xref: /freebsd/sys/powerpc/booke/pmap.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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 - kernel_pdir-1            : message buffer
63   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
64   *             kernel_pp2d - .                        : kernel pointers to page directory
65   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
66   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
67   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
68   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
69   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
70   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
71   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
72   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
73   *                         - 0xffff_ffff_ffff_ffff    : device direct map
74   */
75 
76 #include <sys/cdefs.h>
77 __FBSDID("$FreeBSD$");
78 
79 #include "opt_ddb.h"
80 #include "opt_kstack_pages.h"
81 
82 #include <sys/param.h>
83 #include <sys/conf.h>
84 #include <sys/malloc.h>
85 #include <sys/ktr.h>
86 #include <sys/proc.h>
87 #include <sys/user.h>
88 #include <sys/queue.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/kerneldump.h>
92 #include <sys/linker.h>
93 #include <sys/msgbuf.h>
94 #include <sys/lock.h>
95 #include <sys/mutex.h>
96 #include <sys/rwlock.h>
97 #include <sys/sched.h>
98 #include <sys/smp.h>
99 #include <sys/vmmeter.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_kern.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_param.h>
108 #include <vm/vm_map.h>
109 #include <vm/vm_pager.h>
110 #include <vm/vm_phys.h>
111 #include <vm/vm_pagequeue.h>
112 #include <vm/uma.h>
113 
114 #include <machine/_inttypes.h>
115 #include <machine/cpu.h>
116 #include <machine/pcb.h>
117 #include <machine/platform.h>
118 
119 #include <machine/tlb.h>
120 #include <machine/spr.h>
121 #include <machine/md_var.h>
122 #include <machine/mmuvar.h>
123 #include <machine/pmap.h>
124 #include <machine/pte.h>
125 
126 #include <ddb/ddb.h>
127 
128 #include "mmu_if.h"
129 
130 #define	SPARSE_MAPDEV
131 #ifdef  DEBUG
132 #define debugf(fmt, args...) printf(fmt, ##args)
133 #else
134 #define debugf(fmt, args...)
135 #endif
136 
137 #ifdef __powerpc64__
138 #define	PRI0ptrX	"016lx"
139 #else
140 #define	PRI0ptrX	"08x"
141 #endif
142 
143 #define TODO			panic("%s: not implemented", __func__);
144 
145 extern unsigned char _etext[];
146 extern unsigned char _end[];
147 
148 extern uint32_t *bootinfo;
149 
150 vm_paddr_t kernload;
151 vm_offset_t kernstart;
152 vm_size_t kernsize;
153 
154 /* Message buffer and tables. */
155 static vm_offset_t data_start;
156 static vm_size_t data_end;
157 
158 /* Phys/avail memory regions. */
159 static struct mem_region *availmem_regions;
160 static int availmem_regions_sz;
161 static struct mem_region *physmem_regions;
162 static int physmem_regions_sz;
163 
164 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
165 static vm_offset_t zero_page_va;
166 static struct mtx zero_page_mutex;
167 
168 static struct mtx tlbivax_mutex;
169 
170 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
171 static vm_offset_t copy_page_src_va;
172 static vm_offset_t copy_page_dst_va;
173 static struct mtx copy_page_mutex;
174 
175 /**************************************************************************/
176 /* PMAP */
177 /**************************************************************************/
178 
179 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
180     vm_prot_t, u_int flags, int8_t psind);
181 
182 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
183 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
184 #ifdef __powerpc64__
185 unsigned int kernel_pdirs;
186 #endif
187 
188 /*
189  * If user pmap is processed with mmu_booke_remove and the resident count
190  * drops to 0, there are no more pages to remove, so we need not continue.
191  */
192 #define PMAP_REMOVE_DONE(pmap) \
193 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
194 
195 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
196 extern int elf32_nxstack;
197 #endif
198 
199 /**************************************************************************/
200 /* TLB and TID handling */
201 /**************************************************************************/
202 
203 /* Translation ID busy table */
204 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
205 
206 /*
207  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
208  * core revisions and should be read from h/w registers during early config.
209  */
210 uint32_t tlb0_entries;
211 uint32_t tlb0_ways;
212 uint32_t tlb0_entries_per_way;
213 uint32_t tlb1_entries;
214 
215 #define TLB0_ENTRIES		(tlb0_entries)
216 #define TLB0_WAYS		(tlb0_ways)
217 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
218 
219 #define TLB1_ENTRIES (tlb1_entries)
220 
221 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
222 
223 static tlbtid_t tid_alloc(struct pmap *);
224 static void tid_flush(tlbtid_t tid);
225 
226 #ifdef DDB
227 #ifdef __powerpc64__
228 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
229 #else
230 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
231 #endif
232 #endif
233 
234 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
235 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
236 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
237 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
238 
239 static vm_size_t tsize2size(unsigned int);
240 static unsigned int size2tsize(vm_size_t);
241 static unsigned int ilog2(unsigned long);
242 
243 static void set_mas4_defaults(void);
244 
245 static inline void tlb0_flush_entry(vm_offset_t);
246 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
247 
248 /**************************************************************************/
249 /* Page table management */
250 /**************************************************************************/
251 
252 static struct rwlock_padalign pvh_global_lock;
253 
254 /* Data for the pv entry allocation mechanism */
255 static uma_zone_t pvzone;
256 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
257 
258 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
259 
260 #ifndef PMAP_SHPGPERPROC
261 #define PMAP_SHPGPERPROC	200
262 #endif
263 
264 #ifdef __powerpc64__
265 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
266 			 unsigned int, boolean_t);
267 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int);
268 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
269 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
270 #else
271 static void ptbl_init(void);
272 static struct ptbl_buf *ptbl_buf_alloc(void);
273 static void ptbl_buf_free(struct ptbl_buf *);
274 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
275 
276 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
277 static void ptbl_free(mmu_t, pmap_t, unsigned int);
278 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
279 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
280 #endif
281 
282 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
283 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
284 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
285 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
286 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
287 
288 static pv_entry_t pv_alloc(void);
289 static void pv_free(pv_entry_t);
290 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
291 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
292 
293 static void booke_pmap_init_qpages(void);
294 
295 struct ptbl_buf {
296 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
297 	vm_offset_t kva;		/* va of mapping */
298 };
299 
300 #ifndef __powerpc64__
301 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
302 #define PTBL_BUFS		(128 * 16)
303 
304 /* ptbl free list and a lock used for access synchronization. */
305 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
306 static struct mtx ptbl_buf_freelist_lock;
307 
308 /* Base address of kva space allocated fot ptbl bufs. */
309 static vm_offset_t ptbl_buf_pool_vabase;
310 
311 /* Pointer to ptbl_buf structures. */
312 static struct ptbl_buf *ptbl_bufs;
313 #endif
314 
315 #ifdef SMP
316 extern tlb_entry_t __boot_tlb1[];
317 void pmap_bootstrap_ap(volatile uint32_t *);
318 #endif
319 
320 /*
321  * Kernel MMU interface
322  */
323 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
324 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
325     vm_size_t, vm_offset_t);
326 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
327 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
328     vm_offset_t, vm_page_t *, vm_offset_t, int);
329 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
330     vm_prot_t, u_int flags, int8_t psind);
331 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
332     vm_page_t, vm_prot_t);
333 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
334     vm_prot_t);
335 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
336 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
337     vm_prot_t);
338 static void		mmu_booke_init(mmu_t);
339 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
340 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
341 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
342 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
343 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
344     int);
345 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
346     vm_paddr_t *);
347 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
348     vm_object_t, vm_pindex_t, vm_size_t);
349 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
350 static void		mmu_booke_page_init(mmu_t, vm_page_t);
351 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
352 static void		mmu_booke_pinit(mmu_t, pmap_t);
353 static void		mmu_booke_pinit0(mmu_t, pmap_t);
