xref: /freebsd/sys/powerpc/booke/pmap.c (revision 44d314f704764f0247a540648a4b4fc3e8012133)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/param.h>
55 #include <sys/conf.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
66 #include <sys/lock.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
70 #include <sys/smp.h>
71 #include <sys/vmmeter.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
82 #include <vm/uma.h>
83 
84 #include <machine/cpu.h>
85 #include <machine/pcb.h>
86 #include <machine/platform.h>
87 
88 #include <machine/tlb.h>
89 #include <machine/spr.h>
90 #include <machine/md_var.h>
91 #include <machine/mmuvar.h>
92 #include <machine/pmap.h>
93 #include <machine/pte.h>
94 
95 #include "mmu_if.h"
96 
97 #ifdef  DEBUG
98 #define debugf(fmt, args...) printf(fmt, ##args)
99 #else
100 #define debugf(fmt, args...)
101 #endif
102 
103 #define TODO			panic("%s: not implemented", __func__);
104 
105 extern unsigned char _etext[];
106 extern unsigned char _end[];
107 
108 extern uint32_t *bootinfo;
109 
110 #ifdef SMP
111 extern uint32_t bp_ntlb1s;
112 #endif
113 
114 vm_paddr_t kernload;
115 vm_offset_t kernstart;
116 vm_size_t kernsize;
117 
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
121 
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
127 
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
131 
132 static struct mtx tlbivax_mutex;
133 
134 /*
135  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136  * by idle thred only, no lock required.
137  */
138 static vm_offset_t zero_page_idle_va;
139 
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
144 
145 /**************************************************************************/
146 /* PMAP */
147 /**************************************************************************/
148 
149 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, u_int flags, int8_t psind);
151 
152 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
154 
155 /*
156  * If user pmap is processed with mmu_booke_remove and the resident count
157  * drops to 0, there are no more pages to remove, so we need not continue.
158  */
159 #define PMAP_REMOVE_DONE(pmap) \
160 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
161 
162 extern void tid_flush(tlbtid_t tid, int tlb0_ways, int tlb0_entries_per_way);
163 extern int elf32_nxstack;
164 
165 /**************************************************************************/
166 /* TLB and TID handling */
167 /**************************************************************************/
168 
169 /* Translation ID busy table */
170 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
171 
172 /*
173  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
174  * core revisions and should be read from h/w registers during early config.
175  */
176 uint32_t tlb0_entries;
177 uint32_t tlb0_ways;
178 uint32_t tlb0_entries_per_way;
179 
180 #define TLB0_ENTRIES		(tlb0_entries)
181 #define TLB0_WAYS		(tlb0_ways)
182 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
183 
184 #define TLB1_ENTRIES 16
185 
186 /* In-ram copy of the TLB1 */
187 static tlb_entry_t tlb1[TLB1_ENTRIES];
188 
189 /* Next free entry in the TLB1 */
190 static unsigned int tlb1_idx;
191 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
192 
193 static tlbtid_t tid_alloc(struct pmap *);
194 
195 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
196 
197 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
198 static void tlb1_write_entry(unsigned int);
199 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
200 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
201 
202 static vm_size_t tsize2size(unsigned int);
203 static unsigned int size2tsize(vm_size_t);
204 static unsigned int ilog2(unsigned int);
205 
206 static void set_mas4_defaults(void);
207 
208 static inline void tlb0_flush_entry(vm_offset_t);
209 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
210 
211 /**************************************************************************/
212 /* Page table management */
213 /**************************************************************************/
214 
215 static struct rwlock_padalign pvh_global_lock;
216 
217 /* Data for the pv entry allocation mechanism */
218 static uma_zone_t pvzone;
219 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
220 
221 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
222 
223 #ifndef PMAP_SHPGPERPROC
224 #define PMAP_SHPGPERPROC	200
225 #endif
226 
227 static void ptbl_init(void);
228 static struct ptbl_buf *ptbl_buf_alloc(void);
229 static void ptbl_buf_free(struct ptbl_buf *);
230 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
231 
232 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
233 static void ptbl_free(mmu_t, pmap_t, unsigned int);
234 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
235 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
236 
237 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
238 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
239 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
240 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
241 
242 static pv_entry_t pv_alloc(void);
243 static void pv_free(pv_entry_t);
244 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
245 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
246 
247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
248 #define PTBL_BUFS		(128 * 16)
249 
250 struct ptbl_buf {
251 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
252 	vm_offset_t kva;		/* va of mapping */
253 };
254 
255 /* ptbl free list and a lock used for access synchronization. */
256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
257 static struct mtx ptbl_buf_freelist_lock;
258 
259 /* Base address of kva space allocated fot ptbl bufs. */
260 static vm_offset_t ptbl_buf_pool_vabase;
261 
262 /* Pointer to ptbl_buf structures. */
263 static struct ptbl_buf *ptbl_bufs;
264 
265 #ifdef SMP
266 void pmap_bootstrap_ap(volatile uint32_t *);
267 #endif
268 
269 /*
270  * Kernel MMU interface
271  */
272 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
273 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
274     vm_size_t, vm_offset_t);
275 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
276 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
277     vm_offset_t, vm_page_t *, vm_offset_t, int);
278 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
279     vm_prot_t, u_int flags, int8_t psind);
280 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
281     vm_page_t, vm_prot_t);
282 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
283     vm_prot_t);
284 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
285 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
286     vm_prot_t);
287 static void		mmu_booke_init(mmu_t);
288 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
289 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
290 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
291 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
292 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
293     int);
294 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
295     vm_paddr_t *);
296 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
297     vm_object_t, vm_pindex_t, vm_size_t);
298 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
299 static void		mmu_booke_page_init(mmu_t, vm_page_t);
300 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
301 static void		mmu_booke_pinit(mmu_t, pmap_t);
302 static void		mmu_booke_pinit0(mmu_t, pmap_t);
303 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
304     vm_prot_t);
305 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
306 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
307 static void		mmu_booke_release(mmu_t, pmap_t);
308 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
309 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
310 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
311 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
312 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
313 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
314 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
315 static void		mmu_booke_activate(mmu_t, struct thread *);
316 static void		mmu_booke_deactivate(mmu_t, struct thread *);
317 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
318 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
319 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
320 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
321 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
322 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
323 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
324 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
325 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
326 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
327     vm_size_t);
328 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
329     void **);
330 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
331     void *);
332 static void		mmu_booke_scan_init(mmu_t);
333 
334 static mmu_method_t mmu_booke_methods[] = {
335 	/* pmap dispatcher interface */
336 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
337 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
338 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
339 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
340 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
341 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
342 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
343 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
344 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
345 	MMUMETHOD(mmu_init,		mmu_booke_init),
346 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
347 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
348 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
349 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
350 	MMUMETHOD(mmu_map,		mmu_booke_map),
351 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
352 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
353 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
354 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
355 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
356 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
357 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
358 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
359 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
360 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
361 	MMUMETHOD(mmu_release,		mmu_booke_release),
362 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
363 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
364 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
365 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
366 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
367 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
368 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
369 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
370 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
371 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
372 
373 	/* Internal interfaces */
374 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
375 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
376 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
377 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
378 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
379 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
380 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
381 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
382 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
383 
384 	/* dumpsys() support */
385 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
386 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
387 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
388 
389 	{ 0, 0 }
390 };
391 
392 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
393 
394 static __inline uint32_t
395 tlb_calc_wimg(vm_offset_t pa, vm_memattr_t ma)
396 {
397 	uint32_t attrib;
398 	int i;
399 
400 	if (ma != VM_MEMATTR_DEFAULT) {
401 		switch (ma) {
402 		case VM_MEMATTR_UNCACHEABLE:
403 			return (PTE_I | PTE_G);
404 		case VM_MEMATTR_WRITE_COMBINING:
405 		case VM_MEMATTR_WRITE_BACK:
406 		case VM_MEMATTR_PREFETCHABLE:
407 			return (PTE_I);
408 		case VM_MEMATTR_WRITE_THROUGH:
409 			return (PTE_W | PTE_M);
410 		}
411 	}
412 
413 	/*
414 	 * Assume the page is cache inhibited and access is guarded unless
415 	 * it's in our available memory array.
416 	 */
417 	attrib = _TLB_ENTRY_IO;
418 	for (i = 0; i < physmem_regions_sz; i++) {
419 		if ((pa >= physmem_regions[i].mr_start) &&
420 		    (pa < (physmem_regions[i].mr_start +
421 		     physmem_regions[i].mr_size))) {
422 			attrib = _TLB_ENTRY_MEM;
423 			break;
424 		}
425 	}
426 
427 	return (attrib);
428 }
429 
430 static inline void
431 tlb_miss_lock(void)
432 {
433 #ifdef SMP
434 	struct pcpu *pc;
435 
436 	if (!smp_started)
437 		return;
438 
439 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
440 		if (pc != pcpup) {
441 
442 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
443 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
444 
445 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
446 			    ("tlb_miss_lock: tried to lock self"));
447 
448 			tlb_lock(pc->pc_booke_tlb_lock);
449 
450 			CTR1(KTR_PMAP, "%s: locked", __func__);
451 		}
452 	}
453 #endif
454 }
455 
456 static inline void
457 tlb_miss_unlock(void)
458 {
459 #ifdef SMP
460 	struct pcpu *pc;
461 
462 	if (!smp_started)
463 		return;
464 
465 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
466 		if (pc != pcpup) {
467 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
468 			    __func__, pc->pc_cpuid);
469 
470 			tlb_unlock(pc->pc_booke_tlb_lock);
471 
472 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
473 		}
474 	}
475 #endif
476 }
477 
478 /* Return number of entries in TLB0. */
479 static __inline void
480 tlb0_get_tlbconf(void)
481 {
482 	uint32_t tlb0_cfg;
483 
484 	tlb0_cfg = mfspr(SPR_TLB0CFG);
485 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
486 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
487 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
488 }
489 
490 /* Initialize pool of kva ptbl buffers. */
491 static void
492 ptbl_init(void)
493 {
494 	int i;
495 
496 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
497 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
498 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
499 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
500 
501 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
502 	TAILQ_INIT(&ptbl_buf_freelist);
503 
504 	for (i = 0; i < PTBL_BUFS; i++) {
505 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
506 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
507 	}
508 }
509 
510 /* Get a ptbl_buf from the freelist. */
511 static struct ptbl_buf *
512 ptbl_buf_alloc(void)
513 {
514 	struct ptbl_buf *buf;
515 
516 	mtx_lock(&ptbl_buf_freelist_lock);
517 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
518 	if (buf != NULL)
519 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
520 	mtx_unlock(&ptbl_buf_freelist_lock);
521 
522 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
523 
524 	return (buf);
525 }
526 
527 /* Return ptbl buff to free pool. */
528 static void
529 ptbl_buf_free(struct ptbl_buf *buf)
530 {
531 
532 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
533 
534 	mtx_lock(&ptbl_buf_freelist_lock);
535 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
536 	mtx_unlock(&ptbl_buf_freelist_lock);
537 }
538 
539 /*
540  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
541  */
542 static void
543 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
544 {
545 	struct ptbl_buf *pbuf;
546 
547 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
548 
549 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
550 
551 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
552 		if (pbuf->kva == (vm_offset_t)ptbl) {
553 			/* Remove from pmap ptbl buf list. */
554 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
555 
556 			/* Free corresponding ptbl buf. */
557 			ptbl_buf_free(pbuf);
558 			break;
559 		}
560 }
561 
562 /* Allocate page table. */
563 static pte_t *
564 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
565 {
566 	vm_page_t mtbl[PTBL_PAGES];
567 	vm_page_t m;
568 	struct ptbl_buf *pbuf;
569 	unsigned int pidx;
570 	pte_t *ptbl;
571 	int i, j;
572 
573 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
574 	    (pmap == kernel_pmap), pdir_idx);
575 
576 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
577 	    ("ptbl_alloc: invalid pdir_idx"));
578 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
579 	    ("pte_alloc: valid ptbl entry exists!"));
580 
581 	pbuf = ptbl_buf_alloc();
582 	if (pbuf == NULL)
583 		panic("pte_alloc: couldn't alloc kernel virtual memory");
584 
585 	ptbl = (pte_t *)pbuf->kva;
586 
587 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
588 
589 	/* Allocate ptbl pages, this will sleep! */
590 	for (i = 0; i < PTBL_PAGES; i++) {
591 		pidx = (PTBL_PAGES * pdir_idx) + i;
592 		while ((m = vm_page_alloc(NULL, pidx,
593 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
594 			PMAP_UNLOCK(pmap);
595 			rw_wunlock(&pvh_global_lock);
596 			if (nosleep) {
597 				ptbl_free_pmap_ptbl(pmap, ptbl);
598 				for (j = 0; j < i; j++)
599 					vm_page_free(mtbl[j]);
600 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
601 				return (NULL);
602 			}
603 			VM_WAIT;
604 			rw_wlock(&pvh_global_lock);
605 			PMAP_LOCK(pmap);
606 		}
607 		mtbl[i] = m;
608 	}
609 
610 	/* Map allocated pages into kernel_pmap. */
611 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
612 
613 	/* Zero whole ptbl. */
614 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
615 
616 	/* Add pbuf to the pmap ptbl bufs list. */
617 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
618 
619 	return (ptbl);
620 }
621 
622 /* Free ptbl pages and invalidate pdir entry. */
623 static void
624 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
625 {
626 	pte_t *ptbl;
627 	vm_paddr_t pa;
628 	vm_offset_t va;
629 	vm_page_t m;
630 	int i;
631 
632 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
633 	    (pmap == kernel_pmap), pdir_idx);
634 
635 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
636 	    ("ptbl_free: invalid pdir_idx"));
637 
638 	ptbl = pmap->pm_pdir[pdir_idx];
639 
640 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
641 
642 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
643 
644 	/*
645 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
646 	 * don't attempt to look up the page tables we are releasing.
