xref: /freebsd/sys/powerpc/booke/pmap.c (revision 21a4258d89a4e27632cfd87e5ad6e8538a6e77a2)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include "opt_kstack_pages.h"
55 
56 #include <sys/param.h>
57 #include <sys/conf.h>
58 #include <sys/malloc.h>
59 #include <sys/ktr.h>
60 #include <sys/proc.h>
61 #include <sys/user.h>
62 #include <sys/queue.h>
63 #include <sys/systm.h>
64 #include <sys/kernel.h>
65 #include <sys/kerneldump.h>
66 #include <sys/linker.h>
67 #include <sys/msgbuf.h>
68 #include <sys/lock.h>
69 #include <sys/mutex.h>
70 #include <sys/rwlock.h>
71 #include <sys/sched.h>
72 #include <sys/smp.h>
73 #include <sys/vmmeter.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_pager.h>
84 #include <vm/uma.h>
85 
86 #include <machine/cpu.h>
87 #include <machine/pcb.h>
88 #include <machine/platform.h>
89 
90 #include <machine/tlb.h>
91 #include <machine/spr.h>
92 #include <machine/md_var.h>
93 #include <machine/mmuvar.h>
94 #include <machine/pmap.h>
95 #include <machine/pte.h>
96 
97 #include "mmu_if.h"
98 
99 #define	SPARSE_MAPDEV
100 #ifdef  DEBUG
101 #define debugf(fmt, args...) printf(fmt, ##args)
102 #else
103 #define debugf(fmt, args...)
104 #endif
105 
106 #define TODO			panic("%s: not implemented", __func__);
107 
108 extern unsigned char _etext[];
109 extern unsigned char _end[];
110 
111 extern uint32_t *bootinfo;
112 
113 vm_paddr_t kernload;
114 vm_offset_t kernstart;
115 vm_size_t kernsize;
116 
117 /* Message buffer and tables. */
118 static vm_offset_t data_start;
119 static vm_size_t data_end;
120 
121 /* Phys/avail memory regions. */
122 static struct mem_region *availmem_regions;
123 static int availmem_regions_sz;
124 static struct mem_region *physmem_regions;
125 static int physmem_regions_sz;
126 
127 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
128 static vm_offset_t zero_page_va;
129 static struct mtx zero_page_mutex;
130 
131 static struct mtx tlbivax_mutex;
132 
133 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
134 static vm_offset_t copy_page_src_va;
135 static vm_offset_t copy_page_dst_va;
136 static struct mtx copy_page_mutex;
137 
138 /**************************************************************************/
139 /* PMAP */
140 /**************************************************************************/
141 
142 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
143     vm_prot_t, u_int flags, int8_t psind);
144 
145 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
146 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
147 
148 /*
149  * If user pmap is processed with mmu_booke_remove and the resident count
150  * drops to 0, there are no more pages to remove, so we need not continue.
151  */
152 #define PMAP_REMOVE_DONE(pmap) \
153 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
154 
155 extern int elf32_nxstack;
156 
157 /**************************************************************************/
158 /* TLB and TID handling */
159 /**************************************************************************/
160 
161 /* Translation ID busy table */
162 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
163 
164 /*
165  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
166  * core revisions and should be read from h/w registers during early config.
167  */
168 uint32_t tlb0_entries;
169 uint32_t tlb0_ways;
170 uint32_t tlb0_entries_per_way;
171 uint32_t tlb1_entries;
172 
173 #define TLB0_ENTRIES		(tlb0_entries)
174 #define TLB0_WAYS		(tlb0_ways)
175 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
176 
177 #define TLB1_ENTRIES (tlb1_entries)
178 #define TLB1_MAXENTRIES	64
179 
180 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
181 
182 static tlbtid_t tid_alloc(struct pmap *);
183 static void tid_flush(tlbtid_t tid);
184 
185 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
186 
187 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
188 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
189 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
190 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
191 
192 static vm_size_t tsize2size(unsigned int);
193 static unsigned int size2tsize(vm_size_t);
194 static unsigned int ilog2(unsigned int);
195 
196 static void set_mas4_defaults(void);
197 
198 static inline void tlb0_flush_entry(vm_offset_t);
199 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
200 
201 /**************************************************************************/
202 /* Page table management */
203 /**************************************************************************/
204 
205 static struct rwlock_padalign pvh_global_lock;
206 
207 /* Data for the pv entry allocation mechanism */
208 static uma_zone_t pvzone;
209 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
210 
211 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
212 
213 #ifndef PMAP_SHPGPERPROC
214 #define PMAP_SHPGPERPROC	200
215 #endif
216 
217 static void ptbl_init(void);
218 static struct ptbl_buf *ptbl_buf_alloc(void);
219 static void ptbl_buf_free(struct ptbl_buf *);
220 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
221 
222 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
223 static void ptbl_free(mmu_t, pmap_t, unsigned int);
224 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
225 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
226 
227 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
228 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
229 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
230 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
231 static void kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr,
232 			     vm_offset_t pdir);
233 
234 static pv_entry_t pv_alloc(void);
235 static void pv_free(pv_entry_t);
236 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
237 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
238 
239 static void booke_pmap_init_qpages(void);
240 
241 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
242 #define PTBL_BUFS		(128 * 16)
243 
244 struct ptbl_buf {
245 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
246 	vm_offset_t kva;		/* va of mapping */
247 };
248 
249 /* ptbl free list and a lock used for access synchronization. */
250 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
251 static struct mtx ptbl_buf_freelist_lock;
252 
253 /* Base address of kva space allocated fot ptbl bufs. */
254 static vm_offset_t ptbl_buf_pool_vabase;
255 
256 /* Pointer to ptbl_buf structures. */
257 static struct ptbl_buf *ptbl_bufs;
258 
259 #ifdef SMP
260 extern tlb_entry_t __boot_tlb1[];
261 void pmap_bootstrap_ap(volatile uint32_t *);
262 #endif
263 
264 /*
265  * Kernel MMU interface
266  */
267 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
268 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
269     vm_size_t, vm_offset_t);
270 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
271 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
272     vm_offset_t, vm_page_t *, vm_offset_t, int);
273 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
274     vm_prot_t, u_int flags, int8_t psind);
275 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
276     vm_page_t, vm_prot_t);
277 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
278     vm_prot_t);
279 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
280 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
281     vm_prot_t);
282 static void		mmu_booke_init(mmu_t);
283 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
284 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
285 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
286 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
287 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
288     int);
289 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
290     vm_paddr_t *);
291 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
292     vm_object_t, vm_pindex_t, vm_size_t);
293 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
294 static void		mmu_booke_page_init(mmu_t, vm_page_t);
295 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
296 static void		mmu_booke_pinit(mmu_t, pmap_t);
297 static void		mmu_booke_pinit0(mmu_t, pmap_t);
298 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
299     vm_prot_t);
300 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
301 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
302 static void		mmu_booke_release(mmu_t, pmap_t);
303 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
304 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
305 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
306 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
308 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
309 static void		mmu_booke_activate(mmu_t, struct thread *);
310 static void		mmu_booke_deactivate(mmu_t, struct thread *);
311 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
312 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
313 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
314 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
315 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
316 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
317 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
318 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
319 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
320 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
321     vm_size_t);
322 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
323     void **);
324 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
325     void *);
326 static void		mmu_booke_scan_init(mmu_t);
327 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
328 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
329 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
330     vm_size_t sz, vm_memattr_t mode);
331 
332 static mmu_method_t mmu_booke_methods[] = {
333 	/* pmap dispatcher interface */
334 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
335 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
336 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
337 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
338 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
339 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
340 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
341 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
342 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
343 	MMUMETHOD(mmu_init,		mmu_booke_init),
344 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
345 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
346 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
347 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
348 	MMUMETHOD(mmu_map,		mmu_booke_map),
349 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
350 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
351 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
352 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
353 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
354 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
355 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
356 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
357 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
358 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
359 	MMUMETHOD(mmu_release,		mmu_booke_release),
360 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
361 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
362 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
363 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
364 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
365 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
366 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
367 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
368 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
369 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
370 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
371 
372 	/* Internal interfaces */
373 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
374 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
375 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
376 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
377 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
378 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
379 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
380 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
381 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
382 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
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_paddr_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 (MAS2_I | MAS2_G);
404 		case VM_MEMATTR_WRITE_COMBINING:
405 		case VM_MEMATTR_WRITE_BACK:
406 		case VM_MEMATTR_PREFETCHABLE:
407 			return (MAS2_I);
408 		case VM_MEMATTR_WRITE_THROUGH:
409 			return (MAS2_W | MAS2_M);
410 		case VM_MEMATTR_CACHEABLE:
411 			return (MAS2_M);
412 		}
413 	}
414 
415 	/*
416 	 * Assume the page is cache inhibited and access is guarded unless
417 	 * it's in our available memory array.
418 	 */
419 	attrib = _TLB_ENTRY_IO;
420 	for (i = 0; i < physmem_regions_sz; i++) {
421 		if ((pa >= physmem_regions[i].mr_start) &&
422 		    (pa < (physmem_regions[i].mr_start +
423 		     physmem_regions[i].mr_size))) {
424 			attrib = _TLB_ENTRY_MEM;
425 			break;
426 		}
427 	}
428 
429 	return (attrib);
430 }
431 
432 static inline void
433 tlb_miss_lock(void)
434 {
435 #ifdef SMP
436 	struct pcpu *pc;
437 
438 	if (!smp_started)
439 		return;
440 
441 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
442 		if (pc != pcpup) {
443 
444 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
445 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
446 
447 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
448 			    ("tlb_miss_lock: tried to lock self"));
449 
450 			tlb_lock(pc->pc_booke_tlb_lock);
451 
452 			CTR1(KTR_PMAP, "%s: locked", __func__);
453 		}
454 	}
455 #endif
456 }
457 
458 static inline void
459 tlb_miss_unlock(void)
460 {
461 #ifdef SMP
462 	struct pcpu *pc;
463 
464 	if (!smp_started)
465 		return;
466 
467 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
468 		if (pc != pcpup) {
469 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
470 			    __func__, pc->pc_cpuid);
471 
472 			tlb_unlock(pc->pc_booke_tlb_lock);
473 
474 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
475 		}
476 	}
477 #endif
478 }
479 
480 /* Return number of entries in TLB0. */
481 static __inline void
482 tlb0_get_tlbconf(void)
483 {
484 	uint32_t tlb0_cfg;
485 
486 	tlb0_cfg = mfspr(SPR_TLB0CFG);
487 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
488 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
489 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
490 }
491 
492 /* Return number of entries in TLB1. */
493 static __inline void
494 tlb1_get_tlbconf(void)
495 {
496 	uint32_t tlb1_cfg;
497 
498 	tlb1_cfg = mfspr(SPR_TLB1CFG);
499 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
500 }
501 
502 /**************************************************************************/
503 /* Page table related */
504 /**************************************************************************/
505 
506 /* Initialize pool of kva ptbl buffers. */
507 static void
508 ptbl_init(void)
509 {
510 	int i;
511 
512 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
513 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
514 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
515 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
516 
517 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
518 	TAILQ_INIT(&ptbl_buf_freelist);
519 
520 	for (i = 0; i < PTBL_BUFS; i++) {
521 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
522 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
523 	}
524 }
525 
526 /* Get a ptbl_buf from the freelist. */
527 static struct ptbl_buf *
528 ptbl_buf_alloc(void)
529 {
530 	struct ptbl_buf *buf;
531 
532 	mtx_lock(&ptbl_buf_freelist_lock);
533 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
534 	if (buf != NULL)
535 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
536 	mtx_unlock(&ptbl_buf_freelist_lock);
537 
538 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
539 
540 	return (buf);
541 }
542 
543 /* Return ptbl buff to free pool. */
544 static void
545 ptbl_buf_free(struct ptbl_buf *buf)
546 {
547 
548 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
549 
550 	mtx_lock(&ptbl_buf_freelist_lock);
551 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
552 	mtx_unlock(&ptbl_buf_freelist_lock);
553 }
554 
555 /*
556  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
557  */
558 static void
559 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
560 {
561 	struct ptbl_buf *pbuf;
562 
563 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
564 
565 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
566 
567 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
568 		if (pbuf->kva == (vm_offset_t)ptbl) {
569 			/* Remove from pmap ptbl buf list. */
570 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
571 
572 			/* Free corresponding ptbl buf. */
573 			ptbl_buf_free(pbuf);
574 			break;
575 		}
576 }
577 
578 /* Allocate page table. */
579 static pte_t *
580 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
581 {
582 	vm_page_t mtbl[PTBL_PAGES];
583 	vm_page_t m;
584 	struct ptbl_buf *pbuf;
585 	unsigned int pidx;
586 	pte_t *ptbl;
587 	int i, j;
588 
589 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
590 	    (pmap == kernel_pmap), pdir_idx);
591 
592 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
593 	    ("ptbl_alloc: invalid pdir_idx"));
594 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
595 	    ("pte_alloc: valid ptbl entry exists!"));
596 
597 	pbuf = ptbl_buf_alloc();
598 	if (pbuf == NULL)
599 		panic("pte_alloc: couldn't alloc kernel virtual memory");
600 
601 	ptbl = (pte_t *)pbuf->kva;
602 
603 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
604 
605 	/* Allocate ptbl pages, this will sleep! */
606 	for (i = 0; i < PTBL_PAGES; i++) {
607 		pidx = (PTBL_PAGES * pdir_idx) + i;
608 		while ((m = vm_page_alloc(NULL, pidx,
609 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
610 			PMAP_UNLOCK(pmap);
611 			rw_wunlock(&pvh_global_lock);
612 			if (nosleep) {
613 				ptbl_free_pmap_ptbl(pmap, ptbl);
614 				for (j = 0; j < i; j++)
615 					vm_page_free(mtbl[j]);
616 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
617 				return (NULL);
618 			}
619 			VM_WAIT;
620 			rw_wlock(&pvh_global_lock);
621 			PMAP_LOCK(pmap);
622 		}
623 		mtbl[i] = m;
624 	}
625 
626 	/* Map allocated pages into kernel_pmap. */
627 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
628 
629 	/* Zero whole ptbl. */
630 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
631 
632 	/* Add pbuf to the pmap ptbl bufs list. */
633 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
634 
635 	return (ptbl);
636 }
637 
638 /* Free ptbl pages and invalidate pdir entry. */
639 static void
640 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
641 {
642 	pte_t *ptbl;
643 	vm_paddr_t pa;
644 	vm_offset_t va;
645 	vm_page_t m;
646 	int i;
647 
648 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
649 	    (pmap == kernel_pmap), pdir_idx);
650 
651 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
652 	    ("ptbl_free: invalid pdir_idx"));
653 
654 	ptbl = pmap->pm_pdir[pdir_idx];
655 
656 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
657 
658 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
659 
660 	/*
661 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
662 	 * don't attempt to look up the page tables we are releasing.
