xref: /freebsd/sys/powerpc/booke/pmap.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - ptbl_bufs-1              : message buffer
63   *               ptbl_bufs - kernel_pdir-1            : kernel page tables
64   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
65   *             kernel_pp2d - .                        : kernel pointers to page directory
66   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
67   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
68   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
69   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
70   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
71   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
72   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
73   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
74   *                         - 0xffff_ffff_ffff_ffff    : device direct map
75   */
76 
77 #include <sys/cdefs.h>
78 __FBSDID("$FreeBSD$");
79 
80 #include "opt_ddb.h"
81 #include "opt_kstack_pages.h"
82 
83 #include <sys/param.h>
84 #include <sys/conf.h>
85 #include <sys/malloc.h>
86 #include <sys/ktr.h>
87 #include <sys/proc.h>
88 #include <sys/user.h>
89 #include <sys/queue.h>
90 #include <sys/systm.h>
91 #include <sys/kernel.h>
92 #include <sys/kerneldump.h>
93 #include <sys/linker.h>
94 #include <sys/msgbuf.h>
95 #include <sys/lock.h>
96 #include <sys/mutex.h>
97 #include <sys/rwlock.h>
98 #include <sys/sched.h>
99 #include <sys/smp.h>
100 #include <sys/vmmeter.h>
101 
102 #include <vm/vm.h>
103 #include <vm/vm_page.h>
104 #include <vm/vm_kern.h>
105 #include <vm/vm_pageout.h>
106 #include <vm/vm_extern.h>
107 #include <vm/vm_object.h>
108 #include <vm/vm_param.h>
109 #include <vm/vm_map.h>
110 #include <vm/vm_pager.h>
111 #include <vm/vm_phys.h>
112 #include <vm/vm_pagequeue.h>
113 #include <vm/uma.h>
114 
115 #include <machine/_inttypes.h>
116 #include <machine/cpu.h>
117 #include <machine/pcb.h>
118 #include <machine/platform.h>
119 
120 #include <machine/tlb.h>
121 #include <machine/spr.h>
122 #include <machine/md_var.h>
123 #include <machine/mmuvar.h>
124 #include <machine/pmap.h>
125 #include <machine/pte.h>
126 
127 #include <ddb/ddb.h>
128 
129 #include "mmu_if.h"
130 
131 #define	SPARSE_MAPDEV
132 #ifdef  DEBUG
133 #define debugf(fmt, args...) printf(fmt, ##args)
134 #else
135 #define debugf(fmt, args...)
136 #endif
137 
138 #ifdef __powerpc64__
139 #define	PRI0ptrX	"016lx"
140 #else
141 #define	PRI0ptrX	"08x"
142 #endif
143 
144 #define TODO			panic("%s: not implemented", __func__);
145 
146 extern unsigned char _etext[];
147 extern unsigned char _end[];
148 
149 extern uint32_t *bootinfo;
150 
151 vm_paddr_t kernload;
152 vm_offset_t kernstart;
153 vm_size_t kernsize;
154 
155 /* Message buffer and tables. */
156 static vm_offset_t data_start;
157 static vm_size_t data_end;
158 
159 /* Phys/avail memory regions. */
160 static struct mem_region *availmem_regions;
161 static int availmem_regions_sz;
162 static struct mem_region *physmem_regions;
163 static int physmem_regions_sz;
164 
165 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
166 static vm_offset_t zero_page_va;
167 static struct mtx zero_page_mutex;
168 
169 static struct mtx tlbivax_mutex;
170 
171 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
172 static vm_offset_t copy_page_src_va;
173 static vm_offset_t copy_page_dst_va;
174 static struct mtx copy_page_mutex;
175 
176 /**************************************************************************/
177 /* PMAP */
178 /**************************************************************************/
179 
180 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
181     vm_prot_t, u_int flags, int8_t psind);
182 
183 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
184 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
185 #ifdef __powerpc64__
186 unsigned int kernel_pdirs;
187 #endif
188 
189 /*
190  * If user pmap is processed with mmu_booke_remove and the resident count
191  * drops to 0, there are no more pages to remove, so we need not continue.
192  */
193 #define PMAP_REMOVE_DONE(pmap) \
194 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
195 
196 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
197 extern int elf32_nxstack;
198 #endif
199 
200 /**************************************************************************/
201 /* TLB and TID handling */
202 /**************************************************************************/
203 
204 /* Translation ID busy table */
205 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
206 
207 /*
208  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
209  * core revisions and should be read from h/w registers during early config.
210  */
211 uint32_t tlb0_entries;
212 uint32_t tlb0_ways;
213 uint32_t tlb0_entries_per_way;
214 uint32_t tlb1_entries;
215 
216 #define TLB0_ENTRIES		(tlb0_entries)
217 #define TLB0_WAYS		(tlb0_ways)
218 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
219 
220 #define TLB1_ENTRIES (tlb1_entries)
221 
222 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
223 
224 static tlbtid_t tid_alloc(struct pmap *);
225 static void tid_flush(tlbtid_t tid);
226 
227 #ifdef DDB
228 #ifdef __powerpc64__
229 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
230 #else
231 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
232 #endif
233 #endif
234 
235 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
236 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
237 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
238 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
239 
240 static vm_size_t tsize2size(unsigned int);
241 static unsigned int size2tsize(vm_size_t);
242 static unsigned int ilog2(unsigned long);
243 
244 static void set_mas4_defaults(void);
245 
246 static inline void tlb0_flush_entry(vm_offset_t);
247 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
248 
249 /**************************************************************************/
250 /* Page table management */
251 /**************************************************************************/
252 
253 static struct rwlock_padalign pvh_global_lock;
254 
255 /* Data for the pv entry allocation mechanism */
256 static uma_zone_t pvzone;
257 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
258 
259 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
260 
261 #ifndef PMAP_SHPGPERPROC
262 #define PMAP_SHPGPERPROC	200
263 #endif
264 
265 static void ptbl_init(void);
266 static struct ptbl_buf *ptbl_buf_alloc(void);
267 static void ptbl_buf_free(struct ptbl_buf *);
268 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
269 
270 #ifdef __powerpc64__
271 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
272 			 unsigned int, boolean_t);
273 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int);
274 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
275 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
276 #else
277 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
278 static void ptbl_free(mmu_t, pmap_t, unsigned int);
279 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
280 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
281 #endif
282 
283 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
284 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
285 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
286 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
287 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
288 
289 static pv_entry_t pv_alloc(void);
290 static void pv_free(pv_entry_t);
291 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
292 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
293 
294 static void booke_pmap_init_qpages(void);
295 
296 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
297 #ifdef __powerpc64__
298 #define PTBL_BUFS               (16UL * 16 * 16)
299 #else
300 #define PTBL_BUFS		(128 * 16)
301 #endif
302 
303 struct ptbl_buf {
304 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
305 	vm_offset_t kva;		/* va of mapping */
306 };
307 
308 /* ptbl free list and a lock used for access synchronization. */
309 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
310 static struct mtx ptbl_buf_freelist_lock;
311 
312 /* Base address of kva space allocated fot ptbl bufs. */
313 static vm_offset_t ptbl_buf_pool_vabase;
314 
315 /* Pointer to ptbl_buf structures. */
316 static struct ptbl_buf *ptbl_bufs;
317 
318 #ifdef SMP
319 extern tlb_entry_t __boot_tlb1[];
320 void pmap_bootstrap_ap(volatile uint32_t *);
321 #endif
322 
323 /*
324  * Kernel MMU interface
325  */
326 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
327 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
328     vm_size_t, vm_offset_t);
329 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
330 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
331     vm_offset_t, vm_page_t *, vm_offset_t, int);
332 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
333     vm_prot_t, u_int flags, int8_t psind);
334 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
335     vm_page_t, vm_prot_t);
336 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
337     vm_prot_t);
338 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
339 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
340     vm_prot_t);
341 static void		mmu_booke_init(mmu_t);
342 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
343 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
344 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
345 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
346 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
347     int);
348 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
349     vm_paddr_t *);
350 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
351     vm_object_t, vm_pindex_t, vm_size_t);
352 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
353 static void		mmu_booke_page_init(mmu_t, vm_page_t);
354 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
355 static void		mmu_booke_pinit(mmu_t, pmap_t);
356 static void		mmu_booke_pinit0(mmu_t, pmap_t);
357 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
358     vm_prot_t);
359 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
360 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
361 static void		mmu_booke_release(mmu_t, pmap_t);
362 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
363 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
364 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
365 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
366 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
367 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
368 static void		mmu_booke_activate(mmu_t, struct thread *);
369 static void		mmu_booke_deactivate(mmu_t, struct thread *);
370 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
371 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
372 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
373 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
374 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
375 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
376 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
377 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
378 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
379 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
380     vm_size_t);
381 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
382     void **);
383 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
384     void *);
385 static void		mmu_booke_scan_init(mmu_t);
386 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
387 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
388 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
389     vm_size_t sz, vm_memattr_t mode);
390 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
391     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
392 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
393     int *is_user, vm_offset_t *decoded_addr);
394 
395 
396 static mmu_method_t mmu_booke_methods[] = {
397 	/* pmap dispatcher interface */
398 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
399 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
400 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
401 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
402 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
403 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
404 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
405 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
406 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
407 	MMUMETHOD(mmu_init,		mmu_booke_init),
408 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
409 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
410 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
411 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
412 	MMUMETHOD(mmu_map,		mmu_booke_map),
413 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
414 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
415 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
416 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
417 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
418 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
419 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
420 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
421 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
422 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
423 	MMUMETHOD(mmu_release,		mmu_booke_release),
424 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
425 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
426 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
427 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
428 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
429 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
430 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
431 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
432 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
433 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
434 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
435 
436 	/* Internal interfaces */
437 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
438 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
439 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
440 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
441 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
442 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
443 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
444 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
445 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
446 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
447 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
448 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
449 
450 	/* dumpsys() support */
451 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
452 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
453 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
454 
455 	{ 0, 0 }
456 };
457 
458 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
459 
460 static __inline uint32_t
461 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
462 {
463 	uint32_t attrib;
464 	int i;
465 
466 	if (ma != VM_MEMATTR_DEFAULT) {
467 		switch (ma) {
468 		case VM_MEMATTR_UNCACHEABLE:
469 			return (MAS2_I | MAS2_G);
470 		case VM_MEMATTR_WRITE_COMBINING:
471 		case VM_MEMATTR_WRITE_BACK:
472 		case VM_MEMATTR_PREFETCHABLE:
473 			return (MAS2_I);
474 		case VM_MEMATTR_WRITE_THROUGH:
475 			return (MAS2_W | MAS2_M);
476 		case VM_MEMATTR_CACHEABLE:
477 			return (MAS2_M);
478 		}
479 	}
480 
481 	/*
482 	 * Assume the page is cache inhibited and access is guarded unless
483 	 * it's in our available memory array.
484 	 */
485 	attrib = _TLB_ENTRY_IO;
486 	for (i = 0; i < physmem_regions_sz; i++) {
487 		if ((pa >= physmem_regions[i].mr_start) &&
488 		    (pa < (physmem_regions[i].mr_start +
489 		     physmem_regions[i].mr_size))) {
490 			attrib = _TLB_ENTRY_MEM;
491 			break;
492 		}
493 	}
494 
495 	return (attrib);
496 }
497 
498 static inline void
499 tlb_miss_lock(void)
500 {
501 #ifdef SMP
502 	struct pcpu *pc;
503 
504 	if (!smp_started)
505 		return;
506 
507 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
508 		if (pc != pcpup) {
509 
510 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
511 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
512 
513 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
514 			    ("tlb_miss_lock: tried to lock self"));
515 
516 			tlb_lock(pc->pc_booke.tlb_lock);
517 
518 			CTR1(KTR_PMAP, "%s: locked", __func__);
519 		}
520 	}
521 #endif
522 }
523 
524 static inline void
525 tlb_miss_unlock(void)
526 {
527 #ifdef SMP
528 	struct pcpu *pc;
529 
530 	if (!smp_started)
531 		return;
532 
533 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
534 		if (pc != pcpup) {
535 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
536 			    __func__, pc->pc_cpuid);
537 
538 			tlb_unlock(pc->pc_booke.tlb_lock);
539 
540 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
541 		}
542 	}
543 #endif
544 }
545 
546 /* Return number of entries in TLB0. */
547 static __inline void
548 tlb0_get_tlbconf(void)
549 {
550 	uint32_t tlb0_cfg;
551 
552 	tlb0_cfg = mfspr(SPR_TLB0CFG);
553 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
554 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
555 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
556 }
557 
558 /* Return number of entries in TLB1. */
559 static __inline void
560 tlb1_get_tlbconf(void)
561 {
562 	uint32_t tlb1_cfg;
563 
564 	tlb1_cfg = mfspr(SPR_TLB1CFG);
565 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
566 }
567 
568 /**************************************************************************/
569 /* Page table related */
570 /**************************************************************************/
571 
572 #ifdef __powerpc64__
573 /* Initialize pool of kva ptbl buffers. */
574 static void
575 ptbl_init(void)
576 {
577 	int		i;
578 
579 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
580 	TAILQ_INIT(&ptbl_buf_freelist);
581 
582 	for (i = 0; i < PTBL_BUFS; i++) {
583 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase +
584 		    i * MAX(PTBL_PAGES,PDIR_PAGES) * PAGE_SIZE;
585 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
586 	}
587 }
588 
589 /* Get an sf_buf from the freelist. */
590 static struct ptbl_buf *
591 ptbl_buf_alloc(void)
592 {
593 	struct ptbl_buf *buf;
594 
595 	mtx_lock(&ptbl_buf_freelist_lock);
596 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
597 	if (buf != NULL)
598 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
599 	mtx_unlock(&ptbl_buf_freelist_lock);
600 
601 	return (buf);
602 }
603 
604 /* Return ptbl buff to free pool. */
605 static void
606 ptbl_buf_free(struct ptbl_buf *buf)
607 {
608 	mtx_lock(&ptbl_buf_freelist_lock);
609 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
610 	mtx_unlock(&ptbl_buf_freelist_lock);
611 }
612 
613 /*
614  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
615  */
616 static void
617 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t * ptbl)
618 {
619 	struct ptbl_buf *pbuf;
620 
621 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) {
622 		if (pbuf->kva == (vm_offset_t) ptbl) {
623 			/* Remove from pmap ptbl buf list. */
624 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
625 
626 			/* Free corresponding ptbl buf. */
627 			ptbl_buf_free(pbuf);
628 
629 			break;
630 		}
631 	}
632 }
633 
634 /* Get a pointer to a PTE in a page table. */
635 static __inline pte_t *
636 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
637 {
638 	pte_t         **pdir;
639 	pte_t          *ptbl;
640 
641 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
642 
643 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
644 	if (!pdir)
645 		return NULL;
646 	ptbl = pdir[PDIR_IDX(va)];
647 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
648 }
649 
650 /*
651  * Search the list of allocated pdir bufs and find on list of allocated pdirs
652  */
653 static void
654 ptbl_free_pmap_pdir(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
655 {
656 	struct ptbl_buf *pbuf;
657 
658 	TAILQ_FOREACH(pbuf, &pmap->pm_pdir_list, link) {
659 		if (pbuf->kva == (vm_offset_t) pdir) {
660 			/* Remove from pmap ptbl buf list. */
661 			TAILQ_REMOVE(&pmap->pm_pdir_list, pbuf, link);
662 
663 			/* Free corresponding pdir buf. */
664 			ptbl_buf_free(pbuf);
665 
666 			break;
667 		}
668 	}
669 }
670 /* Free pdir pages and invalidate pdir entry. */
671 static void
672 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
673 {
674 	pte_t         **pdir;
675 	vm_paddr_t	pa;
676 	vm_offset_t	va;
677 	vm_page_t	m;
678 	int		i;
679 
680 	pdir = pmap->pm_pp2d[pp2d_idx];
681 
682 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
683 
684 	pmap->pm_pp2d[pp2d_idx] = NULL;
685 
686 	for (i = 0; i < PDIR_PAGES; i++) {
687 		va = ((vm_offset_t) pdir + (i * PAGE_SIZE));
688 		pa = pte_vatopa(mmu, kernel_pmap, va);
689 		m = PHYS_TO_VM_PAGE(pa);
690 		vm_page_free_zero(m);
691 		vm_wire_sub(1);
692 		pmap_kremove(va);
693 	}
694 
695 	ptbl_free_pmap_pdir(mmu, pmap, pdir);
696 }
697 
698 /*
699  * Decrement pdir pages hold count and attempt to free pdir pages. Called
700  * when removing directory entry from pdir.