647 	 */
648 	mtx_lock_spin(&tlbivax_mutex);
649 	tlb_miss_lock();
650 
651 	pmap->pm_pdir[pdir_idx] = NULL;
652 
653 	tlb_miss_unlock();
654 	mtx_unlock_spin(&tlbivax_mutex);
655 
656 	for (i = 0; i < PTBL_PAGES; i++) {
657 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
658 		pa = pte_vatopa(mmu, kernel_pmap, va);
659 		m = PHYS_TO_VM_PAGE(pa);
660 		vm_page_free_zero(m);
661 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
662 		mmu_booke_kremove(mmu, va);
663 	}
664 
665 	ptbl_free_pmap_ptbl(pmap, ptbl);
666 }
667 
668 /*
669  * Decrement ptbl pages hold count and attempt to free ptbl pages.
670  * Called when removing pte entry from ptbl.
671  *
672  * Return 1 if ptbl pages were freed.
673  */
674 static int
675 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
676 {
677 	pte_t *ptbl;
678 	vm_paddr_t pa;
679 	vm_page_t m;
680 	int i;
681 
682 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
683 	    (pmap == kernel_pmap), pdir_idx);
684 
685 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
686 	    ("ptbl_unhold: invalid pdir_idx"));
687 	KASSERT((pmap != kernel_pmap),
688 	    ("ptbl_unhold: unholding kernel ptbl!"));
689 
690 	ptbl = pmap->pm_pdir[pdir_idx];
691 
692 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
693 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
694 	    ("ptbl_unhold: non kva ptbl"));
695 
696 	/* decrement hold count */
697 	for (i = 0; i < PTBL_PAGES; i++) {
698 		pa = pte_vatopa(mmu, kernel_pmap,
699 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
700 		m = PHYS_TO_VM_PAGE(pa);
701 		m->wire_count--;
702 	}
703 
704 	/*
705 	 * Free ptbl pages if there are no pte etries in this ptbl.
706 	 * wire_count has the same value for all ptbl pages, so check the last
707 	 * page.
708 	 */
709 	if (m->wire_count == 0) {
710 		ptbl_free(mmu, pmap, pdir_idx);
711 
712 		//debugf("ptbl_unhold: e (freed ptbl)\n");
713 		return (1);
714 	}
715 
716 	return (0);
717 }
718 
719 /*
720  * Increment hold count for ptbl pages. This routine is used when a new pte
721  * entry is being inserted into the ptbl.
722  */
723 static void
724 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
725 {
726 	vm_paddr_t pa;
727 	pte_t *ptbl;
728 	vm_page_t m;
729 	int i;
730 
731 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
732 	    pdir_idx);
733 
734 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
735 	    ("ptbl_hold: invalid pdir_idx"));
736 	KASSERT((pmap != kernel_pmap),
737 	    ("ptbl_hold: holding kernel ptbl!"));
738 
739 	ptbl = pmap->pm_pdir[pdir_idx];
740 
741 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
742 
743 	for (i = 0; i < PTBL_PAGES; i++) {
744 		pa = pte_vatopa(mmu, kernel_pmap,
745 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
746 		m = PHYS_TO_VM_PAGE(pa);
747 		m->wire_count++;
748 	}
749 }
750 
751 /* Allocate pv_entry structure. */
752 pv_entry_t
753 pv_alloc(void)
754 {
755 	pv_entry_t pv;
756 
757 	pv_entry_count++;
758 	if (pv_entry_count > pv_entry_high_water)
759 		pagedaemon_wakeup();
760 	pv = uma_zalloc(pvzone, M_NOWAIT);
761 
762 	return (pv);
763 }
764 
765 /* Free pv_entry structure. */
766 static __inline void
767 pv_free(pv_entry_t pve)
768 {
769 
770 	pv_entry_count--;
771 	uma_zfree(pvzone, pve);
772 }
773 
774 
775 /* Allocate and initialize pv_entry structure. */
776 static void
777 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
778 {
779 	pv_entry_t pve;
780 
781 	//int su = (pmap == kernel_pmap);
782 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
783 	//	(u_int32_t)pmap, va, (u_int32_t)m);
784 
785 	pve = pv_alloc();
786 	if (pve == NULL)
787 		panic("pv_insert: no pv entries!");
788 
789 	pve->pv_pmap = pmap;
790 	pve->pv_va = va;
791 
792 	/* add to pv_list */
793 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
794 	rw_assert(&pvh_global_lock, RA_WLOCKED);
795 
796 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
797 
798 	//debugf("pv_insert: e\n");
799 }
800 
801 /* Destroy pv entry. */
802 static void
803 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
804 {
805 	pv_entry_t pve;
806 
807 	//int su = (pmap == kernel_pmap);
808 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
809 
810 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
811 	rw_assert(&pvh_global_lock, RA_WLOCKED);
812 
813 	/* find pv entry */
814 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
815 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
816 			/* remove from pv_list */
817 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
818 			if (TAILQ_EMPTY(&m->md.pv_list))
819 				vm_page_aflag_clear(m, PGA_WRITEABLE);
820 
821 			/* free pv entry struct */
822 			pv_free(pve);
823 			break;
824 		}
825 	}
826 
827 	//debugf("pv_remove: e\n");
828 }
829 
830 /*
831  * Clean pte entry, try to free page table page if requested.
832  *
833  * Return 1 if ptbl pages were freed, otherwise return 0.
834  */
835 static int
836 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
837 {
838 	unsigned int pdir_idx = PDIR_IDX(va);
839 	unsigned int ptbl_idx = PTBL_IDX(va);
840 	vm_page_t m;
841 	pte_t *ptbl;
842 	pte_t *pte;
843 
844 	//int su = (pmap == kernel_pmap);
845 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
846 	//		su, (u_int32_t)pmap, va, flags);
847 
848 	ptbl = pmap->pm_pdir[pdir_idx];
849 	KASSERT(ptbl, ("pte_remove: null ptbl"));
850 
851 	pte = &ptbl[ptbl_idx];
852 
853 	if (pte == NULL || !PTE_ISVALID(pte))
854 		return (0);
855 
856 	if (PTE_ISWIRED(pte))
857 		pmap->pm_stats.wired_count--;
858 
859 	/* Handle managed entry. */
860 	if (PTE_ISMANAGED(pte)) {
861 		/* Get vm_page_t for mapped pte. */
862 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
863 
864 		if (PTE_ISMODIFIED(pte))
865 			vm_page_dirty(m);
866 
867 		if (PTE_ISREFERENCED(pte))
868 			vm_page_aflag_set(m, PGA_REFERENCED);
869 
870 		pv_remove(pmap, va, m);
871 	}
872 
873 	mtx_lock_spin(&tlbivax_mutex);
874 	tlb_miss_lock();
875 
876 	tlb0_flush_entry(va);
877 	pte->flags = 0;
878 	pte->rpn = 0;
879 
880 	tlb_miss_unlock();
881 	mtx_unlock_spin(&tlbivax_mutex);
882 
883 	pmap->pm_stats.resident_count--;
884 
885 	if (flags & PTBL_UNHOLD) {
886 		//debugf("pte_remove: e (unhold)\n");
887 		return (ptbl_unhold(mmu, pmap, pdir_idx));
888 	}
889 
890 	//debugf("pte_remove: e\n");
891 	return (0);
892 }
893 
894 /*
895  * Insert PTE for a given page and virtual address.
896  */
897 static int
898 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
899     boolean_t nosleep)
900 {
901 	unsigned int pdir_idx = PDIR_IDX(va);
902 	unsigned int ptbl_idx = PTBL_IDX(va);
903 	pte_t *ptbl, *pte;
904 
905 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
906 	    pmap == kernel_pmap, pmap, va);
907 
908 	/* Get the page table pointer. */
909 	ptbl = pmap->pm_pdir[pdir_idx];
910 
911 	if (ptbl == NULL) {
912 		/* Allocate page table pages. */
913 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
914 		if (ptbl == NULL) {
915 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
916 			return (ENOMEM);
917 		}
918 	} else {
919 		/*
920 		 * Check if there is valid mapping for requested
921 		 * va, if there is, remove it.
922 		 */
923 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
924 		if (PTE_ISVALID(pte)) {
925 			pte_remove(mmu, pmap, va, PTBL_HOLD);
926 		} else {
927 			/*
928 			 * pte is not used, increment hold count
929 			 * for ptbl pages.
930 			 */
931 			if (pmap != kernel_pmap)
932 				ptbl_hold(mmu, pmap, pdir_idx);
933 		}
934 	}
935 
936 	/*
937 	 * Insert pv_entry into pv_list for mapped page if part of managed
938 	 * memory.
939 	 */
940 	if ((m->oflags & VPO_UNMANAGED) == 0) {
941 		flags |= PTE_MANAGED;
942 
943 		/* Create and insert pv entry. */
944 		pv_insert(pmap, va, m);
945 	}
946 
947 	pmap->pm_stats.resident_count++;
948 
949 	mtx_lock_spin(&tlbivax_mutex);
950 	tlb_miss_lock();
951 
952 	tlb0_flush_entry(va);
953 	if (pmap->pm_pdir[pdir_idx] == NULL) {
954 		/*
955 		 * If we just allocated a new page table, hook it in
956 		 * the pdir.
