFERENCED(pte))
2345 				rv = TRUE;
2346 		}
2347 		PMAP_UNLOCK(pv->pv_pmap);
2348 		if (rv)
2349 			break;
2350 	}
2351 	rw_wunlock(&pvh_global_lock);
2352 	return (rv);
2353 }
2354 
2355 /*
2356  * Clear the modify bits on the specified physical page.
2357  */
2358 static void
2359 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2360 {
2361 	pte_t *pte;
2362 	pv_entry_t pv;
2363 
2364 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2365 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2366 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2367 	KASSERT(!vm_page_xbusied(m),
2368 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2369 
2370 	/*
2371 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2372 	 * If the object containing the page is locked and the page is not
2373 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2374 	 */
2375 	if ((m->aflags & PGA_WRITEABLE) == 0)
2376 		return;
2377 	rw_wlock(&pvh_global_lock);
2378 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2379 		PMAP_LOCK(pv->pv_pmap);
2380 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2381 		    PTE_ISVALID(pte)) {
2382 			mtx_lock_spin(&tlbivax_mutex);
2383 			tlb_miss_lock();
2384 
2385 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2386 				tlb0_flush_entry(pv->pv_va);
2387 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2388 				    PTE_REFERENCED);
2389 			}
2390 
2391 			tlb_miss_unlock();
2392 			mtx_unlock_spin(&tlbivax_mutex);
2393 		}
2394 		PMAP_UNLOCK(pv->pv_pmap);
2395 	}
2396 	rw_wunlock(&pvh_global_lock);
2397 }
2398 
2399 /*
2400  * Return a count of reference bits for a page, clearing those bits.
2401  * It is not necessary for every reference bit to be cleared, but it
2402  * is necessary that 0 only be returned when there are truly no
2403  * reference bits set.
2404  *
2405  * XXX: The exact number of bits to check and clear is a matter that
2406  * should be tested and standardized at some point in the future for
2407  * optimal aging of shared pages.
2408  */
2409 static int
2410 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2411 {
2412 	pte_t *pte;
2413 	pv_entry_t pv;
2414 	int count;
2415 
2416 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2417 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2418 	count = 0;
2419 	rw_wlock(&pvh_global_lock);
2420 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2421 		PMAP_LOCK(pv->pv_pmap);
2422 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2423 		    PTE_ISVALID(pte)) {
2424 			if (PTE_ISREFERENCED(pte)) {
2425 				mtx_lock_spin(&tlbivax_mutex);
2426 				tlb_miss_lock();
2427 
2428 				tlb0_flush_entry(pv->pv_va);
2429 				pte->flags &= ~PTE_REFERENCED;
2430 
2431 				tlb_miss_unlock();
2432 				mtx_unlock_spin(&tlbivax_mutex);
2433 
2434 				if (++count > 4) {
2435 					PMAP_UNLOCK(pv->pv_pmap);
2436 					break;
2437 				}
2438 			}
2439 		}
2440 		PMAP_UNLOCK(pv->pv_pmap);
2441 	}
2442 	rw_wunlock(&pvh_global_lock);
2443 	return (count);
2444 }
2445 
2446 /*
2447  * Clear the wired attribute from the mappings for the specified range of
2448  * addresses in the given pmap.  Every valid mapping within that range must
2449  * have the wired attribute set.  In contrast, invalid mappings cannot have
2450  * the wired attribute set, so they are ignored.
2451  *
2452  * The wired attribute of the page table entry is not a hardware feature, so
2453  * there is no need to invalidate any TLB entries.
2454  */
2455 static void
2456 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2457 {
2458 	vm_offset_t va;
2459 	pte_t *pte;
2460 
2461 	PMAP_LOCK(pmap);
2462 	for (va = sva; va < eva; va += PAGE_SIZE) {
2463 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2464 		    PTE_ISVALID(pte)) {
2465 			if (!PTE_ISWIRED(pte))
2466 				panic("mmu_booke_unwire: pte %p isn't wired",
2467 				    pte);
2468 			pte->flags &= ~PTE_WIRED;
2469 			pmap->pm_stats.wired_count--;
2470 		}
2471 	}
2472 	PMAP_UNLOCK(pmap);
2473 
2474 }
2475 
2476 /*
2477  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2478  * page.  This count may be changed upwards or downwards in the future; it is
2479  * only necessary that true be returned for a small subset of pmaps for proper
2480  * page aging.