354 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
355     vm_prot_t);
356 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
357 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
358 static void		mmu_booke_release(mmu_t, pmap_t);
359 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
360 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
361 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
362 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
363 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
364 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
365 static void		mmu_booke_activate(mmu_t, struct thread *);
366 static void		mmu_booke_deactivate(mmu_t, struct thread *);
367 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
368 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
369 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
370 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
371 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
372 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
373 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
374 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
375 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
376 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
377     vm_size_t);
378 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
379     void **);
380 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
381     void *);
382 static void		mmu_booke_scan_init(mmu_t);
383 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
384 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
385 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
386     vm_size_t sz, vm_memattr_t mode);
387 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
388     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
389 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
390     int *is_user, vm_offset_t *decoded_addr);
391 
392 
393 static mmu_method_t mmu_booke_methods[] = {
394 	/* pmap dispatcher interface */
395 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
396 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
397 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
398 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
399 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
400 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
401 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
402 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
403 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
404 	MMUMETHOD(mmu_init,		mmu_booke_init),
405 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
406 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
407 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
408 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
409 	MMUMETHOD(mmu_map,		mmu_booke_map),
410 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
411 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
412 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
413 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
414 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
415 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
416 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
417 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
418 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
419 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
420 	MMUMETHOD(mmu_release,		mmu_booke_release),
421 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
422 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
423 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
424 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
425 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
426 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
427 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
428 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
429 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
430 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
431 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
432 
433 	/* Internal interfaces */
434 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
435 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
436 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
437 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
438 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
439 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
440 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
441 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
442 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
443 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
444 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
445 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
446 
447 	/* dumpsys() support */
448 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
449 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
450 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
451 
452 	{ 0, 0 }
453 };
454 
455 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
456 
457 static __inline uint32_t
458 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
459 {
460 	uint32_t attrib;
461 	int i;
462 
463 	if (ma != VM_MEMATTR_DEFAULT) {
464 		switch (ma) {
465 		case VM_MEMATTR_UNCACHEABLE:
466 			return (MAS2_I | MAS2_G);
467 		case VM_MEMATTR_WRITE_COMBINING:
468 		case VM_MEMATTR_WRITE_BACK:
469 		case VM_MEMATTR_PREFETCHABLE:
470 			return (MAS2_I);
471 		case VM_MEMATTR_WRITE_THROUGH:
472 			return (MAS2_W | MAS2_M);
473 		case VM_MEMATTR_CACHEABLE:
474 			return (MAS2_M);
475 		}
476 	}
477 
478 	/*
479 	 * Assume the page is cache inhibited and access is guarded unless
480 	 * it's in our available memory array.
481 	 */
482 	attrib = _TLB_ENTRY_IO;
483 	for (i = 0; i < physmem_regions_sz; i++) {
484 		if ((pa >= physmem_regions[i].mr_start) &&
485 		    (pa < (physmem_regions[i].mr_start +
486 		     physmem_regions[i].mr_size))) {
487 			attrib = _TLB_ENTRY_MEM;
488 			break;
489 		}
490 	}
491 
492 	return (attrib);
493 }
494 
495 static inline void
496 tlb_miss_lock(void)
497 {
498 #ifdef SMP
499 	struct pcpu *pc;
500 
501 	if (!smp_started)
502 		return;
503 
504 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
505 		if (pc != pcpup) {
506 
507 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
508 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
509 
510 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
511 			    ("tlb_miss_lock: tried to lock self"));
512 
513 			tlb_lock(pc->pc_booke.tlb_lock);
514 
515 			CTR1(KTR_PMAP, "%s: locked", __func__);
516 		}
517 	}
518 #endif
519 }
520 
521 static inline void
522 tlb_miss_unlock(void)
523 {
524 #ifdef SMP
525 	struct pcpu *pc;
526 
527 	if (!smp_started)
528 		return;
529 
530 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
531 		if (pc != pcpup) {
532 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
533 			    __func__, pc->pc_cpuid);
534 
535 			tlb_unlock(pc->pc_booke.tlb_lock);
536 
537 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
538 		}
539 	}
540 #endif
541 }
542 
543 /* Return number of entries in TLB0. */
544 static __inline void
545 tlb0_get_tlbconf(void)
546 {
547 	uint32_t tlb0_cfg;
548 
549 	tlb0_cfg = mfspr(SPR_TLB0CFG);
550 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
551 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
552 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
553 }
554 
555 /* Return number of entries in TLB1. */
556 static __inline void
557 tlb1_get_tlbconf(void)
558 {
559 	uint32_t tlb1_cfg;
560 
561 	tlb1_cfg = mfspr(SPR_TLB1CFG);
562 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
563 }
564 
565 /**************************************************************************/
566 /* Page table related */
567 /**************************************************************************/
568 
569 #ifdef __powerpc64__
570 /* Initialize pool of kva ptbl buffers. */
571 static void
572 ptbl_init(void)
573 {
574 }
575 
576 /* Get a pointer to a PTE in a page table. */
577 static __inline pte_t *
578 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
579 {
580 	pte_t         **pdir;
581 	pte_t          *ptbl;
582 
583 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
584 
585 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
586 	if (!pdir)
587 		return NULL;
588 	ptbl = pdir[PDIR_IDX(va)];
589 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
590 }
591 
592 /*
593  * allocate a page of pointers to page directories, do not preallocate the
594  * page tables
595  */
596 static pte_t  **
597 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
598 {
599 	vm_page_t	m;
600 	pte_t          **pdir;
601 	int		req;
602 
603 	KASSERT((pdir[pp2d_idx] == NULL),
604 		("%s: valid pdir entry exists!", __func__));
605 
606 	req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
607 	while ((m = vm_page_alloc(NULL, pp2d_idx, req)) == NULL) {
608 		PMAP_UNLOCK(pmap);
609 		if (nosleep) {
610 			return (NULL);
611 		}
612 		vm_wait(NULL);
613 		PMAP_LOCK(pmap);
614 	}
615 
616 	/* Zero whole ptbl. */
617 	pdir = (pte_t **)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
618 	bzero(pdir, PAGE_SIZE);
619 
620 	return (pdir);
621 }
622 
623 /* Free pdir pages and invalidate pdir entry. */
624 static void
625 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
626 {
627 	pte_t         **pdir;
628 	vm_paddr_t	pa;
629 	vm_offset_t	va;
630 	vm_page_t	m;
631 
632 	pdir = pmap->pm_pp2d[pp2d_idx];
633 
634 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
635 
636 	pmap->pm_pp2d[pp2d_idx] = NULL;
637 
638 	va = (vm_offset_t) pdir;
639 	pa = DMAP_TO_PHYS(va);
640 	m = PHYS_TO_VM_PAGE(pa);
641 	vm_page_free_zero(m);
642 }
643 
644 /*
645  * Decrement pdir pages hold count and attempt to free pdir pages. Called
646  * when removing directory entry from pdir.