957 		 */
958 		pmap->pm_pdir[pdir_idx] = ptbl;
959 	}
960 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
961 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
962 	pte->flags |= (PTE_VALID | flags);
963 
964 	tlb_miss_unlock();
965 	mtx_unlock_spin(&tlbivax_mutex);
966 	return (0);
967 }
968 
969 /* Return the pa for the given pmap/va. */
970 static vm_paddr_t
971 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
972 {
973 	vm_paddr_t pa = 0;
974 	pte_t *pte;
975 
976 	pte = pte_find(mmu, pmap, va);
977 	if ((pte != NULL) && PTE_ISVALID(pte))
978 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
979 	return (pa);
980 }
981 
982 /* Get a pointer to a PTE in a page table. */
983 static pte_t *
984 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
985 {
986 	unsigned int pdir_idx = PDIR_IDX(va);
987 	unsigned int ptbl_idx = PTBL_IDX(va);
988 
989 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
990 
991 	if (pmap->pm_pdir[pdir_idx])
992 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
993 
994 	return (NULL);
995 }
996 
997 /**************************************************************************/
998 /* PMAP related */
999 /**************************************************************************/
1000 
1001 /*
1002  * This is called during booke_init, before the system is really initialized.
1003  */
1004 static void
1005 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1006 {
1007 	vm_offset_t phys_kernelend;
1008 	struct mem_region *mp, *mp1;
1009 	int cnt, i, j;
1010 	u_int s, e, sz;
1011 	u_int phys_avail_count;
1012 	vm_size_t physsz, hwphyssz, kstack0_sz;
1013 	vm_offset_t kernel_pdir, kstack0, va;
1014 	vm_paddr_t kstack0_phys;
1015 	void *dpcpu;
1016 	pte_t *pte;
1017 
1018 	debugf("mmu_booke_bootstrap: entered\n");
1019 
1020 	/* Set interesting system properties */
1021 	hw_direct_map = 0;
1022 	elf32_nxstack = 1;
1023 
1024 	/* Initialize invalidation mutex */
1025 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1026 
1027 	/* Read TLB0 size and associativity. */
1028 	tlb0_get_tlbconf();
1029 
1030 	/*
1031 	 * Align kernel start and end address (kernel image).
1032 	 * Note that kernel end does not necessarily relate to kernsize.
1033 	 * kernsize is the size of the kernel that is actually mapped.
1034 	 */
1035 	kernstart = trunc_page(start);
1036 	data_start = round_page(kernelend);
1037 	data_end = data_start;
1038 
1039 	/*
1040 	 * Addresses of preloaded modules (like file systems) use
1041 	 * physical addresses. Make sure we relocate those into
1042 	 * virtual addresses.
1043 	 */
1044 	preload_addr_relocate = kernstart - kernload;
1045 
1046 	/* Allocate the dynamic per-cpu area. */
1047 	dpcpu = (void *)data_end;
1048 	data_end += DPCPU_SIZE;
1049 
1050 	/* Allocate space for the message buffer. */
1051 	msgbufp = (struct msgbuf *)data_end;
1052 	data_end += msgbufsize;
1053 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1054 	    data_end);
1055 
1056 	data_end = round_page(data_end);
1057 
1058 	/* Allocate space for ptbl_bufs. */
1059 	ptbl_bufs = (struct ptbl_buf *)data_end;
1060 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1061 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1062 	    data_end);
1063 
1064 	data_end = round_page(data_end);
1065 
1066 	/* Allocate PTE tables for kernel KVA. */
1067 	kernel_pdir = data_end;
1068 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1069 	    PDIR_SIZE - 1) / PDIR_SIZE;
1070 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1071 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1072 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1073 
1074 	debugf(" data_end: 0x%08x\n", data_end);
1075 	if (data_end - kernstart > kernsize) {
1076 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1077 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1078 	}
1079 	data_end = kernstart + kernsize;
1080 	debugf(" updated data_end: 0x%08x\n", data_end);
1081 
1082 	/*
1083 	 * Clear the structures - note we can only do it safely after the
1084 	 * possible additional TLB1 translations are in place (above) so that
1085 	 * all range up to the currently calculated 'data_end' is covered.
1086 	 */
1087 	dpcpu_init(dpcpu, 0);
1088 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1089 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1090 
1091 	/*******************************************************/
1092 	/* Set the start and end of kva. */
1093 	/*******************************************************/
1094 	virtual_avail = round_page(data_end);
1095 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1096 
1097 	/* Allocate KVA space for page zero/copy operations. */
1098 	zero_page_va = virtual_avail;
1099 	virtual_avail += PAGE_SIZE;
1100 	zero_page_idle_va = virtual_avail;
1101 	virtual_avail += PAGE_SIZE;
1102 	copy_page_src_va = virtual_avail;
1103 	virtual_avail += PAGE_SIZE;
1104 	copy_page_dst_va = virtual_avail;
1105 	virtual_avail += PAGE_SIZE;
1106 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1107 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1108 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1109 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1110 
1111 	/* Initialize page zero/copy mutexes. */
1112 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1113 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1114 
1115 	/* Allocate KVA space for ptbl bufs. */
1116 	ptbl_buf_pool_vabase = virtual_avail;
1117 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1118 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1119 	    ptbl_buf_pool_vabase, virtual_avail);
1120 
1121 	/* Calculate corresponding physical addresses for the kernel region. */
1122 	phys_kernelend = kernload + kernsize;
1123 	debugf("kernel image and allocated data:\n");
1124 	debugf(" kernload    = 0x%08x\n", kernload);
1125 	debugf(" kernstart   = 0x%08x\n", kernstart);
1126 	debugf(" kernsize    = 0x%08x\n", kernsize);
1127 
1128 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1129 		panic("mmu_booke_bootstrap: phys_avail too small");
1130 
1131 	/*
1132 	 * Remove kernel physical address range from avail regions list. Page
1133 	 * align all regions.  Non-page aligned memory isn't very interesting
1134 	 * to us.  Also, sort the entries for ascending addresses.
1135 	 */
1136 
1137 	/* Retrieve phys/avail mem regions */
1138 	mem_regions(&physmem_regions, &physmem_regions_sz,
1139 	    &availmem_regions, &availmem_regions_sz);
1140 	sz = 0;
1141 	cnt = availmem_regions_sz;
1142 	debugf("processing avail regions:\n");
1143 	for (mp = availmem_regions; mp->mr_size; mp++) {
1144 		s = mp->mr_start;
1145 		e = mp->mr_start + mp->mr_size;
1146 		debugf(" %08x-%08x -> ", s, e);
1147 		/* Check whether this region holds all of the kernel. */
1148 		if (s < kernload && e > phys_kernelend) {
1149 			availmem_regions[cnt].mr_start = phys_kernelend;
1150 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1151 			e = kernload;
1152 		}
1153 		/* Look whether this regions starts within the kernel. */
1154 		if (s >= kernload && s < phys_kernelend) {
1155 			if (e <= phys_kernelend)
1156 				goto empty;
1157 			s = phys_kernelend;
1158 		}
1159 		/* Now look whether this region ends within the kernel. */
1160 		if (e > kernload && e <= phys_kernelend) {
1161 			if (s >= kernload)
1162 				goto empty;
1163 			e = kernload;
1164 		}
1165 		/* Now page align the start and size of the region. */
1166 		s = round_page(s);
1167 		e = trunc_page(e);
1168 		if (e < s)
1169 			e = s;
1170 		sz = e - s;
1171 		debugf("%08x-%08x = %x\n", s, e, sz);
1172 
1173 		/* Check whether some memory is left here. */
1174 		if (sz == 0) {
1175 		empty:
1176 			memmove(mp, mp + 1,
1177 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1178 			cnt--;
1179 			mp--;
1180 			continue;
1181 		}
1182 
1183 		/* Do an insertion sort. */
1184 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1185 			if (s < mp1->mr_start)
1186 				break;
1187 		if (mp1 < mp) {
1188 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1189 			mp1->mr_start = s;
1190 			mp1->mr_size = sz;
1191 		} else {
1192 			mp->mr_start = s;
1193 			mp->mr_size = sz;
1194 		}
1195 	}
1196 	availmem_regions_sz = cnt;
1197 
1198 	/*******************************************************/
1199 	/* Steal physical memory for kernel stack from the end */
1200 	/* of the first avail region                           */
1201 	/*******************************************************/
1202 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1203 	kstack0_phys = availmem_regions[0].mr_start +
1204 	    availmem_regions[0].mr_size;
1205 	kstack0_phys -= kstack0_sz;
1206 	availmem_regions[0].mr_size -= kstack0_sz;
1207 
1208 	/*******************************************************/
1209 	/* Fill in phys_avail table, based on availmem_regions */
1210 	/*******************************************************/
1211 	phys_avail_count = 0;
1212 	physsz = 0;
1213 	hwphyssz = 0;
1214 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1215 
1216 	debugf("fill in phys_avail:\n");
1217 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1218 
1219 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1220 		    availmem_regions[i].mr_start,
1221 		    availmem_regions[i].mr_start +
1222 		        availmem_regions[i].mr_size,
1223 		    availmem_regions[i].mr_size);
1224 
1225 		if (hwphyssz != 0 &&
1226 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1227 			debugf(" hw.physmem adjust\n");
1228 			if (physsz < hwphyssz) {
1229 				phys_avail[j] = availmem_regions[i].mr_start;
1230 				phys_avail[j + 1] =
1231 				    availmem_regions[i].mr_start +
1232 				    hwphyssz - physsz;
1233 				physsz = hwphyssz;
1234 				phys_avail_count++;
1235 			}
1236 			break;
1237 		}
1238 
1239 		phys_avail[j] = availmem_regions[i].mr_start;
1240 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1241 		    availmem_regions[i].mr_size;
1242 		phys_avail_count++;
1243 		physsz += availmem_regions[i].mr_size;
1244 	}
1245 	physmem = btoc(physsz);
1246 
1247 	/* Calculate the last available physical address. */
1248 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1249 		;
1250 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1251 
1252 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1253 	debugf("phys_avail_count = %d\n", phys_avail_count);
1254 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1255 	    physmem);
1256 
1257 	/*******************************************************/
1258 	/* Initialize (statically allocated) kernel pmap. */
1259 	/*******************************************************/
1260 	PMAP_LOCK_INIT(kernel_pmap);
1261 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1262 
1263 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1264 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1265 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1266 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1267 
1268 	/* Initialize kernel pdir */
1269 	for (i = 0; i < kernel_ptbls; i++)
1270 		kernel_pmap->pm_pdir[kptbl_min + i] =
1271 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1272 
1273 	for (i = 0; i < MAXCPU; i++) {
1274 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1275 
1276 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1277 		tidbusy[i][0] = kernel_pmap;
1278 	}
1279 
1280 	/*
1281 	 * Fill in PTEs covering kernel code and data. They are not required
1282 	 * for address translation, as this area is covered by static TLB1
1283 	 * entries, but for pte_vatopa() to work correctly with kernel area
1284 	 * addresses.
1285 	 */
1286 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1287 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1288 		pte->rpn = kernload + (va - kernstart);
1289 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1290 		    PTE_VALID;
1291 	}
1292 	/* Mark kernel_pmap active on all CPUs */
1293 	CPU_FILL(&kernel_pmap->pm_active);
1294 
1295  	/*
1296 	 * Initialize the global pv list lock.