701  *
702  * Return 1 if pdir pages were freed.
703  */
704 static int
705 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
706 {
707 	pte_t         **pdir;
708 	vm_paddr_t	pa;
709 	vm_page_t	m;
710 	int		i;
711 
712 	KASSERT((pmap != kernel_pmap),
713 		("pdir_unhold: unholding kernel pdir!"));
714 
715 	pdir = pmap->pm_pp2d[pp2d_idx];
716 
717 	KASSERT(((vm_offset_t) pdir >= VM_MIN_KERNEL_ADDRESS),
718 	    ("pdir_unhold: non kva pdir"));
719 
720 	/* decrement hold count */
721 	for (i = 0; i < PDIR_PAGES; i++) {
722 		pa = pte_vatopa(mmu, kernel_pmap,
723 		    (vm_offset_t) pdir + (i * PAGE_SIZE));
724 		m = PHYS_TO_VM_PAGE(pa);
725 		m->wire_count--;
726 	}
727 
728 	/*
729 	 * Free pdir pages if there are no dir entries in this pdir.
730 	 * wire_count has the same value for all ptbl pages, so check the
731 	 * last page.
732 	 */
733 	if (m->wire_count == 0) {
734 		pdir_free(mmu, pmap, pp2d_idx);
735 		return (1);
736 	}
737 	return (0);
738 }
739 
740 /*
741  * Increment hold count for pdir pages. This routine is used when new ptlb
742  * entry is being inserted into pdir.
743  */
744 static void
745 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
746 {
747 	vm_paddr_t	pa;
748 	vm_page_t	m;
749 	int		i;
750 
751 	KASSERT((pmap != kernel_pmap),
752 		("pdir_hold: holding kernel pdir!"));
753 
754 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
755 
756 	for (i = 0; i < PDIR_PAGES; i++) {
757 		pa = pte_vatopa(mmu, kernel_pmap,
758 				(vm_offset_t) pdir + (i * PAGE_SIZE));
759 		m = PHYS_TO_VM_PAGE(pa);
760 		m->wire_count++;
761 	}
762 }
763 
764 /* Allocate page table. */
765 static pte_t   *
766 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
767     boolean_t nosleep)
768 {
769 	vm_page_t	mtbl  [PTBL_PAGES];
770 	vm_page_t	m;
771 	struct ptbl_buf *pbuf;
772 	unsigned int	pidx;
773 	pte_t          *ptbl;
774 	int		i, j;
775 	int		req;
776 
777 	KASSERT((pdir[pdir_idx] == NULL),
778 		("%s: valid ptbl entry exists!", __func__));
779 
780 	pbuf = ptbl_buf_alloc();
781 	if (pbuf == NULL)
782 		panic("%s: couldn't alloc kernel virtual memory", __func__);
783 
784 	ptbl = (pte_t *) pbuf->kva;
785 
786 	for (i = 0; i < PTBL_PAGES; i++) {
787 		pidx = (PTBL_PAGES * pdir_idx) + i;
788 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
789 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
790 			PMAP_UNLOCK(pmap);
791 			rw_wunlock(&pvh_global_lock);
792 			if (nosleep) {
793 				ptbl_free_pmap_ptbl(pmap, ptbl);
794 				for (j = 0; j < i; j++)
795 					vm_page_free(mtbl[j]);
796 				vm_wire_sub(i);
797 				return (NULL);
798 			}
799 			vm_wait(NULL);
800 			rw_wlock(&pvh_global_lock);
801 			PMAP_LOCK(pmap);
802 		}
803 		mtbl[i] = m;
804 	}
805 
806 	/* Mapin allocated pages into kernel_pmap. */
807 	mmu_booke_qenter(mmu, (vm_offset_t) ptbl, mtbl, PTBL_PAGES);
808 	/* Zero whole ptbl. */
809 	bzero((caddr_t) ptbl, PTBL_PAGES * PAGE_SIZE);
810 
811 	/* Add pbuf to the pmap ptbl bufs list. */
812 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
813 
814 	return (ptbl);
815 }
816 
817 /* Free ptbl pages and invalidate pdir entry. */
818 static void
819 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
820 {
821 	pte_t          *ptbl;
822 	vm_paddr_t	pa;
823 	vm_offset_t	va;
824 	vm_page_t	m;
825 	int		i;
826 
827 	ptbl = pdir[pdir_idx];
828 
829 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
830 
831 	pdir[pdir_idx] = NULL;
832 
833 	for (i = 0; i < PTBL_PAGES; i++) {
834 		va = ((vm_offset_t) ptbl + (i * PAGE_SIZE));
835 		pa = pte_vatopa(mmu, kernel_pmap, va);
836 		m = PHYS_TO_VM_PAGE(pa);
837 		vm_page_free_zero(m);
838 		vm_wire_sub(1);
839 		pmap_kremove(va);
840 	}
841 
842 	ptbl_free_pmap_ptbl(pmap, ptbl);
843 }
844 
845 /*
846  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
847  * when removing pte entry from ptbl.
848  *
849  * Return 1 if ptbl pages were freed.
850  */
851 static int
852 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
853 {
854 	pte_t          *ptbl;
855 	vm_paddr_t	pa;
856 	vm_page_t	m;
857 	u_int		pp2d_idx;
858 	pte_t         **pdir;
859 	u_int		pdir_idx;
860 	int		i;
861 
862 	pp2d_idx = PP2D_IDX(va);
863 	pdir_idx = PDIR_IDX(va);
864 
865 	KASSERT((pmap != kernel_pmap),
866 		("ptbl_unhold: unholding kernel ptbl!"));
867 
868 	pdir = pmap->pm_pp2d[pp2d_idx];
869 	ptbl = pdir[pdir_idx];
870 
871 	KASSERT(((vm_offset_t) ptbl >= VM_MIN_KERNEL_ADDRESS),
872 	    ("ptbl_unhold: non kva ptbl"));
873 
874 	/* decrement hold count */
875 	for (i = 0; i < PTBL_PAGES; i++) {
876 		pa = pte_vatopa(mmu, kernel_pmap,
877 		    (vm_offset_t) ptbl + (i * PAGE_SIZE));
878 		m = PHYS_TO_VM_PAGE(pa);
879 		m->wire_count--;
880 	}
881 
882 	/*
883 	 * Free ptbl pages if there are no pte entries in this ptbl.
884 	 * wire_count has the same value for all ptbl pages, so check the
885 	 * last page.
886 	 */
887 	if (m->wire_count == 0) {
888 		/* A pair of indirect entries might point to this ptbl page */
889 #if 0
890 		tlb_flush_entry(pmap, va & ~((2UL * PAGE_SIZE_1M) - 1),
891 				TLB_SIZE_1M, MAS6_SIND);
892 		tlb_flush_entry(pmap, (va & ~((2UL * PAGE_SIZE_1M) - 1)) | PAGE_SIZE_1M,
893 				TLB_SIZE_1M, MAS6_SIND);
894 #endif
895 		ptbl_free(mmu, pmap, pdir, pdir_idx);
896 		pdir_unhold(mmu, pmap, pp2d_idx);
897 		return (1);
898 	}
899 	return (0);
900 }
901 
902 /*
903  * Increment hold count for ptbl pages. This routine is used when new pte
904  * entry is being inserted into ptbl.
905  */
906 static void
907 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
908 {
909 	vm_paddr_t	pa;
910 	pte_t          *ptbl;
911 	vm_page_t	m;
912 	int		i;
913 
914 	KASSERT((pmap != kernel_pmap),
915 		("ptbl_hold: holding kernel ptbl!"));
916 
917 	ptbl = pdir[pdir_idx];
918 
919 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
920 
921 	for (i = 0; i < PTBL_PAGES; i++) {
922 		pa = pte_vatopa(mmu, kernel_pmap,
923 				(vm_offset_t) ptbl + (i * PAGE_SIZE));
924 		m = PHYS_TO_VM_PAGE(pa);
925 		m->wire_count++;
926 	}
927 }
928 #else
929 
930 /* Initialize pool of kva ptbl buffers. */
931 static void
932 ptbl_init(void)
933 {
934 	int i;
935 
936 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
937 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
938 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
939 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
940 
941 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
942 	TAILQ_INIT(&ptbl_buf_freelist);
943 
944 	for (i = 0; i < PTBL_BUFS; i++) {
945 		ptbl_bufs[i].kva =
946 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
947 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
948 	}
949 }
950 
951 /* Get a ptbl_buf from the freelist. */
952 static struct ptbl_buf *
953 ptbl_buf_alloc(void)
954 {
955 	struct ptbl_buf *buf;
956 
957 	mtx_lock(&ptbl_buf_freelist_lock);
958 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
959 	if (buf != NULL)
960 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
961 	mtx_unlock(&ptbl_buf_freelist_lock);
962 
963 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
964 
965 	return (buf);
966 }
967 
968 /* Return ptbl buff to free pool. */
969 static void
970 ptbl_buf_free(struct ptbl_buf *buf)
971 {
972 
973 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
974 
975 	mtx_lock(&ptbl_buf_freelist_lock);
976 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
977 	mtx_unlock(&ptbl_buf_freelist_lock);
978 }
979 
980 /*
981  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
982  */
983 static void
984 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
985 {
986 	struct ptbl_buf *pbuf;
987 
988 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
989 
990 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
991 
992 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
993 		if (pbuf->kva == (vm_offset_t)ptbl) {
994 			/* Remove from pmap ptbl buf list. */
995 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
996 
997 			/* Free corresponding ptbl buf. */
998 			ptbl_buf_free(pbuf);
999 			break;
1000 		}
1001 }
1002 
1003 /* Allocate page table. */
1004 static pte_t *
1005 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
1006 {
1007 	vm_page_t mtbl[PTBL_PAGES];
1008 	vm_page_t m;
1009 	struct ptbl_buf *pbuf;
1010 	unsigned int pidx;
1011 	pte_t *ptbl;
1012 	int i, j;
1013 
1014 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1015 	    (pmap == kernel_pmap), pdir_idx);
1016 
1017 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1018 	    ("ptbl_alloc: invalid pdir_idx"));
1019 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
1020 	    ("pte_alloc: valid ptbl entry exists!"));
1021 
1022 	pbuf = ptbl_buf_alloc();
1023 	if (pbuf == NULL)
1024 		panic("pte_alloc: couldn't alloc kernel virtual memory");
1025 
1026 	ptbl = (pte_t *)pbuf->kva;
1027 
1028 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
1029 
1030 	for (i = 0; i < PTBL_PAGES; i++) {
1031 		pidx = (PTBL_PAGES * pdir_idx) + i;
1032 		while ((m = vm_page_alloc(NULL, pidx,
1033 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
1034 			PMAP_UNLOCK(pmap);
1035 			rw_wunlock(&pvh_global_lock);
1036 			if (nosleep) {
1037 				ptbl_free_pmap_ptbl(pmap, ptbl);
1038 				for (j = 0; j < i; j++)
1039 					vm_page_free(mtbl[j]);
1040 				vm_wire_sub(i);
1041 				return (NULL);
1042 			}
1043 			vm_wait(NULL);
1044 			rw_wlock(&pvh_global_lock);
1045 			PMAP_LOCK(pmap);
1046 		}
1047 		mtbl[i] = m;
1048 	}
1049 
1050 	/* Map allocated pages into kernel_pmap. */
1051 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
1052 
1053 	/* Zero whole ptbl. */
1054 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
1055 
1056 	/* Add pbuf to the pmap ptbl bufs list. */
1057 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
1058 
1059 	return (ptbl);
1060 }
1061 
1062 /* Free ptbl pages and invalidate pdir entry. */
1063 static void
1064 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1065 {
1066 	pte_t *ptbl;
1067 	vm_paddr_t pa;
1068 	vm_offset_t va;
1069 	vm_page_t m;
1070 	int i;
1071 
1072 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1073 	    (pmap == kernel_pmap), pdir_idx);
1074 
1075 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1076 	    ("ptbl_free: invalid pdir_idx"));
1077 
1078 	ptbl = pmap->pm_pdir[pdir_idx];
1079 
1080 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
1081 
1082 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
1083 
1084 	/*
1085 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
1086 	 * don't attempt to look up the page tables we are releasing.
1087 	 */
1088 	mtx_lock_spin(&tlbivax_mutex);
1089 	tlb_miss_lock();
1090 
1091 	pmap->pm_pdir[pdir_idx] = NULL;
1092 
1093 	tlb_miss_unlock();
1094 	mtx_unlock_spin(&tlbivax_mutex);
1095 
1096 	for (i = 0; i < PTBL_PAGES; i++) {
1097 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
1098 		pa = pte_vatopa(mmu, kernel_pmap, va);
1099 		m = PHYS_TO_VM_PAGE(pa);
1100 		vm_page_free_zero(m);
1101 		vm_wire_sub(1);
1102 		mmu_booke_kremove(mmu, va);
1103 	}
1104 
1105 	ptbl_free_pmap_ptbl(pmap, ptbl);
1106 }
1107 
1108 /*
1109  * Decrement ptbl pages hold count and attempt to free ptbl pages.
1110  * Called when removing pte entry from ptbl.
1111  *
1112  * Return 1 if ptbl pages were freed.
1113  */
1114 static int
1115 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1116 {
1117 	pte_t *ptbl;
1118 	vm_paddr_t pa;
1119 	vm_page_t m;
1120 	int i;
1121 
1122 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1123 	    (pmap == kernel_pmap), pdir_idx);
1124 
1125 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1126 	    ("ptbl_unhold: invalid pdir_idx"));
1127 	KASSERT((pmap != kernel_pmap),
1128 	    ("ptbl_unhold: unholding kernel ptbl!"));
1129 
1130 	ptbl = pmap->pm_pdir[pdir_idx];
1131 
1132 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1133 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1134 	    ("ptbl_unhold: non kva ptbl"));
1135 
1136 	/* decrement hold count */
1137 	for (i = 0; i < PTBL_PAGES; i++) {
1138 		pa = pte_vatopa(mmu, kernel_pmap,
1139 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1140 		m = PHYS_TO_VM_PAGE(pa);
1141 		m->wire_count--;
1142 	}
1143 
1144 	/*
1145 	 * Free ptbl pages if there are no pte etries in this ptbl.
1146 	 * wire_count has the same value for all ptbl pages, so check the last
1147 	 * page.
1148 	 */
1149 	if (m->wire_count == 0) {
1150 		ptbl_free(mmu, pmap, pdir_idx);
1151 
1152 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1153 		return (1);
1154 	}
1155 
1156 	return (0);
1157 }
1158 
1159 /*
1160  * Increment hold count for ptbl pages. This routine is used when a new pte
1161  * entry is being inserted into the ptbl.