647  *
648  * Return 1 if pdir pages were freed.
649  */
650 static int
651 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
652 {
653 	pte_t         **pdir;
654 	vm_paddr_t	pa;
655 	vm_page_t	m;
656 
657 	KASSERT((pmap != kernel_pmap),
658 		("pdir_unhold: unholding kernel pdir!"));
659 
660 	pdir = pmap->pm_pp2d[pp2d_idx];
661 
662 	/* decrement hold count */
663 	pa = DMAP_TO_PHYS((vm_offset_t) pdir);
664 	m = PHYS_TO_VM_PAGE(pa);
665 
666 	/*
667 	 * Free pdir page if there are no dir entries in this pdir.
668 	 */
669 	if (vm_page_unwire_noq(m)) {
670 		pdir_free(mmu, pmap, pp2d_idx);
671 		return (1);
672 	}
673 	return (0);
674 }
675 
676 /*
677  * Increment hold count for pdir pages. This routine is used when new ptlb
678  * entry is being inserted into pdir.
679  */
680 static void
681 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
682 {
683 	vm_page_t	m;
684 
685 	KASSERT((pmap != kernel_pmap),
686 		("pdir_hold: holding kernel pdir!"));
687 
688 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
689 
690 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pdir));
691 	vm_page_wire(m);
692 }
693 
694 /* Allocate page table. */
695 static pte_t   *
696 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
697     boolean_t nosleep)
698 {
699 	vm_page_t	m;
700 	pte_t          *ptbl;
701 	int		req;
702 
703 	KASSERT((pdir[pdir_idx] == NULL),
704 		("%s: valid ptbl entry exists!", __func__));
705 
706 	req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
707 	while ((m = vm_page_alloc(NULL, pdir_idx, req)) == NULL) {
708 		PMAP_UNLOCK(pmap);
709 		rw_wunlock(&pvh_global_lock);
710 		if (nosleep) {
711 			return (NULL);
712 		}
713 		vm_wait(NULL);
714 		rw_wlock(&pvh_global_lock);
715 		PMAP_LOCK(pmap);
716 	}
717 
718 	/* Zero whole ptbl. */
719 	ptbl = (pte_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
720 	bzero(ptbl, PAGE_SIZE);
721 
722 	return (ptbl);
723 }
724 
725 /* Free ptbl pages and invalidate pdir entry. */
726 static void
727 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
728 {
729 	pte_t          *ptbl;
730 	vm_paddr_t	pa;
731 	vm_offset_t	va;
732 	vm_page_t	m;
733 
734 	ptbl = pdir[pdir_idx];
735 
736 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
737 
738 	pdir[pdir_idx] = NULL;
739 
740 	va = (vm_offset_t) ptbl;
741 	pa = DMAP_TO_PHYS(va);
742 	m = PHYS_TO_VM_PAGE(pa);
743 	vm_page_free_zero(m);
744 }
745 
746 /*
747  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
748  * when removing pte entry from ptbl.
749  *
750  * Return 1 if ptbl pages were freed.
751  */
752 static int
753 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
754 {
755 	pte_t          *ptbl;
756 	vm_page_t	m;
757 	u_int		pp2d_idx;
758 	pte_t         **pdir;
759 	u_int		pdir_idx;
760 
761 	pp2d_idx = PP2D_IDX(va);
762 	pdir_idx = PDIR_IDX(va);
763 
764 	KASSERT((pmap != kernel_pmap),
765 		("ptbl_unhold: unholding kernel ptbl!"));
766 
767 	pdir = pmap->pm_pp2d[pp2d_idx];
768 	ptbl = pdir[pdir_idx];
769 
770 	/* decrement hold count */
771 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
772 
773 	/*
774 	 * Free ptbl pages if there are no pte entries in this ptbl.
775 	 * wire_count has the same value for all ptbl pages, so check the
776 	 * last page.
777 	 */
778 	if (vm_page_unwire_noq(m)) {
779 		ptbl_free(mmu, pmap, pdir, pdir_idx);
780 		pdir_unhold(mmu, pmap, pp2d_idx);
781 		return (1);
782 	}
783 	return (0);
784 }
785 
786 /*
787  * Increment hold count for ptbl pages. This routine is used when new pte
788  * entry is being inserted into ptbl.