2481  */
2482 static boolean_t
2483 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2484 {
2485 	pv_entry_t pv;
2486 	int loops;
2487 	boolean_t rv;
2488 
2489 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2490 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2491 	loops = 0;
2492 	rv = FALSE;
2493 	rw_wlock(&pvh_global_lock);
2494 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2495 		if (pv->pv_pmap == pmap) {
2496 			rv = TRUE;
2497 			break;
2498 		}
2499 		if (++loops >= 16)
2500 			break;
2501 	}
2502 	rw_wunlock(&pvh_global_lock);
2503 	return (rv);
2504 }
2505 
2506 /*
2507  * Return the number of managed mappings to the given physical page that are
2508  * wired.
2509  */
2510 static int
2511 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2512 {
2513 	pv_entry_t pv;
2514 	pte_t *pte;
2515 	int count = 0;
2516 
2517 	if ((m->oflags & VPO_UNMANAGED) != 0)
2518 		return (count);
2519 	rw_wlock(&pvh_global_lock);
2520 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2521 		PMAP_LOCK(pv->pv_pmap);
2522 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2523 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2524 				count++;
2525 		PMAP_UNLOCK(pv->pv_pmap);
2526 	}
2527 	rw_wunlock(&pvh_global_lock);
2528 	return (count);
2529 }
2530 
2531 static int
2532 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2533 {
2534 	int i;
2535 	vm_offset_t va;
2536 
2537 	/*
2538 	 * This currently does not work for entries that
2539 	 * overlap TLB1 entries.
2540 	 */
2541 	for (i = 0; i < tlb1_idx; i ++) {
2542 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2543 			return (0);
2544 	}
2545 
2546 	return (EFAULT);
2547 }
2548 
2549 void
2550 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2551 {
2552 	vm_paddr_t ppa;
2553 	vm_offset_t ofs;
2554 	vm_size_t gran;
2555 
2556 	/* Minidumps are based on virtual memory addresses. */
2557 	if (do_minidump) {
2558 		*va = (void *)pa;
2559 		return;
2560 	}
2561 
2562 	/* Raw physical memory dumps don't have a virtual address. */
2563 	/* We always map a 256MB page at 256M. */
2564 	gran = 256 * 1024 * 1024;
2565 	ppa = pa & ~(gran - 1);
2566 	ofs = pa - ppa;
2567 	*va = (void *)gran;
2568 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2569 
2570 	if (sz > (gran - ofs))
2571 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2572 		    _TLB_ENTRY_IO);
2573 }
2574 
2575 void
2576 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2577 {
2578 	vm_paddr_t ppa;
2579 	vm_offset_t ofs;
2580 	vm_size_t gran;
2581 
2582 	/* Minidumps are based on virtual memory addresses. */
2583 	/* Nothing to do... */
2584 	if (do_minidump)
2585 		return;
2586 
2587 	/* Raw physical memory dumps don't have a virtual address. */
2588 	tlb1_idx--;
2589 	tlb1[tlb1_idx].mas1 = 0;
2590 	tlb1[tlb1_idx].mas2 = 0;
2591 	tlb1[tlb1_idx].mas3 = 0;
2592 	tlb1_write_entry(tlb1_idx);
2593 
2594 	gran = 256 * 1024 * 1024;
2595 	ppa = pa & ~(gran - 1);
2596 	ofs = pa - ppa;
2597 	if (sz > (gran - ofs)) {
2598 		tlb1_idx--;
2599 		tlb1[tlb1_idx].mas1 = 0;
2600 		tlb1[tlb1_idx].mas2 = 0;
2601 		tlb1[tlb1_idx].mas3 = 0;
2602 		tlb1_write_entry(tlb1_idx);
2603 	}
2604 }
2605 
2606 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2607 
2608 void
2609 mmu_booke_scan_init(mmu_t mmu)
2610 {
2611 	vm_offset_t va;
2612 	pte_t *pte;
2613 	int i;
2614 
2615 	if (!do_minidump) {
2616 		/* Initialize phys. segments for dumpsys(). */
2617 		memset(&dump_map, 0, sizeof(dump_map));
2618 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2619 		    &availmem_regions_sz);
2620 		for (i = 0; i < physmem_regions_sz; i++) {
2621 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2622 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2623 		}
2624 		return;
2625 	}
2626 
2627 	/* Virtual segments for minidumps: */