1297 	 */
1298 	rw_init(&pvh_global_lock, "pmap pv global");
1299 
1300 	/*******************************************************/
1301 	/* Final setup */
1302 	/*******************************************************/
1303 
1304 	/* Enter kstack0 into kernel map, provide guard page */
1305 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1306 	thread0.td_kstack = kstack0;
1307 	thread0.td_kstack_pages = KSTACK_PAGES;
1308 
1309 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1310 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1311 	    kstack0_phys, kstack0_phys + kstack0_sz);
1312 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1313 
1314 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1315 	for (i = 0; i < KSTACK_PAGES; i++) {
1316 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1317 		kstack0 += PAGE_SIZE;
1318 		kstack0_phys += PAGE_SIZE;
1319 	}
1320 
1321 	pmap_bootstrapped = 1;
1322 
1323 	debugf("virtual_avail = %08x\n", virtual_avail);
1324 	debugf("virtual_end   = %08x\n", virtual_end);
1325 
1326 	debugf("mmu_booke_bootstrap: exit\n");
1327 }
1328 
1329 #ifdef SMP
1330 void
1331 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1332 {
1333 	int i;
1334 
1335 	/*
1336 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1337 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1338 	 * these values directly to (re)program AP's TLB1 hardware.
1339 	 */
1340 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1341 		/* Skip invalid entries */
1342 		if (!(tlb1[i].mas1 & MAS1_VALID))
1343 			continue;
1344 
1345 		tlb1_write_entry(i);
1346 	}
1347 
1348 	set_mas4_defaults();
1349 }
1350 #endif
1351 
1352 /*
1353  * Get the physical page address for the given pmap/virtual address.
1354  */
1355 static vm_paddr_t
1356 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1357 {
1358 	vm_paddr_t pa;
1359 
1360 	PMAP_LOCK(pmap);
1361 	pa = pte_vatopa(mmu, pmap, va);
1362 	PMAP_UNLOCK(pmap);
1363 
1364 	return (pa);
1365 }
1366 
1367 /*
1368  * Extract the physical page address associated with the given
1369  * kernel virtual address.
1370  */
1371 static vm_paddr_t
1372 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1373 {
1374 	int i;
1375 
1376 	/* Check TLB1 mappings */
1377 	for (i = 0; i < tlb1_idx; i++) {
1378 		if (!(tlb1[i].mas1 & MAS1_VALID))
1379 			continue;
1380 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1381 			return (tlb1[i].phys + (va - tlb1[i].virt));
1382 	}
1383 
1384 	return (pte_vatopa(mmu, kernel_pmap, va));
1385 }
1386 
1387 /*
1388  * Initialize the pmap module.
1389  * Called by vm_init, to initialize any structures that the pmap
1390  * system needs to map virtual memory.
1391  */
1392 static void
1393 mmu_booke_init(mmu_t mmu)
1394 {
1395 	int shpgperproc = PMAP_SHPGPERPROC;
1396 
1397 	/*
1398 	 * Initialize the address space (zone) for the pv entries.  Set a
1399 	 * high water mark so that the system can recover from excessive
1400 	 * numbers of pv entries.
1401 	 */
1402 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1403 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1404 
1405 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1406 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1407 
1408 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1409 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1410 
1411 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1412 
1413 	/* Pre-fill pvzone with initial number of pv entries. */
1414 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1415 
1416 	/* Initialize ptbl allocation. */
1417 	ptbl_init();
1418 }
1419 
1420 /*
1421  * Map a list of wired pages into kernel virtual address space.  This is
1422  * intended for temporary mappings which do not need page modification or
1423  * references recorded.  Existing mappings in the region are overwritten.
1424  */
1425 static void
1426 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1427 {
1428 	vm_offset_t va;
1429 
1430 	va = sva;
1431 	while (count-- > 0) {
1432 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1433 		va += PAGE_SIZE;
1434 		m++;
1435 	}
1436 }
1437 
1438 /*
1439  * Remove page mappings from kernel virtual address space.  Intended for
1440  * temporary mappings entered by mmu_booke_qenter.
1441  */
1442 static void
1443 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1444 {
1445 	vm_offset_t va;
1446 
1447 	va = sva;
1448 	while (count-- > 0) {
1449 		mmu_booke_kremove(mmu, va);
1450 		va += PAGE_SIZE;
1451 	}
1452 }
1453 
1454 /*
1455  * Map a wired page into kernel virtual address space.
1456  */
1457 static void
1458 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1459 {
1460 
1461 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1462 }
1463 
1464 static void
1465 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1466 {
1467 	unsigned int pdir_idx = PDIR_IDX(va);
1468 	unsigned int ptbl_idx = PTBL_IDX(va);
1469 	uint32_t flags;
1470 	pte_t *pte;
1471 
1472 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1473 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1474 
1475 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1476 	flags |= tlb_calc_wimg(pa, ma);
1477 
1478 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1479 
1480 	mtx_lock_spin(&tlbivax_mutex);
1481 	tlb_miss_lock();
1482 
1483 	if (PTE_ISVALID(pte)) {
1484 
1485 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1486 
1487 		/* Flush entry from TLB0 */
1488 		tlb0_flush_entry(va);
1489 	}
1490 
1491 	pte->rpn = pa & ~PTE_PA_MASK;
1492 	pte->flags = flags;
1493 
1494 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1495 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1496 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1497 
1498 	/* Flush the real memory from the instruction cache. */
1499 	if ((flags & (PTE_I | PTE_G)) == 0) {
1500 		__syncicache((void *)va, PAGE_SIZE);
1501 	}
1502 
1503 	tlb_miss_unlock();
1504 	mtx_unlock_spin(&tlbivax_mutex);
1505 }
1506 
1507 /*
1508  * Remove a page from kernel page table.
1509  */
1510 static void
1511 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1512 {
1513 	unsigned int pdir_idx = PDIR_IDX(va);
1514 	unsigned int ptbl_idx = PTBL_IDX(va);
1515 	pte_t *pte;
1516 
1517 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1518 
1519 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1520 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1521 	    ("mmu_booke_kremove: invalid va"));
1522 
1523 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1524 
1525 	if (!PTE_ISVALID(pte)) {
1526 
1527 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1528 
1529 		return;
1530 	}
1531 
1532 	mtx_lock_spin(&tlbivax_mutex);
1533 	tlb_miss_lock();
1534 
1535 	/* Invalidate entry in TLB0, update PTE. */
1536 	tlb0_flush_entry(va);
1537 	pte->flags = 0;
1538 	pte->rpn = 0;
1539 
1540 	tlb_miss_unlock();
1541 	mtx_unlock_spin(&tlbivax_mutex);
1542 }
1543 
1544 /*
1545  * Initialize pmap associated with process 0.
1546  */
1547 static void
1548 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1549 {
1550 
1551 	PMAP_LOCK_INIT(pmap);
1552 	mmu_booke_pinit(mmu, pmap);
1553 	PCPU_SET(curpmap, pmap);
1554 }
1555 
1556 /*
1557  * Initialize a preallocated and zeroed pmap structure,
1558  * such as one in a vmspace structure.
1559  */
1560 static void
1561 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1562 {
1563 	int i;
1564 
1565 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1566 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1567 
1568 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1569 
1570 	for (i = 0; i < MAXCPU; i++)
1571 		pmap->pm_tid[i] = TID_NONE;
1572 	CPU_ZERO(&kernel_pmap->pm_active);
1573 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1574 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1575 	TAILQ_INIT(&pmap->pm_ptbl_list);
1576 }
1577 
1578 /*
1579  * Release any resources held by the given physical map.
1580  * Called when a pmap initialized by mmu_booke_pinit is being released.
1581  * Should only be called if the map contains no valid mappings.
1582  */
1583 static void
1584 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1585 {
1586 
1587 	KASSERT(pmap->pm_stats.resident_count == 0,
1588 	    ("pmap_release: pmap resident count %ld != 0",
1589 	    pmap->pm_stats.resident_count));
1590 }
1591 
1592 /*
1593  * Insert the given physical page at the specified virtual address in the
1594  * target physical map with the protection requested. If specified the page
1595  * will be wired down.
1596  */
1597 static int
1598 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1599     vm_prot_t prot, u_int flags, int8_t psind)
1600 {
1601 	int error;
1602 
1603 	rw_wlock(&pvh_global_lock);
1604 	PMAP_LOCK(pmap);
1605 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1606 	rw_wunlock(&pvh_global_lock);
1607 	PMAP_UNLOCK(pmap);
1608 	return (error);
1609 }
1610 
1611 static int
1612 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1613     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1614 {
1615 	pte_t *pte;
1616 	vm_paddr_t pa;
1617 	uint32_t flags;
1618 	int error, su, sync;
1619 
1620 	pa = VM_PAGE_TO_PHYS(m);
1621 	su = (pmap == kernel_pmap);
1622 	sync = 0;
1623 
1624 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1625 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1626 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1627 	//		(u_int32_t)m, va, pa, prot, flags);
1628 
1629 	if (su) {
1630 		KASSERT(((va >= virtual_avail) &&
1631 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1632 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1633 	} else {
1634 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1635 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1636 	}
1637 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1638 		VM_OBJECT_ASSERT_LOCKED(m->object);
1639 
1640 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1641 
1642 	/*
1643 	 * If there is an existing mapping, and the physical address has not
1644 	 * changed, must be protection or wiring change.
1645 	 */
1646 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1647 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1648 
1649 		/*
1650 		 * Before actually updating pte->flags we calculate and
1651 		 * prepare its new value in a helper var.
1652 		 */
1653 		flags = pte->flags;
1654 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1655 
1656 		/* Wiring change, just update stats. */
1657 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1658 			if (!PTE_ISWIRED(pte)) {
1659 				flags |= PTE_WIRED;
1660 				pmap->pm_stats.wired_count++;
1661 			}
1662 		} else {
1663 			if (PTE_ISWIRED(pte)) {
1664 				flags &= ~PTE_WIRED;
1665 				pmap->pm_stats.wired_count--;
1666 			}
1667 		}
1668 
1669 		if (prot & VM_PROT_WRITE) {
1670 			/* Add write permissions. */
1671 			flags |= PTE_SW;
1672 			if (!su)
1673 				flags |= PTE_UW;
1674 
1675 			if ((flags & PTE_MANAGED) != 0)
1676 				vm_page_aflag_set(m, PGA_WRITEABLE);
1677 		} else {
1678 			/* Handle modified pages, sense modify status. */
1679 
1680 			/*
1681 			 * The PTE_MODIFIED flag could be set by underlying
1682 			 * TLB misses since we last read it (above), possibly
1683 			 * other CPUs could update it so we check in the PTE
1684 			 * directly rather than rely on that saved local flags
1685 			 * copy.
1686 			 */
1687 			if (PTE_ISMODIFIED(pte))
1688 				vm_page_dirty(m);
1689 		}
1690 
1691 		if (prot & VM_PROT_EXECUTE) {
1692 			flags |= PTE_SX;
1693 			if (!su)
1694 				flags |= PTE_UX;
1695 
1696 			/*
1697 			 * Check existing flags for execute permissions: if we
1698 			 * are turning execute permissions on, icache should
1699 			 * be flushed.
1700 			 */
1701 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1702 				sync++;
1703 		}
1704 
1705 		flags &= ~PTE_REFERENCED;
1706 
1707 		/*
1708 		 * The new flags value is all calculated -- only now actually
1709 		 * update the PTE.
1710 		 */
1711 		mtx_lock_spin(&tlbivax_mutex);
1712 		tlb_miss_lock();
1713 
1714 		tlb0_flush_entry(va);
1715 		pte->flags = flags;
1716 
1717 		tlb_miss_unlock();
1718 		mtx_unlock_spin(&tlbivax_mutex);
1719 
1720 	} else {
1721 		/*
1722 		 * If there is an existing mapping, but it's for a different
1723 		 * physical address, pte_enter() will delete the old mapping.