789  */
790 static void
791 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
792 {
793 	pte_t          *ptbl;
794 	vm_page_t	m;
795 
796 	KASSERT((pmap != kernel_pmap),
797 		("ptbl_hold: holding kernel ptbl!"));
798 
799 	ptbl = pdir[pdir_idx];
800 
801 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
802 
803 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
804 	vm_page_wire(m);
805 }
806 #else
807 
808 /* Initialize pool of kva ptbl buffers. */
809 static void
810 ptbl_init(void)
811 {
812 	int i;
813 
814 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
815 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
816 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
817 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
818 
819 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
820 	TAILQ_INIT(&ptbl_buf_freelist);
821 
822 	for (i = 0; i < PTBL_BUFS; i++) {
823 		ptbl_bufs[i].kva =
824 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
825 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
826 	}
827 }
828 
829 /* Get a ptbl_buf from the freelist. */
830 static struct ptbl_buf *
831 ptbl_buf_alloc(void)
832 {
833 	struct ptbl_buf *buf;
834 
835 	mtx_lock(&ptbl_buf_freelist_lock);
836 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
837 	if (buf != NULL)
838 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
839 	mtx_unlock(&ptbl_buf_freelist_lock);
840 
841 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
842 
843 	return (buf);
844 }
845 
846 /* Return ptbl buff to free pool. */
847 static void
848 ptbl_buf_free(struct ptbl_buf *buf)
849 {
850 
851 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
852 
853 	mtx_lock(&ptbl_buf_freelist_lock);
854 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
855 	mtx_unlock(&ptbl_buf_freelist_lock);
856 }
857 
858 /*
859  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
860  */
861 static void
862 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
863 {
864 	struct ptbl_buf *pbuf;
865 
866 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
867 
868 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
869 
870 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
871 		if (pbuf->kva == (vm_offset_t)ptbl) {
872 			/* Remove from pmap ptbl buf list. */
873 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
874 
875 			/* Free corresponding ptbl buf. */
876 			ptbl_buf_free(pbuf);
877 			break;
878 		}
879 }
880 
881 /* Allocate page table. */
882 static pte_t *
883 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
884 {
885 	vm_page_t mtbl[PTBL_PAGES];
886 	vm_page_t m;
887 	struct ptbl_buf *pbuf;
888 	unsigned int pidx;
889 	pte_t *ptbl;
890 	int i, j;
891 
892 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
893 	    (pmap == kernel_pmap), pdir_idx);
894 
895 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
896 	    ("ptbl_alloc: invalid pdir_idx"));
897 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
898 	    ("pte_alloc: valid ptbl entry exists!"));
899 
900 	pbuf = ptbl_buf_alloc();
901 	if (pbuf == NULL)
902 		panic("pte_alloc: couldn't alloc kernel virtual memory");
903 
904 	ptbl = (pte_t *)pbuf->kva;
905 
906 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
907 
908 	for (i = 0; i < PTBL_PAGES; i++) {
909 		pidx = (PTBL_PAGES * pdir_idx) + i;
910 		while ((m = vm_page_alloc(NULL, pidx,
911 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
912 			PMAP_UNLOCK(pmap);
913 			rw_wunlock(&pvh_global_lock);
914 			if (nosleep) {
915 				ptbl_free_pmap_ptbl(pmap, ptbl);
916 				for (j = 0; j < i; j++)
917 					vm_page_free(mtbl[j]);
918 				vm_wire_sub(i);
919 				return (NULL);
920 			}
921 			vm_wait(NULL);
922 			rw_wlock(&pvh_global_lock);
923 			PMAP_LOCK(pmap);
924 		}
925 		mtbl[i] = m;
926 	}
927 
928 	/* Map allocated pages into kernel_pmap. */
929 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
930 
931 	/* Zero whole ptbl. */
932 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
933 
934 	/* Add pbuf to the pmap ptbl bufs list. */
935 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
936 
937 	return (ptbl);
938 }
939 
940 /* Free ptbl pages and invalidate pdir entry. */
941 static void
942 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
943 {
944 	pte_t *ptbl;
945 	vm_paddr_t pa;
946 	vm_offset_t va;
947 	vm_page_t m;
948 	int i;
949 
950 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
951 	    (pmap == kernel_pmap), pdir_idx);
952 
953 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
954 	    ("ptbl_free: invalid pdir_idx"));
955 
956 	ptbl = pmap->pm_pdir[pdir_idx];
957 
958 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
959 
960 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
961 
962 	/*
963 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
964 	 * don't attempt to look up the page tables we are releasing.
965 	 */
966 	mtx_lock_spin(&tlbivax_mutex);
967 	tlb_miss_lock();
968 
969 	pmap->pm_pdir[pdir_idx] = NULL;
970 
971 	tlb_miss_unlock();
972 	mtx_unlock_spin(&tlbivax_mutex);
973 
974 	for (i = 0; i < PTBL_PAGES; i++) {
975 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
976 		pa = pte_vatopa(mmu, kernel_pmap, va);
977 		m = PHYS_TO_VM_PAGE(pa);
978 		vm_page_free_zero(m);
979 		vm_wire_sub(1);
980 		mmu_booke_kremove(mmu, va);
981 	}
982 
983 	ptbl_free_pmap_ptbl(pmap, ptbl);
984 }
985 
986 /*
987  * Decrement ptbl pages hold count and attempt to free ptbl pages.
988  * Called when removing pte entry from ptbl.
989  *
990  * Return 1 if ptbl pages were freed.
991  */
992 static int
993 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
994 {
995 	pte_t *ptbl;
996 	vm_paddr_t pa;
997 	vm_page_t m;
998 	int i;
999 
1000 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1001 	    (pmap == kernel_pmap), pdir_idx);
1002 
1003 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1004 	    ("ptbl_unhold: invalid pdir_idx"));
1005 	KASSERT((pmap != kernel_pmap),
1006 	    ("ptbl_unhold: unholding kernel ptbl!"));
1007 
1008 	ptbl = pmap->pm_pdir[pdir_idx];
1009 
1010 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1011 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1012 	    ("ptbl_unhold: non kva ptbl"));
1013 
1014 	/* decrement hold count */
1015 	for (i = 0; i < PTBL_PAGES; i++) {
1016 		pa = pte_vatopa(mmu, kernel_pmap,
1017 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1018 		m = PHYS_TO_VM_PAGE(pa);
1019 		m->wire_count--;
1020 	}
1021 
1022 	/*
1023 	 * Free ptbl pages if there are no pte etries in this ptbl.
1024 	 * wire_count has the same value for all ptbl pages, so check the last
1025 	 * page.
1026 	 */
1027 	if (m->wire_count == 0) {
1028 		ptbl_free(mmu, pmap, pdir_idx);
1029 
1030 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1031 		return (1);
1032 	}
1033 
1034 	return (0);
1035 }
1036 
1037 /*
1038  * Increment hold count for ptbl pages. This routine is used when a new pte
1039  * entry is being inserted into the ptbl.