663 	 */
664 	mtx_lock_spin(&tlbivax_mutex);
665 	tlb_miss_lock();
666 
667 	pmap->pm_pdir[pdir_idx] = NULL;
668 
669 	tlb_miss_unlock();
670 	mtx_unlock_spin(&tlbivax_mutex);
671 
672 	for (i = 0; i < PTBL_PAGES; i++) {
673 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
674 		pa = pte_vatopa(mmu, kernel_pmap, va);
675 		m = PHYS_TO_VM_PAGE(pa);
676 		vm_page_free_zero(m);
677 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
678 		mmu_booke_kremove(mmu, va);
679 	}
680 
681 	ptbl_free_pmap_ptbl(pmap, ptbl);
682 }
683 
684 /*
685  * Decrement ptbl pages hold count and attempt to free ptbl pages.
686  * Called when removing pte entry from ptbl.
687  *
688  * Return 1 if ptbl pages were freed.
689  */
690 static int
691 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
692 {
693 	pte_t *ptbl;
694 	vm_paddr_t pa;
695 	vm_page_t m;
696 	int i;
697 
698 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
699 	    (pmap == kernel_pmap), pdir_idx);
700 
701 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
702 	    ("ptbl_unhold: invalid pdir_idx"));
703 	KASSERT((pmap != kernel_pmap),
704 	    ("ptbl_unhold: unholding kernel ptbl!"));
705 
706 	ptbl = pmap->pm_pdir[pdir_idx];
707 
708 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
709 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
710 	    ("ptbl_unhold: non kva ptbl"));
711 
712 	/* decrement hold count */
713 	for (i = 0; i < PTBL_PAGES; i++) {
714 		pa = pte_vatopa(mmu, kernel_pmap,
715 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
716 		m = PHYS_TO_VM_PAGE(pa);
717 		m->wire_count--;
718 	}
719 
720 	/*
721 	 * Free ptbl pages if there are no pte etries in this ptbl.
722 	 * wire_count has the same value for all ptbl pages, so check the last
723 	 * page.
724 	 */
725 	if (m->wire_count == 0) {
726 		ptbl_free(mmu, pmap, pdir_idx);
727 
728 		//debugf("ptbl_unhold: e (freed ptbl)\n");
729 		return (1);
730 	}
731 
732 	return (0);
733 }
734 
735 /*
736  * Increment hold count for ptbl pages. This routine is used when a new pte
737  * entry is being inserted into the ptbl.
738  */
739 static void
740 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
741 {
742 	vm_paddr_t pa;
743 	pte_t *ptbl;
744 	vm_page_t m;
745 	int i;
746 
747 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
748 	    pdir_idx);
749 
750 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
751 	    ("ptbl_hold: invalid pdir_idx"));
752 	KASSERT((pmap != kernel_pmap),
753 	    ("ptbl_hold: holding kernel ptbl!"));
754 
755 	ptbl = pmap->pm_pdir[pdir_idx];
756 
757 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
758 
759 	for (i = 0; i < PTBL_PAGES; i++) {
760 		pa = pte_vatopa(mmu, kernel_pmap,
761 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
762 		m = PHYS_TO_VM_PAGE(pa);
763 		m->wire_count++;
764 	}
765 }
766 
767 /* Allocate pv_entry structure. */
768 pv_entry_t
769 pv_alloc(void)
770 {
771 	pv_entry_t pv;
772 
773 	pv_entry_count++;
774 	if (pv_entry_count > pv_entry_high_water)
775 		pagedaemon_wakeup();
776 	pv = uma_zalloc(pvzone, M_NOWAIT);
777 
778 	return (pv);
779 }
780 
781 /* Free pv_entry structure. */
782 static __inline void
783 pv_free(pv_entry_t pve)
784 {
785 
786 	pv_entry_count--;
787 	uma_zfree(pvzone, pve);
788 }
789 
790 
791 /* Allocate and initialize pv_entry structure. */
792 static void
793 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
794 {
795 	pv_entry_t pve;
796 
797 	//int su = (pmap == kernel_pmap);
798 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
799 	//	(u_int32_t)pmap, va, (u_int32_t)m);
800 
801 	pve = pv_alloc();
802 	if (pve == NULL)
803 		panic("pv_insert: no pv entries!");
804 
805 	pve->pv_pmap = pmap;
806 	pve->pv_va = va;
807 
808 	/* add to pv_list */
809 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
810 	rw_assert(&pvh_global_lock, RA_WLOCKED);
811 
812 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
813 
814 	//debugf("pv_insert: e\n");
815 }
816 
817 /* Destroy pv entry. */
818 static void
819 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
820 {
821 	pv_entry_t pve;
822 
823 	//int su = (pmap == kernel_pmap);
824 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
825 
826 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
827 	rw_assert(&pvh_global_lock, RA_WLOCKED);
828 
829 	/* find pv entry */
830 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
831 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
832 			/* remove from pv_list */
833 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
834 			if (TAILQ_EMPTY(&m->md.pv_list))
835 				vm_page_aflag_clear(m, PGA_WRITEABLE);
836 
837 			/* free pv entry struct */
838 			pv_free(pve);
839 			break;
840 		}
841 	}
842 
843 	//debugf("pv_remove: e\n");
844 }
845 
846 /*
847  * Clean pte entry, try to free page table page if requested.
848  *
849  * Return 1 if ptbl pages were freed, otherwise return 0.
850  */
851 static int
852 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
853 {
854 	unsigned int pdir_idx = PDIR_IDX(va);
855 	unsigned int ptbl_idx = PTBL_IDX(va);
856 	vm_page_t m;
857 	pte_t *ptbl;
858 	pte_t *pte;
859 
860 	//int su = (pmap == kernel_pmap);
861 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
862 	//		su, (u_int32_t)pmap, va, flags);
863 
864 	ptbl = pmap->pm_pdir[pdir_idx];
865 	KASSERT(ptbl, ("pte_remove: null ptbl"));
866 
867 	pte = &ptbl[ptbl_idx];
868 
869 	if (pte == NULL || !PTE_ISVALID(pte))
870 		return (0);
871 
872 	if (PTE_ISWIRED(pte))
873 		pmap->pm_stats.wired_count--;
874 
875 	/* Handle managed entry. */
876 	if (PTE_ISMANAGED(pte)) {
877 		/* Get vm_page_t for mapped pte. */
878 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
879 
880 		if (PTE_ISMODIFIED(pte))
881 			vm_page_dirty(m);
882 
883 		if (PTE_ISREFERENCED(pte))
884 			vm_page_aflag_set(m, PGA_REFERENCED);
885 
886 		pv_remove(pmap, va, m);
887 	}
888 
889 	mtx_lock_spin(&tlbivax_mutex);
890 	tlb_miss_lock();
891 
892 	tlb0_flush_entry(va);
893 	*pte = 0;
894 
895 	tlb_miss_unlock();
896 	mtx_unlock_spin(&tlbivax_mutex);
897 
898 	pmap->pm_stats.resident_count--;
899 
900 	if (flags & PTBL_UNHOLD) {
901 		//debugf("pte_remove: e (unhold)\n");
902 		return (ptbl_unhold(mmu, pmap, pdir_idx));
903 	}
904 
905 	//debugf("pte_remove: e\n");
906 	return (0);
907 }
908 
909 /*
910  * Insert PTE for a given page and virtual address.
911  */
912 static int
913 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
914     boolean_t nosleep)
915 {
916 	unsigned int pdir_idx = PDIR_IDX(va);
917 	unsigned int ptbl_idx = PTBL_IDX(va);
918 	pte_t *ptbl, *pte;
919 
920 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
921 	    pmap == kernel_pmap, pmap, va);
922 
923 	/* Get the page table pointer. */
924 	ptbl = pmap->pm_pdir[pdir_idx];
925 
926 	if (ptbl == NULL) {
927 		/* Allocate page table pages. */
928 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
929 		if (ptbl == NULL) {
930 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
931 			return (ENOMEM);
932 		}
933 	} else {
934 		/*
935 		 * Check if there is valid mapping for requested
936 		 * va, if there is, remove it.
937 		 */
938 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
939 		if (PTE_ISVALID(pte)) {
940 			pte_remove(mmu, pmap, va, PTBL_HOLD);
941 		} else {
942 			/*
943 			 * pte is not used, increment hold count
944 			 * for ptbl pages.
945 			 */
946 			if (pmap != kernel_pmap)
947 				ptbl_hold(mmu, pmap, pdir_idx);
948 		}
949 	}
950 
951 	/*
952 	 * Insert pv_entry into pv_list for mapped page if part of managed
953 	 * memory.
954 	 */
955 	if ((m->oflags & VPO_UNMANAGED) == 0) {
956 		flags |= PTE_MANAGED;
957 
958 		/* Create and insert pv entry. */
959 		pv_insert(pmap, va, m);
960 	}
961 
962 	pmap->pm_stats.resident_count++;
963 
964 	mtx_lock_spin(&tlbivax_mutex);
965 	tlb_miss_lock();
966 
967 	tlb0_flush_entry(va);
968 	if (pmap->pm_pdir[pdir_idx] == NULL) {
969 		/*
970 		 * If we just allocated a new page table, hook it in
971 		 * the pdir.
972 		 */
973 		pmap->pm_pdir[pdir_idx] = ptbl;
974 	}
975 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
976 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
977 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
978 
979 	tlb_miss_unlock();
980 	mtx_unlock_spin(&tlbivax_mutex);
981 	return (0);
982 }
983 
984 /* Return the pa for the given pmap/va. */
985 static vm_paddr_t
986 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
987 {
988 	vm_paddr_t pa = 0;
989 	pte_t *pte;
990 
991 	pte = pte_find(mmu, pmap, va);
992 	if ((pte != NULL) && PTE_ISVALID(pte))
993 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
994 	return (pa);
995 }
996 
997 /* Get a pointer to a PTE in a page table. */
998 static pte_t *
999 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1000 {
1001 	unsigned int pdir_idx = PDIR_IDX(va);
1002 	unsigned int ptbl_idx = PTBL_IDX(va);
1003 
1004 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1005 
1006 	if (pmap->pm_pdir[pdir_idx])
1007 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1008 
1009 	return (NULL);
1010 }
1011 
1012 /* Set up kernel page tables. */
1013 static void
1014 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1015 {
1016 	int		i;
1017 	vm_offset_t	va;
1018 	pte_t		*pte;
1019 
1020 	/* Initialize kernel pdir */
1021 	for (i = 0; i < kernel_ptbls; i++)
1022 		kernel_pmap->pm_pdir[kptbl_min + i] =
1023 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1024 
1025 	/*
1026 	 * Fill in PTEs covering kernel code and data. They are not required
1027 	 * for address translation, as this area is covered by static TLB1
1028 	 * entries, but for pte_vatopa() to work correctly with kernel area
1029 	 * addresses.