1162  */
1163 static void
1164 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1165 {
1166 	vm_paddr_t pa;
1167 	pte_t *ptbl;
1168 	vm_page_t m;
1169 	int i;
1170 
1171 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1172 	    pdir_idx);
1173 
1174 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1175 	    ("ptbl_hold: invalid pdir_idx"));
1176 	KASSERT((pmap != kernel_pmap),
1177 	    ("ptbl_hold: holding kernel ptbl!"));
1178 
1179 	ptbl = pmap->pm_pdir[pdir_idx];
1180 
1181 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1182 
1183 	for (i = 0; i < PTBL_PAGES; i++) {
1184 		pa = pte_vatopa(mmu, kernel_pmap,
1185 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1186 		m = PHYS_TO_VM_PAGE(pa);
1187 		m->wire_count++;
1188 	}
1189 }
1190 #endif
1191 
1192 /* Allocate pv_entry structure. */
1193 pv_entry_t
1194 pv_alloc(void)
1195 {
1196 	pv_entry_t pv;
1197 
1198 	pv_entry_count++;
1199 	if (pv_entry_count > pv_entry_high_water)
1200 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1201 	pv = uma_zalloc(pvzone, M_NOWAIT);
1202 
1203 	return (pv);
1204 }
1205 
1206 /* Free pv_entry structure. */
1207 static __inline void
1208 pv_free(pv_entry_t pve)
1209 {
1210 
1211 	pv_entry_count--;
1212 	uma_zfree(pvzone, pve);
1213 }
1214 
1215 
1216 /* Allocate and initialize pv_entry structure. */
1217 static void
1218 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1219 {
1220 	pv_entry_t pve;
1221 
1222 	//int su = (pmap == kernel_pmap);
1223 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1224 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1225 
1226 	pve = pv_alloc();
1227 	if (pve == NULL)
1228 		panic("pv_insert: no pv entries!");
1229 
1230 	pve->pv_pmap = pmap;
1231 	pve->pv_va = va;
1232 
1233 	/* add to pv_list */
1234 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1235 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1236 
1237 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1238 
1239 	//debugf("pv_insert: e\n");
1240 }
1241 
1242 /* Destroy pv entry. */
1243 static void
1244 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1245 {
1246 	pv_entry_t pve;
1247 
1248 	//int su = (pmap == kernel_pmap);
1249 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1250 
1251 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1252 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1253 
1254 	/* find pv entry */
1255 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1256 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1257 			/* remove from pv_list */
1258 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1259 			if (TAILQ_EMPTY(&m->md.pv_list))
1260 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1261 
1262 			/* free pv entry struct */
1263 			pv_free(pve);
1264 			break;
1265 		}
1266 	}
1267 
1268 	//debugf("pv_remove: e\n");
1269 }
1270 
1271 #ifdef __powerpc64__
1272 /*
1273  * Clean pte entry, try to free page table page if requested.
1274  *
1275  * Return 1 if ptbl pages were freed, otherwise return 0.
1276  */
1277 static int
1278 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1279 {
1280 	vm_page_t	m;
1281 	pte_t          *pte;
1282 
1283 	pte = pte_find(mmu, pmap, va);
1284 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1285 
1286 	if (!PTE_ISVALID(pte))
1287 		return (0);
1288 
1289 	/* Get vm_page_t for mapped pte. */
1290 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1291 
1292 	if (PTE_ISWIRED(pte))
1293 		pmap->pm_stats.wired_count--;
1294 
1295 	/* Handle managed entry. */
1296 	if (PTE_ISMANAGED(pte)) {
1297 
1298 		/* Handle modified pages. */
1299 		if (PTE_ISMODIFIED(pte))
1300 			vm_page_dirty(m);
1301 
1302 		/* Referenced pages. */
1303 		if (PTE_ISREFERENCED(pte))
1304 			vm_page_aflag_set(m, PGA_REFERENCED);
1305 
1306 		/* Remove pv_entry from pv_list. */
1307 		pv_remove(pmap, va, m);
1308 	} else if (m->md.pv_tracked) {
1309 		pv_remove(pmap, va, m);
1310 		if (TAILQ_EMPTY(&m->md.pv_list))
1311 			m->md.pv_tracked = false;
1312 	}
1313 	mtx_lock_spin(&tlbivax_mutex);
1314 	tlb_miss_lock();
1315 
1316 	tlb0_flush_entry(va);
1317 	*pte = 0;
1318 
1319 	tlb_miss_unlock();
1320 	mtx_unlock_spin(&tlbivax_mutex);
1321 
1322 	pmap->pm_stats.resident_count--;
1323 
1324 	if (flags & PTBL_UNHOLD) {
1325 		return (ptbl_unhold(mmu, pmap, va));
1326 	}
1327 	return (0);
1328 }
1329 
1330 /*
1331  * allocate a page of pointers to page directories, do not preallocate the
1332  * page tables
1333  */
1334 static pte_t  **
1335 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
1336 {
1337 	vm_page_t	mtbl  [PDIR_PAGES];
1338 	vm_page_t	m;
1339 	struct ptbl_buf *pbuf;
1340 	pte_t         **pdir;
1341 	unsigned int	pidx;
1342 	int		i;
1343 	int		req;
1344 
1345 	pbuf = ptbl_buf_alloc();
1346 
1347 	if (pbuf == NULL)
1348 		panic("%s: couldn't alloc kernel virtual memory", __func__);
1349 
1350 	/* Allocate pdir pages, this will sleep! */
1351 	for (i = 0; i < PDIR_PAGES; i++) {
1352 		pidx = (PDIR_PAGES * pp2d_idx) + i;
1353 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
1354 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
1355 			PMAP_UNLOCK(pmap);
1356 			vm_wait(NULL);
1357 			PMAP_LOCK(pmap);
1358 		}
1359 		mtbl[i] = m;
1360 	}
1361 
1362 	/* Mapin allocated pages into kernel_pmap. */
1363 	pdir = (pte_t **) pbuf->kva;
1364 	pmap_qenter((vm_offset_t) pdir, mtbl, PDIR_PAGES);
1365 
1366 	/* Zero whole pdir. */
1367 	bzero((caddr_t) pdir, PDIR_PAGES * PAGE_SIZE);
1368 
1369 	/* Add pdir to the pmap pdir bufs list. */
1370 	TAILQ_INSERT_TAIL(&pmap->pm_pdir_list, pbuf, link);
1371 
1372 	return pdir;
1373 }
1374 
1375 /*
1376  * Insert PTE for a given page and virtual address.
1377  */
1378 static int
1379 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1380     boolean_t nosleep)
1381 {
1382 	unsigned int	pp2d_idx = PP2D_IDX(va);
1383 	unsigned int	pdir_idx = PDIR_IDX(va);
1384 	unsigned int	ptbl_idx = PTBL_IDX(va);
1385 	pte_t          *ptbl, *pte, pte_tmp;
1386 	pte_t         **pdir;
1387 
1388 	/* Get the page directory pointer. */
1389 	pdir = pmap->pm_pp2d[pp2d_idx];
1390 	if (pdir == NULL)
1391 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1392 
1393 	/* Get the page table pointer. */
1394 	ptbl = pdir[pdir_idx];
1395 
1396 	if (ptbl == NULL) {
1397 		/* Allocate page table pages. */
1398 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1399 		if (ptbl == NULL) {
1400 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1401 			return (ENOMEM);
1402 		}
1403 		pte = &ptbl[ptbl_idx];
1404 	} else {
1405 		/*
1406 		 * Check if there is valid mapping for requested va, if there
1407 		 * is, remove it.
1408 		 */
1409 		pte = &ptbl[ptbl_idx];
1410 		if (PTE_ISVALID(pte)) {
1411 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1412 		} else {
1413 			/*
1414 			 * pte is not used, increment hold count for ptbl
1415 			 * pages.
1416 			 */
1417 			if (pmap != kernel_pmap)
1418 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1419 		}
1420 	}
1421 
1422 	if (pdir[pdir_idx] == NULL) {
1423 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1424 			pdir_hold(mmu, pmap, pdir);
1425 		pdir[pdir_idx] = ptbl;
1426 	}
1427 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1428 		pmap->pm_pp2d[pp2d_idx] = pdir;
1429 
1430 	/*
1431 	 * Insert pv_entry into pv_list for mapped page if part of managed
1432 	 * memory.
1433 	 */
1434 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1435 		flags |= PTE_MANAGED;
1436 
1437 		/* Create and insert pv entry. */
1438 		pv_insert(pmap, va, m);
1439 	}
1440 
1441 	pmap->pm_stats.resident_count++;
1442 
1443 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1444 	pte_tmp |= (PTE_VALID | flags);
1445 
1446 	mtx_lock_spin(&tlbivax_mutex);
1447 	tlb_miss_lock();
1448 
1449 	tlb0_flush_entry(va);
1450 	*pte = pte_tmp;
1451 
1452 	tlb_miss_unlock();
1453 	mtx_unlock_spin(&tlbivax_mutex);
1454 
1455 	return (0);
1456 }
1457 
1458 /* Return the pa for the given pmap/va. */
1459 static	vm_paddr_t
1460 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1461 {
1462 	vm_paddr_t	pa = 0;
1463 	pte_t          *pte;
1464 
1465 	pte = pte_find(mmu, pmap, va);
1466 	if ((pte != NULL) && PTE_ISVALID(pte))
1467 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1468 	return (pa);
1469 }
1470 
1471 
1472 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1473 static void
1474 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1475 {
1476 	int		i, j;
1477 	vm_offset_t	va;
1478 	pte_t		*pte;
1479 
1480 	va = addr;
1481 	/* Initialize kernel pdir */
1482 	for (i = 0; i < kernel_pdirs; i++) {
1483 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1484 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1485 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1486 		    j < PDIR_NENTRIES; j++) {
1487 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1488 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE * PDIR_PAGES) +
1489 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE * PTBL_PAGES));
1490 		}
1491 	}
1492 
1493 	/*
1494 	 * Fill in PTEs covering kernel code and data. They are not required
1495 	 * for address translation, as this area is covered by static TLB1
1496 	 * entries, but for pte_vatopa() to work correctly with kernel area
1497 	 * addresses.
1498 	 */
1499 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1500 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1501 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1502 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1503 		    PTE_VALID | PTE_PS_4KB;
1504 	}
1505 }
1506 #else
1507 /*
1508  * Clean pte entry, try to free page table page if requested.
1509  *
1510  * Return 1 if ptbl pages were freed, otherwise return 0.
1511  */
1512 static int
1513 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1514 {
1515 	unsigned int pdir_idx = PDIR_IDX(va);
1516 	unsigned int ptbl_idx = PTBL_IDX(va);
1517 	vm_page_t m;
1518 	pte_t *ptbl;
1519 	pte_t *pte;
1520 
1521 	//int su = (pmap == kernel_pmap);
1522 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1523 	//		su, (u_int32_t)pmap, va, flags);
1524 
1525 	ptbl = pmap->pm_pdir[pdir_idx];
1526 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1527 
1528 	pte = &ptbl[ptbl_idx];
1529 
1530 	if (pte == NULL || !PTE_ISVALID(pte))
1531 		return (0);
1532 
1533 	if (PTE_ISWIRED(pte))
1534 		pmap->pm_stats.wired_count--;
1535 
1536 	/* Get vm_page_t for mapped pte. */
1537 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1538 
1539 	/* Handle managed entry. */
1540 	if (PTE_ISMANAGED(pte)) {
1541 
1542 		if (PTE_ISMODIFIED(pte))
1543 			vm_page_dirty(m);
1544 
1545 		if (PTE_ISREFERENCED(pte))
1546 			vm_page_aflag_set(m, PGA_REFERENCED);
1547 
1548 		pv_remove(pmap, va, m);
1549 	} else if (m->md.pv_tracked) {
1550 		/*
1551 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1552 		 * used.  This is needed by the NCSW support code for fast
1553 		 * VA<->PA translation.
1554 		 */
1555 		pv_remove(pmap, va, m);
1556 		if (TAILQ_EMPTY(&m->md.pv_list))
1557 			m->md.pv_tracked = false;
1558 	}
1559 
1560 	mtx_lock_spin(&tlbivax_mutex);
1561 	tlb_miss_lock();
1562 
1563 	tlb0_flush_entry(va);
1564 	*pte = 0;
1565 
1566 	tlb_miss_unlock();
1567 	mtx_unlock_spin(&tlbivax_mutex);
1568 
1569 	pmap->pm_stats.resident_count--;
1570 
1571 	if (flags & PTBL_UNHOLD) {
1572 		//debugf("pte_remove: e (unhold)\n");
1573 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1574 	}
1575 
1576 	//debugf("pte_remove: e\n");
1577 	return (0);
1578 }
1579 
1580 /*
1581  * Insert PTE for a given page and virtual address.
1582  */
1583 static int
1584 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1585     boolean_t nosleep)
1586 {
1587 	unsigned int pdir_idx = PDIR_IDX(va);
1588 	unsigned int ptbl_idx = PTBL_IDX(va);
1589 	pte_t *ptbl, *pte, pte_tmp;
1590 
1591 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1592 	    pmap == kernel_pmap, pmap, va);
1593 
1594 	/* Get the page table pointer. */
1595 	ptbl = pmap->pm_pdir[pdir_idx];
1596 
1597 	if (ptbl == NULL) {
1598 		/* Allocate page table pages. */
1599 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1600 		if (ptbl == NULL) {
1601 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1602 			return (ENOMEM);
1603 		}
1604 		pmap->pm_pdir[pdir_idx] = ptbl;
1605 		pte = &ptbl[ptbl_idx];
1606 	} else {
1607 		/*
1608 		 * Check if there is valid mapping for requested
1609 		 * va, if there is, remove it.
1610 		 */
1611 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1612 		if (PTE_ISVALID(pte)) {
1613 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1614 		} else {
1615 			/*
1616 			 * pte is not used, increment hold count
1617 			 * for ptbl pages.
1618 			 */
1619 			if (pmap != kernel_pmap)
1620 				ptbl_hold(mmu, pmap, pdir_idx);
1621 		}
1622 	}
1623 
1624 	/*
1625 	 * Insert pv_entry into pv_list for mapped page if part of managed
1626 	 * memory.
1627 	 */
1628 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1629 		flags |= PTE_MANAGED;
1630 
1631 		/* Create and insert pv entry. */
1632 		pv_insert(pmap, va, m);
1633 	}
1634 
1635 	pmap->pm_stats.resident_count++;
1636 
1637 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1638 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1639 
1640 	mtx_lock_spin(&tlbivax_mutex);
1641 	tlb_miss_lock();
1642 
1643 	tlb0_flush_entry(va);
1644 	*pte = pte_tmp;
1645 
1646 	tlb_miss_unlock();
1647 	mtx_unlock_spin(&tlbivax_mutex);
1648 	return (0);
1649 }
1650 
1651 /* Return the pa for the given pmap/va. */
1652 static vm_paddr_t
1653 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1654 {
1655 	vm_paddr_t pa = 0;
1656 	pte_t *pte;
1657 
1658 	pte = pte_find(mmu, pmap, va);
1659 	if ((pte != NULL) && PTE_ISVALID(pte))
1660 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1661 	return (pa);
1662 }
1663 
1664 /* Get a pointer to a PTE in a page table. */
1665 static pte_t *
1666 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1667 {
1668 	unsigned int pdir_idx = PDIR_IDX(va);
1669 	unsigned int ptbl_idx = PTBL_IDX(va);
1670 
1671 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1672 
1673 	if (pmap->pm_pdir[pdir_idx])
1674 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1675 
1676 	return (NULL);
1677 }
1678 
1679 /* Set up kernel page tables. */
1680 static void
1681 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1682 {
1683 	int		i;
1684 	vm_offset_t	va;
1685 	pte_t		*pte;
1686 
1687 	/* Initialize kernel pdir */
1688 	for (i = 0; i < kernel_ptbls; i++)
1689 		kernel_pmap->pm_pdir[kptbl_min + i] =
1690 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1691 
1692 	/*
1693 	 * Fill in PTEs covering kernel code and data. They are not required
1694 	 * for address translation, as this area is covered by static TLB1
1695 	 * entries, but for pte_vatopa() to work correctly with kernel area
1696 	 * addresses.
1697 	 */
1698 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1699 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1700 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1701 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1702 		    PTE_VALID | PTE_PS_4KB;
1703 	}
1704 }
1705 #endif
1706 
1707 /**************************************************************************/
1708 /* PMAP related */
1709 /**************************************************************************/
1710 
1711 /*
1712  * This is called during booke_init, before the system is really initialized.
1713  */
1714 static void
1715 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1716 {
1717 	vm_paddr_t phys_kernelend;
1718 	struct mem_region *mp, *mp1;
1719 	int cnt, i, j;
1720 	vm_paddr_t s, e, sz;
1721 	vm_paddr_t physsz, hwphyssz;
1722 	u_int phys_avail_count;
1723 	vm_size_t kstack0_sz;
1724 	vm_offset_t kernel_pdir, kstack0;
1725 	vm_paddr_t kstack0_phys;
1726 	void *dpcpu;
1727 
1728 	debugf("mmu_booke_bootstrap: entered\n");
1729 
1730 	/* Set interesting system properties */
1731 #ifdef __powerpc64__
1732 	hw_direct_map = 1;
1733 #else
1734 	hw_direct_map = 0;
1735 #endif
1736 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1737 	elf32_nxstack = 1;
1738 #endif
1739 
1740 	/* Initialize invalidation mutex */
1741 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1742 
1743 	/* Read TLB0 size and associativity. */
1744 	tlb0_get_tlbconf();
1745 
1746 	/*
1747 	 * Align kernel start and end address (kernel image).