1040  */
1041 static void
1042 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1043 {
1044 	vm_paddr_t pa;
1045 	pte_t *ptbl;
1046 	vm_page_t m;
1047 	int i;
1048 
1049 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1050 	    pdir_idx);
1051 
1052 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1053 	    ("ptbl_hold: invalid pdir_idx"));
1054 	KASSERT((pmap != kernel_pmap),
1055 	    ("ptbl_hold: holding kernel ptbl!"));
1056 
1057 	ptbl = pmap->pm_pdir[pdir_idx];
1058 
1059 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1060 
1061 	for (i = 0; i < PTBL_PAGES; i++) {
1062 		pa = pte_vatopa(mmu, kernel_pmap,
1063 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1064 		m = PHYS_TO_VM_PAGE(pa);
1065 		m->wire_count++;
1066 	}
1067 }
1068 #endif
1069 
1070 /* Allocate pv_entry structure. */
1071 pv_entry_t
1072 pv_alloc(void)
1073 {
1074 	pv_entry_t pv;
1075 
1076 	pv_entry_count++;
1077 	if (pv_entry_count > pv_entry_high_water)
1078 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1079 	pv = uma_zalloc(pvzone, M_NOWAIT);
1080 
1081 	return (pv);
1082 }
1083 
1084 /* Free pv_entry structure. */
1085 static __inline void
1086 pv_free(pv_entry_t pve)
1087 {
1088 
1089 	pv_entry_count--;
1090 	uma_zfree(pvzone, pve);
1091 }
1092 
1093 
1094 /* Allocate and initialize pv_entry structure. */
1095 static void
1096 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1097 {
1098 	pv_entry_t pve;
1099 
1100 	//int su = (pmap == kernel_pmap);
1101 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1102 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1103 
1104 	pve = pv_alloc();
1105 	if (pve == NULL)
1106 		panic("pv_insert: no pv entries!");
1107 
1108 	pve->pv_pmap = pmap;
1109 	pve->pv_va = va;
1110 
1111 	/* add to pv_list */
1112 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1113 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1114 
1115 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1116 
1117 	//debugf("pv_insert: e\n");
1118 }
1119 
1120 /* Destroy pv entry. */
1121 static void
1122 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1123 {
1124 	pv_entry_t pve;
1125 
1126 	//int su = (pmap == kernel_pmap);
1127 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1128 
1129 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1130 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1131 
1132 	/* find pv entry */
1133 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1134 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1135 			/* remove from pv_list */
1136 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1137 			if (TAILQ_EMPTY(&m->md.pv_list))
1138 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1139 
1140 			/* free pv entry struct */
1141 			pv_free(pve);
1142 			break;
1143 		}
1144 	}
1145 
1146 	//debugf("pv_remove: e\n");
1147 }
1148 
1149 #ifdef __powerpc64__
1150 /*
1151  * Clean pte entry, try to free page table page if requested.
1152  *
1153  * Return 1 if ptbl pages were freed, otherwise return 0.
1154  */
1155 static int
1156 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1157 {
1158 	vm_page_t	m;
1159 	pte_t          *pte;
1160 
1161 	pte = pte_find(mmu, pmap, va);
1162 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1163 
1164 	if (!PTE_ISVALID(pte))
1165 		return (0);
1166 
1167 	/* Get vm_page_t for mapped pte. */
1168 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1169 
1170 	if (PTE_ISWIRED(pte))
1171 		pmap->pm_stats.wired_count--;
1172 
1173 	/* Handle managed entry. */
1174 	if (PTE_ISMANAGED(pte)) {
1175 
1176 		/* Handle modified pages. */
1177 		if (PTE_ISMODIFIED(pte))
1178 			vm_page_dirty(m);
1179 
1180 		/* Referenced pages. */
1181 		if (PTE_ISREFERENCED(pte))
1182 			vm_page_aflag_set(m, PGA_REFERENCED);
1183 
1184 		/* Remove pv_entry from pv_list. */
1185 		pv_remove(pmap, va, m);
1186 	} else if (m->md.pv_tracked) {
1187 		pv_remove(pmap, va, m);
1188 		if (TAILQ_EMPTY(&m->md.pv_list))
1189 			m->md.pv_tracked = false;
1190 	}
1191 	mtx_lock_spin(&tlbivax_mutex);
1192 	tlb_miss_lock();
1193 
1194 	tlb0_flush_entry(va);
1195 	*pte = 0;
1196 
1197 	tlb_miss_unlock();
1198 	mtx_unlock_spin(&tlbivax_mutex);
1199 
1200 	pmap->pm_stats.resident_count--;
1201 
1202 	if (flags & PTBL_UNHOLD) {
1203 		return (ptbl_unhold(mmu, pmap, va));
1204 	}
1205 	return (0);
1206 }
1207 
1208 /*
1209  * Insert PTE for a given page and virtual address.
1210  */
1211 static int
1212 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1213     boolean_t nosleep)
1214 {
1215 	unsigned int	pp2d_idx = PP2D_IDX(va);
1216 	unsigned int	pdir_idx = PDIR_IDX(va);
1217 	unsigned int	ptbl_idx = PTBL_IDX(va);
1218 	pte_t          *ptbl, *pte, pte_tmp;
1219 	pte_t         **pdir;
1220 
1221 	/* Get the page directory pointer. */
1222 	pdir = pmap->pm_pp2d[pp2d_idx];
1223 	if (pdir == NULL)
1224 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1225 
1226 	/* Get the page table pointer. */
1227 	ptbl = pdir[pdir_idx];
1228 
1229 	if (ptbl == NULL) {
1230 		/* Allocate page table pages. */
1231 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1232 		if (ptbl == NULL) {
1233 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1234 			return (ENOMEM);
1235 		}
1236 		pte = &ptbl[ptbl_idx];
1237 	} else {
1238 		/*
1239 		 * Check if there is valid mapping for requested va, if there
1240 		 * is, remove it.
1241 		 */
1242 		pte = &ptbl[ptbl_idx];
1243 		if (PTE_ISVALID(pte)) {
1244 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1245 		} else {
1246 			/*
1247 			 * pte is not used, increment hold count for ptbl
1248 			 * pages.
1249 			 */
1250 			if (pmap != kernel_pmap)
1251 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1252 		}
1253 	}
1254 
1255 	if (pdir[pdir_idx] == NULL) {
1256 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1257 			pdir_hold(mmu, pmap, pdir);
1258 		pdir[pdir_idx] = ptbl;
1259 	}
1260 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1261 		pmap->pm_pp2d[pp2d_idx] = pdir;
1262 
1263 	/*
1264 	 * Insert pv_entry into pv_list for mapped page if part of managed
1265 	 * memory.