2628 	memset(&dump_map, 0, sizeof(dump_map));
2629 
2630 	/* 1st: kernel .data and .bss. */
2631 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2632 	dump_map[0].pa_size =
2633 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2634 
2635 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2636 	dump_map[1].pa_start = data_start;
2637 	dump_map[1].pa_size = data_end - data_start;
2638 
2639 	/* 3rd: kernel VM. */
2640 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2641 	/* Find start of next chunk (from va). */
2642 	while (va < virtual_end) {
2643 		/* Don't dump the buffer cache. */
2644 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2645 			va = kmi.buffer_eva;
2646 			continue;
2647 		}
2648 		pte = pte_find(mmu, kernel_pmap, va);
2649 		if (pte != NULL && PTE_ISVALID(pte))
2650 			break;
2651 		va += PAGE_SIZE;
2652 	}
2653 	if (va < virtual_end) {
2654 		dump_map[2].pa_start = va;
2655 		va += PAGE_SIZE;
2656 		/* Find last page in chunk. */
2657 		while (va < virtual_end) {
2658 			/* Don't run into the buffer cache. */
2659 			if (va == kmi.buffer_sva)
2660 				break;
2661 			pte = pte_find(mmu, kernel_pmap, va);
2662 			if (pte == NULL || !PTE_ISVALID(pte))
2663 				break;
2664 			va += PAGE_SIZE;
2665 		}
2666 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2667 	}
2668 }
2669 
2670 /*
2671  * Map a set of physical memory pages into the kernel virtual address space.
2672  * Return a pointer to where it is mapped. This routine is intended to be used
2673  * for mapping device memory, NOT real memory.
2674  */
2675 static void *
2676 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2677 {
2678 
2679 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2680 }
2681 
2682 static void *
2683 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2684 {
2685 	void *res;
2686 	uintptr_t va;
2687 	vm_size_t sz;
2688 	int i;
2689 
2690 	/*
2691 	 * Check if this is premapped in TLB1. Note: this should probably also
2692 	 * check whether a sequence of TLB1 entries exist that match the
2693 	 * requirement, but now only checks the easy case.
2694 	 */
2695 	if (ma == VM_MEMATTR_DEFAULT) {
2696 		for (i = 0; i < tlb1_idx; i++) {
2697 			if (!(tlb1[i].mas1 & MAS1_VALID))
2698 				continue;
2699 			if (pa >= tlb1[i].phys &&
2700 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2701 				return (void *)(tlb1[i].virt +
2702 				    (pa - tlb1[i].phys));
2703 		}
2704 	}
2705 
2706 	size = roundup(size, PAGE_SIZE);
2707 
2708 	/*
2709 	 * We leave a hole for device direct mapping between the maximum user
2710 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2711 	 * devices are in there, just map them 1:1. If not, map them to the
2712 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2713 	 * addresses should be pulled from an allocator, but since we do not
2714 	 * ever free TLB1 entries, it is safe just to increment a counter.
2715 	 * Note that there isn't a lot of address space here (128 MB) and it
2716 	 * is not at all difficult to imagine running out, since that is a 4:1
2717 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2718 	 * mapped there.
2719 	 */
2720 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2721 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2722 		va = pa;
2723 	else
2724 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2725 	res = (void *)va;
2726 
2727 	do {
2728 		sz = 1 << (ilog2(size) & ~1);
2729 		if (bootverbose)
2730 			printf("Wiring VA=%x to PA=%x (size=%x), "
2731 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2732 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2733 		size -= sz;
2734 		pa += sz;
2735 		va += sz;
2736 	} while (size > 0);
2737 
2738 	return (res);
2739 }
2740 
2741 /*
2742  * 'Unmap' a range mapped by mmu_booke_mapdev().