1724 		 */
1725 		//if ((pte != NULL) && PTE_ISVALID(pte))
1726 		//	debugf("mmu_booke_enter_locked: replace\n");
1727 		//else
1728 		//	debugf("mmu_booke_enter_locked: new\n");
1729 
1730 		/* Now set up the flags and install the new mapping. */
1731 		flags = (PTE_SR | PTE_VALID);
1732 		flags |= PTE_M;
1733 
1734 		if (!su)
1735 			flags |= PTE_UR;
1736 
1737 		if (prot & VM_PROT_WRITE) {
1738 			flags |= PTE_SW;
1739 			if (!su)
1740 				flags |= PTE_UW;
1741 
1742 			if ((m->oflags & VPO_UNMANAGED) == 0)
1743 				vm_page_aflag_set(m, PGA_WRITEABLE);
1744 		}
1745 
1746 		if (prot & VM_PROT_EXECUTE) {
1747 			flags |= PTE_SX;
1748 			if (!su)
1749 				flags |= PTE_UX;
1750 		}
1751 
1752 		/* If its wired update stats. */
1753 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1754 			flags |= PTE_WIRED;
1755 
1756 		error = pte_enter(mmu, pmap, m, va, flags,
1757 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1758 		if (error != 0)
1759 			return (KERN_RESOURCE_SHORTAGE);
1760 
1761 		if ((flags & PMAP_ENTER_WIRED) != 0)
1762 			pmap->pm_stats.wired_count++;
1763 
1764 		/* Flush the real memory from the instruction cache. */
1765 		if (prot & VM_PROT_EXECUTE)
1766 			sync++;
1767 	}
1768 
1769 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1770 		__syncicache((void *)va, PAGE_SIZE);
1771 		sync = 0;
1772 	}
1773 
1774 	return (KERN_SUCCESS);
1775 }
1776 
1777 /*
1778  * Maps a sequence of resident pages belonging to the same object.
1779  * The sequence begins with the given page m_start.  This page is
1780  * mapped at the given virtual address start.  Each subsequent page is
1781  * mapped at a virtual address that is offset from start by the same
1782  * amount as the page is offset from m_start within the object.  The
1783  * last page in the sequence is the page with the largest offset from
1784  * m_start that can be mapped at a virtual address less than the given
1785  * virtual address end.  Not every virtual page between start and end
1786  * is mapped; only those for which a resident page exists with the
1787  * corresponding offset from m_start are mapped.
1788  */
1789 static void
1790 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1791     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1792 {
1793 	vm_page_t m;
1794 	vm_pindex_t diff, psize;
1795 
1796 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1797 
1798 	psize = atop(end - start);
1799 	m = m_start;
1800 	rw_wlock(&pvh_global_lock);
1801 	PMAP_LOCK(pmap);
1802 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1803 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1804 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1805 		    PMAP_ENTER_NOSLEEP, 0);
1806 		m = TAILQ_NEXT(m, listq);
1807 	}
1808 	rw_wunlock(&pvh_global_lock);
1809 	PMAP_UNLOCK(pmap);
1810 }
1811 
1812 static void
1813 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1814     vm_prot_t prot)
1815 {
1816 
1817 	rw_wlock(&pvh_global_lock);
1818 	PMAP_LOCK(pmap);
1819 	mmu_booke_enter_locked(mmu, pmap, va, m,
1820 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1821 	    0);
1822 	rw_wunlock(&pvh_global_lock);
1823 	PMAP_UNLOCK(pmap);
1824 }
1825 
1826 /*
1827  * Remove the given range of addresses from the specified map.
1828  *
1829  * It is assumed that the start and end are properly rounded to the page size.
1830  */
1831 static void
1832 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1833 {
1834 	pte_t *pte;
1835 	uint8_t hold_flag;
1836 
1837 	int su = (pmap == kernel_pmap);
1838 
1839 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1840 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1841 
1842 	if (su) {
1843 		KASSERT(((va >= virtual_avail) &&
1844 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1845 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1846 	} else {
1847 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1848 		    ("mmu_booke_remove: user pmap, non user va"));
1849 	}
1850 
1851 	if (PMAP_REMOVE_DONE(pmap)) {
1852 		//debugf("mmu_booke_remove: e (empty)\n");
1853 		return;
1854 	}
1855 
1856 	hold_flag = PTBL_HOLD_FLAG(pmap);
1857 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1858 
1859 	rw_wlock(&pvh_global_lock);
1860 	PMAP_LOCK(pmap);
1861 	for (; va < endva; va += PAGE_SIZE) {
1862 		pte = pte_find(mmu, pmap, va);
1863 		if ((pte != NULL) && PTE_ISVALID(pte))
1864 			pte_remove(mmu, pmap, va, hold_flag);
1865 	}
1866 	PMAP_UNLOCK(pmap);
1867 	rw_wunlock(&pvh_global_lock);
1868 
1869 	//debugf("mmu_booke_remove: e\n");
1870 }
1871 
1872 /*
1873  * Remove physical page from all pmaps in which it resides.
1874  */
1875 static void
1876 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1877 {
1878 	pv_entry_t pv, pvn;
1879 	uint8_t hold_flag;
1880 
1881 	rw_wlock(&pvh_global_lock);
1882 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1883 		pvn = TAILQ_NEXT(pv, pv_link);
1884 
1885 		PMAP_LOCK(pv->pv_pmap);
1886 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1887 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1888 		PMAP_UNLOCK(pv->pv_pmap);
1889 	}
1890 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1891 	rw_wunlock(&pvh_global_lock);
1892 }
1893 
1894 /*
1895  * Map a range of physical addresses into kernel virtual address space.
1896  */
1897 static vm_offset_t
1898 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1899     vm_paddr_t pa_end, int prot)
1900 {
1901 	vm_offset_t sva = *virt;
1902 	vm_offset_t va = sva;
1903 
1904 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1905 	//		sva, pa_start, pa_end);
1906 
1907 	while (pa_start < pa_end) {
1908 		mmu_booke_kenter(mmu, va, pa_start);
1909 		va += PAGE_SIZE;
1910 		pa_start += PAGE_SIZE;
1911 	}
1912 	*virt = va;
1913 
1914 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1915 	return (sva);
1916 }
1917 
1918 /*
1919  * The pmap must be activated before it's address space can be accessed in any
1920  * way.
1921  */
1922 static void
1923 mmu_booke_activate(mmu_t mmu, struct thread *td)
1924 {
1925 	pmap_t pmap;
1926 	u_int cpuid;
1927 
1928 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1929 
1930 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1931 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1932 
1933 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1934 
1935 	sched_pin();
1936 
1937 	cpuid = PCPU_GET(cpuid);
1938 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1939 	PCPU_SET(curpmap, pmap);
1940 
1941 	if (pmap->pm_tid[cpuid] == TID_NONE)
1942 		tid_alloc(pmap);
1943 
1944 	/* Load PID0 register with pmap tid value. */
1945 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1946 	__asm __volatile("isync");
1947 
1948 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1949 
1950 	sched_unpin();
1951 
1952 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1953 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1954 }
1955 
1956 /*
1957  * Deactivate the specified process's address space.
1958  */
1959 static void
1960 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1961 {
1962 	pmap_t pmap;
1963 
1964 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1965 
1966 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1967 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1968 
1969 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1970 
1971 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1972 	PCPU_SET(curpmap, NULL);
1973 }
1974 
1975 /*
1976  * Copy the range specified by src_addr/len
1977  * from the source map to the range dst_addr/len
1978  * in the destination map.
1979  *
1980  * This routine is only advisory and need not do anything.
1981  */
1982 static void
1983 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1984     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1985 {
1986 
1987 }
1988 
1989 /*
1990  * Set the physical protection on the specified range of this map as requested.
1991  */
1992 static void
1993 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1994     vm_prot_t prot)
1995 {
1996 	vm_offset_t va;
1997 	vm_page_t m;
1998 	pte_t *pte;
1999 
2000 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2001 		mmu_booke_remove(mmu, pmap, sva, eva);
2002 		return;
2003 	}
2004 
2005 	if (prot & VM_PROT_WRITE)
2006 		return;
2007 
2008 	PMAP_LOCK(pmap);
2009 	for (va = sva; va < eva; va += PAGE_SIZE) {
2010 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2011 			if (PTE_ISVALID(pte)) {
2012 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2013 
2014 				mtx_lock_spin(&tlbivax_mutex);
2015 				tlb_miss_lock();
2016 
2017 				/* Handle modified pages. */
2018 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2019 					vm_page_dirty(m);
2020 
2021 				tlb0_flush_entry(va);
2022 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2023 
2024 				tlb_miss_unlock();
2025 				mtx_unlock_spin(&tlbivax_mutex);
2026 			}
2027 		}
2028 	}
2029 	PMAP_UNLOCK(pmap);
2030 }
2031 
2032 /*
2033  * Clear the write and modified bits in each of the given page's mappings.
2034  */
2035 static void
2036 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2037 {
2038 	pv_entry_t pv;
2039 	pte_t *pte;
2040 
2041 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2042 	    ("mmu_booke_remove_write: page %p is not managed", m));
2043 
2044 	/*
2045 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2046 	 * set by another thread while the object is locked.  Thus,
2047 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2048 	 */
2049 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2050 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2051 		return;
2052 	rw_wlock(&pvh_global_lock);
2053 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2054 		PMAP_LOCK(pv->pv_pmap);
2055 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2056 			if (PTE_ISVALID(pte)) {
2057 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2058 
2059 				mtx_lock_spin(&tlbivax_mutex);
2060 				tlb_miss_lock();
2061 
2062 				/* Handle modified pages. */
2063 				if (PTE_ISMODIFIED(pte))
2064 					vm_page_dirty(m);
2065 
2066 				/* Flush mapping from TLB0. */
2067 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2068 
2069 				tlb_miss_unlock();
2070 				mtx_unlock_spin(&tlbivax_mutex);
2071 			}
2072 		}
2073 		PMAP_UNLOCK(pv->pv_pmap);
2074 	}
2075 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2076 	rw_wunlock(&pvh_global_lock);
2077 }
2078 
2079 static void
2080 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2081 {
2082 	pte_t *pte;
2083 	pmap_t pmap;
2084 	vm_page_t m;
2085 	vm_offset_t addr;
2086 	vm_paddr_t pa = 0;
2087 	int active, valid;
2088 
2089 	va = trunc_page(va);
2090 	sz = round_page(sz);
2091 
2092 	rw_wlock(&pvh_global_lock);
2093 	pmap = PCPU_GET(curpmap);
2094 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2095 	while (sz > 0) {
2096 		PMAP_LOCK(pm);
2097 		pte = pte_find(mmu, pm, va);
2098 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2099 		if (valid)
2100 			pa = PTE_PA(pte);
2101 		PMAP_UNLOCK(pm);
2102 		if (valid) {
2103 			if (!active) {
2104 				/* Create a mapping in the active pmap. */
2105 				addr = 0;
2106 				m = PHYS_TO_VM_PAGE(pa);
2107 				PMAP_LOCK(pmap);
2108 				pte_enter(mmu, pmap, m, addr,
2109 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2110 				__syncicache((void *)addr, PAGE_SIZE);
2111 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2112 				PMAP_UNLOCK(pmap);
2113 			} else
2114 				__syncicache((void *)va, PAGE_SIZE);
2115 		}
2116 		va += PAGE_SIZE;
2117 		sz -= PAGE_SIZE;
2118 	}
2119 	rw_wunlock(&pvh_global_lock);
2120 }
2121 
2122 /*
2123  * Atomically extract and hold the physical page with the given
2124  * pmap and virtual address pair if that mapping permits the given
2125  * protection.