1030 	 */
1031 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1032 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1033 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1034 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1035 		    PTE_VALID | PTE_PS_4KB;
1036 	}
1037 }
1038 
1039 /**************************************************************************/
1040 /* PMAP related */
1041 /**************************************************************************/
1042 
1043 /*
1044  * This is called during booke_init, before the system is really initialized.
1045  */
1046 static void
1047 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1048 {
1049 	vm_paddr_t phys_kernelend;
1050 	struct mem_region *mp, *mp1;
1051 	int cnt, i, j;
1052 	vm_paddr_t s, e, sz;
1053 	vm_paddr_t physsz, hwphyssz;
1054 	u_int phys_avail_count;
1055 	vm_size_t kstack0_sz;
1056 	vm_offset_t kernel_pdir, kstack0;
1057 	vm_paddr_t kstack0_phys;
1058 	void *dpcpu;
1059 
1060 	debugf("mmu_booke_bootstrap: entered\n");
1061 
1062 	/* Set interesting system properties */
1063 	hw_direct_map = 0;
1064 	elf32_nxstack = 1;
1065 
1066 	/* Initialize invalidation mutex */
1067 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1068 
1069 	/* Read TLB0 size and associativity. */
1070 	tlb0_get_tlbconf();
1071 
1072 	/*
1073 	 * Align kernel start and end address (kernel image).
1074 	 * Note that kernel end does not necessarily relate to kernsize.
1075 	 * kernsize is the size of the kernel that is actually mapped.
1076 	 */
1077 	kernstart = trunc_page(start);
1078 	data_start = round_page(kernelend);
1079 	data_end = data_start;
1080 
1081 	/*
1082 	 * Addresses of preloaded modules (like file systems) use
1083 	 * physical addresses. Make sure we relocate those into
1084 	 * virtual addresses.
1085 	 */
1086 	preload_addr_relocate = kernstart - kernload;
1087 
1088 	/* Allocate the dynamic per-cpu area. */
1089 	dpcpu = (void *)data_end;
1090 	data_end += DPCPU_SIZE;
1091 
1092 	/* Allocate space for the message buffer. */
1093 	msgbufp = (struct msgbuf *)data_end;
1094 	data_end += msgbufsize;
1095 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1096 	    data_end);
1097 
1098 	data_end = round_page(data_end);
1099 
1100 	/* Allocate space for ptbl_bufs. */
1101 	ptbl_bufs = (struct ptbl_buf *)data_end;
1102 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1103 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1104 	    data_end);
1105 
1106 	data_end = round_page(data_end);
1107 
1108 	/* Allocate PTE tables for kernel KVA. */
1109 	kernel_pdir = data_end;
1110 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1111 	    PDIR_SIZE);
1112 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1113 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1114 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1115 
1116 	debugf(" data_end: 0x%08x\n", data_end);
1117 	if (data_end - kernstart > kernsize) {
1118 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1119 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1120 	}
1121 	data_end = kernstart + kernsize;
1122 	debugf(" updated data_end: 0x%08x\n", data_end);
1123 
1124 	/*
1125 	 * Clear the structures - note we can only do it safely after the
1126 	 * possible additional TLB1 translations are in place (above) so that
1127 	 * all range up to the currently calculated 'data_end' is covered.
1128 	 */
1129 	dpcpu_init(dpcpu, 0);
1130 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1131 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1132 
1133 	/*******************************************************/
1134 	/* Set the start and end of kva. */
1135 	/*******************************************************/
1136 	virtual_avail = round_page(data_end);
1137 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1138 
1139 	/* Allocate KVA space for page zero/copy operations. */
1140 	zero_page_va = virtual_avail;
1141 	virtual_avail += PAGE_SIZE;
1142 	copy_page_src_va = virtual_avail;
1143 	virtual_avail += PAGE_SIZE;
1144 	copy_page_dst_va = virtual_avail;
1145 	virtual_avail += PAGE_SIZE;
1146 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1147 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1148 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1149 
1150 	/* Initialize page zero/copy mutexes. */
1151 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1152 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1153 
1154 	/* Allocate KVA space for ptbl bufs. */
1155 	ptbl_buf_pool_vabase = virtual_avail;
1156 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1157 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1158 	    ptbl_buf_pool_vabase, virtual_avail);
1159 
1160 	/* Calculate corresponding physical addresses for the kernel region. */
1161 	phys_kernelend = kernload + kernsize;
1162 	debugf("kernel image and allocated data:\n");
1163 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1164 	debugf(" kernstart   = 0x%08x\n", kernstart);
1165 	debugf(" kernsize    = 0x%08x\n", kernsize);
1166 
1167 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1168 		panic("mmu_booke_bootstrap: phys_avail too small");
1169 
1170 	/*
1171 	 * Remove kernel physical address range from avail regions list. Page
1172 	 * align all regions.  Non-page aligned memory isn't very interesting
1173 	 * to us.  Also, sort the entries for ascending addresses.
1174 	 */
1175 
1176 	/* Retrieve phys/avail mem regions */
1177 	mem_regions(&physmem_regions, &physmem_regions_sz,
1178 	    &availmem_regions, &availmem_regions_sz);
1179 	sz = 0;
1180 	cnt = availmem_regions_sz;
1181 	debugf("processing avail regions:\n");
1182 	for (mp = availmem_regions; mp->mr_size; mp++) {
1183 		s = mp->mr_start;
1184 		e = mp->mr_start + mp->mr_size;
1185 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1186 		/* Check whether this region holds all of the kernel. */
1187 		if (s < kernload && e > phys_kernelend) {
1188 			availmem_regions[cnt].mr_start = phys_kernelend;
1189 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1190 			e = kernload;
1191 		}
1192 		/* Look whether this regions starts within the kernel. */
1193 		if (s >= kernload && s < phys_kernelend) {
1194 			if (e <= phys_kernelend)
1195 				goto empty;
1196 			s = phys_kernelend;
1197 		}
1198 		/* Now look whether this region ends within the kernel. */
1199 		if (e > kernload && e <= phys_kernelend) {
1200 			if (s >= kernload)
1201 				goto empty;
1202 			e = kernload;
1203 		}
1204 		/* Now page align the start and size of the region. */
1205 		s = round_page(s);
1206 		e = trunc_page(e);
1207 		if (e < s)
1208 			e = s;
1209 		sz = e - s;
1210 		debugf("%09jx-%09jx = %jx\n",
1211 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1212 
1213 		/* Check whether some memory is left here. */
1214 		if (sz == 0) {
1215 		empty:
1216 			memmove(mp, mp + 1,
1217 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1218 			cnt--;
1219 			mp--;
1220 			continue;
1221 		}
1222 
1223 		/* Do an insertion sort. */
1224 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1225 			if (s < mp1->mr_start)
1226 				break;
1227 		if (mp1 < mp) {
1228 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1229 			mp1->mr_start = s;
1230 			mp1->mr_size = sz;
1231 		} else {
1232 			mp->mr_start = s;
1233 			mp->mr_size = sz;
1234 		}
1235 	}
1236 	availmem_regions_sz = cnt;
1237 
1238 	/*******************************************************/
1239 	/* Steal physical memory for kernel stack from the end */
1240 	/* of the first avail region                           */
1241 	/*******************************************************/
1242 	kstack0_sz = kstack_pages * PAGE_SIZE;
1243 	kstack0_phys = availmem_regions[0].mr_start +
1244 	    availmem_regions[0].mr_size;
1245 	kstack0_phys -= kstack0_sz;
1246 	availmem_regions[0].mr_size -= kstack0_sz;
1247 
1248 	/*******************************************************/
1249 	/* Fill in phys_avail table, based on availmem_regions */
1250 	/*******************************************************/
1251 	phys_avail_count = 0;
1252 	physsz = 0;
1253 	hwphyssz = 0;
1254 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1255 
1256 	debugf("fill in phys_avail:\n");
1257 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1258 
1259 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1260 		    (uintmax_t)availmem_regions[i].mr_start,
1261 		    (uintmax_t)availmem_regions[i].mr_start +
1262 		        availmem_regions[i].mr_size,
1263 		    (uintmax_t)availmem_regions[i].mr_size);
1264 
1265 		if (hwphyssz != 0 &&
1266 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1267 			debugf(" hw.physmem adjust\n");
1268 			if (physsz < hwphyssz) {
1269 				phys_avail[j] = availmem_regions[i].mr_start;
1270 				phys_avail[j + 1] =
1271 				    availmem_regions[i].mr_start +
1272 				    hwphyssz - physsz;
1273 				physsz = hwphyssz;
1274 				phys_avail_count++;
1275 			}
1276 			break;
1277 		}
1278 
1279 		phys_avail[j] = availmem_regions[i].mr_start;
1280 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1281 		    availmem_regions[i].mr_size;
1282 		phys_avail_count++;
1283 		physsz += availmem_regions[i].mr_size;
1284 	}
1285 	physmem = btoc(physsz);
1286 
1287 	/* Calculate the last available physical address. */
1288 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1289 		;
1290 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1291 
1292 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1293 	debugf("phys_avail_count = %d\n", phys_avail_count);
1294 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1295 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1296 
1297 	/*******************************************************/
1298 	/* Initialize (statically allocated) kernel pmap. */
1299 	/*******************************************************/
1300 	PMAP_LOCK_INIT(kernel_pmap);
1301 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1302 
1303 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1304 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1305 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1306 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1307 
1308 	kernel_pte_alloc(data_end, kernstart, kernel_pdir);
1309 	for (i = 0; i < MAXCPU; i++) {
1310 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1311 
1312 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1313 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1314 	}
1315 
1316 	/* Mark kernel_pmap active on all CPUs */
1317 	CPU_FILL(&kernel_pmap->pm_active);
1318 
1319  	/*
1320 	 * Initialize the global pv list lock.
1321 	 */
1322 	rw_init(&pvh_global_lock, "pmap pv global");
1323 
1324 	/*******************************************************/
1325 	/* Final setup */
1326 	/*******************************************************/
1327 
1328 	/* Enter kstack0 into kernel map, provide guard page */
1329 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1330 	thread0.td_kstack = kstack0;
1331 	thread0.td_kstack_pages = kstack_pages;
1332 
1333 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1334 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1335 	    kstack0_phys, kstack0_phys + kstack0_sz);
1336 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1337 
1338 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1339 	for (i = 0; i < kstack_pages; i++) {
1340 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1341 		kstack0 += PAGE_SIZE;
1342 		kstack0_phys += PAGE_SIZE;
1343 	}
1344 
1345 	pmap_bootstrapped = 1;
1346 
1347 	debugf("virtual_avail = %08x\n", virtual_avail);
1348 	debugf("virtual_end   = %08x\n", virtual_end);
1349 
1350 	debugf("mmu_booke_bootstrap: exit\n");
1351 }
1352 
1353 #ifdef SMP
1354  void
1355 tlb1_ap_prep(void)
1356 {
1357 	tlb_entry_t *e, tmp;
1358 	unsigned int i;
1359 
1360 	/* Prepare TLB1 image for AP processors */
1361 	e = __boot_tlb1;
1362 	for (i = 0; i < TLB1_ENTRIES; i++) {
1363 		tlb1_read_entry(&tmp, i);
1364 
1365 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1366 			memcpy(e++, &tmp, sizeof(tmp));
1367 	}
1368 }
1369 
1370 void
1371 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1372 {
1373 	int i;
1374 
1375 	/*
1376 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1377 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1378 	 * created by tlb1_ap_prep(), so use these values directly to
1379 	 * (re)program AP's TLB1 hardware.
1380 	 *
1381 	 * Start at index 1 because index 0 has the kernel map.
1382 	 */
1383 	for (i = 1; i < TLB1_ENTRIES; i++) {
1384 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1385 			tlb1_write_entry(&__boot_tlb1[i], i);
1386 	}
1387 
1388 	set_mas4_defaults();
1389 }
1390 #endif
1391 
1392 static void
1393 booke_pmap_init_qpages(void)
1394 {
1395 	struct pcpu *pc;
1396 	int i;
1397 
1398 	CPU_FOREACH(i) {
1399 		pc = pcpu_find(i);
1400 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1401 		if (pc->pc_qmap_addr == 0)
1402 			panic("pmap_init_qpages: unable to allocate KVA");
1403 	}
1404 }
1405 
1406 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1407 
1408 /*
1409  * Get the physical page address for the given pmap/virtual address.
1410  */
1411 static vm_paddr_t
1412 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1413 {
1414 	vm_paddr_t pa;
1415 
1416 	PMAP_LOCK(pmap);
1417 	pa = pte_vatopa(mmu, pmap, va);
1418 	PMAP_UNLOCK(pmap);
1419 
1420 	return (pa);
1421 }
1422 
1423 /*
1424  * Extract the physical page address associated with the given
1425  * kernel virtual address.