1266 	 */
1267 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1268 		flags |= PTE_MANAGED;
1269 
1270 		/* Create and insert pv entry. */
1271 		pv_insert(pmap, va, m);
1272 	}
1273 
1274 	pmap->pm_stats.resident_count++;
1275 
1276 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1277 	pte_tmp |= (PTE_VALID | flags);
1278 
1279 	mtx_lock_spin(&tlbivax_mutex);
1280 	tlb_miss_lock();
1281 
1282 	tlb0_flush_entry(va);
1283 	*pte = pte_tmp;
1284 
1285 	tlb_miss_unlock();
1286 	mtx_unlock_spin(&tlbivax_mutex);
1287 
1288 	return (0);
1289 }
1290 
1291 /* Return the pa for the given pmap/va. */
1292 static	vm_paddr_t
1293 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1294 {
1295 	vm_paddr_t	pa = 0;
1296 	pte_t          *pte;
1297 
1298 	pte = pte_find(mmu, pmap, va);
1299 	if ((pte != NULL) && PTE_ISVALID(pte))
1300 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1301 	return (pa);
1302 }
1303 
1304 
1305 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1306 static void
1307 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1308 {
1309 	int		i, j;
1310 	vm_offset_t	va;
1311 	pte_t		*pte;
1312 
1313 	va = addr;
1314 	/* Initialize kernel pdir */
1315 	for (i = 0; i < kernel_pdirs; i++) {
1316 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1317 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1318 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1319 		    j < PDIR_NENTRIES; j++) {
1320 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1321 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE) +
1322 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE));
1323 		}
1324 	}
1325 
1326 	/*
1327 	 * Fill in PTEs covering kernel code and data. They are not required
1328 	 * for address translation, as this area is covered by static TLB1
1329 	 * entries, but for pte_vatopa() to work correctly with kernel area
1330 	 * addresses.
1331 	 */
1332 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1333 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1334 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1335 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1336 		    PTE_VALID | PTE_PS_4KB;
1337 	}
1338 }
1339 #else
1340 /*
1341  * Clean pte entry, try to free page table page if requested.
1342  *
1343  * Return 1 if ptbl pages were freed, otherwise return 0.
1344  */
1345 static int
1346 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1347 {
1348 	unsigned int pdir_idx = PDIR_IDX(va);
1349 	unsigned int ptbl_idx = PTBL_IDX(va);
1350 	vm_page_t m;
1351 	pte_t *ptbl;
1352 	pte_t *pte;
1353 
1354 	//int su = (pmap == kernel_pmap);
1355 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1356 	//		su, (u_int32_t)pmap, va, flags);
1357 
1358 	ptbl = pmap->pm_pdir[pdir_idx];
1359 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1360 
1361 	pte = &ptbl[ptbl_idx];
1362 
1363 	if (pte == NULL || !PTE_ISVALID(pte))
1364 		return (0);
1365 
1366 	if (PTE_ISWIRED(pte))
1367 		pmap->pm_stats.wired_count--;
1368 
1369 	/* Get vm_page_t for mapped pte. */
1370 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1371 
1372 	/* Handle managed entry. */
1373 	if (PTE_ISMANAGED(pte)) {
1374 
1375 		if (PTE_ISMODIFIED(pte))
1376 			vm_page_dirty(m);
1377 
1378 		if (PTE_ISREFERENCED(pte))
1379 			vm_page_aflag_set(m, PGA_REFERENCED);
1380 
1381 		pv_remove(pmap, va, m);
1382 	} else if (m->md.pv_tracked) {
1383 		/*
1384 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1385 		 * used.  This is needed by the NCSW support code for fast
1386 		 * VA<->PA translation.
1387 		 */
1388 		pv_remove(pmap, va, m);
1389 		if (TAILQ_EMPTY(&m->md.pv_list))
1390 			m->md.pv_tracked = false;
1391 	}
1392 
1393 	mtx_lock_spin(&tlbivax_mutex);
1394 	tlb_miss_lock();
1395 
1396 	tlb0_flush_entry(va);
1397 	*pte = 0;
1398 
1399 	tlb_miss_unlock();
1400 	mtx_unlock_spin(&tlbivax_mutex);
1401 
1402 	pmap->pm_stats.resident_count--;
1403 
1404 	if (flags & PTBL_UNHOLD) {
1405 		//debugf("pte_remove: e (unhold)\n");
1406 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1407 	}
1408 
1409 	//debugf("pte_remove: e\n");
1410 	return (0);
1411 }
1412 
1413 /*
1414  * Insert PTE for a given page and virtual address.
1415  */
1416 static int
1417 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1418     boolean_t nosleep)
1419 {
1420 	unsigned int pdir_idx = PDIR_IDX(va);
1421 	unsigned int ptbl_idx = PTBL_IDX(va);
1422 	pte_t *ptbl, *pte, pte_tmp;
1423 
1424 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1425 	    pmap == kernel_pmap, pmap, va);
1426 
1427 	/* Get the page table pointer. */
1428 	ptbl = pmap->pm_pdir[pdir_idx];
1429 
1430 	if (ptbl == NULL) {
1431 		/* Allocate page table pages. */
1432 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1433 		if (ptbl == NULL) {
1434 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1435 			return (ENOMEM);
1436 		}
1437 		pmap->pm_pdir[pdir_idx] = ptbl;
1438 		pte = &ptbl[ptbl_idx];
1439 	} else {
1440 		/*
1441 		 * Check if there is valid mapping for requested
1442 		 * va, if there is, remove it.
1443 		 */
1444 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1445 		if (PTE_ISVALID(pte)) {
1446 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1447 		} else {
1448 			/*
1449 			 * pte is not used, increment hold count
1450 			 * for ptbl pages.
1451 			 */
1452 			if (pmap != kernel_pmap)
1453 				ptbl_hold(mmu, pmap, pdir_idx);
1454 		}
1455 	}
1456 
1457 	/*
1458 	 * Insert pv_entry into pv_list for mapped page if part of managed
1459 	 * memory.
1460 	 */
1461 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1462 		flags |= PTE_MANAGED;
1463 
1464 		/* Create and insert pv entry. */
1465 		pv_insert(pmap, va, m);
1466 	}
1467 
1468 	pmap->pm_stats.resident_count++;
1469 
1470 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1471 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1472 
1473 	mtx_lock_spin(&tlbivax_mutex);
1474 	tlb_miss_lock();
1475 
1476 	tlb0_flush_entry(va);
1477 	*pte = pte_tmp;
1478 
1479 	tlb_miss_unlock();
1480 	mtx_unlock_spin(&tlbivax_mutex);
1481 	return (0);
1482 }
1483 
1484 /* Return the pa for the given pmap/va. */
1485 static vm_paddr_t
1486 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1487 {
1488 	vm_paddr_t pa = 0;
1489 	pte_t *pte;
1490 
1491 	pte = pte_find(mmu, pmap, va);
1492 	if ((pte != NULL) && PTE_ISVALID(pte))
1493 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1494 	return (pa);
1495 }
1496 
1497 /* Get a pointer to a PTE in a page table. */
1498 static pte_t *
1499 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1500 {
1501 	unsigned int pdir_idx = PDIR_IDX(va);
1502 	unsigned int ptbl_idx = PTBL_IDX(va);
1503 
1504 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1505 
1506 	if (pmap->pm_pdir[pdir_idx])
1507 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1508 
1509 	return (NULL);
1510 }
1511 
1512 /* Set up kernel page tables. */
1513 static void
1514 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1515 {
1516 	int		i;
1517 	vm_offset_t	va;
1518 	pte_t		*pte;
1519 
1520 	/* Initialize kernel pdir */
1521 	for (i = 0; i < kernel_ptbls; i++)
1522 		kernel_pmap->pm_pdir[kptbl_min + i] =
1523 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1524 
1525 	/*
1526 	 * Fill in PTEs covering kernel code and data. They are not required
1527 	 * for address translation, as this area is covered by static TLB1
1528 	 * entries, but for pte_vatopa() to work correctly with kernel area
1529 	 * addresses.