2126  */
2127 static vm_page_t
2128 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2129     vm_prot_t prot)
2130 {
2131 	pte_t *pte;
2132 	vm_page_t m;
2133 	uint32_t pte_wbit;
2134 	vm_paddr_t pa;
2135 
2136 	m = NULL;
2137 	pa = 0;
2138 	PMAP_LOCK(pmap);
2139 retry:
2140 	pte = pte_find(mmu, pmap, va);
2141 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2142 		if (pmap == kernel_pmap)
2143 			pte_wbit = PTE_SW;
2144 		else
2145 			pte_wbit = PTE_UW;
2146 
2147 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2148 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2149 				goto retry;
2150 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2151 			vm_page_hold(m);
2152 		}
2153 	}
2154 
2155 	PA_UNLOCK_COND(pa);
2156 	PMAP_UNLOCK(pmap);
2157 	return (m);
2158 }
2159 
2160 /*
2161  * Initialize a vm_page's machine-dependent fields.
2162  */
2163 static void
2164 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2165 {
2166 
2167 	TAILQ_INIT(&m->md.pv_list);
2168 }
2169 
2170 /*
2171  * mmu_booke_zero_page_area zeros the specified hardware page by
2172  * mapping it into virtual memory and using bzero to clear
2173  * its contents.
2174  *
2175  * off and size must reside within a single page.
2176  */
2177 static void
2178 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2179 {
2180 	vm_offset_t va;
2181 
2182 	/* XXX KASSERT off and size are within a single page? */
2183 
2184 	mtx_lock(&zero_page_mutex);
2185 	va = zero_page_va;
2186 
2187 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2188 	bzero((caddr_t)va + off, size);
2189 	mmu_booke_kremove(mmu, va);
2190 
2191 	mtx_unlock(&zero_page_mutex);
2192 }
2193 
2194 /*
2195  * mmu_booke_zero_page zeros the specified hardware page.
2196  */
2197 static void
2198 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2199 {
2200 
2201 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2202 }
2203 
2204 /*
2205  * mmu_booke_copy_page copies the specified (machine independent) page by
2206  * mapping the page into virtual memory and using memcopy to copy the page,
2207  * one machine dependent page at a time.
2208  */
2209 static void
2210 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2211 {
2212 	vm_offset_t sva, dva;
2213 
2214 	sva = copy_page_src_va;
2215 	dva = copy_page_dst_va;
2216 
2217 	mtx_lock(&copy_page_mutex);
2218 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2219 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2220 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2221 	mmu_booke_kremove(mmu, dva);
2222 	mmu_booke_kremove(mmu, sva);
2223 	mtx_unlock(&copy_page_mutex);
2224 }
2225 
2226 static inline void
2227 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2228     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2229 {
2230 	void *a_cp, *b_cp;
2231 	vm_offset_t a_pg_offset, b_pg_offset;
2232 	int cnt;
2233 
2234 	mtx_lock(&copy_page_mutex);
2235 	while (xfersize > 0) {
2236 		a_pg_offset = a_offset & PAGE_MASK;
2237 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2238 		mmu_booke_kenter(mmu, copy_page_src_va,
2239 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2240 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2241 		b_pg_offset = b_offset & PAGE_MASK;
2242 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2243 		mmu_booke_kenter(mmu, copy_page_dst_va,
2244 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2245 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2246 		bcopy(a_cp, b_cp, cnt);
2247 		mmu_booke_kremove(mmu, copy_page_dst_va);
2248 		mmu_booke_kremove(mmu, copy_page_src_va);
2249 		a_offset += cnt;
2250 		b_offset += cnt;
2251 		xfersize -= cnt;
2252 	}
2253 	mtx_unlock(&copy_page_mutex);
2254 }
2255 
2256 /*
2257  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2258  * into virtual memory and using bzero to clear its contents. This is intended
2259  * to be called from the vm_pagezero process only and outside of Giant. No
2260  * lock is required.
2261  */
2262 static void
2263 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2264 {
2265 	vm_offset_t va;
2266 
2267 	va = zero_page_idle_va;
2268 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2269 	bzero((caddr_t)va, PAGE_SIZE);
2270 	mmu_booke_kremove(mmu, va);
2271 }
2272 
2273 /*
2274  * Return whether or not the specified physical page was modified
2275  * in any of physical maps.
2276  */
2277 static boolean_t
2278 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2279 {
2280 	pte_t *pte;
2281 	pv_entry_t pv;
2282 	boolean_t rv;
2283 
2284 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2285 	    ("mmu_booke_is_modified: page %p is not managed", m));
2286 	rv = FALSE;
2287 
2288 	/*
2289 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2290 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2291 	 * is clear, no PTEs can be modified.
2292 	 */
2293 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2294 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2295 		return (rv);
2296 	rw_wlock(&pvh_global_lock);
2297 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2298 		PMAP_LOCK(pv->pv_pmap);
2299 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2300 		    PTE_ISVALID(pte)) {
2301 			if (PTE_ISMODIFIED(pte))
2302 				rv = TRUE;
2303 		}
2304 		PMAP_UNLOCK(pv->pv_pmap);
2305 		if (rv)
2306 			break;
2307 	}
2308 	rw_wunlock(&pvh_global_lock);
2309 	return (rv);
2310 }
2311 
2312 /*
2313  * Return whether or not the specified virtual address is eligible
2314  * for prefault.
2315  */
2316 static boolean_t
2317 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2318 {
2319 
2320 	return (FALSE);
2321 }
2322 
2323 /*
2324  * Return whether or not the specified physical page was referenced
2325  * in any physical maps.
2326  */
2327 static boolean_t
2328 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2329 {
2330 	pte_t *pte;
2331 	pv_entry_t pv;
2332 	boolean_t rv;
2333 
2334 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2335 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2336 	rv = FALSE;
2337 	rw_wlock(&pvh_global_lock);
2338 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2339 		PMAP_LOCK(pv->pv_pmap);
2340 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2341 		    PTE_ISVALID(pte)) {
2342 			if (PTE_ISREFERENCED(pte))
2343 				rv = TRUE;
2344 		}
2345 		PMAP_UNLOCK(pv->pv_pmap);
2346 		if (rv)
2347 			break;
2348 	}
2349 	rw_wunlock(&pvh_global_lock);
2350 	return (rv);
2351 }
2352 
2353 /*
2354  * Clear the modify bits on the specified physical page.
2355  */
2356 static void
2357 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2358 {
2359 	pte_t *pte;
2360 	pv_entry_t pv;
2361 
2362 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2363 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2364 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2365 	KASSERT(!vm_page_xbusied(m),
2366 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2367 
2368 	/*
2369 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2370 	 * If the object containing the page is locked and the page is not
2371 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2372 	 */
2373 	if ((m->aflags & PGA_WRITEABLE) == 0)
2374 		return;
2375 	rw_wlock(&pvh_global_lock);
2376 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2377 		PMAP_LOCK(pv->pv_pmap);
2378 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2379 		    PTE_ISVALID(pte)) {
2380 			mtx_lock_spin(&tlbivax_mutex);
2381 			tlb_miss_lock();
2382 
2383 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2384 				tlb0_flush_entry(pv->pv_va);
2385 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2386 				    PTE_REFERENCED);
2387 			}
2388 
2389 			tlb_miss_unlock();
2390 			mtx_unlock_spin(&tlbivax_mutex);
2391 		}
2392 		PMAP_UNLOCK(pv->pv_pmap);
2393 	}
2394 	rw_wunlock(&pvh_global_lock);
2395 }
2396 
2397 /*
2398  * Return a count of reference bits for a page, clearing those bits.
2399  * It is not necessary for every reference bit to be cleared, but it
2400  * is necessary that 0 only be returned when there are truly no
2401  * reference bits set.
2402  *
2403  * XXX: The exact number of bits to check and clear is a matter that
2404  * should be tested and standardized at some point in the future for
2405  * optimal aging of shared pages.
2406  */
2407 static int
2408 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2409 {
2410 	pte_t *pte;
2411 	pv_entry_t pv;
2412 	int count;
2413 
2414 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2415 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2416 	count = 0;
2417 	rw_wlock(&pvh_global_lock);
2418 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2419 		PMAP_LOCK(pv->pv_pmap);
2420 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2421 		    PTE_ISVALID(pte)) {
2422 			if (PTE_ISREFERENCED(pte)) {
2423 				mtx_lock_spin(&tlbivax_mutex);
2424 				tlb_miss_lock();
2425 
2426 				tlb0_flush_entry(pv->pv_va);
2427 				pte->flags &= ~PTE_REFERENCED;
2428 
2429 				tlb_miss_unlock();
2430 				mtx_unlock_spin(&tlbivax_mutex);
2431 
2432 				if (++count > 4) {
2433 					PMAP_UNLOCK(pv->pv_pmap);
2434 					break;
2435 				}
2436 			}
2437 		}
2438 		PMAP_UNLOCK(pv->pv_pmap);
2439 	}
2440 	rw_wunlock(&pvh_global_lock);
2441 	return (count);
2442 }
2443 
2444 /*
2445  * Clear the wired attribute from the mappings for the specified range of
2446  * addresses in the given pmap.  Every valid mapping within that range must
2447  * have the wired attribute set.  In contrast, invalid mappings cannot have
2448  * the wired attribute set, so they are ignored.
2449  *
2450  * The wired attribute of the page table entry is not a hardware feature, so
2451  * there is no need to invalidate any TLB entries.
2452  */
2453 static void
2454 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2455 {
2456 	vm_offset_t va;
2457 	pte_t *pte;
2458 
2459 	PMAP_LOCK(pmap);
2460 	for (va = sva; va < eva; va += PAGE_SIZE) {
2461 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2462 		    PTE_ISVALID(pte)) {
2463 			if (!PTE_ISWIRED(pte))
2464 				panic("mmu_booke_unwire: pte %p isn't wired",
2465 				    pte);
2466 			pte->flags &= ~PTE_WIRED;
2467 			pmap->pm_stats.wired_count--;
2468 		}
2469 	}
2470 	PMAP_UNLOCK(pmap);
2471 
2472 }
2473 
2474 /*
2475  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2476  * page.  This count may be changed upwards or downwards in the future; it is
2477  * only necessary that true be returned for a small subset of pmaps for proper
2478  * page aging.
2479  */
2480 static boolean_t
2481 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2482 {
2483 	pv_entry_t pv;
2484 	int loops;
2485 	boolean_t rv;
2486 
2487 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2488 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2489 	loops = 0;
2490 	rv = FALSE;
2491 	rw_wlock(&pvh_global_lock);
2492 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2493 		if (pv->pv_pmap == pmap) {
2494 			rv = TRUE;
2495 			break;
2496 		}
2497 		if (++loops >= 16)
2498 			break;
2499 	}
2500 	rw_wunlock(&pvh_global_lock);
2501 	return (rv);
2502 }
2503 
2504 /*
2505  * Return the number of managed mappings to the given physical page that are
2506  * wired.
2507  */
2508 static int
2509 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2510 {
2511 	pv_entry_t pv;
2512 	pte_t *pte;
2513 	int count = 0;
2514 
2515 	if ((m->oflags & VPO_UNMANAGED) != 0)
2516 		return (count);
2517 	rw_wlock(&pvh_global_lock);
2518 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2519 		PMAP_LOCK(pv->pv_pmap);
2520 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2521 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2522 				count++;
2523 		PMAP_UNLOCK(pv->pv_pmap);
2524 	}
2525 	rw_wunlock(&pvh_global_lock);
2526 	return (count);
2527 }
2528 
2529 static int
2530 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2531 {
2532 	int i;
2533 	vm_offset_t va;
2534 
2535 	/*
2536 	 * This currently does not work for entries that
2537 	 * overlap TLB1 entries.