1426  */
1427 static vm_paddr_t
1428 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1429 {
1430 	tlb_entry_t e;
1431 	int i;
1432 
1433 	/* Check TLB1 mappings */
1434 	for (i = 0; i < TLB1_ENTRIES; i++) {
1435 		tlb1_read_entry(&e, i);
1436 		if (!(e.mas1 & MAS1_VALID))
1437 			continue;
1438 		if (va >= e.virt && va < e.virt + e.size)
1439 			return (e.phys + (va - e.virt));
1440 	}
1441 
1442 	return (pte_vatopa(mmu, kernel_pmap, va));
1443 }
1444 
1445 /*
1446  * Initialize the pmap module.
1447  * Called by vm_init, to initialize any structures that the pmap
1448  * system needs to map virtual memory.
1449  */
1450 static void
1451 mmu_booke_init(mmu_t mmu)
1452 {
1453 	int shpgperproc = PMAP_SHPGPERPROC;
1454 
1455 	/*
1456 	 * Initialize the address space (zone) for the pv entries.  Set a
1457 	 * high water mark so that the system can recover from excessive
1458 	 * numbers of pv entries.
1459 	 */
1460 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1461 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1462 
1463 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1464 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1465 
1466 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1467 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1468 
1469 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1470 
1471 	/* Pre-fill pvzone with initial number of pv entries. */
1472 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1473 
1474 	/* Initialize ptbl allocation. */
1475 	ptbl_init();
1476 }
1477 
1478 /*
1479  * Map a list of wired pages into kernel virtual address space.  This is
1480  * intended for temporary mappings which do not need page modification or
1481  * references recorded.  Existing mappings in the region are overwritten.
1482  */
1483 static void
1484 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1485 {
1486 	vm_offset_t va;
1487 
1488 	va = sva;
1489 	while (count-- > 0) {
1490 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1491 		va += PAGE_SIZE;
1492 		m++;
1493 	}
1494 }
1495 
1496 /*
1497  * Remove page mappings from kernel virtual address space.  Intended for
1498  * temporary mappings entered by mmu_booke_qenter.
1499  */
1500 static void
1501 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1502 {
1503 	vm_offset_t va;
1504 
1505 	va = sva;
1506 	while (count-- > 0) {
1507 		mmu_booke_kremove(mmu, va);
1508 		va += PAGE_SIZE;
1509 	}
1510 }
1511 
1512 /*
1513  * Map a wired page into kernel virtual address space.
1514  */
1515 static void
1516 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1517 {
1518 
1519 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1520 }
1521 
1522 static void
1523 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1524 {
1525 	uint32_t flags;
1526 	pte_t *pte;
1527 
1528 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1529 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1530 
1531 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1532 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1533 	flags |= PTE_PS_4KB;
1534 
1535 	pte = pte_find(mmu, kernel_pmap, va);
1536 
1537 	mtx_lock_spin(&tlbivax_mutex);
1538 	tlb_miss_lock();
1539 
1540 	if (PTE_ISVALID(pte)) {
1541 
1542 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1543 
1544 		/* Flush entry from TLB0 */
1545 		tlb0_flush_entry(va);
1546 	}
1547 
1548 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1549 
1550 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1551 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1552 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1553 
1554 	/* Flush the real memory from the instruction cache. */
1555 	if ((flags & (PTE_I | PTE_G)) == 0)
1556 		__syncicache((void *)va, PAGE_SIZE);
1557 
1558 	tlb_miss_unlock();
1559 	mtx_unlock_spin(&tlbivax_mutex);
1560 }
1561 
1562 /*
1563  * Remove a page from kernel page table.
1564  */
1565 static void
1566 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1567 {
1568 	pte_t *pte;
1569 
1570 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
1571 
1572 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1573 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1574 	    ("mmu_booke_kremove: invalid va"));
1575 
1576 	pte = pte_find(mmu, kernel_pmap, va);
1577 
1578 	if (!PTE_ISVALID(pte)) {
1579 
1580 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1581 
1582 		return;
1583 	}
1584 
1585 	mtx_lock_spin(&tlbivax_mutex);
1586 	tlb_miss_lock();
1587 
1588 	/* Invalidate entry in TLB0, update PTE. */
1589 	tlb0_flush_entry(va);
1590 	*pte = 0;
1591 
1592 	tlb_miss_unlock();
1593 	mtx_unlock_spin(&tlbivax_mutex);
1594 }
1595 
1596 /*
1597  * Initialize pmap associated with process 0.
1598  */
1599 static void
1600 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1601 {
1602 
1603 	PMAP_LOCK_INIT(pmap);
1604 	mmu_booke_pinit(mmu, pmap);
1605 	PCPU_SET(curpmap, pmap);
1606 }
1607 
1608 /*
1609  * Initialize a preallocated and zeroed pmap structure,
1610  * such as one in a vmspace structure.
1611  */
1612 static void
1613 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1614 {
1615 	int i;
1616 
1617 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1618 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1619 
1620 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1621 
1622 	for (i = 0; i < MAXCPU; i++)
1623 		pmap->pm_tid[i] = TID_NONE;
1624 	CPU_ZERO(&kernel_pmap->pm_active);
1625 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1626 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1627 	TAILQ_INIT(&pmap->pm_ptbl_list);
1628 }
1629 
1630 /*
1631  * Release any resources held by the given physical map.
1632  * Called when a pmap initialized by mmu_booke_pinit is being released.
1633  * Should only be called if the map contains no valid mappings.
1634  */
1635 static void
1636 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1637 {
1638 
1639 	KASSERT(pmap->pm_stats.resident_count == 0,
1640 	    ("pmap_release: pmap resident count %ld != 0",
1641 	    pmap->pm_stats.resident_count));
1642 }
1643 
1644 /*
1645  * Insert the given physical page at the specified virtual address in the
1646  * target physical map with the protection requested. If specified the page
1647  * will be wired down.
1648  */
1649 static int
1650 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1651     vm_prot_t prot, u_int flags, int8_t psind)
1652 {
1653 	int error;
1654 
1655 	rw_wlock(&pvh_global_lock);
1656 	PMAP_LOCK(pmap);
1657 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1658 	rw_wunlock(&pvh_global_lock);
1659 	PMAP_UNLOCK(pmap);
1660 	return (error);
1661 }
1662 
1663 static int
1664 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1665     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1666 {
1667 	pte_t *pte;
1668 	vm_paddr_t pa;
1669 	uint32_t flags;
1670 	int error, su, sync;
1671 
1672 	pa = VM_PAGE_TO_PHYS(m);
1673 	su = (pmap == kernel_pmap);
1674 	sync = 0;
1675 
1676 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1677 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1678 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1679 	//		(u_int32_t)m, va, pa, prot, flags);
1680 
1681 	if (su) {
1682 		KASSERT(((va >= virtual_avail) &&
1683 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1684 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1685 	} else {
1686 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1687 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1688 	}
1689 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1690 		VM_OBJECT_ASSERT_LOCKED(m->object);
1691 
1692 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1693 
1694 	/*
1695 	 * If there is an existing mapping, and the physical address has not
1696 	 * changed, must be protection or wiring change.
1697 	 */
1698 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1699 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1700 
1701 		/*
1702 		 * Before actually updating pte->flags we calculate and
1703 		 * prepare its new value in a helper var.
1704 		 */
1705 		flags = *pte;
1706 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1707 
1708 		/* Wiring change, just update stats. */
1709 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1710 			if (!PTE_ISWIRED(pte)) {
1711 				flags |= PTE_WIRED;
1712 				pmap->pm_stats.wired_count++;
1713 			}
1714 		} else {
1715 			if (PTE_ISWIRED(pte)) {
1716 				flags &= ~PTE_WIRED;
1717 				pmap->pm_stats.wired_count--;
1718 			}
1719 		}
1720 
1721 		if (prot & VM_PROT_WRITE) {
1722 			/* Add write permissions. */
1723 			flags |= PTE_SW;
1724 			if (!su)
1725 				flags |= PTE_UW;
1726 
1727 			if ((flags & PTE_MANAGED) != 0)
1728 				vm_page_aflag_set(m, PGA_WRITEABLE);
1729 		} else {
1730 			/* Handle modified pages, sense modify status. */
1731 
1732 			/*
1733 			 * The PTE_MODIFIED flag could be set by underlying
1734 			 * TLB misses since we last read it (above), possibly
1735 			 * other CPUs could update it so we check in the PTE
1736 			 * directly rather than rely on that saved local flags
1737 			 * copy.
1738 			 */
1739 			if (PTE_ISMODIFIED(pte))
1740 				vm_page_dirty(m);
1741 		}
1742 
1743 		if (prot & VM_PROT_EXECUTE) {
1744 			flags |= PTE_SX;
1745 			if (!su)
1746 				flags |= PTE_UX;
1747 
1748 			/*
1749 			 * Check existing flags for execute permissions: if we
1750 			 * are turning execute permissions on, icache should
1751 			 * be flushed.
1752 			 */
1753 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1754 				sync++;
1755 		}
1756 
1757 		flags &= ~PTE_REFERENCED;
1758 
1759 		/*
1760 		 * The new flags value is all calculated -- only now actually
1761 		 * update the PTE.
1762 		 */
1763 		mtx_lock_spin(&tlbivax_mutex);
1764 		tlb_miss_lock();
1765 
1766 		tlb0_flush_entry(va);
1767 		*pte &= ~PTE_FLAGS_MASK;
1768 		*pte |= flags;
1769 
1770 		tlb_miss_unlock();
1771 		mtx_unlock_spin(&tlbivax_mutex);
1772 
1773 	} else {
1774 		/*
1775 		 * If there is an existing mapping, but it's for a different
1776 		 * physical address, pte_enter() will delete the old mapping.
1777 		 */
1778 		//if ((pte != NULL) && PTE_ISVALID(pte))
1779 		//	debugf("mmu_booke_enter_locked: replace\n");
1780 		//else
1781 		//	debugf("mmu_booke_enter_locked: new\n");
1782 
1783 		/* Now set up the flags and install the new mapping. */
1784 		flags = (PTE_SR | PTE_VALID);
1785 		flags |= PTE_M;
1786 
1787 		if (!su)
1788 			flags |= PTE_UR;
1789 
1790 		if (prot & VM_PROT_WRITE) {
1791 			flags |= PTE_SW;
1792 			if (!su)
1793 				flags |= PTE_UW;
1794 
1795 			if ((m->oflags & VPO_UNMANAGED) == 0)
1796 				vm_page_aflag_set(m, PGA_WRITEABLE);
1797 		}
1798 
1799 		if (prot & VM_PROT_EXECUTE) {
1800 			flags |= PTE_SX;
1801 			if (!su)
1802 				flags |= PTE_UX;
1803 		}
1804 
1805 		/* If its wired update stats. */
1806 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1807 			flags |= PTE_WIRED;
1808 
1809 		error = pte_enter(mmu, pmap, m, va, flags,
1810 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1811 		if (error != 0)
1812 			return (KERN_RESOURCE_SHORTAGE);
1813 
1814 		if ((flags & PMAP_ENTER_WIRED) != 0)
1815 			pmap->pm_stats.wired_count++;
1816 
1817 		/* Flush the real memory from the instruction cache. */
1818 		if (prot & VM_PROT_EXECUTE)
1819 			sync++;
1820 	}
1821 
1822 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1823 		__syncicache((void *)va, PAGE_SIZE);
1824 		sync = 0;
1825 	}
1826 
1827 	return (KERN_SUCCESS);
1828 }
1829 
1830 /*
1831  * Maps a sequence of resident pages belonging to the same object.
1832  * The sequence begins with the given page m_start.  This page is
1833  * mapped at the given virtual address start.  Each subsequent page is
1834  * mapped at a virtual address that is offset from start by the same
1835  * amount as the page is offset from m_start within the object.  The
1836  * last page in the sequence is the page with the largest offset from
1837  * m_start that can be mapped at a virtual address less than the given
1838  * virtual address end.  Not every virtual page between start and end
1839  * is mapped; only those for which a resident page exists with the
1840  * corresponding offset from m_start are mapped.
1841  */
1842 static void
1843 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1844     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1845 {
1846 	vm_page_t m;
1847 	vm_pindex_t diff, psize;
1848 
1849 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1850 
1851 	psize = atop(end - start);
1852 	m = m_start;
1853 	rw_wlock(&pvh_global_lock);
1854 	PMAP_LOCK(pmap);
1855 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1856 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1857 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1858 		    PMAP_ENTER_NOSLEEP, 0);
1859 		m = TAILQ_NEXT(m, listq);
1860 	}
1861 	rw_wunlock(&pvh_global_lock);
1862 	PMAP_UNLOCK(pmap);
1863 }
1864 
1865 static void
1866 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1867     vm_prot_t prot)
1868 {
1869 
1870 	rw_wlock(&pvh_global_lock);
1871 	PMAP_LOCK(pmap);
1872 	mmu_booke_enter_locked(mmu, pmap, va, m,
1873 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1874 	    0);
1875 	rw_wunlock(&pvh_global_lock);
1876 	PMAP_UNLOCK(pmap);
1877 }
1878 
1879 /*
1880  * Remove the given range of addresses from the specified map.