1748 	 * Note that kernel end does not necessarily relate to kernsize.
1749 	 * kernsize is the size of the kernel that is actually mapped.
1750 	 */
1751 	kernstart = trunc_page(start);
1752 	data_start = round_page(kernelend);
1753 	data_end = data_start;
1754 
1755 	/* Allocate the dynamic per-cpu area. */
1756 	dpcpu = (void *)data_end;
1757 	data_end += DPCPU_SIZE;
1758 
1759 	/* Allocate space for the message buffer. */
1760 	msgbufp = (struct msgbuf *)data_end;
1761 	data_end += msgbufsize;
1762 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1763 	    (uintptr_t)msgbufp, data_end);
1764 
1765 	data_end = round_page(data_end);
1766 
1767 	/* Allocate space for ptbl_bufs. */
1768 	ptbl_bufs = (struct ptbl_buf *)data_end;
1769 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1770 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1771 	    (uintptr_t)ptbl_bufs, data_end);
1772 
1773 	data_end = round_page(data_end);
1774 
1775 	/* Allocate PTE tables for kernel KVA. */
1776 	kernel_pdir = data_end;
1777 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1778 	    PDIR_SIZE);
1779 #ifdef __powerpc64__
1780 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1781 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1782 #endif
1783 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1784 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1785 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1786 	    kernel_pdir, data_end);
1787 
1788 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1789 	if (data_end - kernstart > kernsize) {
1790 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1791 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1792 	}
1793 	data_end = kernstart + kernsize;
1794 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1795 
1796 	/*
1797 	 * Clear the structures - note we can only do it safely after the
1798 	 * possible additional TLB1 translations are in place (above) so that
1799 	 * all range up to the currently calculated 'data_end' is covered.
1800 	 */
1801 	dpcpu_init(dpcpu, 0);
1802 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1803 #ifdef __powerpc64__
1804 	memset((void *)kernel_pdir, 0,
1805 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1806 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1807 #else
1808 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1809 #endif
1810 
1811 	/*******************************************************/
1812 	/* Set the start and end of kva. */
1813 	/*******************************************************/
1814 	virtual_avail = round_page(data_end);
1815 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1816 
1817 	/* Allocate KVA space for page zero/copy operations. */
1818 	zero_page_va = virtual_avail;
1819 	virtual_avail += PAGE_SIZE;
1820 	copy_page_src_va = virtual_avail;
1821 	virtual_avail += PAGE_SIZE;
1822 	copy_page_dst_va = virtual_avail;
1823 	virtual_avail += PAGE_SIZE;
1824 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
1825 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
1826 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
1827 
1828 	/* Initialize page zero/copy mutexes. */
1829 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1830 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1831 
1832 	/* Allocate KVA space for ptbl bufs. */
1833 	ptbl_buf_pool_vabase = virtual_avail;
1834 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1835 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1836 	    ptbl_buf_pool_vabase, virtual_avail);
1837 
1838 	/* Calculate corresponding physical addresses for the kernel region. */
1839 	phys_kernelend = kernload + kernsize;
1840 	debugf("kernel image and allocated data:\n");
1841 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1842 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
1843 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
1844 
1845 	/*
1846 	 * Remove kernel physical address range from avail regions list. Page
1847 	 * align all regions.  Non-page aligned memory isn't very interesting
1848 	 * to us.  Also, sort the entries for ascending addresses.
1849 	 */
1850 
1851 	/* Retrieve phys/avail mem regions */
1852 	mem_regions(&physmem_regions, &physmem_regions_sz,
1853 	    &availmem_regions, &availmem_regions_sz);
1854 
1855 	if (nitems(phys_avail) < availmem_regions_sz)
1856 		panic("mmu_booke_bootstrap: phys_avail too small");
1857 
1858 	sz = 0;
1859 	cnt = availmem_regions_sz;
1860 	debugf("processing avail regions:\n");
1861 	for (mp = availmem_regions; mp->mr_size; mp++) {
1862 		s = mp->mr_start;
1863 		e = mp->mr_start + mp->mr_size;
1864 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1865 		/* Check whether this region holds all of the kernel. */
1866 		if (s < kernload && e > phys_kernelend) {
1867 			availmem_regions[cnt].mr_start = phys_kernelend;
1868 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1869 			e = kernload;
1870 		}
1871 		/* Look whether this regions starts within the kernel. */
1872 		if (s >= kernload && s < phys_kernelend) {
1873 			if (e <= phys_kernelend)
1874 				goto empty;
1875 			s = phys_kernelend;
1876 		}
1877 		/* Now look whether this region ends within the kernel. */
1878 		if (e > kernload && e <= phys_kernelend) {
1879 			if (s >= kernload)
1880 				goto empty;
1881 			e = kernload;
1882 		}
1883 		/* Now page align the start and size of the region. */
1884 		s = round_page(s);
1885 		e = trunc_page(e);
1886 		if (e < s)
1887 			e = s;
1888 		sz = e - s;
1889 		debugf("%09jx-%09jx = %jx\n",
1890 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1891 
1892 		/* Check whether some memory is left here. */
1893 		if (sz == 0) {
1894 		empty:
1895 			memmove(mp, mp + 1,
1896 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1897 			cnt--;
1898 			mp--;
1899 			continue;
1900 		}
1901 
1902 		/* Do an insertion sort. */
1903 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1904 			if (s < mp1->mr_start)
1905 				break;
1906 		if (mp1 < mp) {
1907 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1908 			mp1->mr_start = s;
1909 			mp1->mr_size = sz;
1910 		} else {
1911 			mp->mr_start = s;
1912 			mp->mr_size = sz;
1913 		}
1914 	}
1915 	availmem_regions_sz = cnt;
1916 
1917 	/*******************************************************/
1918 	/* Steal physical memory for kernel stack from the end */
1919 	/* of the first avail region                           */
1920 	/*******************************************************/
1921 	kstack0_sz = kstack_pages * PAGE_SIZE;
1922 	kstack0_phys = availmem_regions[0].mr_start +
1923 	    availmem_regions[0].mr_size;
1924 	kstack0_phys -= kstack0_sz;
1925 	availmem_regions[0].mr_size -= kstack0_sz;
1926 
1927 	/*******************************************************/
1928 	/* Fill in phys_avail table, based on availmem_regions */
1929 	/*******************************************************/
1930 	phys_avail_count = 0;
1931 	physsz = 0;
1932 	hwphyssz = 0;
1933 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1934 
1935 	debugf("fill in phys_avail:\n");
1936 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1937 
1938 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1939 		    (uintmax_t)availmem_regions[i].mr_start,
1940 		    (uintmax_t)availmem_regions[i].mr_start +
1941 		        availmem_regions[i].mr_size,
1942 		    (uintmax_t)availmem_regions[i].mr_size);
1943 
1944 		if (hwphyssz != 0 &&
1945 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1946 			debugf(" hw.physmem adjust\n");
1947 			if (physsz < hwphyssz) {
1948 				phys_avail[j] = availmem_regions[i].mr_start;
1949 				phys_avail[j + 1] =
1950 				    availmem_regions[i].mr_start +
1951 				    hwphyssz - physsz;
1952 				physsz = hwphyssz;
1953 				phys_avail_count++;
1954 			}
1955 			break;
1956 		}
1957 
1958 		phys_avail[j] = availmem_regions[i].mr_start;
1959 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1960 		    availmem_regions[i].mr_size;
1961 		phys_avail_count++;
1962 		physsz += availmem_regions[i].mr_size;
1963 	}
1964 	physmem = btoc(physsz);
1965 
1966 	/* Calculate the last available physical address. */
1967 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1968 		;
1969 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1970 
1971 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1972 	debugf("phys_avail_count = %d\n", phys_avail_count);
1973 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1974 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1975 
1976 #ifdef __powerpc64__
1977 	/*
1978 	 * Map the physical memory contiguously in TLB1.
1979 	 * Round so it fits into a single mapping.
1980 	 */
1981 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
1982 	    phys_avail[i + 1]);
1983 #endif
1984 
1985 	/*******************************************************/
1986 	/* Initialize (statically allocated) kernel pmap. */
1987 	/*******************************************************/
1988 	PMAP_LOCK_INIT(kernel_pmap);
1989 #ifndef __powerpc64__
1990 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1991 #endif
1992 
1993 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1994 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1995 	for (i = 0; i < MAXCPU; i++) {
1996 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1997 
1998 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1999 		tidbusy[i][TID_KERNEL] = kernel_pmap;
2000 	}
2001 
2002 	/* Mark kernel_pmap active on all CPUs */
2003 	CPU_FILL(&kernel_pmap->pm_active);
2004 
2005  	/*
2006 	 * Initialize the global pv list lock.
2007 	 */
2008 	rw_init(&pvh_global_lock, "pmap pv global");
2009 
2010 	/*******************************************************/
2011 	/* Final setup */
2012 	/*******************************************************/
2013 
2014 	/* Enter kstack0 into kernel map, provide guard page */
2015 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
2016 	thread0.td_kstack = kstack0;
2017 	thread0.td_kstack_pages = kstack_pages;
2018 
2019 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
2020 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
2021 	    kstack0_phys, kstack0_phys + kstack0_sz);
2022 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
2023 	    kstack0, kstack0 + kstack0_sz);
2024 
2025 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
2026 	for (i = 0; i < kstack_pages; i++) {
2027 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
2028 		kstack0 += PAGE_SIZE;
2029 		kstack0_phys += PAGE_SIZE;
2030 	}
2031 
2032 	pmap_bootstrapped = 1;
2033 
2034 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
2035 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
2036 
2037 	debugf("mmu_booke_bootstrap: exit\n");
2038 }
2039 
2040 #ifdef SMP
2041  void
2042 tlb1_ap_prep(void)
2043 {
2044 	tlb_entry_t *e, tmp;
2045 	unsigned int i;
2046 
2047 	/* Prepare TLB1 image for AP processors */
2048 	e = __boot_tlb1;
2049 	for (i = 0; i < TLB1_ENTRIES; i++) {
2050 		tlb1_read_entry(&tmp, i);
2051 
2052 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
2053 			memcpy(e++, &tmp, sizeof(tmp));
2054 	}
2055 }
2056 
2057 void
2058 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
2059 {
2060 	int i;
2061 
2062 	/*
2063 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
2064 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
2065 	 * created by tlb1_ap_prep(), so use these values directly to
2066 	 * (re)program AP's TLB1 hardware.
2067 	 *
2068 	 * Start at index 1 because index 0 has the kernel map.
2069 	 */
2070 	for (i = 1; i < TLB1_ENTRIES; i++) {
2071 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
2072 			tlb1_write_entry(&__boot_tlb1[i], i);
2073 	}
2074 
2075 	set_mas4_defaults();
2076 }
2077 #endif
2078 
2079 static void
2080 booke_pmap_init_qpages(void)
2081 {
2082 	struct pcpu *pc;
2083 	int i;
2084 
2085 	CPU_FOREACH(i) {
2086 		pc = pcpu_find(i);
2087 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
2088 		if (pc->pc_qmap_addr == 0)
2089 			panic("pmap_init_qpages: unable to allocate KVA");
2090 	}
2091 }
2092 
2093 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
2094 
2095 /*
2096  * Get the physical page address for the given pmap/virtual address.
2097  */
2098 static vm_paddr_t
2099 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
2100 {
2101 	vm_paddr_t pa;
2102 
2103 	PMAP_LOCK(pmap);
2104 	pa = pte_vatopa(mmu, pmap, va);
2105 	PMAP_UNLOCK(pmap);
2106 
2107 	return (pa);
2108 }
2109 
2110 /*
2111  * Extract the physical page address associated with the given
2112  * kernel virtual address.
2113  */
2114 static vm_paddr_t
2115 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
2116 {
2117 	tlb_entry_t e;
2118 	vm_paddr_t p = 0;
2119 	int i;
2120 
2121 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
2122 		p = pte_vatopa(mmu, kernel_pmap, va);
2123 
2124 	if (p == 0) {
2125 		/* Check TLB1 mappings */
2126 		for (i = 0; i < TLB1_ENTRIES; i++) {
2127 			tlb1_read_entry(&e, i);
2128 			if (!(e.mas1 & MAS1_VALID))
2129 				continue;
2130 			if (va >= e.virt && va < e.virt + e.size)
2131 				return (e.phys + (va - e.virt));
2132 		}
2133 	}
2134 
2135 	return (p);
2136 }
2137 
2138 /*
2139  * Initialize the pmap module.
2140  * Called by vm_init, to initialize any structures that the pmap
2141  * system needs to map virtual memory.
2142  */
2143 static void
2144 mmu_booke_init(mmu_t mmu)
2145 {
2146 	int shpgperproc = PMAP_SHPGPERPROC;
2147 
2148 	/*
2149 	 * Initialize the address space (zone) for the pv entries.  Set a
2150 	 * high water mark so that the system can recover from excessive
2151 	 * numbers of pv entries.
2152 	 */
2153 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2154 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2155 
2156 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2157 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2158 
2159 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2160 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2161 
2162 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2163 
2164 	/* Pre-fill pvzone with initial number of pv entries. */
2165 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2166 
2167 	/* Initialize ptbl allocation. */
2168 	ptbl_init();
2169 }
2170 
2171 /*
2172  * Map a list of wired pages into kernel virtual address space.  This is
2173  * intended for temporary mappings which do not need page modification or
2174  * references recorded.  Existing mappings in the region are overwritten.
2175  */
2176 static void
2177 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2178 {
2179 	vm_offset_t va;
2180 
2181 	va = sva;
2182 	while (count-- > 0) {
2183 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2184 		va += PAGE_SIZE;
2185 		m++;
2186 	}
2187 }
2188 
2189 /*
2190  * Remove page mappings from kernel virtual address space.  Intended for
2191  * temporary mappings entered by mmu_booke_qenter.
2192  */
2193 static void
2194 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2195 {
2196 	vm_offset_t va;
2197 
2198 	va = sva;
2199 	while (count-- > 0) {
2200 		mmu_booke_kremove(mmu, va);
2201 		va += PAGE_SIZE;
2202 	}
2203 }
2204 
2205 /*
2206  * Map a wired page into kernel virtual address space.
2207  */
2208 static void
2209 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2210 {
2211 
2212 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2213 }
2214 
2215 static void
2216 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2217 {
2218 	uint32_t flags;
2219 	pte_t *pte;
2220 
2221 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2222 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2223 
2224 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2225 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2226 	flags |= PTE_PS_4KB;
2227 
2228 	pte = pte_find(mmu, kernel_pmap, va);
2229 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2230 
2231 	mtx_lock_spin(&tlbivax_mutex);
2232 	tlb_miss_lock();
2233 
2234 	if (PTE_ISVALID(pte)) {
2235 
2236 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2237 
2238 		/* Flush entry from TLB0 */
2239 		tlb0_flush_entry(va);
2240 	}
2241 
2242 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2243 
2244 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2245 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2246 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2247 
2248 	/* Flush the real memory from the instruction cache. */
2249 	if ((flags & (PTE_I | PTE_G)) == 0)
2250 		__syncicache((void *)va, PAGE_SIZE);
2251 
2252 	tlb_miss_unlock();
2253 	mtx_unlock_spin(&tlbivax_mutex);
2254 }
2255 
2256 /*
2257  * Remove a page from kernel page table.
2258  */
2259 static void
2260 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2261 {
2262 	pte_t *pte;
2263 
2264 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
2265 
2266 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2267 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2268 	    ("mmu_booke_kremove: invalid va"));
2269 
2270 	pte = pte_find(mmu, kernel_pmap, va);
2271 
2272 	if (!PTE_ISVALID(pte)) {
2273 
2274 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2275 
2276 		return;
2277 	}
2278 
2279 	mtx_lock_spin(&tlbivax_mutex);
2280 	tlb_miss_lock();
2281 
2282 	/* Invalidate entry in TLB0, update PTE. */
2283 	tlb0_flush_entry(va);
2284 	*pte = 0;
2285 
2286 	tlb_miss_unlock();
2287 	mtx_unlock_spin(&tlbivax_mutex);
2288 }
2289 
2290 /*
2291  * Provide a kernel pointer corresponding to a given userland pointer.
2292  * The returned pointer is valid until the next time this function is
2293  * called in this thread. This is used internally in copyin/copyout.