2538 	 */
2539 	for (i = 0; i < tlb1_idx; i ++) {
2540 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2541 			return (0);
2542 	}
2543 
2544 	return (EFAULT);
2545 }
2546 
2547 void
2548 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2549 {
2550 	vm_paddr_t ppa;
2551 	vm_offset_t ofs;
2552 	vm_size_t gran;
2553 
2554 	/* Minidumps are based on virtual memory addresses. */
2555 	if (do_minidump) {
2556 		*va = (void *)pa;
2557 		return;
2558 	}
2559 
2560 	/* Raw physical memory dumps don't have a virtual address. */
2561 	/* We always map a 256MB page at 256M. */
2562 	gran = 256 * 1024 * 1024;
2563 	ppa = pa & ~(gran - 1);
2564 	ofs = pa - ppa;
2565 	*va = (void *)gran;
2566 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2567 
2568 	if (sz > (gran - ofs))
2569 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2570 		    _TLB_ENTRY_IO);
2571 }
2572 
2573 void
2574 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2575 {
2576 	vm_paddr_t ppa;
2577 	vm_offset_t ofs;
2578 	vm_size_t gran;
2579 
2580 	/* Minidumps are based on virtual memory addresses. */
2581 	/* Nothing to do... */
2582 	if (do_minidump)
2583 		return;
2584 
2585 	/* Raw physical memory dumps don't have a virtual address. */
2586 	tlb1_idx--;
2587 	tlb1[tlb1_idx].mas1 = 0;
2588 	tlb1[tlb1_idx].mas2 = 0;
2589 	tlb1[tlb1_idx].mas3 = 0;
2590 	tlb1_write_entry(tlb1_idx);
2591 
2592 	gran = 256 * 1024 * 1024;
2593 	ppa = pa & ~(gran - 1);
2594 	ofs = pa - ppa;
2595 	if (sz > (gran - ofs)) {
2596 		tlb1_idx--;
2597 		tlb1[tlb1_idx].mas1 = 0;
2598 		tlb1[tlb1_idx].mas2 = 0;
2599 		tlb1[tlb1_idx].mas3 = 0;
2600 		tlb1_write_entry(tlb1_idx);
2601 	}
2602 }
2603 
2604 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2605 
2606 void
2607 mmu_booke_scan_init(mmu_t mmu)
2608 {
2609 	vm_offset_t va;
2610 	pte_t *pte;
2611 	int i;
2612 
2613 	if (!do_minidump) {
2614 		/* Initialize phys. segments for dumpsys(). */
2615 		memset(&dump_map, 0, sizeof(dump_map));
2616 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2617 		    &availmem_regions_sz);
2618 		for (i = 0; i < physmem_regions_sz; i++) {
2619 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2620 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2621 		}
2622 		return;
2623 	}
2624 
2625 	/* Virtual segments for minidumps: */
2626 	memset(&dump_map, 0, sizeof(dump_map));
2627 
2628 	/* 1st: kernel .data and .bss. */
2629 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2630 	dump_map[0].pa_size =
2631 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2632 
2633 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2634 	dump_map[1].pa_start = data_start;
2635 	dump_map[1].pa_size = data_end - data_start;
2636 
2637 	/* 3rd: kernel VM. */
2638 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2639 	/* Find start of next chunk (from va). */
2640 	while (va < virtual_end) {
2641 		/* Don't dump the buffer cache. */
2642 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2643 			va = kmi.buffer_eva;
2644 			continue;
2645 		}
2646 		pte = pte_find(mmu, kernel_pmap, va);
2647 		if (pte != NULL && PTE_ISVALID(pte))
2648 			break;
2649 		va += PAGE_SIZE;
2650 	}
2651 	if (va < virtual_end) {
2652 		dump_map[2].pa_start = va;
2653 		va += PAGE_SIZE;
2654 		/* Find last page in chunk. */
2655 		while (va < virtual_end) {
2656 			/* Don't run into the buffer cache. */
2657 			if (va == kmi.buffer_sva)
2658 				break;
2659 			pte = pte_find(mmu, kernel_pmap, va);
2660 			if (pte == NULL || !PTE_ISVALID(pte))
2661 				break;
2662 			va += PAGE_SIZE;
2663 		}
2664 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2665 	}
2666 }
2667 
2668 /*
2669  * Map a set of physical memory pages into the kernel virtual address space.
2670  * Return a pointer to where it is mapped. This routine is intended to be used
2671  * for mapping device memory, NOT real memory.
2672  */
2673 static void *
2674 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2675 {
2676 
2677 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2678 }
2679 
2680 static void *
2681 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2682 {
2683 	void *res;
2684 	uintptr_t va;
2685 	vm_size_t sz;
2686 	int i;
2687 
2688 	/*
2689 	 * Check if this is premapped in TLB1. Note: this should probably also
2690 	 * check whether a sequence of TLB1 entries exist that match the
2691 	 * requirement, but now only checks the easy case.
2692 	 */
2693 	if (ma == VM_MEMATTR_DEFAULT) {
2694 		for (i = 0; i < tlb1_idx; i++) {
2695 			if (!(tlb1[i].mas1 & MAS1_VALID))
2696 				continue;
2697 			if (pa >= tlb1[i].phys &&
2698 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2699 				return (void *)(tlb1[i].virt +
2700 				    (pa - tlb1[i].phys));
2701 		}
2702 	}
2703 
2704 	size = roundup(size, PAGE_SIZE);
2705 
2706 	/*
2707 	 * We leave a hole for device direct mapping between the maximum user
2708 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2709 	 * devices are in there, just map them 1:1. If not, map them to the
2710 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2711 	 * addresses should be pulled from an allocator, but since we do not
2712 	 * ever free TLB1 entries, it is safe just to increment a counter.
2713 	 * Note that there isn't a lot of address space here (128 MB) and it
2714 	 * is not at all difficult to imagine running out, since that is a 4:1
2715 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2716 	 * mapped there.
2717 	 */
2718 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2719 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2720 		va = pa;
2721 	else
2722 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2723 	res = (void *)va;
2724 
2725 	do {
2726 		sz = 1 << (ilog2(size) & ~1);
2727 		if (bootverbose)
2728 			printf("Wiring VA=%x to PA=%x (size=%x), "
2729 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2730 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2731 		size -= sz;
2732 		pa += sz;
2733 		va += sz;
2734 	} while (size > 0);
2735 
2736 	return (res);
2737 }
2738 
2739 /*
2740  * 'Unmap' a range mapped by mmu_booke_mapdev().
2741  */
2742 static void
2743 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2744 {
2745 #ifdef SUPPORTS_SHRINKING_TLB1
2746 	vm_offset_t base, offset;
2747 
2748 	/*
2749 	 * Unmap only if this is inside kernel virtual space.
2750 	 */
2751 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2752 		base = trunc_page(va);
2753 		offset = va & PAGE_MASK;
2754 		size = roundup(offset + size, PAGE_SIZE);
2755 		kva_free(base, size);
2756 	}
2757 #endif
2758 }
2759 
2760 /*
2761  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2762  * specified pmap. This eliminates the blast of soft faults on process startup
2763  * and immediately after an mmap.
2764  */
2765 static void
2766 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2767     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2768 {
2769 
2770 	VM_OBJECT_ASSERT_WLOCKED(object);
2771 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2772 	    ("mmu_booke_object_init_pt: non-device object"));
2773 }
2774 
2775 /*
2776  * Perform the pmap work for mincore.
2777  */
2778 static int
2779 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2780     vm_paddr_t *locked_pa)
2781 {
2782 
2783 	/* XXX: this should be implemented at some point */
2784 	return (0);
2785 }
2786 
2787 /**************************************************************************/
2788 /* TID handling */
2789 /**************************************************************************/
2790 
2791 /*
2792  * Allocate a TID. If necessary, steal one from someone else.
2793  * The new TID is flushed from the TLB before returning.
2794  */
2795 static tlbtid_t
2796 tid_alloc(pmap_t pmap)
2797 {
2798 	tlbtid_t tid;
2799 	int thiscpu;
2800 
2801 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2802 
2803 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2804 
2805 	thiscpu = PCPU_GET(cpuid);
2806 
2807 	tid = PCPU_GET(tid_next);
2808 	if (tid > TID_MAX)
2809 		tid = TID_MIN;
2810 	PCPU_SET(tid_next, tid + 1);
2811 
2812 	/* If we are stealing TID then clear the relevant pmap's field */
2813 	if (tidbusy[thiscpu][tid] != NULL) {
2814 
2815 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2816 
2817 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2818 
2819 		/* Flush all entries from TLB0 matching this TID. */
2820 		tid_flush(tid, tlb0_ways, tlb0_entries_per_way);
2821 	}
2822 
2823 	tidbusy[thiscpu][tid] = pmap;
2824 	pmap->pm_tid[thiscpu] = tid;
2825 	__asm __volatile("msync; isync");
2826 
2827 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2828 	    PCPU_GET(tid_next));
2829 
2830 	return (tid);
2831 }
2832 
2833 /**************************************************************************/
2834 /* TLB0 handling */
2835 /**************************************************************************/
2836 
2837 static void
2838 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2839     uint32_t mas7)
2840 {
2841 	int as;
2842 	char desc[3];
2843 	tlbtid_t tid;
2844 	vm_size_t size;
2845 	unsigned int tsize;
2846 
2847 	desc[2] = '\0';
2848 	if (mas1 & MAS1_VALID)
2849 		desc[0] = 'V';
2850 	else
2851 		desc[0] = ' ';
2852 
2853 	if (mas1 & MAS1_IPROT)
2854 		desc[1] = 'P';
2855 	else
2856 		desc[1] = ' ';
2857 
2858 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2859 	tid = MAS1_GETTID(mas1);
2860 
2861 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2862 	size = 0;
2863 	if (tsize)
2864 		size = tsize2size(tsize);
2865 
2866 	debugf("%3d: (%s) [AS=%d] "
2867 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2868 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2869 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2870 }
2871 
2872 /* Convert TLB0 va and way number to tlb0[] table index. */
2873 static inline unsigned int
2874 tlb0_tableidx(vm_offset_t va, unsigned int way)
2875 {
2876 	unsigned int idx;
2877 
2878 	idx = (way * TLB0_ENTRIES_PER_WAY);
2879 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2880 	return (idx);
2881 }
2882 
2883 /*
2884  * Invalidate TLB0 entry.
2885  */
2886 static inline void
2887 tlb0_flush_entry(vm_offset_t va)
2888 {
2889 
2890 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2891 
2892 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2893 
2894 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2895 	__asm __volatile("isync; msync");
2896 	__asm __volatile("tlbsync; msync");
2897 
2898 	CTR1(KTR_PMAP, "%s: e", __func__);
2899 }
2900 
2901 /* Print out contents of the MAS registers for each TLB0 entry */
2902 void
2903 tlb0_print_tlbentries(void)
2904 {
2905 	uint32_t mas0, mas1, mas2, mas3, mas7;
2906 	int entryidx, way, idx;
2907 
2908 	debugf("TLB0 entries:\n");
2909 	for (way = 0; way < TLB0_WAYS; way ++)
2910 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2911 
2912 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2913 			mtspr(SPR_MAS0, mas0);
2914 			__asm __volatile("isync");
2915 
2916 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2917 			mtspr(SPR_MAS2, mas2);
2918 
2919 			__asm __volatile("isync; tlbre");
2920 
2921 			mas1 = mfspr(SPR_MAS1);
2922 			mas2 = mfspr(SPR_MAS2);
2923 			mas3 = mfspr(SPR_MAS3);
2924 			mas7 = mfspr(SPR_MAS7);
2925 
2926 			idx = tlb0_tableidx(mas2, way);
2927 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2928 		}
2929 }
2930 
2931 /**************************************************************************/
2932 /* TLB1 handling */
2933 /**************************************************************************/
2934 
2935 /*
2936  * TLB1 mapping notes:
2937  *
2938  * TLB1[0]	Kernel text and data.