2743  */
2744 static void
2745 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2746 {
2747 #ifdef SUPPORTS_SHRINKING_TLB1
2748 	vm_offset_t base, offset;
2749 
2750 	/*
2751 	 * Unmap only if this is inside kernel virtual space.
2752 	 */
2753 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2754 		base = trunc_page(va);
2755 		offset = va & PAGE_MASK;
2756 		size = roundup(offset + size, PAGE_SIZE);
2757 		kva_free(base, size);
2758 	}
2759 #endif
2760 }
2761 
2762 /*
2763  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2764  * specified pmap. This eliminates the blast of soft faults on process startup
2765  * and immediately after an mmap.
2766  */
2767 static void
2768 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2769     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2770 {
2771 
2772 	VM_OBJECT_ASSERT_WLOCKED(object);
2773 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2774 	    ("mmu_booke_object_init_pt: non-device object"));
2775 }
2776 
2777 /*
2778  * Perform the pmap work for mincore.
2779  */
2780 static int
2781 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2782     vm_paddr_t *locked_pa)
2783 {
2784 
2785 	/* XXX: this should be implemented at some point */
2786 	return (0);
2787 }
2788 
2789 /**************************************************************************/
2790 /* TID handling */
2791 /**************************************************************************/
2792 
2793 /*
2794  * Allocate a TID. If necessary, steal one from someone else.
2795  * The new TID is flushed from the TLB before returning.
2796  */
2797 static tlbtid_t
2798 tid_alloc(pmap_t pmap)
2799 {
2800 	tlbtid_t tid;
2801 	int thiscpu;
2802 
2803 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2804 
2805 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2806 
2807 	thiscpu = PCPU_GET(cpuid);
2808 
2809 	tid = PCPU_GET(tid_next);
2810 	if (tid > TID_MAX)
2811 		tid = TID_MIN;
2812 	PCPU_SET(tid_next, tid + 1);
2813 
2814 	/* If we are stealing TID then clear the relevant pmap's field */
2815 	if (tidbusy[thiscpu][tid] != NULL) {
2816 
2817 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2818 
2819 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2820 
2821 		/* Flush all entries from TLB0 matching this TID. */
2822 		tid_flush(tid, tlb0_ways, tlb0_entries_per_way);
2823 	}
2824 
2825 	tidbusy[thiscpu][tid] = pmap;
2826 	pmap->pm_tid[thiscpu] = tid;
2827 	__asm __volatile("msync; isync");
2828 
2829 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2830 	    PCPU_GET(tid_next));
2831 
2832 	return (tid);
2833 }
2834 
2835 /**************************************************************************/
2836 /* TLB0 handling */
2837 /**************************************************************************/
2838 
2839 static void
2840 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2841     uint32_t mas7)
2842 {
2843 	int as;
2844 	char desc[3];
2845 	tlbtid_t tid;
2846 	vm_size_t size;
2847 	unsigned int tsize;
2848 
2849 	desc[2] = '\0';
2850 	if (mas1 & MAS1_VALID)
2851 		desc[0] = 'V';
2852 	else
2853 		desc[0] = ' ';
2854 
2855 	if (mas1 & MAS1_IPROT)
2856 		desc[1] = 'P';
2857 	else
2858 		desc[1] = ' ';
2859 
2860 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2861 	tid = MAS1_GETTID(mas1);
2862 
2863 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2864 	size = 0;
2865 	if (tsize)
2866 		size = tsize2size(tsize);
2867 
2868 	debugf("%3d: (%s) [AS=%d] "
2869 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2870 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2871 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2872 }
2873 
2874 /* Convert TLB0 va and way number to tlb0[] table index. */
2875 static inline unsigned int
2876 tlb0_tableidx(vm_offset_t va, unsigned int way)
2877 {
2878 	unsigned int idx;
2879 
2880 	idx = (way * TLB0_ENTRIES_PER_WAY);
2881 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2882 	return (idx);
2883 }
2884 
2885 /*
2886  * Invalidate TLB0 entry.