1530 	 */
1531 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1532 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1533 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1534 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1535 		    PTE_VALID | PTE_PS_4KB;
1536 	}
1537 }
1538 #endif
1539 
1540 /**************************************************************************/
1541 /* PMAP related */
1542 /**************************************************************************/
1543 
1544 /*
1545  * This is called during booke_init, before the system is really initialized.
1546  */
1547 static void
1548 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1549 {
1550 	vm_paddr_t phys_kernelend;
1551 	struct mem_region *mp, *mp1;
1552 	int cnt, i, j;
1553 	vm_paddr_t s, e, sz;
1554 	vm_paddr_t physsz, hwphyssz;
1555 	u_int phys_avail_count;
1556 	vm_size_t kstack0_sz;
1557 	vm_offset_t kernel_pdir, kstack0;
1558 	vm_paddr_t kstack0_phys;
1559 	void *dpcpu;
1560 
1561 	debugf("mmu_booke_bootstrap: entered\n");
1562 
1563 	/* Set interesting system properties */
1564 #ifdef __powerpc64__
1565 	hw_direct_map = 1;
1566 #else
1567 	hw_direct_map = 0;
1568 #endif
1569 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1570 	elf32_nxstack = 1;
1571 #endif
1572 
1573 	/* Initialize invalidation mutex */
1574 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1575 
1576 	/* Read TLB0 size and associativity. */
1577 	tlb0_get_tlbconf();
1578 
1579 	/*
1580 	 * Align kernel start and end address (kernel image).
1581 	 * Note that kernel end does not necessarily relate to kernsize.
1582 	 * kernsize is the size of the kernel that is actually mapped.
1583 	 */
1584 	kernstart = trunc_page(start);
1585 	data_start = round_page(kernelend);
1586 	data_end = data_start;
1587 
1588 	/* Allocate the dynamic per-cpu area. */
1589 	dpcpu = (void *)data_end;
1590 	data_end += DPCPU_SIZE;
1591 
1592 	/* Allocate space for the message buffer. */
1593 	msgbufp = (struct msgbuf *)data_end;
1594 	data_end += msgbufsize;
1595 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1596 	    (uintptr_t)msgbufp, data_end);
1597 
1598 	data_end = round_page(data_end);
1599 
1600 #ifndef __powerpc64__
1601 	/* Allocate space for ptbl_bufs. */
1602 	ptbl_bufs = (struct ptbl_buf *)data_end;
1603 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1604 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1605 	    (uintptr_t)ptbl_bufs, data_end);
1606 
1607 	data_end = round_page(data_end);
1608 #endif
1609 
1610 	/* Allocate PTE tables for kernel KVA. */
1611 	kernel_pdir = data_end;
1612 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1613 	    PDIR_SIZE);
1614 #ifdef __powerpc64__
1615 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1616 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1617 #endif
1618 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1619 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1620 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1621 	    kernel_pdir, data_end);
1622 
1623 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1624 	if (data_end - kernstart > kernsize) {
1625 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1626 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1627 	}
1628 	data_end = kernstart + kernsize;
1629 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1630 
1631 	/*
1632 	 * Clear the structures - note we can only do it safely after the
1633 	 * possible additional TLB1 translations are in place (above) so that
1634 	 * all range up to the currently calculated 'data_end' is covered.
1635 	 */
1636 	dpcpu_init(dpcpu, 0);
1637 #ifdef __powerpc64__
1638 	memset((void *)kernel_pdir, 0,
1639 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1640 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1641 #else
1642 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1643 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1644 #endif
1645 
1646 	/*******************************************************/
1647 	/* Set the start and end of kva. */
1648 	/*******************************************************/
1649 	virtual_avail = round_page(data_end);
1650 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1651 
1652 	/* Allocate KVA space for page zero/copy operations. */
1653 	zero_page_va = virtual_avail;
1654 	virtual_avail += PAGE_SIZE;
1655 	copy_page_src_va = virtual_avail;
1656 	virtual_avail += PAGE_SIZE;
1657 	copy_page_dst_va = virtual_avail;
1658 	virtual_avail += PAGE_SIZE;
1659 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
1660 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
1661 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
1662 
1663 	/* Initialize page zero/copy mutexes. */
1664 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1665 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1666 
1667 #ifndef __powerpc64__
1668 	/* Allocate KVA space for ptbl bufs. */
1669 	ptbl_buf_pool_vabase = virtual_avail;
1670 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1671 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1672 	    ptbl_buf_pool_vabase, virtual_avail);
1673 #endif
1674 
1675 	/* Calculate corresponding physical addresses for the kernel region. */
1676 	phys_kernelend = kernload + kernsize;
1677 	debugf("kernel image and allocated data:\n");
1678 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1679 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
1680 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
1681 
1682 	/*
1683 	 * Remove kernel physical address range from avail regions list. Page
1684 	 * align all regions.  Non-page aligned memory isn't very interesting
1685 	 * to us.  Also, sort the entries for ascending addresses.