2294  */
2295 int
2296 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2297     void **kaddr, size_t ulen, size_t *klen)
2298 {
2299 
2300 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2301 		return (EFAULT);
2302 
2303 	*kaddr = (void *)(uintptr_t)uaddr;
2304 	if (klen)
2305 		*klen = ulen;
2306 
2307 	return (0);
2308 }
2309 
2310 /*
2311  * Figure out where a given kernel pointer (usually in a fault) points
2312  * to from the VM's perspective, potentially remapping into userland's
2313  * address space.
2314  */
2315 static int
2316 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2317     vm_offset_t *decoded_addr)
2318 {
2319 
2320 	if (addr < VM_MAXUSER_ADDRESS)
2321 		*is_user = 1;
2322 	else
2323 		*is_user = 0;
2324 
2325 	*decoded_addr = addr;
2326 	return (0);
2327 }
2328 
2329 /*
2330  * Initialize pmap associated with process 0.
2331  */
2332 static void
2333 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2334 {
2335 
2336 	PMAP_LOCK_INIT(pmap);
2337 	mmu_booke_pinit(mmu, pmap);
2338 	PCPU_SET(curpmap, pmap);
2339 }
2340 
2341 /*
2342  * Initialize a preallocated and zeroed pmap structure,
2343  * such as one in a vmspace structure.
2344  */
2345 static void
2346 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2347 {
2348 	int i;
2349 
2350 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2351 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2352 
2353 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2354 
2355 	for (i = 0; i < MAXCPU; i++)
2356 		pmap->pm_tid[i] = TID_NONE;
2357 	CPU_ZERO(&kernel_pmap->pm_active);
2358 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2359 #ifdef __powerpc64__
2360 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2361 	TAILQ_INIT(&pmap->pm_pdir_list);
2362 #else
2363 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2364 #endif
2365 	TAILQ_INIT(&pmap->pm_ptbl_list);
2366 }
2367 
2368 /*
2369  * Release any resources held by the given physical map.
2370  * Called when a pmap initialized by mmu_booke_pinit is being released.
2371  * Should only be called if the map contains no valid mappings.
2372  */
2373 static void
2374 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2375 {
2376 
2377 	KASSERT(pmap->pm_stats.resident_count == 0,
2378 	    ("pmap_release: pmap resident count %ld != 0",
2379 	    pmap->pm_stats.resident_count));
2380 }
2381 
2382 /*
2383  * Insert the given physical page at the specified virtual address in the
2384  * target physical map with the protection requested. If specified the page
2385  * will be wired down.
2386  */
2387 static int
2388 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2389     vm_prot_t prot, u_int flags, int8_t psind)
2390 {
2391 	int error;
2392 
2393 	rw_wlock(&pvh_global_lock);
2394 	PMAP_LOCK(pmap);
2395 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2396 	PMAP_UNLOCK(pmap);
2397 	rw_wunlock(&pvh_global_lock);
2398 	return (error);
2399 }
2400 
2401 static int
2402 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2403     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2404 {
2405 	pte_t *pte;
2406 	vm_paddr_t pa;
2407 	uint32_t flags;
2408 	int error, su, sync;
2409 
2410 	pa = VM_PAGE_TO_PHYS(m);
2411 	su = (pmap == kernel_pmap);
2412 	sync = 0;
2413 
2414 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2415 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2416 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2417 	//		(u_int32_t)m, va, pa, prot, flags);
2418 
2419 	if (su) {
2420 		KASSERT(((va >= virtual_avail) &&
2421 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2422 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2423 	} else {
2424 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2425 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2426 	}
2427 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2428 		VM_OBJECT_ASSERT_LOCKED(m->object);
2429 
2430 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2431 
2432 	/*
2433 	 * If there is an existing mapping, and the physical address has not
2434 	 * changed, must be protection or wiring change.
2435 	 */
2436 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2437 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2438 
2439 		/*
2440 		 * Before actually updating pte->flags we calculate and
2441 		 * prepare its new value in a helper var.
2442 		 */
2443 		flags = *pte;
2444 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2445 
2446 		/* Wiring change, just update stats. */
2447 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2448 			if (!PTE_ISWIRED(pte)) {
2449 				flags |= PTE_WIRED;
2450 				pmap->pm_stats.wired_count++;
2451 			}
2452 		} else {
2453 			if (PTE_ISWIRED(pte)) {
2454 				flags &= ~PTE_WIRED;
2455 				pmap->pm_stats.wired_count--;
2456 			}
2457 		}
2458 
2459 		if (prot & VM_PROT_WRITE) {
2460 			/* Add write permissions. */
2461 			flags |= PTE_SW;
2462 			if (!su)
2463 				flags |= PTE_UW;
2464 
2465 			if ((flags & PTE_MANAGED) != 0)
2466 				vm_page_aflag_set(m, PGA_WRITEABLE);
2467 		} else {
2468 			/* Handle modified pages, sense modify status. */
2469 
2470 			/*
2471 			 * The PTE_MODIFIED flag could be set by underlying
2472 			 * TLB misses since we last read it (above), possibly
2473 			 * other CPUs could update it so we check in the PTE
2474 			 * directly rather than rely on that saved local flags
2475 			 * copy.
2476 			 */
2477 			if (PTE_ISMODIFIED(pte))
2478 				vm_page_dirty(m);
2479 		}
2480 
2481 		if (prot & VM_PROT_EXECUTE) {
2482 			flags |= PTE_SX;
2483 			if (!su)
2484 				flags |= PTE_UX;
2485 
2486 			/*
2487 			 * Check existing flags for execute permissions: if we
2488 			 * are turning execute permissions on, icache should
2489 			 * be flushed.
2490 			 */
2491 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2492 				sync++;
2493 		}
2494 
2495 		flags &= ~PTE_REFERENCED;
2496 
2497 		/*
2498 		 * The new flags value is all calculated -- only now actually
2499 		 * update the PTE.
2500 		 */
2501 		mtx_lock_spin(&tlbivax_mutex);
2502 		tlb_miss_lock();
2503 
2504 		tlb0_flush_entry(va);
2505 		*pte &= ~PTE_FLAGS_MASK;
2506 		*pte |= flags;
2507 
2508 		tlb_miss_unlock();
2509 		mtx_unlock_spin(&tlbivax_mutex);
2510 
2511 	} else {
2512 		/*
2513 		 * If there is an existing mapping, but it's for a different
2514 		 * physical address, pte_enter() will delete the old mapping.
2515 		 */
2516 		//if ((pte != NULL) && PTE_ISVALID(pte))
2517 		//	debugf("mmu_booke_enter_locked: replace\n");
2518 		//else
2519 		//	debugf("mmu_booke_enter_locked: new\n");
2520 
2521 		/* Now set up the flags and install the new mapping. */
2522 		flags = (PTE_SR | PTE_VALID);
2523 		flags |= PTE_M;
2524 
2525 		if (!su)
2526 			flags |= PTE_UR;
2527 
2528 		if (prot & VM_PROT_WRITE) {
2529 			flags |= PTE_SW;
2530 			if (!su)
2531 				flags |= PTE_UW;
2532 
2533 			if ((m->oflags & VPO_UNMANAGED) == 0)
2534 				vm_page_aflag_set(m, PGA_WRITEABLE);
2535 		}
2536 
2537 		if (prot & VM_PROT_EXECUTE) {
2538 			flags |= PTE_SX;
2539 			if (!su)
2540 				flags |= PTE_UX;
2541 		}
2542 
2543 		/* If its wired update stats. */
2544 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2545 			flags |= PTE_WIRED;
2546 
2547 		error = pte_enter(mmu, pmap, m, va, flags,
2548 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2549 		if (error != 0)
2550 			return (KERN_RESOURCE_SHORTAGE);
2551 
2552 		if ((flags & PMAP_ENTER_WIRED) != 0)
2553 			pmap->pm_stats.wired_count++;
2554 
2555 		/* Flush the real memory from the instruction cache. */
2556 		if (prot & VM_PROT_EXECUTE)
2557 			sync++;
2558 	}
2559 
2560 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2561 		__syncicache((void *)va, PAGE_SIZE);
2562 		sync = 0;
2563 	}
2564 
2565 	return (KERN_SUCCESS);
2566 }
2567 
2568 /*
2569  * Maps a sequence of resident pages belonging to the same object.
2570  * The sequence begins with the given page m_start.  This page is
2571  * mapped at the given virtual address start.  Each subsequent page is
2572  * mapped at a virtual address that is offset from start by the same
2573  * amount as the page is offset from m_start within the object.  The
2574  * last page in the sequence is the page with the largest offset from
2575  * m_start that can be mapped at a virtual address less than the given
2576  * virtual address end.  Not every virtual page between start and end
2577  * is mapped; only those for which a resident page exists with the
2578  * corresponding offset from m_start are mapped.
2579  */
2580 static void
2581 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2582     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2583 {
2584 	vm_page_t m;
2585 	vm_pindex_t diff, psize;
2586 
2587 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2588 
2589 	psize = atop(end - start);
2590 	m = m_start;
2591 	rw_wlock(&pvh_global_lock);
2592 	PMAP_LOCK(pmap);
2593 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2594 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2595 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2596 		    PMAP_ENTER_NOSLEEP, 0);
2597 		m = TAILQ_NEXT(m, listq);
2598 	}
2599 	rw_wunlock(&pvh_global_lock);
2600 	PMAP_UNLOCK(pmap);
2601 }
2602 
2603 static void
2604 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2605     vm_prot_t prot)
2606 {
2607 
2608 	rw_wlock(&pvh_global_lock);
2609 	PMAP_LOCK(pmap);
2610 	mmu_booke_enter_locked(mmu, pmap, va, m,
2611 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2612 	    0);
2613 	rw_wunlock(&pvh_global_lock);
2614 	PMAP_UNLOCK(pmap);
2615 }
2616 
2617 /*
2618  * Remove the given range of addresses from the specified map.
2619  *
2620  * It is assumed that the start and end are properly rounded to the page size.
2621  */
2622 static void
2623 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2624 {
2625 	pte_t *pte;
2626 	uint8_t hold_flag;
2627 
2628 	int su = (pmap == kernel_pmap);
2629 
2630 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2631 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2632 
2633 	if (su) {
2634 		KASSERT(((va >= virtual_avail) &&
2635 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2636 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2637 	} else {
2638 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2639 		    ("mmu_booke_remove: user pmap, non user va"));
2640 	}
2641 
2642 	if (PMAP_REMOVE_DONE(pmap)) {
2643 		//debugf("mmu_booke_remove: e (empty)\n");
2644 		return;
2645 	}
2646 
2647 	hold_flag = PTBL_HOLD_FLAG(pmap);
2648 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2649 
2650 	rw_wlock(&pvh_global_lock);
2651 	PMAP_LOCK(pmap);
2652 	for (; va < endva; va += PAGE_SIZE) {
2653 		pte = pte_find(mmu, pmap, va);
2654 		if ((pte != NULL) && PTE_ISVALID(pte))
2655 			pte_remove(mmu, pmap, va, hold_flag);
2656 	}
2657 	PMAP_UNLOCK(pmap);
2658 	rw_wunlock(&pvh_global_lock);
2659 
2660 	//debugf("mmu_booke_remove: e\n");
2661 }
2662 
2663 /*
2664  * Remove physical page from all pmaps in which it resides.
2665  */
2666 static void
2667 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2668 {
2669 	pv_entry_t pv, pvn;
2670 	uint8_t hold_flag;
2671 
2672 	rw_wlock(&pvh_global_lock);
2673 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2674 		pvn = TAILQ_NEXT(pv, pv_link);
2675 
2676 		PMAP_LOCK(pv->pv_pmap);
2677 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2678 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2679 		PMAP_UNLOCK(pv->pv_pmap);
2680 	}
2681 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2682 	rw_wunlock(&pvh_global_lock);
2683 }
2684 
2685 /*
2686  * Map a range of physical addresses into kernel virtual address space.
2687  */
2688 static vm_offset_t
2689 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2690     vm_paddr_t pa_end, int prot)
2691 {
2692 	vm_offset_t sva = *virt;
2693 	vm_offset_t va = sva;
2694 
2695 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2696 	//		sva, pa_start, pa_end);
2697 
2698 	while (pa_start < pa_end) {
2699 		mmu_booke_kenter(mmu, va, pa_start);
2700 		va += PAGE_SIZE;
2701 		pa_start += PAGE_SIZE;
2702 	}
2703 	*virt = va;
2704 
2705 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2706 	return (sva);
2707 }
2708 
2709 /*
2710  * The pmap must be activated before it's address space can be accessed in any
2711  * way.
2712  */
2713 static void
2714 mmu_booke_activate(mmu_t mmu, struct thread *td)
2715 {
2716 	pmap_t pmap;
2717 	u_int cpuid;
2718 
2719 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2720 
2721 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
2722 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2723 
2724 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2725 
2726 	sched_pin();
2727 
2728 	cpuid = PCPU_GET(cpuid);
2729 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2730 	PCPU_SET(curpmap, pmap);
2731 
2732 	if (pmap->pm_tid[cpuid] == TID_NONE)
2733 		tid_alloc(pmap);
2734 
2735 	/* Load PID0 register with pmap tid value. */
2736 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2737 	__asm __volatile("isync");
2738 
2739 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2740 
2741 	sched_unpin();
2742 
2743 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2744 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2745 }
2746 
2747 /*
2748  * Deactivate the specified process's address space.
2749  */
2750 static void
2751 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2752 {
2753 	pmap_t pmap;
2754 
2755 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2756 
2757 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
2758 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2759 
2760 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2761 
2762 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2763 	PCPU_SET(curpmap, NULL);
2764 }
2765 
2766 /*
2767  * Copy the range specified by src_addr/len
2768  * from the source map to the range dst_addr/len
2769  * in the destination map.
2770  *
2771  * This routine is only advisory and need not do anything.
2772  */
2773 static void
2774 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2775     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2776 {
2777 
2778 }
2779 
2780 /*
2781  * Set the physical protection on the specified range of this map as requested.
2782  */
2783 static void
2784 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2785     vm_prot_t prot)
2786 {
2787 	vm_offset_t va;
2788 	vm_page_t m;
2789 	pte_t *pte;
2790 
2791 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2792 		mmu_booke_remove(mmu, pmap, sva, eva);
2793 		return;
2794 	}
2795 
2796 	if (prot & VM_PROT_WRITE)
2797 		return;
2798 
2799 	PMAP_LOCK(pmap);
2800 	for (va = sva; va < eva; va += PAGE_SIZE) {
2801 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2802 			if (PTE_ISVALID(pte)) {
2803 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2804 
2805 				mtx_lock_spin(&tlbivax_mutex);
2806 				tlb_miss_lock();
2807 
2808 				/* Handle modified pages. */
2809 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2810 					vm_page_dirty(m);
2811 
2812 				tlb0_flush_entry(va);
2813 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2814 
2815 				tlb_miss_unlock();
2816 				mtx_unlock_spin(&tlbivax_mutex);
2817 			}
2818 		}
2819 	}
2820 	PMAP_UNLOCK(pmap);
2821 }
2822 
2823 /*
2824  * Clear the write and modified bits in each of the given page's mappings.