2887  */
2888 static inline void
2889 tlb0_flush_entry(vm_offset_t va)
2890 {
2891 
2892 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2893 
2894 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2895 
2896 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2897 	__asm __volatile("isync; msync");
2898 	__asm __volatile("tlbsync; msync");
2899 
2900 	CTR1(KTR_PMAP, "%s: e", __func__);
2901 }
2902 
2903 /* Print out contents of the MAS registers for each TLB0 entry */
2904 void
2905 tlb0_print_tlbentries(void)
2906 {
2907 	uint32_t mas0, mas1, mas2, mas3, mas7;
2908 	int entryidx, way, idx;
2909 
2910 	debugf("TLB0 entries:\n");
2911 	for (way = 0; way < TLB0_WAYS; way ++)
2912 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2913 
2914 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2915 			mtspr(SPR_MAS0, mas0);
2916 			__asm __volatile("isync");
2917 
2918 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2919 			mtspr(SPR_MAS2, mas2);
2920 
2921 			__asm __volatile("isync; tlbre");
2922 
2923 			mas1 = mfspr(SPR_MAS1);
2924 			mas2 = mfspr(SPR_MAS2);
2925 			mas3 = mfspr(SPR_MAS3);
2926 			mas7 = mfspr(SPR_MAS7);
2927 
2928 			idx = tlb0_tableidx(mas2, way);
2929 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2930 		}
2931 }
2932 
2933 /**************************************************************************/
2934 /* TLB1 handling */
2935 /**************************************************************************/
2936 
2937 /*
2938  * TLB1 mapping notes:
2939  *
2940  * TLB1[0]	Kernel text and data.
2941  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2942  *		windows, other devices mappings.
2943  */
2944 
2945 /*
2946  * Write given entry to TLB1 hardware.
2947  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2948  */
2949 static void
2950 tlb1_write_entry(unsigned int idx)
2951 {
2952 	uint32_t mas0, mas7;
2953 
2954 	//debugf("tlb1_write_entry: s\n");
2955 
2956 	/* Clear high order RPN bits */
2957 	mas7 = 0;
2958 
2959 	/* Select entry */
2960 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2961 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2962 
2963 	mtspr(SPR_MAS0, mas0);
2964 	__asm __volatile("isync");
2965 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2966 	__asm __volatile("isync");
2967 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2968 	__asm __volatile("isync");
2969 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2970 	__asm __volatile("isync");
2971 	mtspr(SPR_MAS7, mas7);
2972 	__asm __volatile("isync; tlbwe; isync; msync");
2973 
2974 	//debugf("tlb1_write_entry: e\n");
2975 }
2976 
2977 /*
2978  * Return the largest uint value log such that 2^log <= num.
2979  */
2980 static unsigned int
2981 ilog2(unsigned int num)
2982 {
2983 	int lz;
2984 
2985 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2986 	return (31 - lz);
2987 }
2988 
2989 /*
2990  * Convert TLB TSIZE value to mapped region size.
2991  */
2992 static vm_size_t
2993 tsize2size(unsigned int tsize)
2994 {
2995 
2996 	/*
2997 	 * size = 4^tsize KB
2998 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2999 	 */
3000 
3001 	return ((1 << (2 * tsize)) * 1024);
3002 }
3003 
3004 /*
3005  * Convert region size (must be power of 4) to TLB TSIZE value.
3006  */
3007 static unsigned int
3008 size2tsize(vm_size_t size)
3009 {
3010 
3011 	return (ilog2(size) / 2 - 5);
3012 }
3013 
3014 /*
3015  * Register permanent kernel mapping in TLB1.
3016  *
3017  * Entries are created starting from index 0 (current free entry is
3018  * kept in tlb1_idx) and are not supposed to be invalidated.
3019  */
3020 static int
3021 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3022     uint32_t flags)
3023 {
3024 	uint32_t ts, tid;
3025 	int tsize, index;
3026 
3027 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3028 	if (index >= TLB1_ENTRIES) {
3029 		printf("tlb1_set_entry: TLB1 full!\n");
3030 		return (-1);
3031 	}
3032 
3033 	/* Convert size to TSIZE */
3034 	tsize = size2tsize(size);
3035 
3036 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3037 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3038 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3039 
3040 	/*
3041 	 * Atomicity is preserved by the atomic increment above since nothing
3042 	 * is ever removed from tlb1.