1881  *
1882  * It is assumed that the start and end are properly rounded to the page size.
1883  */
1884 static void
1885 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1886 {
1887 	pte_t *pte;
1888 	uint8_t hold_flag;
1889 
1890 	int su = (pmap == kernel_pmap);
1891 
1892 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1893 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1894 
1895 	if (su) {
1896 		KASSERT(((va >= virtual_avail) &&
1897 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1898 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1899 	} else {
1900 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1901 		    ("mmu_booke_remove: user pmap, non user va"));
1902 	}
1903 
1904 	if (PMAP_REMOVE_DONE(pmap)) {
1905 		//debugf("mmu_booke_remove: e (empty)\n");
1906 		return;
1907 	}
1908 
1909 	hold_flag = PTBL_HOLD_FLAG(pmap);
1910 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1911 
1912 	rw_wlock(&pvh_global_lock);
1913 	PMAP_LOCK(pmap);
1914 	for (; va < endva; va += PAGE_SIZE) {
1915 		pte = pte_find(mmu, pmap, va);
1916 		if ((pte != NULL) && PTE_ISVALID(pte))
1917 			pte_remove(mmu, pmap, va, hold_flag);
1918 	}
1919 	PMAP_UNLOCK(pmap);
1920 	rw_wunlock(&pvh_global_lock);
1921 
1922 	//debugf("mmu_booke_remove: e\n");
1923 }
1924 
1925 /*
1926  * Remove physical page from all pmaps in which it resides.
1927  */
1928 static void
1929 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1930 {
1931 	pv_entry_t pv, pvn;
1932 	uint8_t hold_flag;
1933 
1934 	rw_wlock(&pvh_global_lock);
1935 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1936 		pvn = TAILQ_NEXT(pv, pv_link);
1937 
1938 		PMAP_LOCK(pv->pv_pmap);
1939 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1940 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1941 		PMAP_UNLOCK(pv->pv_pmap);
1942 	}
1943 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1944 	rw_wunlock(&pvh_global_lock);
1945 }
1946 
1947 /*
1948  * Map a range of physical addresses into kernel virtual address space.
1949  */
1950 static vm_offset_t
1951 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1952     vm_paddr_t pa_end, int prot)
1953 {
1954 	vm_offset_t sva = *virt;
1955 	vm_offset_t va = sva;
1956 
1957 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1958 	//		sva, pa_start, pa_end);
1959 
1960 	while (pa_start < pa_end) {
1961 		mmu_booke_kenter(mmu, va, pa_start);
1962 		va += PAGE_SIZE;
1963 		pa_start += PAGE_SIZE;
1964 	}
1965 	*virt = va;
1966 
1967 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1968 	return (sva);
1969 }
1970 
1971 /*
1972  * The pmap must be activated before it's address space can be accessed in any
1973  * way.
1974  */
1975 static void
1976 mmu_booke_activate(mmu_t mmu, struct thread *td)
1977 {
1978 	pmap_t pmap;
1979 	u_int cpuid;
1980 
1981 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1982 
1983 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1984 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1985 
1986 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1987 
1988 	sched_pin();
1989 
1990 	cpuid = PCPU_GET(cpuid);
1991 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1992 	PCPU_SET(curpmap, pmap);
1993 
1994 	if (pmap->pm_tid[cpuid] == TID_NONE)
1995 		tid_alloc(pmap);
1996 
1997 	/* Load PID0 register with pmap tid value. */
1998 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1999 	__asm __volatile("isync");
2000 
2001 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2002 
2003 	sched_unpin();
2004 
2005 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2006 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2007 }
2008 
2009 /*
2010  * Deactivate the specified process's address space.
2011  */
2012 static void
2013 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2014 {
2015 	pmap_t pmap;
2016 
2017 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2018 
2019 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2020 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2021 
2022 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2023 
2024 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2025 	PCPU_SET(curpmap, NULL);
2026 }
2027 
2028 /*
2029  * Copy the range specified by src_addr/len
2030  * from the source map to the range dst_addr/len
2031  * in the destination map.
2032  *
2033  * This routine is only advisory and need not do anything.
2034  */
2035 static void
2036 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2037     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2038 {
2039 
2040 }
2041 
2042 /*
2043  * Set the physical protection on the specified range of this map as requested.
2044  */
2045 static void
2046 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2047     vm_prot_t prot)
2048 {
2049 	vm_offset_t va;
2050 	vm_page_t m;
2051 	pte_t *pte;
2052 
2053 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2054 		mmu_booke_remove(mmu, pmap, sva, eva);
2055 		return;
2056 	}
2057 
2058 	if (prot & VM_PROT_WRITE)
2059 		return;
2060 
2061 	PMAP_LOCK(pmap);
2062 	for (va = sva; va < eva; va += PAGE_SIZE) {
2063 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2064 			if (PTE_ISVALID(pte)) {
2065 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2066 
2067 				mtx_lock_spin(&tlbivax_mutex);
2068 				tlb_miss_lock();
2069 
2070 				/* Handle modified pages. */
2071 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2072 					vm_page_dirty(m);
2073 
2074 				tlb0_flush_entry(va);
2075 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2076 
2077 				tlb_miss_unlock();
2078 				mtx_unlock_spin(&tlbivax_mutex);
2079 			}
2080 		}
2081 	}
2082 	PMAP_UNLOCK(pmap);
2083 }
2084 
2085 /*
2086  * Clear the write and modified bits in each of the given page's mappings.
2087  */
2088 static void
2089 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2090 {
2091 	pv_entry_t pv;
2092 	pte_t *pte;
2093 
2094 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2095 	    ("mmu_booke_remove_write: page %p is not managed", m));
2096 
2097 	/*
2098 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2099 	 * set by another thread while the object is locked.  Thus,
2100 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2101 	 */
2102 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2103 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2104 		return;
2105 	rw_wlock(&pvh_global_lock);
2106 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2107 		PMAP_LOCK(pv->pv_pmap);
2108 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2109 			if (PTE_ISVALID(pte)) {
2110 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2111 
2112 				mtx_lock_spin(&tlbivax_mutex);
2113 				tlb_miss_lock();
2114 
2115 				/* Handle modified pages. */
2116 				if (PTE_ISMODIFIED(pte))
2117 					vm_page_dirty(m);
2118 
2119 				/* Flush mapping from TLB0. */
2120 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2121 
2122 				tlb_miss_unlock();
2123 				mtx_unlock_spin(&tlbivax_mutex);
2124 			}
2125 		}
2126 		PMAP_UNLOCK(pv->pv_pmap);
2127 	}
2128 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2129 	rw_wunlock(&pvh_global_lock);
2130 }
2131 
2132 static void
2133 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2134 {
2135 	pte_t *pte;
2136 	pmap_t pmap;
2137 	vm_page_t m;
2138 	vm_offset_t addr;
2139 	vm_paddr_t pa = 0;
2140 	int active, valid;
2141 
2142 	va = trunc_page(va);
2143 	sz = round_page(sz);
2144 
2145 	rw_wlock(&pvh_global_lock);
2146 	pmap = PCPU_GET(curpmap);
2147 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2148 	while (sz > 0) {
2149 		PMAP_LOCK(pm);
2150 		pte = pte_find(mmu, pm, va);
2151 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2152 		if (valid)
2153 			pa = PTE_PA(pte);
2154 		PMAP_UNLOCK(pm);
2155 		if (valid) {
2156 			if (!active) {
2157 				/* Create a mapping in the active pmap. */
2158 				addr = 0;
2159 				m = PHYS_TO_VM_PAGE(pa);
2160 				PMAP_LOCK(pmap);
2161 				pte_enter(mmu, pmap, m, addr,
2162 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2163 				__syncicache((void *)addr, PAGE_SIZE);
2164 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2165 				PMAP_UNLOCK(pmap);
2166 			} else
2167 				__syncicache((void *)va, PAGE_SIZE);
2168 		}
2169 		va += PAGE_SIZE;
2170 		sz -= PAGE_SIZE;
2171 	}
2172 	rw_wunlock(&pvh_global_lock);
2173 }
2174 
2175 /*
2176  * Atomically extract and hold the physical page with the given
2177  * pmap and virtual address pair if that mapping permits the given
2178  * protection.
2179  */
2180 static vm_page_t
2181 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2182     vm_prot_t prot)
2183 {
2184 	pte_t *pte;
2185 	vm_page_t m;
2186 	uint32_t pte_wbit;
2187 	vm_paddr_t pa;
2188 
2189 	m = NULL;
2190 	pa = 0;
2191 	PMAP_LOCK(pmap);
2192 retry:
2193 	pte = pte_find(mmu, pmap, va);
2194 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2195 		if (pmap == kernel_pmap)
2196 			pte_wbit = PTE_SW;
2197 		else
2198 			pte_wbit = PTE_UW;
2199 
2200 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2201 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2202 				goto retry;
2203 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2204 			vm_page_hold(m);
2205 		}
2206 	}
2207 
2208 	PA_UNLOCK_COND(pa);
2209 	PMAP_UNLOCK(pmap);
2210 	return (m);
2211 }
2212 
2213 /*
2214  * Initialize a vm_page's machine-dependent fields.
2215  */
2216 static void
2217 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2218 {
2219 
2220 	TAILQ_INIT(&m->md.pv_list);
2221 }
2222 
2223 /*
2224  * mmu_booke_zero_page_area zeros the specified hardware page by
2225  * mapping it into virtual memory and using bzero to clear
2226  * its contents.
2227  *
2228  * off and size must reside within a single page.
2229  */
2230 static void
2231 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2232 {
2233 	vm_offset_t va;
2234 
2235 	/* XXX KASSERT off and size are within a single page? */
2236 
2237 	mtx_lock(&zero_page_mutex);
2238 	va = zero_page_va;
2239 
2240 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2241 	bzero((caddr_t)va + off, size);
2242 	mmu_booke_kremove(mmu, va);
2243 
2244 	mtx_unlock(&zero_page_mutex);
2245 }
2246 
2247 /*
2248  * mmu_booke_zero_page zeros the specified hardware page.
2249  */
2250 static void
2251 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2252 {
2253 	vm_offset_t off, va;
2254 
2255 	mtx_lock(&zero_page_mutex);
2256 	va = zero_page_va;
2257 
2258 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2259 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2260 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2261 	mmu_booke_kremove(mmu, va);
2262 
2263 	mtx_unlock(&zero_page_mutex);
2264 }
2265 
2266 /*
2267  * mmu_booke_copy_page copies the specified (machine independent) page by
2268  * mapping the page into virtual memory and using memcopy to copy the page,
2269  * one machine dependent page at a time.
2270  */
2271 static void
2272 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2273 {
2274 	vm_offset_t sva, dva;
2275 
2276 	sva = copy_page_src_va;
2277 	dva = copy_page_dst_va;
2278 
2279 	mtx_lock(&copy_page_mutex);
2280 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2281 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2282 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2283 	mmu_booke_kremove(mmu, dva);
2284 	mmu_booke_kremove(mmu, sva);
2285 	mtx_unlock(&copy_page_mutex);
2286 }
2287 
2288 static inline void
2289 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2290     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2291 {
2292 	void *a_cp, *b_cp;
2293 	vm_offset_t a_pg_offset, b_pg_offset;
2294 	int cnt;
2295 
2296 	mtx_lock(&copy_page_mutex);
2297 	while (xfersize > 0) {
2298 		a_pg_offset = a_offset & PAGE_MASK;
2299 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2300 		mmu_booke_kenter(mmu, copy_page_src_va,
2301 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2302 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2303 		b_pg_offset = b_offset & PAGE_MASK;
2304 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2305 		mmu_booke_kenter(mmu, copy_page_dst_va,
2306 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2307 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2308 		bcopy(a_cp, b_cp, cnt);
2309 		mmu_booke_kremove(mmu, copy_page_dst_va);
2310 		mmu_booke_kremove(mmu, copy_page_src_va);
2311 		a_offset += cnt;
2312 		b_offset += cnt;
2313 		xfersize -= cnt;
2314 	}
2315 	mtx_unlock(&copy_page_mutex);
2316 }
2317 
2318 static vm_offset_t
2319 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2320 {
2321 	vm_paddr_t paddr;
2322 	vm_offset_t qaddr;
2323 	uint32_t flags;
2324 	pte_t *pte;
2325 
2326 	paddr = VM_PAGE_TO_PHYS(m);
2327 
2328 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2329 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2330 	flags |= PTE_PS_4KB;
2331 
2332 	critical_enter();
2333 	qaddr = PCPU_GET(qmap_addr);
2334 
2335 	pte = pte_find(mmu, kernel_pmap, qaddr);
2336 
2337 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2338 
2339 	/*
2340 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2341  	 * not be present in other TLBs.  Is there a better instruction
2342 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2343 	 */
2344 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2345 	__asm __volatile("isync; msync");
2346 
2347 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
2348 
2349 	/* Flush the real memory from the instruction cache. */
2350 	if ((flags & (PTE_I | PTE_G)) == 0)
2351 		__syncicache((void *)qaddr, PAGE_SIZE);
2352 
2353 	return (qaddr);
2354 }
2355 
2356 static void
2357 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2358 {
2359 	pte_t *pte;
2360 
2361 	pte = pte_find(mmu, kernel_pmap, addr);
2362 
2363 	KASSERT(PCPU_GET(qmap_addr) == addr,
2364 	    ("mmu_booke_quick_remove_page: invalid address"));
2365 	KASSERT(*pte != 0,
2366 	    ("mmu_booke_quick_remove_page: PTE not in use"));
2367 
2368 	*pte = 0;
2369 	critical_exit();
2370 }
2371 
2372 /*
2373  * Return whether or not the specified physical page was modified
2374  * in any of physical maps.