2825  */
2826 static void
2827 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2828 {
2829 	pv_entry_t pv;
2830 	pte_t *pte;
2831 
2832 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2833 	    ("mmu_booke_remove_write: page %p is not managed", m));
2834 
2835 	/*
2836 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2837 	 * set by another thread while the object is locked.  Thus,
2838 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2839 	 */
2840 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2841 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2842 		return;
2843 	rw_wlock(&pvh_global_lock);
2844 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2845 		PMAP_LOCK(pv->pv_pmap);
2846 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2847 			if (PTE_ISVALID(pte)) {
2848 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2849 
2850 				mtx_lock_spin(&tlbivax_mutex);
2851 				tlb_miss_lock();
2852 
2853 				/* Handle modified pages. */
2854 				if (PTE_ISMODIFIED(pte))
2855 					vm_page_dirty(m);
2856 
2857 				/* Flush mapping from TLB0. */
2858 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2859 
2860 				tlb_miss_unlock();
2861 				mtx_unlock_spin(&tlbivax_mutex);
2862 			}
2863 		}
2864 		PMAP_UNLOCK(pv->pv_pmap);
2865 	}
2866 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2867 	rw_wunlock(&pvh_global_lock);
2868 }
2869 
2870 static void
2871 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2872 {
2873 	pte_t *pte;
2874 	vm_paddr_t pa = 0;
2875 	int sync_sz, valid;
2876 #ifndef __powerpc64__
2877 	pmap_t pmap;
2878 	vm_page_t m;
2879 	vm_offset_t addr;
2880 	int active;
2881 #endif
2882 
2883 #ifndef __powerpc64__
2884 	rw_wlock(&pvh_global_lock);
2885 	pmap = PCPU_GET(curpmap);
2886 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2887 #endif
2888 	while (sz > 0) {
2889 		PMAP_LOCK(pm);
2890 		pte = pte_find(mmu, pm, va);
2891 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2892 		if (valid)
2893 			pa = PTE_PA(pte);
2894 		PMAP_UNLOCK(pm);
2895 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
2896 		sync_sz = min(sync_sz, sz);
2897 		if (valid) {
2898 #ifdef __powerpc64__
2899 			pa += (va & PAGE_MASK);
2900 			__syncicache((void *)PHYS_TO_DMAP(pa), sync_sz);
2901 #else
2902 			if (!active) {
2903 				/* Create a mapping in the active pmap. */
2904 				addr = 0;
2905 				m = PHYS_TO_VM_PAGE(pa);
2906 				PMAP_LOCK(pmap);
2907 				pte_enter(mmu, pmap, m, addr,
2908 				    PTE_SR | PTE_VALID, FALSE);
2909 				addr += (va & PAGE_MASK);
2910 				__syncicache((void *)addr, sync_sz);
2911 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2912 				PMAP_UNLOCK(pmap);
2913 			} else
2914 				__syncicache((void *)va, sync_sz);
2915 #endif
2916 		}
2917 		va += sync_sz;
2918 		sz -= sync_sz;
2919 	}
2920 #ifndef __powerpc64__
2921 	rw_wunlock(&pvh_global_lock);
2922 #endif
2923 }
2924 
2925 /*
2926  * Atomically extract and hold the physical page with the given
2927  * pmap and virtual address pair if that mapping permits the given
2928  * protection.
2929  */
2930 static vm_page_t
2931 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2932     vm_prot_t prot)
2933 {
2934 	pte_t *pte;
2935 	vm_page_t m;
2936 	uint32_t pte_wbit;
2937 	vm_paddr_t pa;
2938 
2939 	m = NULL;
2940 	pa = 0;
2941 	PMAP_LOCK(pmap);
2942 retry:
2943 	pte = pte_find(mmu, pmap, va);
2944 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2945 		if (pmap == kernel_pmap)
2946 			pte_wbit = PTE_SW;
2947 		else
2948 			pte_wbit = PTE_UW;
2949 
2950 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2951 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2952 				goto retry;
2953 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2954 			vm_page_hold(m);
2955 		}
2956 	}
2957 
2958 	PA_UNLOCK_COND(pa);
2959 	PMAP_UNLOCK(pmap);
2960 	return (m);
2961 }
2962 
2963 /*
2964  * Initialize a vm_page's machine-dependent fields.
2965  */
2966 static void
2967 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2968 {
2969 
2970 	m->md.pv_tracked = 0;
2971 	TAILQ_INIT(&m->md.pv_list);
2972 }
2973 
2974 /*
2975  * mmu_booke_zero_page_area zeros the specified hardware page by
2976  * mapping it into virtual memory and using bzero to clear
2977  * its contents.
2978  *
2979  * off and size must reside within a single page.
2980  */
2981 static void
2982 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2983 {
2984 	vm_offset_t va;
2985 
2986 	/* XXX KASSERT off and size are within a single page? */
2987 
2988 #ifdef __powerpc64__
2989 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2990 	bzero((caddr_t)va + off, size);
2991 #else
2992 	mtx_lock(&zero_page_mutex);
2993 	va = zero_page_va;
2994 
2995 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2996 	bzero((caddr_t)va + off, size);
2997 	mmu_booke_kremove(mmu, va);
2998 
2999 	mtx_unlock(&zero_page_mutex);
3000 #endif
3001 }
3002 
3003 /*
3004  * mmu_booke_zero_page zeros the specified hardware page.
3005  */
3006 static void
3007 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
3008 {
3009 	vm_offset_t off, va;
3010 
3011 #ifdef __powerpc64__
3012 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
3013 
3014 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
3015 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
3016 #else
3017 	va = zero_page_va;
3018 	mtx_lock(&zero_page_mutex);
3019 
3020 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
3021 
3022 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
3023 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
3024 
3025 	mmu_booke_kremove(mmu, va);
3026 
3027 	mtx_unlock(&zero_page_mutex);
3028 #endif
3029 }
3030 
3031 /*
3032  * mmu_booke_copy_page copies the specified (machine independent) page by
3033  * mapping the page into virtual memory and using memcopy to copy the page,
3034  * one machine dependent page at a time.
3035  */
3036 static void
3037 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
3038 {
3039 	vm_offset_t sva, dva;
3040 
3041 #ifdef __powerpc64__
3042 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
3043 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
3044 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3045 #else
3046 	sva = copy_page_src_va;
3047 	dva = copy_page_dst_va;
3048 
3049 	mtx_lock(&copy_page_mutex);
3050 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
3051 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
3052 
3053 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3054 
3055 	mmu_booke_kremove(mmu, dva);
3056 	mmu_booke_kremove(mmu, sva);
3057 	mtx_unlock(&copy_page_mutex);
3058 #endif
3059 }
3060 
3061 static inline void
3062 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
3063     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
3064 {
3065 	void *a_cp, *b_cp;
3066 	vm_offset_t a_pg_offset, b_pg_offset;
3067 	int cnt;
3068 
3069 #ifdef __powerpc64__
3070 	vm_page_t pa, pb;
3071 
3072 	while (xfersize > 0) {
3073 		a_pg_offset = a_offset & PAGE_MASK;
3074 		pa = ma[a_offset >> PAGE_SHIFT];
3075 		b_pg_offset = b_offset & PAGE_MASK;
3076 		pb = mb[b_offset >> PAGE_SHIFT];
3077 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3078 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3079 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
3080 		    a_pg_offset);
3081 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
3082 		    b_pg_offset);
3083 		bcopy(a_cp, b_cp, cnt);
3084 		a_offset += cnt;
3085 		b_offset += cnt;
3086 		xfersize -= cnt;
3087 	}
3088 #else
3089 	mtx_lock(&copy_page_mutex);
3090 	while (xfersize > 0) {
3091 		a_pg_offset = a_offset & PAGE_MASK;
3092 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3093 		mmu_booke_kenter(mmu, copy_page_src_va,
3094 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
3095 		a_cp = (char *)copy_page_src_va + a_pg_offset;
3096 		b_pg_offset = b_offset & PAGE_MASK;
3097 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3098 		mmu_booke_kenter(mmu, copy_page_dst_va,
3099 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
3100 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
3101 		bcopy(a_cp, b_cp, cnt);
3102 		mmu_booke_kremove(mmu, copy_page_dst_va);
3103 		mmu_booke_kremove(mmu, copy_page_src_va);
3104 		a_offset += cnt;
3105 		b_offset += cnt;
3106 		xfersize -= cnt;
3107 	}
3108 	mtx_unlock(&copy_page_mutex);
3109 #endif
3110 }
3111 
3112 static vm_offset_t
3113 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
3114 {
3115 #ifdef __powerpc64__
3116 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
3117 #else
3118 	vm_paddr_t paddr;
3119 	vm_offset_t qaddr;
3120 	uint32_t flags;
3121 	pte_t *pte;
3122 
3123 	paddr = VM_PAGE_TO_PHYS(m);
3124 
3125 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3126 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3127 	flags |= PTE_PS_4KB;
3128 
3129 	critical_enter();
3130 	qaddr = PCPU_GET(qmap_addr);
3131 
3132 	pte = pte_find(mmu, kernel_pmap, qaddr);
3133 
3134 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3135 
3136 	/*
3137 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3138  	 * not be present in other TLBs.  Is there a better instruction
3139 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3140 	 */
3141 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3142 	__asm __volatile("isync; msync");
3143 
3144 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3145 
3146 	/* Flush the real memory from the instruction cache. */
3147 	if ((flags & (PTE_I | PTE_G)) == 0)
3148 		__syncicache((void *)qaddr, PAGE_SIZE);
3149 
3150 	return (qaddr);
3151 #endif
3152 }
3153 
3154 static void
3155 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3156 {
3157 #ifndef __powerpc64__
3158 	pte_t *pte;
3159 
3160 	pte = pte_find(mmu, kernel_pmap, addr);
3161 
3162 	KASSERT(PCPU_GET(qmap_addr) == addr,
3163 	    ("mmu_booke_quick_remove_page: invalid address"));
3164 	KASSERT(*pte != 0,
3165 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3166 
3167 	*pte = 0;
3168 	critical_exit();
3169 #endif
3170 }
3171 
3172 /*
3173  * Return whether or not the specified physical page was modified
3174  * in any of physical maps.
3175  */
3176 static boolean_t
3177 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3178 {
3179 	pte_t *pte;
3180 	pv_entry_t pv;
3181 	boolean_t rv;
3182 
3183 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3184 	    ("mmu_booke_is_modified: page %p is not managed", m));
3185 	rv = FALSE;
3186 
3187 	/*
3188 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3189 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3190 	 * is clear, no PTEs can be modified.
3191 	 */
3192 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3193 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3194 		return (rv);
3195 	rw_wlock(&pvh_global_lock);
3196 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3197 		PMAP_LOCK(pv->pv_pmap);
3198 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3199 		    PTE_ISVALID(pte)) {
3200 			if (PTE_ISMODIFIED(pte))
3201 				rv = TRUE;
3202 		}
3203 		PMAP_UNLOCK(pv->pv_pmap);
3204 		if (rv)
3205 			break;
3206 	}
3207 	rw_wunlock(&pvh_global_lock);
3208 	return (rv);
3209 }
3210 
3211 /*
3212  * Return whether or not the specified virtual address is eligible
3213  * for prefault.
3214  */
3215 static boolean_t
3216 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3217 {
3218 
3219 	return (FALSE);
3220 }
3221 
3222 /*
3223  * Return whether or not the specified physical page was referenced
3224  * in any physical maps.
3225  */
3226 static boolean_t
3227 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3228 {
3229 	pte_t *pte;
3230 	pv_entry_t pv;
3231 	boolean_t rv;
3232 
3233 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3234 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3235 	rv = FALSE;
3236 	rw_wlock(&pvh_global_lock);
3237 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3238 		PMAP_LOCK(pv->pv_pmap);
3239 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3240 		    PTE_ISVALID(pte)) {
3241 			if (PTE_ISREFERENCED(pte))
3242 				rv = TRUE;
3243 		}
3244 		PMAP_UNLOCK(pv->pv_pmap);
3245 		if (rv)
3246 			break;
3247 	}
3248 	rw_wunlock(&pvh_global_lock);
3249 	return (rv);
3250 }
3251 
3252 /*
3253  * Clear the modify bits on the specified physical page.
3254  */
3255 static void
3256 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3257 {
3258 	pte_t *pte;
3259 	pv_entry_t pv;
3260 
3261 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3262 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3263 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3264 	KASSERT(!vm_page_xbusied(m),
3265 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3266 
3267 	/*
3268 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3269 	 * If the object containing the page is locked and the page is not
3270 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3271 	 */
3272 	if ((m->aflags & PGA_WRITEABLE) == 0)
3273 		return;
3274 	rw_wlock(&pvh_global_lock);
3275 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3276 		PMAP_LOCK(pv->pv_pmap);
3277 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3278 		    PTE_ISVALID(pte)) {
3279 			mtx_lock_spin(&tlbivax_mutex);
3280 			tlb_miss_lock();
3281 
3282 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3283 				tlb0_flush_entry(pv->pv_va);
3284 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3285 				    PTE_REFERENCED);
3286 			}
3287 
3288 			tlb_miss_unlock();
3289 			mtx_unlock_spin(&tlbivax_mutex);
3290 		}
3291 		PMAP_UNLOCK(pv->pv_pmap);
3292 	}
3293 	rw_wunlock(&pvh_global_lock);
3294 }
3295 
3296 /*
3297  * Return a count of reference bits for a page, clearing those bits.
3298  * It is not necessary for every reference bit to be cleared, but it
3299  * is necessary that 0 only be returned when there are truly no
3300  * reference bits set.
3301  *
3302  * As an optimization, update the page's dirty field if a modified bit is
3303  * found while counting reference bits.  This opportunistic update can be
3304  * performed at low cost and can eliminate the need for some future calls
3305  * to pmap_is_modified().  However, since this function stops after
3306  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3307  * dirty pages.  Those dirty pages will only be detected by a future call
3308  * to pmap_is_modified().
3309  */
3310 static int
3311 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3312 {
3313 	pte_t *pte;
3314 	pv_entry_t pv;
3315 	int count;
3316 
3317 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3318 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3319 	count = 0;
3320 	rw_wlock(&pvh_global_lock);
3321 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3322 		PMAP_LOCK(pv->pv_pmap);
3323 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3324 		    PTE_ISVALID(pte)) {
3325 			if (PTE_ISMODIFIED(pte))
3326 				vm_page_dirty(m);
3327 			if (PTE_ISREFERENCED(pte)) {
3328 				mtx_lock_spin(&tlbivax_mutex);
3329 				tlb_miss_lock();
3330 
3331 				tlb0_flush_entry(pv->pv_va);
3332 				*pte &= ~PTE_REFERENCED;
3333 
3334 				tlb_miss_unlock();
3335 				mtx_unlock_spin(&tlbivax_mutex);
3336 
3337 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3338 					PMAP_UNLOCK(pv->pv_pmap);
3339 					break;
3340 				}
3341 			}
3342 		}
3343 		PMAP_UNLOCK(pv->pv_pmap);
3344 	}
3345 	rw_wunlock(&pvh_global_lock);
3346 	return (count);
3347 }
3348 
3349 /*
3350  * Clear the wired attribute from the mappings for the specified range of
3351  * addresses in the given pmap.  Every valid mapping within that range must
3352  * have the wired attribute set.  In contrast, invalid mappings cannot have
3353  * the wired attribute set, so they are ignored.
3354  *
3355  * The wired attribute of the page table entry is not a hardware feature, so
3356  * there is no need to invalidate any TLB entries.
3357  */
3358 static void
3359 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3360 {
3361 	vm_offset_t va;
3362 	pte_t *pte;
3363 
3364 	PMAP_LOCK(pmap);
3365 	for (va = sva; va < eva; va += PAGE_SIZE) {
3366 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3367 		    PTE_ISVALID(pte)) {
3368 			if (!PTE_ISWIRED(pte))
3369 				panic("mmu_booke_unwire: pte %p isn't wired",
3370 				    pte);
3371 			*pte &= ~PTE_WIRED;
3372 			pmap->pm_stats.wired_count--;
3373 		}
3374 	}
3375 	PMAP_UNLOCK(pmap);
3376 
3377 }
3378 
3379 /*
3380  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3381  * page.  This count may be changed upwards or downwards in the future; it is
3382  * only necessary that true be returned for a small subset of pmaps for proper
3383  * page aging.
3384  */
3385 static boolean_t
3386 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3387 {
3388 	pv_entry_t pv;
3389 	int loops;
3390 	boolean_t rv;
3391 
3392 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3393 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3394 	loops = 0;
3395 	rv = FALSE;
3396 	rw_wlock(&pvh_global_lock);
3397 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3398 		if (pv->pv_pmap == pmap) {
3399 			rv = TRUE;
3400 			break;
3401 		}
3402 		if (++loops >= 16)
3403 			break;
3404 	}
3405 	rw_wunlock(&pvh_global_lock);
3406 	return (rv);
3407 }
3408 
3409 /*
3410  * Return the number of managed mappings to the given physical page that are
3411  * wired.