3043 	 */
3044 
3045 	tlb1[index].phys = pa;
3046 	tlb1[index].virt = va;
3047 	tlb1[index].size = size;
3048 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3049 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3050 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3051 
3052 	/* Set supervisor RWX permission bits */
3053 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3054 
3055 	tlb1_write_entry(index);
3056 
3057 	/*
3058 	 * XXX in general TLB1 updates should be propagated between CPUs,
3059 	 * since current design assumes to have the same TLB1 set-up on all
3060 	 * cores.
3061 	 */
3062 	return (0);
3063 }
3064 
3065 /*
3066  * Map in contiguous RAM region into the TLB1 using maximum of
3067  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3068  *
3069  * If necessary round up last entry size and return total size
3070  * used by all allocated entries.
3071  */
3072 vm_size_t
3073 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3074 {
3075 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3076 	vm_size_t mapped, pgsz, base, mask;
3077 	int idx, nents;
3078 
3079 	/* Round up to the next 1M */
3080 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3081 
3082 	mapped = 0;
3083 	idx = 0;
3084 	base = va;
3085 	pgsz = 64*1024*1024;
3086 	while (mapped < size) {
3087 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3088 			while (pgsz > (size - mapped))
3089 				pgsz >>= 2;
3090 			pgs[idx++] = pgsz;
3091 			mapped += pgsz;
3092 		}
3093 
3094 		/* We under-map. Correct for this. */
3095 		if (mapped < size) {
3096 			while (pgs[idx - 1] == pgsz) {
3097 				idx--;
3098 				mapped -= pgsz;
3099 			}
3100 			/* XXX We may increase beyond out starting point. */
3101 			pgsz <<= 2;
3102 			pgs[idx++] = pgsz;
3103 			mapped += pgsz;
3104 		}
3105 	}
3106 
3107 	nents = idx;
3108 	mask = pgs[0] - 1;
3109 	/* Align address to the boundary */
3110 	if (va & mask) {
3111 		va = (va + mask) & ~mask;
3112 		pa = (pa + mask) & ~mask;
3113 	}
3114 
3115 	for (idx = 0; idx < nents; idx++) {
3116 		pgsz = pgs[idx];
3117 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3118 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3119 		pa += pgsz;
3120 		va += pgsz;
3121 	}
3122 
3123 	mapped = (va - base);
3124 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3125 	    mapped, mapped - size);
3126 	return (mapped);
3127 }
3128 
3129 /*
3130  * TLB1 initialization routine, to be called after the very first
3131  * assembler level setup done in locore.S.
3132  */
3133 void
3134 tlb1_init()
3135 {
3136 	uint32_t mas0, mas1, mas2, mas3;
3137 	uint32_t tsz;
3138 	u_int i;
3139 
3140 	if (bootinfo != NULL && bootinfo[0] != 1) {
3141 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3142 	} else
3143 		tlb1_idx = 1;
3144 
3145 	/* The first entry/entries are used to map the kernel. */
3146 	for (i = 0; i < tlb1_idx; i++) {
3147 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3148 		mtspr(SPR_MAS0, mas0);
3149 		__asm __volatile("isync; tlbre");
3150 
3151 		mas1 = mfspr(SPR_MAS1);
3152 		if ((mas1 & MAS1_VALID) == 0)
3153 			continue;
3154 
3155 		mas2 = mfspr(SPR_MAS2);
3156 		mas3 = mfspr(SPR_MAS3);
3157 
3158 		tlb1[i].mas1 = mas1;
3159 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3160 		tlb1[i].mas3 = mas3;
3161 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3162 		tlb1[i].phys = mas3 & MAS3_RPN;
3163 
3164 		if (i == 0)
3165 			kernload = tlb1[i].phys;
3166 
3167 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3168 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3169 		kernsize += tlb1[i].size;
3170 	}
3171 
3172 #ifdef SMP
3173 	bp_ntlb1s = tlb1_idx;
3174 #endif
3175 
3176 	/* Purge the remaining entries */
3177 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3178 		tlb1_write_entry(i);
3179 
3180 	/* Setup TLB miss defaults */
3181 	set_mas4_defaults();
3182 }
3183 
3184 vm_offset_t
3185 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3186 {
3187 	vm_paddr_t pa_base;
3188 	vm_offset_t va, sz;
3189 	int i;
3190 
3191 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3192 
3193 	for (i = 0; i < tlb1_idx; i++) {
3194 		if (!(tlb1[i].mas1 & MAS1_VALID))
3195 			continue;
3196 		if (pa >= tlb1[i].phys && (pa + size) <=
3197 		    (tlb1[i].phys + tlb1[i].size))
3198 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3199 	}
3200 
3201 	pa_base = trunc_page(pa);
3202 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3203 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3204 	va = tlb1_map_base + (pa - pa_base);
3205 
3206 	do {
3207 		sz = 1 << (ilog2(size) & ~1);
3208 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3209 		size -= sz;
3210 		pa_base += sz;
3211 		tlb1_map_base += sz;
3212 	} while (size > 0);
3213 
3214 #ifdef SMP
3215 	bp_ntlb1s = tlb1_idx;
3216 #endif
3217 
3218 	return (va);
3219 }
3220 
3221 /*
3222  * Setup MAS4 defaults.