2375  */
2376 static boolean_t
2377 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2378 {
2379 	pte_t *pte;
2380 	pv_entry_t pv;
2381 	boolean_t rv;
2382 
2383 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2384 	    ("mmu_booke_is_modified: page %p is not managed", m));
2385 	rv = FALSE;
2386 
2387 	/*
2388 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2389 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2390 	 * is clear, no PTEs can be modified.
2391 	 */
2392 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2393 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2394 		return (rv);
2395 	rw_wlock(&pvh_global_lock);
2396 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2397 		PMAP_LOCK(pv->pv_pmap);
2398 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2399 		    PTE_ISVALID(pte)) {
2400 			if (PTE_ISMODIFIED(pte))
2401 				rv = TRUE;
2402 		}
2403 		PMAP_UNLOCK(pv->pv_pmap);
2404 		if (rv)
2405 			break;
2406 	}
2407 	rw_wunlock(&pvh_global_lock);
2408 	return (rv);
2409 }
2410 
2411 /*
2412  * Return whether or not the specified virtual address is eligible
2413  * for prefault.
2414  */
2415 static boolean_t
2416 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2417 {
2418 
2419 	return (FALSE);
2420 }
2421 
2422 /*
2423  * Return whether or not the specified physical page was referenced
2424  * in any physical maps.
2425  */
2426 static boolean_t
2427 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2428 {
2429 	pte_t *pte;
2430 	pv_entry_t pv;
2431 	boolean_t rv;
2432 
2433 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2434 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2435 	rv = FALSE;
2436 	rw_wlock(&pvh_global_lock);
2437 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2438 		PMAP_LOCK(pv->pv_pmap);
2439 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2440 		    PTE_ISVALID(pte)) {
2441 			if (PTE_ISREFERENCED(pte))
2442 				rv = TRUE;
2443 		}
2444 		PMAP_UNLOCK(pv->pv_pmap);
2445 		if (rv)
2446 			break;
2447 	}
2448 	rw_wunlock(&pvh_global_lock);
2449 	return (rv);
2450 }
2451 
2452 /*
2453  * Clear the modify bits on the specified physical page.
2454  */
2455 static void
2456 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2457 {
2458 	pte_t *pte;
2459 	pv_entry_t pv;
2460 
2461 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2462 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2463 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2464 	KASSERT(!vm_page_xbusied(m),
2465 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2466 
2467 	/*
2468 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2469 	 * If the object containing the page is locked and the page is not
2470 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2471 	 */
2472 	if ((m->aflags & PGA_WRITEABLE) == 0)
2473 		return;
2474 	rw_wlock(&pvh_global_lock);
2475 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2476 		PMAP_LOCK(pv->pv_pmap);
2477 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2478 		    PTE_ISVALID(pte)) {
2479 			mtx_lock_spin(&tlbivax_mutex);
2480 			tlb_miss_lock();
2481 
2482 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2483 				tlb0_flush_entry(pv->pv_va);
2484 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2485 				    PTE_REFERENCED);
2486 			}
2487 
2488 			tlb_miss_unlock();
2489 			mtx_unlock_spin(&tlbivax_mutex);
2490 		}
2491 		PMAP_UNLOCK(pv->pv_pmap);
2492 	}
2493 	rw_wunlock(&pvh_global_lock);
2494 }
2495 
2496 /*
2497  * Return a count of reference bits for a page, clearing those bits.
2498  * It is not necessary for every reference bit to be cleared, but it
2499  * is necessary that 0 only be returned when there are truly no
2500  * reference bits set.
2501  *
2502  * As an optimization, update the page's dirty field if a modified bit is
2503  * found while counting reference bits.  This opportunistic update can be
2504  * performed at low cost and can eliminate the need for some future calls
2505  * to pmap_is_modified().  However, since this function stops after
2506  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
2507  * dirty pages.  Those dirty pages will only be detected by a future call
2508  * to pmap_is_modified().
2509  */
2510 static int
2511 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2512 {
2513 	pte_t *pte;
2514 	pv_entry_t pv;
2515 	int count;
2516 
2517 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2518 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2519 	count = 0;
2520 	rw_wlock(&pvh_global_lock);
2521 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2522 		PMAP_LOCK(pv->pv_pmap);
2523 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2524 		    PTE_ISVALID(pte)) {
2525 			if (PTE_ISMODIFIED(pte))
2526 				vm_page_dirty(m);
2527 			if (PTE_ISREFERENCED(pte)) {
2528 				mtx_lock_spin(&tlbivax_mutex);
2529 				tlb_miss_lock();
2530 
2531 				tlb0_flush_entry(pv->pv_va);
2532 				*pte &= ~PTE_REFERENCED;
2533 
2534 				tlb_miss_unlock();
2535 				mtx_unlock_spin(&tlbivax_mutex);
2536 
2537 				if (++count >= PMAP_TS_REFERENCED_MAX) {
2538 					PMAP_UNLOCK(pv->pv_pmap);
2539 					break;
2540 				}
2541 			}
2542 		}
2543 		PMAP_UNLOCK(pv->pv_pmap);
2544 	}
2545 	rw_wunlock(&pvh_global_lock);
2546 	return (count);
2547 }
2548 
2549 /*
2550  * Clear the wired attribute from the mappings for the specified range of
2551  * addresses in the given pmap.  Every valid mapping within that range must
2552  * have the wired attribute set.  In contrast, invalid mappings cannot have
2553  * the wired attribute set, so they are ignored.
2554  *
2555  * The wired attribute of the page table entry is not a hardware feature, so
2556  * there is no need to invalidate any TLB entries.
2557  */
2558 static void
2559 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2560 {
2561 	vm_offset_t va;
2562 	pte_t *pte;
2563 
2564 	PMAP_LOCK(pmap);
2565 	for (va = sva; va < eva; va += PAGE_SIZE) {
2566 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2567 		    PTE_ISVALID(pte)) {
2568 			if (!PTE_ISWIRED(pte))
2569 				panic("mmu_booke_unwire: pte %p isn't wired",
2570 				    pte);
2571 			*pte &= ~PTE_WIRED;
2572 			pmap->pm_stats.wired_count--;
2573 		}
2574 	}
2575 	PMAP_UNLOCK(pmap);
2576 
2577 }
2578 
2579 /*
2580  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2581  * page.  This count may be changed upwards or downwards in the future; it is
2582  * only necessary that true be returned for a small subset of pmaps for proper
2583  * page aging.
2584  */
2585 static boolean_t
2586 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2587 {
2588 	pv_entry_t pv;
2589 	int loops;
2590 	boolean_t rv;
2591 
2592 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2593 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2594 	loops = 0;
2595 	rv = FALSE;
2596 	rw_wlock(&pvh_global_lock);
2597 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2598 		if (pv->pv_pmap == pmap) {
2599 			rv = TRUE;
2600 			break;
2601 		}
2602 		if (++loops >= 16)
2603 			break;
2604 	}
2605 	rw_wunlock(&pvh_global_lock);
2606 	return (rv);
2607 }
2608 
2609 /*
2610  * Return the number of managed mappings to the given physical page that are
2611  * wired.
2612  */
2613 static int
2614 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2615 {
2616 	pv_entry_t pv;
2617 	pte_t *pte;
2618 	int count = 0;
2619 
2620 	if ((m->oflags & VPO_UNMANAGED) != 0)
2621 		return (count);
2622 	rw_wlock(&pvh_global_lock);
2623 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2624 		PMAP_LOCK(pv->pv_pmap);
2625 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2626 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2627 				count++;
2628 		PMAP_UNLOCK(pv->pv_pmap);
2629 	}
2630 	rw_wunlock(&pvh_global_lock);
2631 	return (count);
2632 }
2633 
2634 static int
2635 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2636 {
2637 	int i;
2638 	vm_offset_t va;
2639 
2640 	/*
2641 	 * This currently does not work for entries that
2642 	 * overlap TLB1 entries.
2643 	 */
2644 	for (i = 0; i < TLB1_ENTRIES; i ++) {
2645 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2646 			return (0);
2647 	}
2648 
2649 	return (EFAULT);
2650 }
2651 
2652 void
2653 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2654 {
2655 	vm_paddr_t ppa;
2656 	vm_offset_t ofs;
2657 	vm_size_t gran;
2658 
2659 	/* Minidumps are based on virtual memory addresses. */
2660 	if (do_minidump) {
2661 		*va = (void *)(vm_offset_t)pa;
2662 		return;
2663 	}
2664 
2665 	/* Raw physical memory dumps don't have a virtual address. */
2666 	/* We always map a 256MB page at 256M. */
2667 	gran = 256 * 1024 * 1024;
2668 	ppa = rounddown2(pa, gran);
2669 	ofs = pa - ppa;
2670 	*va = (void *)gran;
2671 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2672 
2673 	if (sz > (gran - ofs))
2674 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2675 		    _TLB_ENTRY_IO);
2676 }
2677 
2678 void
2679 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2680 {
2681 	vm_paddr_t ppa;
2682 	vm_offset_t ofs;
2683 	vm_size_t gran;
2684 	tlb_entry_t e;
2685 	int i;
2686 
2687 	/* Minidumps are based on virtual memory addresses. */
2688 	/* Nothing to do... */
2689 	if (do_minidump)
2690 		return;
2691 
2692 	for (i = 0; i < TLB1_ENTRIES; i++) {
2693 		tlb1_read_entry(&e, i);
2694 		if (!(e.mas1 & MAS1_VALID))
2695 			break;
2696 	}
2697 
2698 	/* Raw physical memory dumps don't have a virtual address. */
2699 	i--;
2700 	e.mas1 = 0;
2701 	e.mas2 = 0;
2702 	e.mas3 = 0;
2703 	tlb1_write_entry(&e, i);
2704 
2705 	gran = 256 * 1024 * 1024;
2706 	ppa = rounddown2(pa, gran);
2707 	ofs = pa - ppa;
2708 	if (sz > (gran - ofs)) {
2709 		i--;
2710 		e.mas1 = 0;
2711 		e.mas2 = 0;
2712 		e.mas3 = 0;
2713 		tlb1_write_entry(&e, i);
2714 	}
2715 }
2716 
2717 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2718 
2719 void
2720 mmu_booke_scan_init(mmu_t mmu)
2721 {
2722 	vm_offset_t va;
2723 	pte_t *pte;
2724 	int i;
2725 
2726 	if (!do_minidump) {
2727 		/* Initialize phys. segments for dumpsys(). */
2728 		memset(&dump_map, 0, sizeof(dump_map));
2729 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2730 		    &availmem_regions_sz);
2731 		for (i = 0; i < physmem_regions_sz; i++) {
2732 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2733 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2734 		}
2735 		return;
2736 	}
2737 
2738 	/* Virtual segments for minidumps: */
2739 	memset(&dump_map, 0, sizeof(dump_map));
2740 
2741 	/* 1st: kernel .data and .bss. */
2742 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2743 	dump_map[0].pa_size =
2744 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2745 
2746 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2747 	dump_map[1].pa_start = data_start;
2748 	dump_map[1].pa_size = data_end - data_start;
2749 
2750 	/* 3rd: kernel VM. */
2751 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2752 	/* Find start of next chunk (from va). */
2753 	while (va < virtual_end) {
2754 		/* Don't dump the buffer cache. */
2755 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2756 			va = kmi.buffer_eva;
2757 			continue;
2758 		}
2759 		pte = pte_find(mmu, kernel_pmap, va);
2760 		if (pte != NULL && PTE_ISVALID(pte))
2761 			break;
2762 		va += PAGE_SIZE;
2763 	}
2764 	if (va < virtual_end) {
2765 		dump_map[2].pa_start = va;
2766 		va += PAGE_SIZE;
2767 		/* Find last page in chunk. */
2768 		while (va < virtual_end) {
2769 			/* Don't run into the buffer cache. */
2770 			if (va == kmi.buffer_sva)
2771 				break;
2772 			pte = pte_find(mmu, kernel_pmap, va);
2773 			if (pte == NULL || !PTE_ISVALID(pte))
2774 				break;
2775 			va += PAGE_SIZE;
2776 		}
2777 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2778 	}
2779 }
2780 
2781 /*
2782  * Map a set of physical memory pages into the kernel virtual address space.