3412  */
3413 static int
3414 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3415 {
3416 	pv_entry_t pv;
3417 	pte_t *pte;
3418 	int count = 0;
3419 
3420 	if ((m->oflags & VPO_UNMANAGED) != 0)
3421 		return (count);
3422 	rw_wlock(&pvh_global_lock);
3423 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3424 		PMAP_LOCK(pv->pv_pmap);
3425 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3426 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3427 				count++;
3428 		PMAP_UNLOCK(pv->pv_pmap);
3429 	}
3430 	rw_wunlock(&pvh_global_lock);
3431 	return (count);
3432 }
3433 
3434 static int
3435 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3436 {
3437 	int i;
3438 	vm_offset_t va;
3439 
3440 	/*
3441 	 * This currently does not work for entries that
3442 	 * overlap TLB1 entries.
3443 	 */
3444 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3445 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3446 			return (0);
3447 	}
3448 
3449 	return (EFAULT);
3450 }
3451 
3452 void
3453 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3454 {
3455 	vm_paddr_t ppa;
3456 	vm_offset_t ofs;
3457 	vm_size_t gran;
3458 
3459 	/* Minidumps are based on virtual memory addresses. */
3460 	if (do_minidump) {
3461 		*va = (void *)(vm_offset_t)pa;
3462 		return;
3463 	}
3464 
3465 	/* Raw physical memory dumps don't have a virtual address. */
3466 	/* We always map a 256MB page at 256M. */
3467 	gran = 256 * 1024 * 1024;
3468 	ppa = rounddown2(pa, gran);
3469 	ofs = pa - ppa;
3470 	*va = (void *)gran;
3471 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3472 
3473 	if (sz > (gran - ofs))
3474 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3475 		    _TLB_ENTRY_IO);
3476 }
3477 
3478 void
3479 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3480 {
3481 	vm_paddr_t ppa;
3482 	vm_offset_t ofs;
3483 	vm_size_t gran;
3484 	tlb_entry_t e;
3485 	int i;
3486 
3487 	/* Minidumps are based on virtual memory addresses. */
3488 	/* Nothing to do... */
3489 	if (do_minidump)
3490 		return;
3491 
3492 	for (i = 0; i < TLB1_ENTRIES; i++) {
3493 		tlb1_read_entry(&e, i);
3494 		if (!(e.mas1 & MAS1_VALID))
3495 			break;
3496 	}
3497 
3498 	/* Raw physical memory dumps don't have a virtual address. */
3499 	i--;
3500 	e.mas1 = 0;
3501 	e.mas2 = 0;
3502 	e.mas3 = 0;
3503 	tlb1_write_entry(&e, i);
3504 
3505 	gran = 256 * 1024 * 1024;
3506 	ppa = rounddown2(pa, gran);
3507 	ofs = pa - ppa;
3508 	if (sz > (gran - ofs)) {
3509 		i--;
3510 		e.mas1 = 0;
3511 		e.mas2 = 0;
3512 		e.mas3 = 0;
3513 		tlb1_write_entry(&e, i);
3514 	}
3515 }
3516 
3517 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3518 
3519 void
3520 mmu_booke_scan_init(mmu_t mmu)
3521 {
3522 	vm_offset_t va;
3523 	pte_t *pte;
3524 	int i;
3525 
3526 	if (!do_minidump) {
3527 		/* Initialize phys. segments for dumpsys(). */
3528 		memset(&dump_map, 0, sizeof(dump_map));
3529 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3530 		    &availmem_regions_sz);
3531 		for (i = 0; i < physmem_regions_sz; i++) {
3532 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3533 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3534 		}
3535 		return;
3536 	}
3537 
3538 	/* Virtual segments for minidumps: */
3539 	memset(&dump_map, 0, sizeof(dump_map));
3540 
3541 	/* 1st: kernel .data and .bss. */
3542 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3543 	dump_map[0].pa_size =
3544 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3545 
3546 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3547 	dump_map[1].pa_start = data_start;
3548 	dump_map[1].pa_size = data_end - data_start;
3549 
3550 	/* 3rd: kernel VM. */
3551 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3552 	/* Find start of next chunk (from va). */
3553 	while (va < virtual_end) {
3554 		/* Don't dump the buffer cache. */
3555 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3556 			va = kmi.buffer_eva;
3557 			continue;
3558 		}
3559 		pte = pte_find(mmu, kernel_pmap, va);
3560 		if (pte != NULL && PTE_ISVALID(pte))
3561 			break;
3562 		va += PAGE_SIZE;
3563 	}
3564 	if (va < virtual_end) {
3565 		dump_map[2].pa_start = va;
3566 		va += PAGE_SIZE;
3567 		/* Find last page in chunk. */
3568 		while (va < virtual_end) {
3569 			/* Don't run into the buffer cache. */
3570 			if (va == kmi.buffer_sva)
3571 				break;
3572 			pte = pte_find(mmu, kernel_pmap, va);
3573 			if (pte == NULL || !PTE_ISVALID(pte))
3574 				break;
3575 			va += PAGE_SIZE;
3576 		}
3577 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3578 	}
3579 }
3580 
3581 /*
3582  * Map a set of physical memory pages into the kernel virtual address space.
3583  * Return a pointer to where it is mapped. This routine is intended to be used
3584  * for mapping device memory, NOT real memory.
3585  */
3586 static void *
3587 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3588 {
3589 
3590 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3591 }
3592 
3593 static void *
3594 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3595 {
3596 	tlb_entry_t e;
3597 	void *res;
3598 	uintptr_t va, tmpva;
3599 	vm_size_t sz;
3600 	int i;
3601 
3602 	/*
3603 	 * Check if this is premapped in TLB1. Note: this should probably also
3604 	 * check whether a sequence of TLB1 entries exist that match the
3605 	 * requirement, but now only checks the easy case.
3606 	 */
3607 	for (i = 0; i < TLB1_ENTRIES; i++) {
3608 		tlb1_read_entry(&e, i);
3609 		if (!(e.mas1 & MAS1_VALID))
3610 			continue;
3611 		if (pa >= e.phys &&
3612 		    (pa + size) <= (e.phys + e.size) &&
3613 		    (ma == VM_MEMATTR_DEFAULT ||
3614 		     tlb_calc_wimg(pa, ma) ==
3615 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3616 			return (void *)(e.virt +
3617 			    (vm_offset_t)(pa - e.phys));
3618 	}
3619 
3620 	size = roundup(size, PAGE_SIZE);
3621 
3622 	/*
3623 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3624 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3625 	 */
3626 #ifdef SPARSE_MAPDEV
3627 	/*
3628 	 * With a sparse mapdev, align to the largest starting region.  This
3629 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3630 	 * calculation could be very costly.
3631 	 * Align to the smaller of:
3632 	 * - first set bit in overlap of (pa & size mask)
3633 	 * - largest size envelope
3634 	 *
3635 	 * It's possible the device mapping may start at a PA that's not larger
3636 	 * than the size mask, so we need to offset in to maximize the TLB entry
3637 	 * range and minimize the number of used TLB entries.
3638 	 */
3639 	do {
3640 	    tmpva = tlb1_map_base;
3641 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3642 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3643 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3644 #ifdef __powerpc64__
3645 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3646 #else
3647 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3648 #endif
3649 #else
3650 #ifdef __powerpc64__
3651 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3652 #else
3653 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3654 #endif
3655 #endif
3656 	res = (void *)va;
3657 
3658 	do {
3659 		sz = 1 << (ilog2(size) & ~1);
3660 		/* Align size to PA */
3661 		if (pa % sz != 0) {
3662 			do {
3663 				sz >>= 2;
3664 			} while (pa % sz != 0);
3665 		}
3666 		/* Now align from there to VA */
3667 		if (va % sz != 0) {
3668 			do {
3669 				sz >>= 2;
3670 			} while (va % sz != 0);
3671 		}
3672 		if (bootverbose)
3673 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3674 			    va, (uintmax_t)pa, sz);
3675 		if (tlb1_set_entry(va, pa, sz,
3676 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3677 			return (NULL);
3678 		size -= sz;
3679 		pa += sz;
3680 		va += sz;
3681 	} while (size > 0);
3682 
3683 	return (res);
3684 }
3685 
3686 /*
3687  * 'Unmap' a range mapped by mmu_booke_mapdev().
3688  */
3689 static void
3690 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3691 {
3692 #ifdef SUPPORTS_SHRINKING_TLB1
3693 	vm_offset_t base, offset;
3694 
3695 	/*
3696 	 * Unmap only if this is inside kernel virtual space.
3697 	 */
3698 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3699 		base = trunc_page(va);
3700 		offset = va & PAGE_MASK;
3701 		size = roundup(offset + size, PAGE_SIZE);
3702 		kva_free(base, size);
3703 	}
3704 #endif
3705 }
3706 
3707 /*
3708  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3709  * specified pmap. This eliminates the blast of soft faults on process startup
3710  * and immediately after an mmap.
3711  */
3712 static void
3713 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3714     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3715 {
3716 
3717 	VM_OBJECT_ASSERT_WLOCKED(object);
3718 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3719 	    ("mmu_booke_object_init_pt: non-device object"));
3720 }
3721 
3722 /*
3723  * Perform the pmap work for mincore.
3724  */
3725 static int
3726 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3727     vm_paddr_t *locked_pa)
3728 {
3729 
3730 	/* XXX: this should be implemented at some point */
3731 	return (0);
3732 }
3733 
3734 static int
3735 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3736     vm_memattr_t mode)
3737 {
3738 	vm_offset_t va;
3739 	pte_t *pte;
3740 	int i, j;
3741 	tlb_entry_t e;
3742 
3743 	/* Check TLB1 mappings */
3744 	for (i = 0; i < TLB1_ENTRIES; i++) {
3745 		tlb1_read_entry(&e, i);
3746 		if (!(e.mas1 & MAS1_VALID))
3747 			continue;
3748 		if (addr >= e.virt && addr < e.virt + e.size)
3749 			break;
3750 	}
3751 	if (i < TLB1_ENTRIES) {
3752 		/* Only allow full mappings to be modified for now. */
3753 		/* Validate the range. */
3754 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3755 			tlb1_read_entry(&e, j);
3756 			if (va != e.virt || (sz - (va - addr) < e.size))
3757 				return (EINVAL);
3758 		}
3759 		for (va = addr; va < addr + sz; va += e.size, i++) {
3760 			tlb1_read_entry(&e, i);
3761 			e.mas2 &= ~MAS2_WIMGE_MASK;
3762 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3763 
3764 			/*
3765 			 * Write it out to the TLB.  Should really re-sync with other
3766 			 * cores.
3767 			 */
3768 			tlb1_write_entry(&e, i);
3769 		}
3770 		return (0);
3771 	}
3772 
3773 	/* Not in TLB1, try through pmap */
3774 	/* First validate the range. */
3775 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3776 		pte = pte_find(mmu, kernel_pmap, va);
3777 		if (pte == NULL || !PTE_ISVALID(pte))
3778 			return (EINVAL);
3779 	}
3780 
3781 	mtx_lock_spin(&tlbivax_mutex);
3782 	tlb_miss_lock();
3783 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3784 		pte = pte_find(mmu, kernel_pmap, va);
3785 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3786 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3787 		tlb0_flush_entry(va);
3788 	}
3789 	tlb_miss_unlock();
3790 	mtx_unlock_spin(&tlbivax_mutex);
3791 
3792 	return (0);
3793 }
3794 
3795 /**************************************************************************/
3796 /* TID handling */
3797 /**************************************************************************/
3798 
3799 /*
3800  * Allocate a TID. If necessary, steal one from someone else.
3801  * The new TID is flushed from the TLB before returning.
3802  */
3803 static tlbtid_t
3804 tid_alloc(pmap_t pmap)
3805 {
3806 	tlbtid_t tid;
3807 	int thiscpu;
3808 
3809 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3810 
3811 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3812 
3813 	thiscpu = PCPU_GET(cpuid);
3814 
3815 	tid = PCPU_GET(booke.tid_next);
3816 	if (tid > TID_MAX)
3817 		tid = TID_MIN;
3818 	PCPU_SET(booke.tid_next, tid + 1);
3819 
3820 	/* If we are stealing TID then clear the relevant pmap's field */
3821 	if (tidbusy[thiscpu][tid] != NULL) {
3822 
3823 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3824 
3825 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3826 
3827 		/* Flush all entries from TLB0 matching this TID. */
3828 		tid_flush(tid);
3829 	}
3830 
3831 	tidbusy[thiscpu][tid] = pmap;
3832 	pmap->pm_tid[thiscpu] = tid;
3833 	__asm __volatile("msync; isync");
3834 
3835 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3836 	    PCPU_GET(booke.tid_next));
3837 
3838 	return (tid);
3839 }
3840 
3841 /**************************************************************************/
3842 /* TLB0 handling */
3843 /**************************************************************************/
3844 
3845 /* Convert TLB0 va and way number to tlb0[] table index. */
3846 static inline unsigned int
3847 tlb0_tableidx(vm_offset_t va, unsigned int way)
3848 {
3849 	unsigned int idx;
3850 
3851 	idx = (way * TLB0_ENTRIES_PER_WAY);
3852 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3853 	return (idx);
3854 }
3855 
3856 /*
3857  * Invalidate TLB0 entry.
3858  */
3859 static inline void
3860 tlb0_flush_entry(vm_offset_t va)
3861 {
3862 
3863 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3864 
3865 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3866 
3867 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3868 	__asm __volatile("isync; msync");
3869 	__asm __volatile("tlbsync; msync");
3870 
3871 	CTR1(KTR_PMAP, "%s: e", __func__);
3872 }
3873 
3874 
3875 /**************************************************************************/
3876 /* TLB1 handling */
3877 /**************************************************************************/
3878 
3879 /*
3880  * TLB1 mapping notes:
3881  *
3882  * TLB1[0]	Kernel text and data.
3883  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3884  *		windows, other devices mappings.
3885  */
3886 
3887  /*
3888  * Read an entry from given TLB1 slot.
3889  */
3890 void
3891 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3892 {
3893 	register_t msr;
3894 	uint32_t mas0;
3895 
3896 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3897 
3898 	msr = mfmsr();
3899 	__asm __volatile("wrteei 0");
3900 
3901 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3902 	mtspr(SPR_MAS0, mas0);
3903 	__asm __volatile("isync; tlbre");
3904 
3905 	entry->mas1 = mfspr(SPR_MAS1);
3906 	entry->mas2 = mfspr(SPR_MAS2);
3907 	entry->mas3 = mfspr(SPR_MAS3);
3908 
3909 	switch ((mfpvr() >> 16) & 0xFFFF) {
3910 	case FSL_E500v2:
3911 	case FSL_E500mc:
3912 	case FSL_E5500:
3913 	case FSL_E6500:
3914 		entry->mas7 = mfspr(SPR_MAS7);
3915 		break;
3916 	default:
3917 		entry->mas7 = 0;
3918 		break;
3919 	}
3920 	__asm __volatile("wrtee %0" :: "r"(msr));
3921 
3922 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3923 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3924 	    (entry->mas3 & MAS3_RPN);
3925 	entry->size =
3926 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3927 }
3928 
3929 struct tlbwrite_args {
3930 	tlb_entry_t *e;
3931 	unsigned int idx;
3932 };
3933 
3934 static void
3935 tlb1_write_entry_int(void *arg)
3936 {
3937 	struct tlbwrite_args *args = arg;
3938 	uint32_t mas0;
3939 
3940 	/* Select entry */
3941 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3942 
3943 	mtspr(SPR_MAS0, mas0);
3944 	mtspr(SPR_MAS1, args->e->mas1);
3945 	mtspr(SPR_MAS2, args->e->mas2);
3946 	mtspr(SPR_MAS3, args->e->mas3);
3947 	switch ((mfpvr() >> 16) & 0xFFFF) {
3948 	case FSL_E500mc:
3949 	case FSL_E5500:
3950 	case FSL_E6500:
3951 		mtspr(SPR_MAS8, 0);
3952 		/* FALLTHROUGH */
3953 	case FSL_E500v2:
3954 		mtspr(SPR_MAS7, args->e->mas7);
3955 		break;
3956 	default:
3957 		break;
3958 	}
3959 
3960 	__asm __volatile("isync; tlbwe; isync; msync");
3961 
3962 }
3963 
3964 static void
3965 tlb1_write_entry_sync(void *arg)
3966 {
3967 	/* Empty synchronization point for smp_rendezvous(). */
3968 }
3969 
3970 /*
3971  * Write given entry to TLB1 hardware.