3223  * These values are loaded to MAS0-2 on a TLB miss.
3224  */
3225 static void
3226 set_mas4_defaults(void)
3227 {
3228 	uint32_t mas4;
3229 
3230 	/* Defaults: TLB0, PID0, TSIZED=4K */
3231 	mas4 = MAS4_TLBSELD0;
3232 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3233 #ifdef SMP
3234 	mas4 |= MAS4_MD;
3235 #endif
3236 	mtspr(SPR_MAS4, mas4);
3237 	__asm __volatile("isync");
3238 }
3239 
3240 /*
3241  * Print out contents of the MAS registers for each TLB1 entry
3242  */
3243 void
3244 tlb1_print_tlbentries(void)
3245 {
3246 	uint32_t mas0, mas1, mas2, mas3, mas7;
3247 	int i;
3248 
3249 	debugf("TLB1 entries:\n");
3250 	for (i = 0; i < TLB1_ENTRIES; i++) {
3251 
3252 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3253 		mtspr(SPR_MAS0, mas0);
3254 
3255 		__asm __volatile("isync; tlbre");
3256 
3257 		mas1 = mfspr(SPR_MAS1);
3258 		mas2 = mfspr(SPR_MAS2);
3259 		mas3 = mfspr(SPR_MAS3);
3260 		mas7 = mfspr(SPR_MAS7);
3261 
3262 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3263 	}
3264 }
3265 
3266 /*
3267  * Print out contents of the in-ram tlb1 table.
3268  */
3269 void
3270 tlb1_print_entries(void)
3271 {
3272 	int i;
3273 
3274 	debugf("tlb1[] table entries:\n");
3275 	for (i = 0; i < TLB1_ENTRIES; i++)
3276 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3277 }
3278 
3279 /*
3280  * Return 0 if the physical IO range is encompassed by one of the
3281  * the TLB1 entries, otherwise return related error code.
3282  */
3283 static int
3284 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3285 {
3286 	uint32_t prot;
3287 	vm_paddr_t pa_start;
3288 	vm_paddr_t pa_end;
3289 	unsigned int entry_tsize;
3290 	vm_size_t entry_size;
3291 
3292 	*va = (vm_offset_t)NULL;
3293 
3294 	/* Skip invalid entries */
3295 	if (!(tlb1[i].mas1 & MAS1_VALID))
3296 		return (EINVAL);
3297 
3298 	/*
3299 	 * The entry must be cache-inhibited, guarded, and r/w
3300 	 * so it can function as an i/o page
3301 	 */
3302 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3303 	if (prot != (MAS2_I | MAS2_G))
3304 		return (EPERM);
3305 
3306 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3307 	if (prot != (MAS3_SR | MAS3_SW))
3308 		return (EPERM);
3309 
3310 	/* The address should be within the entry range. */
3311 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3312 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3313 
3314 	entry_size = tsize2size(entry_tsize);
3315 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3316 	pa_end = pa_start + entry_size - 1;
3317 
3318 	if ((pa < pa_start) || ((pa + size) > pa_end))
3319 		return (ERANGE);
3320 
3321 	/* Return virtual address of this mapping. */
3322 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3323 	return (0);
3324 }
3325