2783  * Return a pointer to where it is mapped. This routine is intended to be used
2784  * for mapping device memory, NOT real memory.
2785  */
2786 static void *
2787 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2788 {
2789 
2790 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2791 }
2792 
2793 static void *
2794 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2795 {
2796 	tlb_entry_t e;
2797 	void *res;
2798 	uintptr_t va, tmpva;
2799 	vm_size_t sz;
2800 	int i;
2801 
2802 	/*
2803 	 * Check if this is premapped in TLB1. Note: this should probably also
2804 	 * check whether a sequence of TLB1 entries exist that match the
2805 	 * requirement, but now only checks the easy case.
2806 	 */
2807 	if (ma == VM_MEMATTR_DEFAULT) {
2808 		for (i = 0; i < TLB1_ENTRIES; i++) {
2809 			tlb1_read_entry(&e, i);
2810 			if (!(e.mas1 & MAS1_VALID))
2811 				continue;
2812 			if (pa >= e.phys &&
2813 			    (pa + size) <= (e.phys + e.size))
2814 				return (void *)(e.virt +
2815 				    (vm_offset_t)(pa - e.phys));
2816 		}
2817 	}
2818 
2819 	size = roundup(size, PAGE_SIZE);
2820 
2821 	/*
2822 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
2823 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
2824 	 */
2825 #ifdef SPARSE_MAPDEV
2826 	/*
2827 	 * With a sparse mapdev, align to the largest starting region.  This
2828 	 * could feasibly be optimized for a 'best-fit' alignment, but that
2829 	 * calculation could be very costly.
2830 	 */
2831 	do {
2832 	    tmpva = tlb1_map_base;
2833 	    va = roundup(tlb1_map_base, 1 << flsl(size));
2834 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
2835 #else
2836 	va = atomic_fetchadd_int(&tlb1_map_base, size);
2837 #endif
2838 	res = (void *)va;
2839 
2840 	do {
2841 		sz = 1 << (ilog2(size) & ~1);
2842 		if (va % sz != 0) {
2843 			do {
2844 				sz >>= 2;
2845 			} while (va % sz != 0);
2846 		}
2847 		if (bootverbose)
2848 			printf("Wiring VA=%x to PA=%jx (size=%x)\n",
2849 			    va, (uintmax_t)pa, sz);
2850 		tlb1_set_entry(va, pa, sz,
2851 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma));
2852 		size -= sz;
2853 		pa += sz;
2854 		va += sz;
2855 	} while (size > 0);
2856 
2857 	return (res);
2858 }
2859 
2860 /*
2861  * 'Unmap' a range mapped by mmu_booke_mapdev().
2862  */
2863 static void
2864 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2865 {
2866 #ifdef SUPPORTS_SHRINKING_TLB1
2867 	vm_offset_t base, offset;
2868 
2869 	/*
2870 	 * Unmap only if this is inside kernel virtual space.
2871 	 */
2872 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2873 		base = trunc_page(va);
2874 		offset = va & PAGE_MASK;
2875 		size = roundup(offset + size, PAGE_SIZE);
2876 		kva_free(base, size);
2877 	}
2878 #endif
2879 }
2880 
2881 /*
2882  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2883  * specified pmap. This eliminates the blast of soft faults on process startup
2884  * and immediately after an mmap.
2885  */
2886 static void
2887 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2888     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2889 {
2890 
2891 	VM_OBJECT_ASSERT_WLOCKED(object);
2892 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2893 	    ("mmu_booke_object_init_pt: non-device object"));
2894 }
2895 
2896 /*
2897  * Perform the pmap work for mincore.
2898  */
2899 static int
2900 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2901     vm_paddr_t *locked_pa)
2902 {
2903 
2904 	/* XXX: this should be implemented at some point */
2905 	return (0);
2906 }
2907 
2908 static int
2909 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
2910     vm_memattr_t mode)
2911 {
2912 	vm_offset_t va;
2913 	pte_t *pte;
2914 	int i, j;
2915 	tlb_entry_t e;
2916 
2917 	/* Check TLB1 mappings */
2918 	for (i = 0; i < TLB1_ENTRIES; i++) {
2919 		tlb1_read_entry(&e, i);
2920 		if (!(e.mas1 & MAS1_VALID))
2921 			continue;
2922 		if (addr >= e.virt && addr < e.virt + e.size)
2923 			break;
2924 	}
2925 	if (i < TLB1_ENTRIES) {
2926 		/* Only allow full mappings to be modified for now. */
2927 		/* Validate the range. */
2928 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
2929 			tlb1_read_entry(&e, j);
2930 			if (va != e.virt || (sz - (va - addr) < e.size))
2931 				return (EINVAL);
2932 		}
2933 		for (va = addr; va < addr + sz; va += e.size, i++) {
2934 			tlb1_read_entry(&e, i);
2935 			e.mas2 &= ~MAS2_WIMGE_MASK;
2936 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
2937 
2938 			/*
2939 			 * Write it out to the TLB.  Should really re-sync with other
2940 			 * cores.
2941 			 */
2942 			tlb1_write_entry(&e, i);
2943 		}
2944 		return (0);
2945 	}
2946 
2947 	/* Not in TLB1, try through pmap */
2948 	/* First validate the range. */
2949 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2950 		pte = pte_find(mmu, kernel_pmap, va);
2951 		if (pte == NULL || !PTE_ISVALID(pte))
2952 			return (EINVAL);
2953 	}
2954 
2955 	mtx_lock_spin(&tlbivax_mutex);
2956 	tlb_miss_lock();
2957 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2958 		pte = pte_find(mmu, kernel_pmap, va);
2959 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2960 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode << PTE_MAS2_SHIFT);
2961 		tlb0_flush_entry(va);
2962 	}
2963 	tlb_miss_unlock();
2964 	mtx_unlock_spin(&tlbivax_mutex);
2965 
2966 	return (pte_vatopa(mmu, kernel_pmap, va));
2967 }
2968 
2969 /**************************************************************************/
2970 /* TID handling */
2971 /**************************************************************************/
2972 
2973 /*
2974  * Allocate a TID. If necessary, steal one from someone else.
2975  * The new TID is flushed from the TLB before returning.
2976  */
2977 static tlbtid_t
2978 tid_alloc(pmap_t pmap)
2979 {
2980 	tlbtid_t tid;
2981 	int thiscpu;
2982 
2983 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2984 
2985 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2986 
2987 	thiscpu = PCPU_GET(cpuid);
2988 
2989 	tid = PCPU_GET(tid_next);
2990 	if (tid > TID_MAX)
2991 		tid = TID_MIN;
2992 	PCPU_SET(tid_next, tid + 1);
2993 
2994 	/* If we are stealing TID then clear the relevant pmap's field */
2995 	if (tidbusy[thiscpu][tid] != NULL) {
2996 
2997 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2998 
2999 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3000 
3001 		/* Flush all entries from TLB0 matching this TID. */
3002 		tid_flush(tid);
3003 	}
3004 
3005 	tidbusy[thiscpu][tid] = pmap;
3006 	pmap->pm_tid[thiscpu] = tid;
3007 	__asm __volatile("msync; isync");
3008 
3009 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3010 	    PCPU_GET(tid_next));
3011 
3012 	return (tid);
3013 }
3014 
3015 /**************************************************************************/
3016 /* TLB0 handling */
3017 /**************************************************************************/
3018 
3019 static void
3020 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3021     uint32_t mas7)
3022 {
3023 	int as;
3024 	char desc[3];
3025 	tlbtid_t tid;
3026 	vm_size_t size;
3027 	unsigned int tsize;
3028 
3029 	desc[2] = '\0';
3030 	if (mas1 & MAS1_VALID)
3031 		desc[0] = 'V';
3032 	else
3033 		desc[0] = ' ';
3034 
3035 	if (mas1 & MAS1_IPROT)
3036 		desc[1] = 'P';
3037 	else
3038 		desc[1] = ' ';
3039 
3040 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3041 	tid = MAS1_GETTID(mas1);
3042 
3043 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3044 	size = 0;
3045 	if (tsize)
3046 		size = tsize2size(tsize);
3047 
3048 	debugf("%3d: (%s) [AS=%d] "
3049 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3050 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3051 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3052 }
3053 
3054 /* Convert TLB0 va and way number to tlb0[] table index. */
3055 static inline unsigned int
3056 tlb0_tableidx(vm_offset_t va, unsigned int way)
3057 {
3058 	unsigned int idx;
3059 
3060 	idx = (way * TLB0_ENTRIES_PER_WAY);
3061 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3062 	return (idx);
3063 }
3064 
3065 /*
3066  * Invalidate TLB0 entry.
3067  */
3068 static inline void
3069 tlb0_flush_entry(vm_offset_t va)
3070 {
3071 
3072 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3073 
3074 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3075 
3076 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3077 	__asm __volatile("isync; msync");
3078 	__asm __volatile("tlbsync; msync");
3079 
3080 	CTR1(KTR_PMAP, "%s: e", __func__);
3081 }
3082 
3083 /* Print out contents of the MAS registers for each TLB0 entry */
3084 void
3085 tlb0_print_tlbentries(void)
3086 {
3087 	uint32_t mas0, mas1, mas2, mas3, mas7;
3088 	int entryidx, way, idx;
3089 
3090 	debugf("TLB0 entries:\n");
3091 	for (way = 0; way < TLB0_WAYS; way ++)
3092 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3093 
3094 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3095 			mtspr(SPR_MAS0, mas0);
3096 			__asm __volatile("isync");
3097 
3098 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3099 			mtspr(SPR_MAS2, mas2);
3100 
3101 			__asm __volatile("isync; tlbre");
3102 
3103 			mas1 = mfspr(SPR_MAS1);
3104 			mas2 = mfspr(SPR_MAS2);
3105 			mas3 = mfspr(SPR_MAS3);
3106 			mas7 = mfspr(SPR_MAS7);
3107 
3108 			idx = tlb0_tableidx(mas2, way);
3109 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3110 		}
3111 }
3112 
3113 /**************************************************************************/
3114 /* TLB1 handling */
3115 /**************************************************************************/
3116 
3117 /*
3118  * TLB1 mapping notes:
3119  *
3120  * TLB1[0]	Kernel text and data.
3121  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3122  *		windows, other devices mappings.
3123  */
3124 
3125  /*
3126  * Read an entry from given TLB1 slot.
3127  */
3128 void
3129 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3130 {
3131 	uint32_t mas0;
3132 
3133 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3134 
3135 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3136 	mtspr(SPR_MAS0, mas0);
3137 	__asm __volatile("isync; tlbre");
3138 
3139 	entry->mas1 = mfspr(SPR_MAS1);
3140 	entry->mas2 = mfspr(SPR_MAS2);
3141 	entry->mas3 = mfspr(SPR_MAS3);
3142 
3143 	switch ((mfpvr() >> 16) & 0xFFFF) {
3144 	case FSL_E500v2:
3145 	case FSL_E500mc:
3146 	case FSL_E5500:
3147 	case FSL_E6500:
3148 		entry->mas7 = mfspr(SPR_MAS7);
3149 		break;
3150 	default:
3151 		entry->mas7 = 0;
3152 		break;
3153 	}
3154 
3155 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3156 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3157 	    (entry->mas3 & MAS3_RPN);
3158 	entry->size =
3159 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3160 }
3161 
3162 /*
3163  * Write given entry to TLB1 hardware.