3972  */
3973 static void
3974 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3975 {
3976 	struct tlbwrite_args args;
3977 
3978 	args.e = e;
3979 	args.idx = idx;
3980 
3981 #ifdef SMP
3982 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3983 		mb();
3984 		smp_rendezvous(tlb1_write_entry_sync,
3985 		    tlb1_write_entry_int,
3986 		    tlb1_write_entry_sync, &args);
3987 	} else
3988 #endif
3989 	{
3990 		register_t msr;
3991 
3992 		msr = mfmsr();
3993 		__asm __volatile("wrteei 0");
3994 		tlb1_write_entry_int(&args);
3995 		__asm __volatile("wrtee %0" :: "r"(msr));
3996 	}
3997 }
3998 
3999 /*
4000  * Return the largest uint value log such that 2^log <= num.
4001  */
4002 static unsigned int
4003 ilog2(unsigned long num)
4004 {
4005 	long lz;
4006 
4007 #ifdef __powerpc64__
4008 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
4009 	return (63 - lz);
4010 #else
4011 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
4012 	return (31 - lz);
4013 #endif
4014 }
4015 
4016 /*
4017  * Convert TLB TSIZE value to mapped region size.
4018  */
4019 static vm_size_t
4020 tsize2size(unsigned int tsize)
4021 {
4022 
4023 	/*
4024 	 * size = 4^tsize KB
4025 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
4026 	 */
4027 
4028 	return ((1 << (2 * tsize)) * 1024);
4029 }
4030 
4031 /*
4032  * Convert region size (must be power of 4) to TLB TSIZE value.
4033  */
4034 static unsigned int
4035 size2tsize(vm_size_t size)
4036 {
4037 
4038 	return (ilog2(size) / 2 - 5);
4039 }
4040 
4041 /*
4042  * Register permanent kernel mapping in TLB1.
4043  *
4044  * Entries are created starting from index 0 (current free entry is
4045  * kept in tlb1_idx) and are not supposed to be invalidated.
4046  */
4047 int
4048 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
4049     uint32_t flags)
4050 {
4051 	tlb_entry_t e;
4052 	uint32_t ts, tid;
4053 	int tsize, index;
4054 
4055 	for (index = 0; index < TLB1_ENTRIES; index++) {
4056 		tlb1_read_entry(&e, index);
4057 		if ((e.mas1 & MAS1_VALID) == 0)
4058 			break;
4059 		/* Check if we're just updating the flags, and update them. */
4060 		if (e.phys == pa && e.virt == va && e.size == size) {
4061 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
4062 			tlb1_write_entry(&e, index);
4063 			return (0);
4064 		}
4065 	}
4066 	if (index >= TLB1_ENTRIES) {
4067 		printf("tlb1_set_entry: TLB1 full!\n");
4068 		return (-1);
4069 	}
4070 
4071 	/* Convert size to TSIZE */
4072 	tsize = size2tsize(size);
4073 
4074 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
4075 	/* XXX TS is hard coded to 0 for now as we only use single address space */
4076 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
4077 
4078 	e.phys = pa;
4079 	e.virt = va;
4080 	e.size = size;
4081 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
4082 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
4083 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
4084 
4085 	/* Set supervisor RWX permission bits */
4086 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
4087 	e.mas7 = (pa >> 32) & MAS7_RPN;
4088 
4089 	tlb1_write_entry(&e, index);
4090 
4091 	/*
4092 	 * XXX in general TLB1 updates should be propagated between CPUs,
4093 	 * since current design assumes to have the same TLB1 set-up on all
4094 	 * cores.
4095 	 */
4096 	return (0);
4097 }
4098 
4099 /*
4100  * Map in contiguous RAM region into the TLB1 using maximum of
4101  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
4102  *
4103  * If necessary round up last entry size and return total size
4104  * used by all allocated entries.
4105  */
4106 vm_size_t
4107 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
4108 {
4109 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
4110 	vm_size_t mapped, pgsz, base, mask;
4111 	int idx, nents;
4112 
4113 	/* Round up to the next 1M */
4114 	size = roundup2(size, 1 << 20);
4115 
4116 	mapped = 0;
4117 	idx = 0;
4118 	base = va;
4119 	pgsz = 64*1024*1024;
4120 	while (mapped < size) {
4121 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
4122 			while (pgsz > (size - mapped))
4123 				pgsz >>= 2;
4124 			pgs[idx++] = pgsz;
4125 			mapped += pgsz;
4126 		}
4127 
4128 		/* We under-map. Correct for this. */
4129 		if (mapped < size) {
4130 			while (pgs[idx - 1] == pgsz) {
4131 				idx--;
4132 				mapped -= pgsz;
4133 			}
4134 			/* XXX We may increase beyond out starting point. */
4135 			pgsz <<= 2;
4136 			pgs[idx++] = pgsz;
4137 			mapped += pgsz;
4138 		}
4139 	}
4140 
4141 	nents = idx;
4142 	mask = pgs[0] - 1;
4143 	/* Align address to the boundary */
4144 	if (va & mask) {
4145 		va = (va + mask) & ~mask;
4146 		pa = (pa + mask) & ~mask;
4147 	}
4148 
4149 	for (idx = 0; idx < nents; idx++) {
4150 		pgsz = pgs[idx];
4151 		debugf("%u: %llx -> %jx, size=%jx\n", idx, pa,
4152 		    (uintmax_t)va, (uintmax_t)pgsz);
4153 		tlb1_set_entry(va, pa, pgsz,
4154 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4155 		pa += pgsz;
4156 		va += pgsz;
4157 	}
4158 
4159 	mapped = (va - base);
4160 	if (bootverbose)
4161 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
4162 		    mapped, mapped - size);
4163 	return (mapped);
4164 }
4165 
4166 /*
4167  * TLB1 initialization routine, to be called after the very first
4168  * assembler level setup done in locore.S.
4169  */
4170 void
4171 tlb1_init()
4172 {
4173 	uint32_t mas0, mas1, mas2, mas3, mas7;
4174 	uint32_t tsz;
4175 
4176 	tlb1_get_tlbconf();
4177 
4178 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4179 	mtspr(SPR_MAS0, mas0);
4180 	__asm __volatile("isync; tlbre");
4181 
4182 	mas1 = mfspr(SPR_MAS1);
4183 	mas2 = mfspr(SPR_MAS2);
4184 	mas3 = mfspr(SPR_MAS3);
4185 	mas7 = mfspr(SPR_MAS7);
4186 
4187 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4188 	    (mas3 & MAS3_RPN);
4189 
4190 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4191 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4192 
4193 	/* Setup TLB miss defaults */
4194 	set_mas4_defaults();
4195 }
4196 
4197 /*
4198  * pmap_early_io_unmap() should be used in short conjunction with
4199  * pmap_early_io_map(), as in the following snippet:
4200  *
4201  * x = pmap_early_io_map(...);
4202  * <do something with x>
4203  * pmap_early_io_unmap(x, size);
4204  *
4205  * And avoiding more allocations between.
4206  */
4207 void
4208 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4209 {
4210 	int i;
4211 	tlb_entry_t e;
4212 	vm_size_t isize;
4213 
4214 	size = roundup(size, PAGE_SIZE);
4215 	isize = size;
4216 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4217 		tlb1_read_entry(&e, i);
4218 		if (!(e.mas1 & MAS1_VALID))
4219 			continue;
4220 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4221 			size -= e.size;
4222 			e.mas1 &= ~MAS1_VALID;
4223 			tlb1_write_entry(&e, i);
4224 		}
4225 	}
4226 	if (tlb1_map_base == va + isize)
4227 		tlb1_map_base -= isize;
4228 }
4229 
4230 vm_offset_t
4231 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4232 {
4233 	vm_paddr_t pa_base;
4234 	vm_offset_t va, sz;
4235 	int i;
4236 	tlb_entry_t e;
4237 
4238 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4239 
4240 	for (i = 0; i < TLB1_ENTRIES; i++) {
4241 		tlb1_read_entry(&e, i);
4242 		if (!(e.mas1 & MAS1_VALID))
4243 			continue;
4244 		if (pa >= e.phys && (pa + size) <=
4245 		    (e.phys + e.size))
4246 			return (e.virt + (pa - e.phys));
4247 	}
4248 
4249 	pa_base = rounddown(pa, PAGE_SIZE);
4250 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4251 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4252 	va = tlb1_map_base + (pa - pa_base);
4253 
4254 	do {
4255 		sz = 1 << (ilog2(size) & ~1);
4256 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4257 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4258 		size -= sz;
4259 		pa_base += sz;
4260 		tlb1_map_base += sz;
4261 	} while (size > 0);
4262 
4263 	return (va);
4264 }
4265 
4266 void
4267 pmap_track_page(pmap_t pmap, vm_offset_t va)
4268 {
4269 	vm_paddr_t pa;
4270 	vm_page_t page;
4271 	struct pv_entry *pve;
4272 
4273 	va = trunc_page(va);
4274 	pa = pmap_kextract(va);
4275 	page = PHYS_TO_VM_PAGE(pa);
4276 
4277 	rw_wlock(&pvh_global_lock);
4278 	PMAP_LOCK(pmap);
4279 
4280 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4281 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4282 			goto out;
4283 		}
4284 	}
4285 	page->md.pv_tracked = true;
4286 	pv_insert(pmap, va, page);
4287 out:
4288 	PMAP_UNLOCK(pmap);
4289 	rw_wunlock(&pvh_global_lock);
4290 }
4291 
4292 
4293 /*
4294  * Setup MAS4 defaults.
4295  * These values are loaded to MAS0-2 on a TLB miss.
4296  */
4297 static void
4298 set_mas4_defaults(void)
4299 {
4300 	uint32_t mas4;
4301 
4302 	/* Defaults: TLB0, PID0, TSIZED=4K */
4303 	mas4 = MAS4_TLBSELD0;
4304 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4305 #ifdef SMP
4306 	mas4 |= MAS4_MD;
4307 #endif
4308 	mtspr(SPR_MAS4, mas4);
4309 	__asm __volatile("isync");
4310 }
4311 
4312 
4313 /*
4314  * Return 0 if the physical IO range is encompassed by one of the
4315  * the TLB1 entries, otherwise return related error code.
4316  */
4317 static int
4318 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4319 {
4320 	uint32_t prot;
4321 	vm_paddr_t pa_start;
4322 	vm_paddr_t pa_end;
4323 	unsigned int entry_tsize;
4324 	vm_size_t entry_size;
4325 	tlb_entry_t e;
4326 
4327 	*va = (vm_offset_t)NULL;
4328 
4329 	tlb1_read_entry(&e, i);
4330 	/* Skip invalid entries */
4331 	if (!(e.mas1 & MAS1_VALID))
4332 		return (EINVAL);
4333 
4334 	/*
4335 	 * The entry must be cache-inhibited, guarded, and r/w
4336 	 * so it can function as an i/o page
4337 	 */
4338 	prot = e.mas2 & (MAS2_I | MAS2_G);
4339 	if (prot != (MAS2_I | MAS2_G))
4340 		return (EPERM);
4341 
4342 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4343 	if (prot != (MAS3_SR | MAS3_SW))
4344 		return (EPERM);
4345 
4346 	/* The address should be within the entry range. */
4347 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4348 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4349 
4350 	entry_size = tsize2size(entry_tsize);
4351 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4352 	    (e.mas3 & MAS3_RPN);
4353 	pa_end = pa_start + entry_size;
4354 
4355 	if ((pa < pa_start) || ((pa + size) > pa_end))
4356 		return (ERANGE);
4357 
4358 	/* Return virtual address of this mapping. */
4359 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4360 	return (0);
4361 }
4362 
4363 /*
4364  * Invalidate all TLB0 entries which match the given TID. Note this is
4365  * dedicated for cases when invalidations should NOT be propagated to other
4366  * CPUs.
4367  */
4368 static void
4369 tid_flush(tlbtid_t tid)
4370 {
4371 	register_t msr;
4372 	uint32_t mas0, mas1, mas2;
4373 	int entry, way;
4374 
4375 
4376 	/* Don't evict kernel translations */
4377 	if (tid == TID_KERNEL)
4378 		return;
4379 
4380 	msr = mfmsr();
4381 	__asm __volatile("wrteei 0");
4382 
4383 	/*
4384 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
4385 	 * it for PID invalidation.
4386 	 */
4387 	switch ((mfpvr() >> 16) & 0xffff) {
4388 	case FSL_E500mc:
4389 	case FSL_E5500:
4390 	case FSL_E6500:
4391 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
4392 		/* tlbilxpid */
4393 		__asm __volatile("isync; .long 0x7c000024; isync; msync");
4394 		__asm __volatile("wrtee %0" :: "r"(msr));
4395 		return;
4396 	}
4397 
4398 	for (way = 0; way < TLB0_WAYS; way++)
4399 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4400 
4401 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4402 			mtspr(SPR_MAS0, mas0);
4403 
4404 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4405 			mtspr(SPR_MAS2, mas2);
4406 
4407 			__asm __volatile("isync; tlbre");
4408 
4409 			mas1 = mfspr(SPR_MAS1);
4410 
4411 			if (!(mas1 & MAS1_VALID))
4412 				continue;
4413 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4414 				continue;
4415 			mas1 &= ~MAS1_VALID;
4416 			mtspr(SPR_MAS1, mas1);
4417 			__asm __volatile("isync; tlbwe; isync; msync");
4418 		}
4419 	__asm __volatile("wrtee %0" :: "r"(msr));
4420 }
4421 
4422 #ifdef DDB
4423 /* Print out contents of the MAS registers for each TLB0 entry */
4424 static void
4425 #ifdef __powerpc64__
4426 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
4427 #else
4428 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
4429 #endif
4430     uint32_t mas7)
4431 {
4432 	int as;
4433 	char desc[3];
4434 	tlbtid_t tid;
4435 	vm_size_t size;
4436 	unsigned int tsize;
4437 
4438 	desc[2] = '\0';
4439 	if (mas1 & MAS1_VALID)
4440 		desc[0] = 'V';
4441 	else
4442 		desc[0] = ' ';
4443 
4444 	if (mas1 & MAS1_IPROT)
4445 		desc[1] = 'P';
4446 	else
4447 		desc[1] = ' ';
4448 
4449 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
4450 	tid = MAS1_GETTID(mas1);
4451 
4452 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4453 	size = 0;
4454 	if (tsize)
4455 		size = tsize2size(tsize);
4456 
4457 	printf("%3d: (%s) [AS=%d] "
4458 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
4459 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
4460 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
4461 }
4462 
4463 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
4464 {
4465 	uint32_t mas0, mas1, mas3, mas7;
4466 #ifdef __powerpc64__
4467 	uint64_t mas2;
4468 #else
4469 	uint32_t mas2;
4470 #endif
4471 	int entryidx, way, idx;
4472 
4473 	printf("TLB0 entries:\n");
4474 	for (way = 0; way < TLB0_WAYS; way ++)
4475 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
4476 
4477 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4478 			mtspr(SPR_MAS0, mas0);
4479 
4480 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
4481 			mtspr(SPR_MAS2, mas2);
4482 
4483 			__asm __volatile("isync; tlbre");
4484 
4485 			mas1 = mfspr(SPR_MAS1);
4486 			mas2 = mfspr(SPR_MAS2);
4487 			mas3 = mfspr(SPR_MAS3);
4488 			mas7 = mfspr(SPR_MAS7);
4489 
4490 			idx = tlb0_tableidx(mas2, way);
4491 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
4492 		}
4493 }
4494 
4495 /*
4496  * Print out contents of the MAS registers for each TLB1 entry
4497  */
4498 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
4499 {
4500 	uint32_t mas0, mas1, mas3, mas7;
4501 #ifdef __powerpc64__
4502 	uint64_t mas2;
4503 #else
4504 	uint32_t mas2;
4505 #endif
4506 	int i;
4507 
4508 	printf("TLB1 entries:\n");
4509 	for (i = 0; i < TLB1_ENTRIES; i++) {
4510 
4511 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4512 		mtspr(SPR_MAS0, mas0);
4513 
4514 		__asm __volatile("isync; tlbre");
4515 
4516 		mas1 = mfspr(SPR_MAS1);
4517 		mas2 = mfspr(SPR_MAS2);
4518 		mas3 = mfspr(SPR_MAS3);
4519 		mas7 = mfspr(SPR_MAS7);
4520 
4521 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4522 	}
4523 }
4524 #endif
4525