2939  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2940  *		windows, other devices mappings.
2941  */
2942 
2943 /*
2944  * Write given entry to TLB1 hardware.
2945  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2946  */
2947 static void
2948 tlb1_write_entry(unsigned int idx)
2949 {
2950 	uint32_t mas0, mas7;
2951 
2952 	//debugf("tlb1_write_entry: s\n");
2953 
2954 	/* Clear high order RPN bits */
2955 	mas7 = 0;
2956 
2957 	/* Select entry */
2958 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2959 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2960 
2961 	mtspr(SPR_MAS0, mas0);
2962 	__asm __volatile("isync");
2963 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2964 	__asm __volatile("isync");
2965 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2966 	__asm __volatile("isync");
2967 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2968 	__asm __volatile("isync");
2969 	mtspr(SPR_MAS7, mas7);
2970 	__asm __volatile("isync; tlbwe; isync; msync");
2971 
2972 	//debugf("tlb1_write_entry: e\n");
2973 }
2974 
2975 /*
2976  * Return the largest uint value log such that 2^log <= num.
2977  */
2978 static unsigned int
2979 ilog2(unsigned int num)
2980 {
2981 	int lz;
2982 
2983 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2984 	return (31 - lz);
2985 }
2986 
2987 /*
2988  * Convert TLB TSIZE value to mapped region size.
2989  */
2990 static vm_size_t
2991 tsize2size(unsigned int tsize)
2992 {
2993 
2994 	/*
2995 	 * size = 4^tsize KB
2996 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2997 	 */
2998 
2999 	return ((1 << (2 * tsize)) * 1024);
3000 }
3001 
3002 /*
3003  * Convert region size (must be power of 4) to TLB TSIZE value.
3004  */
3005 static unsigned int
3006 size2tsize(vm_size_t size)
3007 {
3008 
3009 	return (ilog2(size) / 2 - 5);
3010 }
3011 
3012 /*
3013  * Register permanent kernel mapping in TLB1.
3014  *
3015  * Entries are created starting from index 0 (current free entry is
3016  * kept in tlb1_idx) and are not supposed to be invalidated.
3017  */
3018 static int
3019 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
3020     uint32_t flags)
3021 {
3022 	uint32_t ts, tid;
3023 	int tsize, index;
3024 
3025 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3026 	if (index >= TLB1_ENTRIES) {
3027 		printf("tlb1_set_entry: TLB1 full!\n");
3028 		return (-1);
3029 	}
3030 
3031 	/* Convert size to TSIZE */
3032 	tsize = size2tsize(size);
3033 
3034 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3035 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3036 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3037 
3038 	/*
3039 	 * Atomicity is preserved by the atomic increment above since nothing
3040 	 * is ever removed from tlb1.
3041 	 */
3042 
3043 	tlb1[index].phys = pa;
3044 	tlb1[index].virt = va;
3045 	tlb1[index].size = size;
3046 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3047 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3048 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3049 
3050 	/* Set supervisor RWX permission bits */
3051 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3052 
3053 	tlb1_write_entry(index);
3054 
3055 	/*
3056 	 * XXX in general TLB1 updates should be propagated between CPUs,
3057 	 * since current design assumes to have the same TLB1 set-up on all
3058 	 * cores.
3059 	 */
3060 	return (0);
3061 }
3062 
3063 /*
3064  * Map in contiguous RAM region into the TLB1 using maximum of
3065  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3066  *
3067  * If necessary round up last entry size and return total size
3068  * used by all allocated entries.
3069  */
3070 vm_size_t
3071 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3072 {
3073 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3074 	vm_size_t mapped, pgsz, base, mask;
3075 	int idx, nents;
3076 
3077 	/* Round up to the next 1M */
3078 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3079 
3080 	mapped = 0;
3081 	idx = 0;
3082 	base = va;
3083 	pgsz = 64*1024*1024;
3084 	while (mapped < size) {
3085 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3086 			while (pgsz > (size - mapped))
3087 				pgsz >>= 2;
3088 			pgs[idx++] = pgsz;
3089 			mapped += pgsz;
3090 		}
3091 
3092 		/* We under-map. Correct for this. */
3093 		if (mapped < size) {
3094 			while (pgs[idx - 1] == pgsz) {
3095 				idx--;
3096 				mapped -= pgsz;
3097 			}
3098 			/* XXX We may increase beyond out starting point. */
3099 			pgsz <<= 2;
3100 			pgs[idx++] = pgsz;
3101 			mapped += pgsz;
3102 		}
3103 	}
3104 
3105 	nents = idx;
3106 	mask = pgs[0] - 1;
3107 	/* Align address to the boundary */
3108 	if (va & mask) {
3109 		va = (va + mask) & ~mask;
3110 		pa = (pa + mask) & ~mask;
3111 	}
3112 
3113 	for (idx = 0; idx < nents; idx++) {
3114 		pgsz = pgs[idx];
3115 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3116 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3117 		pa += pgsz;
3118 		va += pgsz;
3119 	}
3120 
3121 	mapped = (va - base);
3122 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3123 	    mapped, mapped - size);
3124 	return (mapped);
3125 }
3126 
3127 /*
3128  * TLB1 initialization routine, to be called after the very first
3129  * assembler level setup done in locore.S.
3130  */
3131 void
3132 tlb1_init()
3133 {
3134 	uint32_t mas0, mas1, mas2, mas3;
3135 	uint32_t tsz;
3136 	u_int i;
3137 
3138 	if (bootinfo != NULL && bootinfo[0] != 1) {
3139 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3140 	} else
3141 		tlb1_idx = 1;
3142 
3143 	/* The first entry/entries are used to map the kernel. */
3144 	for (i = 0; i < tlb1_idx; i++) {
3145 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3146 		mtspr(SPR_MAS0, mas0);
3147 		__asm __volatile("isync; tlbre");
3148 
3149 		mas1 = mfspr(SPR_MAS1);
3150 		if ((mas1 & MAS1_VALID) == 0)
3151 			continue;
3152 
3153 		mas2 = mfspr(SPR_MAS2);
3154 		mas3 = mfspr(SPR_MAS3);
3155 
3156 		tlb1[i].mas1 = mas1;
3157 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3158 		tlb1[i].mas3 = mas3;
3159 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3160 		tlb1[i].phys = mas3 & MAS3_RPN;
3161 
3162 		if (i == 0)
3163 			kernload = mas3 & MAS3_RPN;
3164 
3165 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3166 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3167 		kernsize += tlb1[i].size;
3168 	}
3169 
3170 #ifdef SMP
3171 	bp_ntlb1s = tlb1_idx;
3172 #endif
3173 
3174 	/* Purge the remaining entries */
3175 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3176 		tlb1_write_entry(i);
3177 
3178 	/* Setup TLB miss defaults */
3179 	set_mas4_defaults();
3180 }
3181 
3182 vm_offset_t
3183 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3184 {
3185 	vm_paddr_t pa_base;
3186 	vm_offset_t va, sz;
3187 	int i;
3188 
3189 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3190 
3191 	for (i = 0; i < tlb1_idx; i++) {
3192 		if (!(tlb1[i].mas1 & MAS1_VALID))
3193 			continue;
3194 		if (pa >= tlb1[i].phys && (pa + size) <=
3195 		    (tlb1[i].phys + tlb1[i].size))
3196 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3197 	}
3198 
3199 	pa_base = trunc_page(pa);
3200 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3201 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3202 	va = tlb1_map_base + (pa - pa_base);
3203 
3204 	do {
3205 		sz = 1 << (ilog2(size) & ~1);
3206 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3207 		size -= sz;
3208 		pa_base += sz;
3209 		tlb1_map_base += sz;
3210 	} while (size > 0);
3211 
3212 #ifdef SMP
3213 	bp_ntlb1s = tlb1_idx;
3214 #endif
3215 
3216 	return (va);
3217 }
3218 
3219 /*
3220  * Setup MAS4 defaults.
3221  * These values are loaded to MAS0-2 on a TLB miss.
3222  */
3223 static void
3224 set_mas4_defaults(void)
3225 {
3226 	uint32_t mas4;
3227 
3228 	/* Defaults: TLB0, PID0, TSIZED=4K */
3229 	mas4 = MAS4_TLBSELD0;
3230 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3231 #ifdef SMP
3232 	mas4 |= MAS4_MD;
3233 #endif
3234 	mtspr(SPR_MAS4, mas4);
3235 	__asm __volatile("isync");
3236 }
3237 
3238 /*
3239  * Print out contents of the MAS registers for each TLB1 entry
3240  */
3241 void
3242 tlb1_print_tlbentries(void)
3243 {
3244 	uint32_t mas0, mas1, mas2, mas3, mas7;
3245 	int i;
3246 
3247 	debugf("TLB1 entries:\n");
3248 	for (i = 0; i < TLB1_ENTRIES; i++) {
3249 
3250 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3251 		mtspr(SPR_MAS0, mas0);
3252 
3253 		__asm __volatile("isync; tlbre");
3254 
3255 		mas1 = mfspr(SPR_MAS1);
3256 		mas2 = mfspr(SPR_MAS2);
3257 		mas3 = mfspr(SPR_MAS3);
3258 		mas7 = mfspr(SPR_MAS7);
3259 
3260 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3261 	}
3262 }
3263 
3264 /*
3265  * Print out contents of the in-ram tlb1 table.
3266  */
3267 void
3268 tlb1_print_entries(void)
3269 {
3270 	int i;
3271 
3272 	debugf("tlb1[] table entries:\n");
3273 	for (i = 0; i < TLB1_ENTRIES; i++)
3274 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3275 }
3276 
3277 /*
3278  * Return 0 if the physical IO range is encompassed by one of the
3279  * the TLB1 entries, otherwise return related error code.
3280  */
3281 static int
3282 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3283 {
3284 	uint32_t prot;
3285 	vm_paddr_t pa_start;
3286 	vm_paddr_t pa_end;
3287 	unsigned int entry_tsize;
3288 	vm_size_t entry_size;
3289 
3290 	*va = (vm_offset_t)NULL;
3291 
3292 	/* Skip invalid entries */
3293 	if (!(tlb1[i].mas1 & MAS1_VALID))
3294 		return (EINVAL);
3295 
3296 	/*
3297 	 * The entry must be cache-inhibited, guarded, and r/w
3298 	 * so it can function as an i/o page
3299 	 */
3300 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3301 	if (prot != (MAS2_I | MAS2_G))
3302 		return (EPERM);
3303 
3304 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3305 	if (prot != (MAS3_SR | MAS3_SW))
3306 		return (EPERM);
3307 
3308 	/* The address should be within the entry range. */
3309 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3310 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3311 
3312 	entry_size = tsize2size(entry_tsize);
3313 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3314 	pa_end = pa_start + entry_size - 1;
3315 
3316 	if ((pa < pa_start) || ((pa + size) > pa_end))
3317 		return (ERANGE);
3318 
3319 	/* Return virtual address of this mapping. */
3320 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3321 	return (0);
3322 }
3323