3164  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
3165  */
3166 static void
3167 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3168 {
3169 	uint32_t mas0;
3170 
3171 	//debugf("tlb1_write_entry: s\n");
3172 
3173 	/* Select entry */
3174 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3175 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
3176 
3177 	mtspr(SPR_MAS0, mas0);
3178 	__asm __volatile("isync");
3179 	mtspr(SPR_MAS1, e->mas1);
3180 	__asm __volatile("isync");
3181 	mtspr(SPR_MAS2, e->mas2);
3182 	__asm __volatile("isync");
3183 	mtspr(SPR_MAS3, e->mas3);
3184 	__asm __volatile("isync");
3185 	switch ((mfpvr() >> 16) & 0xFFFF) {
3186 	case FSL_E500mc:
3187 	case FSL_E5500:
3188 	case FSL_E6500:
3189 		mtspr(SPR_MAS8, 0);
3190 		__asm __volatile("isync");
3191 		/* FALLTHROUGH */
3192 	case FSL_E500v2:
3193 		mtspr(SPR_MAS7, e->mas7);
3194 		__asm __volatile("isync");
3195 		break;
3196 	default:
3197 		break;
3198 	}
3199 
3200 	__asm __volatile("tlbwe; isync; msync");
3201 
3202 	//debugf("tlb1_write_entry: e\n");
3203 }
3204 
3205 /*
3206  * Return the largest uint value log such that 2^log <= num.
3207  */
3208 static unsigned int
3209 ilog2(unsigned int num)
3210 {
3211 	int lz;
3212 
3213 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3214 	return (31 - lz);
3215 }
3216 
3217 /*
3218  * Convert TLB TSIZE value to mapped region size.
3219  */
3220 static vm_size_t
3221 tsize2size(unsigned int tsize)
3222 {
3223 
3224 	/*
3225 	 * size = 4^tsize KB
3226 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3227 	 */
3228 
3229 	return ((1 << (2 * tsize)) * 1024);
3230 }
3231 
3232 /*
3233  * Convert region size (must be power of 4) to TLB TSIZE value.
3234  */
3235 static unsigned int
3236 size2tsize(vm_size_t size)
3237 {
3238 
3239 	return (ilog2(size) / 2 - 5);
3240 }
3241 
3242 /*
3243  * Register permanent kernel mapping in TLB1.
3244  *
3245  * Entries are created starting from index 0 (current free entry is
3246  * kept in tlb1_idx) and are not supposed to be invalidated.
3247  */
3248 int
3249 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3250     uint32_t flags)
3251 {
3252 	tlb_entry_t e;
3253 	uint32_t ts, tid;
3254 	int tsize, index;
3255 
3256 	for (index = 0; index < TLB1_ENTRIES; index++) {
3257 		tlb1_read_entry(&e, index);
3258 		if ((e.mas1 & MAS1_VALID) == 0)
3259 			break;
3260 		/* Check if we're just updating the flags, and update them. */
3261 		if (e.phys == pa && e.virt == va && e.size == size) {
3262 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
3263 			tlb1_write_entry(&e, index);
3264 			return (0);
3265 		}
3266 	}
3267 	if (index >= TLB1_ENTRIES) {
3268 		printf("tlb1_set_entry: TLB1 full!\n");
3269 		return (-1);
3270 	}
3271 
3272 	/* Convert size to TSIZE */
3273 	tsize = size2tsize(size);
3274 
3275 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3276 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3277 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3278 
3279 	e.phys = pa;
3280 	e.virt = va;
3281 	e.size = size;
3282 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3283 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3284 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
3285 
3286 	/* Set supervisor RWX permission bits */
3287 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3288 	e.mas7 = (pa >> 32) & MAS7_RPN;
3289 
3290 	tlb1_write_entry(&e, index);
3291 
3292 	/*
3293 	 * XXX in general TLB1 updates should be propagated between CPUs,
3294 	 * since current design assumes to have the same TLB1 set-up on all
3295 	 * cores.
3296 	 */
3297 	return (0);
3298 }
3299 
3300 /*
3301  * Map in contiguous RAM region into the TLB1 using maximum of
3302  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3303  *
3304  * If necessary round up last entry size and return total size
3305  * used by all allocated entries.
3306  */
3307 vm_size_t
3308 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3309 {
3310 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3311 	vm_size_t mapped, pgsz, base, mask;
3312 	int idx, nents;
3313 
3314 	/* Round up to the next 1M */
3315 	size = roundup2(size, 1 << 20);
3316 
3317 	mapped = 0;
3318 	idx = 0;
3319 	base = va;
3320 	pgsz = 64*1024*1024;
3321 	while (mapped < size) {
3322 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3323 			while (pgsz > (size - mapped))
3324 				pgsz >>= 2;
3325 			pgs[idx++] = pgsz;
3326 			mapped += pgsz;
3327 		}
3328 
3329 		/* We under-map. Correct for this. */
3330 		if (mapped < size) {
3331 			while (pgs[idx - 1] == pgsz) {
3332 				idx--;
3333 				mapped -= pgsz;
3334 			}
3335 			/* XXX We may increase beyond out starting point. */
3336 			pgsz <<= 2;
3337 			pgs[idx++] = pgsz;
3338 			mapped += pgsz;
3339 		}
3340 	}
3341 
3342 	nents = idx;
3343 	mask = pgs[0] - 1;
3344 	/* Align address to the boundary */
3345 	if (va & mask) {
3346 		va = (va + mask) & ~mask;
3347 		pa = (pa + mask) & ~mask;
3348 	}
3349 
3350 	for (idx = 0; idx < nents; idx++) {
3351 		pgsz = pgs[idx];
3352 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
3353 		tlb1_set_entry(va, pa, pgsz,
3354 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
3355 		pa += pgsz;
3356 		va += pgsz;
3357 	}
3358 
3359 	mapped = (va - base);
3360 #ifdef __powerpc64__
3361 	printf("mapped size 0x%016lx (wasted space 0x%16lx)\n",
3362 #else
3363 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3364 #endif
3365 	    mapped, mapped - size);
3366 	return (mapped);
3367 }
3368 
3369 /*
3370  * TLB1 initialization routine, to be called after the very first
3371  * assembler level setup done in locore.S.
3372  */
3373 void
3374 tlb1_init()
3375 {
3376 	uint32_t mas0, mas1, mas2, mas3, mas7;
3377 	uint32_t tsz;
3378 
3379 	tlb1_get_tlbconf();
3380 
3381 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3382 	mtspr(SPR_MAS0, mas0);
3383 	__asm __volatile("isync; tlbre");
3384 
3385 	mas1 = mfspr(SPR_MAS1);
3386 	mas2 = mfspr(SPR_MAS2);
3387 	mas3 = mfspr(SPR_MAS3);
3388 	mas7 = mfspr(SPR_MAS7);
3389 
3390 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
3391 	    (mas3 & MAS3_RPN);
3392 
3393 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3394 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
3395 
3396 	/* Setup TLB miss defaults */
3397 	set_mas4_defaults();
3398 }
3399 
3400 /*
3401  * pmap_early_io_unmap() should be used in short conjunction with
3402  * pmap_early_io_map(), as in the following snippet:
3403  *
3404  * x = pmap_early_io_map(...);
3405  * <do something with x>
3406  * pmap_early_io_unmap(x, size);
3407  *
3408  * And avoiding more allocations between.
3409  */
3410 void
3411 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
3412 {
3413 	int i;
3414 	tlb_entry_t e;
3415 	vm_size_t isize;
3416 
3417 	size = roundup(size, PAGE_SIZE);
3418 	isize = size;
3419 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
3420 		tlb1_read_entry(&e, i);
3421 		if (!(e.mas1 & MAS1_VALID))
3422 			continue;
3423 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
3424 			size -= e.size;
3425 			e.mas1 &= ~MAS1_VALID;
3426 			tlb1_write_entry(&e, i);
3427 		}
3428 	}
3429 	if (tlb1_map_base == va + isize)
3430 		tlb1_map_base -= isize;
3431 }
3432 
3433 vm_offset_t
3434 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3435 {
3436 	vm_paddr_t pa_base;
3437 	vm_offset_t va, sz;
3438 	int i;
3439 	tlb_entry_t e;
3440 
3441 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3442 
3443 	for (i = 0; i < TLB1_ENTRIES; i++) {
3444 		tlb1_read_entry(&e, i);
3445 		if (!(e.mas1 & MAS1_VALID))
3446 			continue;
3447 		if (pa >= e.phys && (pa + size) <=
3448 		    (e.phys + e.size))
3449 			return (e.virt + (pa - e.phys));
3450 	}
3451 
3452 	pa_base = rounddown(pa, PAGE_SIZE);
3453 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3454 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3455 	va = tlb1_map_base + (pa - pa_base);
3456 
3457 	do {
3458 		sz = 1 << (ilog2(size) & ~1);
3459 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
3460 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
3461 		size -= sz;
3462 		pa_base += sz;
3463 		tlb1_map_base += sz;
3464 	} while (size > 0);
3465 
3466 	return (va);
3467 }
3468 
3469 /*
3470  * Setup MAS4 defaults.
3471  * These values are loaded to MAS0-2 on a TLB miss.
3472  */
3473 static void
3474 set_mas4_defaults(void)
3475 {
3476 	uint32_t mas4;
3477 
3478 	/* Defaults: TLB0, PID0, TSIZED=4K */
3479 	mas4 = MAS4_TLBSELD0;
3480 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3481 #ifdef SMP
3482 	mas4 |= MAS4_MD;
3483 #endif
3484 	mtspr(SPR_MAS4, mas4);
3485 	__asm __volatile("isync");
3486 }
3487 
3488 /*
3489  * Print out contents of the MAS registers for each TLB1 entry
3490  */
3491 void
3492 tlb1_print_tlbentries(void)
3493 {
3494 	uint32_t mas0, mas1, mas2, mas3, mas7;
3495 	int i;
3496 
3497 	debugf("TLB1 entries:\n");
3498 	for (i = 0; i < TLB1_ENTRIES; i++) {
3499 
3500 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3501 		mtspr(SPR_MAS0, mas0);
3502 
3503 		__asm __volatile("isync; tlbre");
3504 
3505 		mas1 = mfspr(SPR_MAS1);
3506 		mas2 = mfspr(SPR_MAS2);
3507 		mas3 = mfspr(SPR_MAS3);
3508 		mas7 = mfspr(SPR_MAS7);
3509 
3510 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3511 	}
3512 }
3513 
3514 /*
3515  * Return 0 if the physical IO range is encompassed by one of the
3516  * the TLB1 entries, otherwise return related error code.
3517  */
3518 static int
3519 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3520 {
3521 	uint32_t prot;
3522 	vm_paddr_t pa_start;
3523 	vm_paddr_t pa_end;
3524 	unsigned int entry_tsize;
3525 	vm_size_t entry_size;
3526 	tlb_entry_t e;
3527 
3528 	*va = (vm_offset_t)NULL;
3529 
3530 	tlb1_read_entry(&e, i);
3531 	/* Skip invalid entries */
3532 	if (!(e.mas1 & MAS1_VALID))
3533 		return (EINVAL);
3534 
3535 	/*
3536 	 * The entry must be cache-inhibited, guarded, and r/w
3537 	 * so it can function as an i/o page
3538 	 */
3539 	prot = e.mas2 & (MAS2_I | MAS2_G);
3540 	if (prot != (MAS2_I | MAS2_G))
3541 		return (EPERM);
3542 
3543 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
3544 	if (prot != (MAS3_SR | MAS3_SW))
3545 		return (EPERM);
3546 
3547 	/* The address should be within the entry range. */
3548 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3549 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3550 
3551 	entry_size = tsize2size(entry_tsize);
3552 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
3553 	    (e.mas3 & MAS3_RPN);
3554 	pa_end = pa_start + entry_size;
3555 
3556 	if ((pa < pa_start) || ((pa + size) > pa_end))
3557 		return (ERANGE);
3558 
3559 	/* Return virtual address of this mapping. */
3560 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3561 	return (0);
3562 }
3563 
3564 /*
3565  * Invalidate all TLB0 entries which match the given TID. Note this is
3566  * dedicated for cases when invalidations should NOT be propagated to other
3567  * CPUs.
3568  */
3569 static void
3570 tid_flush(tlbtid_t tid)
3571 {
3572 	register_t msr;
3573 	uint32_t mas0, mas1, mas2;
3574 	int entry, way;
3575 
3576 
3577 	/* Don't evict kernel translations */
3578 	if (tid == TID_KERNEL)
3579 		return;
3580 
3581 	msr = mfmsr();
3582 	__asm __volatile("wrteei 0");
3583 
3584 	for (way = 0; way < TLB0_WAYS; way++)
3585 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
3586 
3587 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3588 			mtspr(SPR_MAS0, mas0);
3589 			__asm __volatile("isync");
3590 
3591 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
3592 			mtspr(SPR_MAS2, mas2);
3593 
3594 			__asm __volatile("isync; tlbre");
3595 
3596 			mas1 = mfspr(SPR_MAS1);
3597 
3598 			if (!(mas1 & MAS1_VALID))
3599 				continue;
3600 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
3601 				continue;
3602 			mas1 &= ~MAS1_VALID;
3603 			mtspr(SPR_MAS1, mas1);
3604 			__asm __volatile("isync; tlbwe; isync; msync");
3605 		}
3606 	mtmsr(msr);
3607 }
3608