1686 	 */
1687 
1688 	/* Retrieve phys/avail mem regions */
1689 	mem_regions(&physmem_regions, &physmem_regions_sz,
1690 	    &availmem_regions, &availmem_regions_sz);
1691 
1692 	if (nitems(phys_avail) < availmem_regions_sz)
1693 		panic("mmu_booke_bootstrap: phys_avail too small");
1694 
1695 	sz = 0;
1696 	cnt = availmem_regions_sz;
1697 	debugf("processing avail regions:\n");
1698 	for (mp = availmem_regions; mp->mr_size; mp++) {
1699 		s = mp->mr_start;
1700 		e = mp->mr_start + mp->mr_size;
1701 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1702 		/* Check whether this region holds all of the kernel. */
1703 		if (s < kernload && e > phys_kernelend) {
1704 			availmem_regions[cnt].mr_start = phys_kernelend;
1705 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1706 			e = kernload;
1707 		}
1708 		/* Look whether this regions starts within the kernel. */
1709 		if (s >= kernload && s < phys_kernelend) {
1710 			if (e <= phys_kernelend)
1711 				goto empty;
1712 			s = phys_kernelend;
1713 		}
1714 		/* Now look whether this region ends within the kernel. */
1715 		if (e > kernload && e <= phys_kernelend) {
1716 			if (s >= kernload)
1717 				goto empty;
1718 			e = kernload;
1719 		}
1720 		/* Now page align the start and size of the region. */
1721 		s = round_page(s);
1722 		e = trunc_page(e);
1723 		if (e < s)
1724 			e = s;
1725 		sz = e - s;
1726 		debugf("%09jx-%09jx = %jx\n",
1727 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1728 
1729 		/* Check whether some memory is left here. */
1730 		if (sz == 0) {
1731 		empty:
1732 			memmove(mp, mp + 1,
1733 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1734 			cnt--;
1735 			mp--;
1736 			continue;
1737 		}
1738 
1739 		/* Do an insertion sort. */
1740 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1741 			if (s < mp1->mr_start)
1742 				break;
1743 		if (mp1 < mp) {
1744 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1745 			mp1->mr_start = s;
1746 			mp1->mr_size = sz;
1747 		} else {
1748 			mp->mr_start = s;
1749 			mp->mr_size = sz;
1750 		}
1751 	}
1752 	availmem_regions_sz = cnt;
1753 
1754 	/*******************************************************/
1755 	/* Steal physical memory for kernel stack from the end */
1756 	/* of the first avail region                           */
1757 	/*******************************************************/
1758 	kstack0_sz = kstack_pages * PAGE_SIZE;
1759 	kstack0_phys = availmem_regions[0].mr_start +
1760 	    availmem_regions[0].mr_size;
1761 	kstack0_phys -= kstack0_sz;
1762 	availmem_regions[0].mr_size -= kstack0_sz;
1763 
1764 	/*******************************************************/
1765 	/* Fill in phys_avail table, based on availmem_regions */
1766 	/*******************************************************/
1767 	phys_avail_count = 0;
1768 	physsz = 0;
1769 	hwphyssz = 0;
1770 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1771 
1772 	debugf("fill in phys_avail:\n");
1773 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1774 
1775 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1776 		    (uintmax_t)availmem_regions[i].mr_start,
1777 		    (uintmax_t)availmem_regions[i].mr_start +
1778 		        availmem_regions[i].mr_size,
1779 		    (uintmax_t)availmem_regions[i].mr_size);
1780 
1781 		if (hwphyssz != 0 &&
1782 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1783 			debugf(" hw.physmem adjust\n");
1784 			if (physsz < hwphyssz) {
1785 				phys_avail[j] = availmem_regions[i].mr_start;
1786 				phys_avail[j + 1] =
1787 				    availmem_regions[i].mr_start +
1788 				    hwphyssz - physsz;
1789 				physsz = hwphyssz;
1790 				phys_avail_count++;
1791 			}
1792 			break;
1793 		}
1794 
1795 		phys_avail[j] = availmem_regions[i].mr_start;
1796 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1797 		    availmem_regions[i].mr_size;
1798 		phys_avail_count++;
1799 		physsz += availmem_regions[i].mr_size;
1800 	}
1801 	physmem = btoc(physsz);
1802 
1803 	/* Calculate the last available physical address. */
1804 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1805 		;
1806 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1807 
1808 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1809 	debugf("phys_avail_count = %d\n", phys_avail_count);
1810 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1811 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1812 
1813 #ifdef __powerpc64__
1814 	/*
1815 	 * Map the physical memory contiguously in TLB1.
1816 	 * Round so it fits into a single mapping.
1817 	 */
1818 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
1819 	    phys_avail[i + 1]);
1820 #endif
1821 
1822 	/*******************************************************/
1823 	/* Initialize (statically allocated) kernel pmap. */
1824 	/*******************************************************/
1825 	PMAP_LOCK_INIT(kernel_pmap);
1826 #ifndef __powerpc64__
1827 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1828 #endif
1829 
1830 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1831 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1832 	for (i = 0; i < MAXCPU; i++) {
1833 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1834 
1835 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1836 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1837 	}
1838 
1839 	/* Mark kernel_pmap active on all CPUs */
1840 	CPU_FILL(&kernel_pmap->pm_active);
1841 
1842  	/*
1843 	 * Initialize the global pv list lock.
1844 	 */
1845 	rw_init(&pvh_global_lock, "pmap pv global");
1846 
1847 	/*******************************************************/
1848 	/* Final setup */
1849 	/*******************************************************/
1850 
1851 	/* Enter kstack0 into kernel map, provide guard page */
1852 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1853 	thread0.td_kstack = kstack0;
1854 	thread0.td_kstack_pages = kstack_pages;
1855 
1856 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1857 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1858 	    kstack0_phys, kstack0_phys + kstack0_sz);
1859 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
1860 	    kstack0, kstack0 + kstack0_sz);
1861 
1862 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1863 	for (i = 0; i < kstack_pages; i++) {
1864 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1865 		kstack0 += PAGE_SIZE;
1866 		kstack0_phys += PAGE_SIZE;
1867 	}
1868 
1869 	pmap_bootstrapped = 1;
1870 
1871 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
1872 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
1873 
1874 	debugf("mmu_booke_bootstrap: exit\n");
1875 }
1876 
1877 #ifdef SMP
1878  void
1879 tlb1_ap_prep(void)
1880 {
1881 	tlb_entry_t *e, tmp;
1882 	unsigned int i;
1883 
1884 	/* Prepare TLB1 image for AP processors */
1885 	e = __boot_tlb1;
1886 	for (i = 0; i < TLB1_ENTRIES; i++) {
1887 		tlb1_read_entry(&tmp, i);
1888 
1889 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1890 			memcpy(e++, &tmp, sizeof(tmp));
1891 	}
1892 }
1893 
1894 void
1895 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1896 {
1897 	int i;
1898 
1899 	/*
1900 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1901 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1902 	 * created by tlb1_ap_prep(), so use these values directly to
1903 	 * (re)program AP's TLB1 hardware.
1904 	 *
1905 	 * Start at index 1 because index 0 has the kernel map.
1906 	 */
1907 	for (i = 1; i < TLB1_ENTRIES; i++) {
1908 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1909 			tlb1_write_entry(&__boot_tlb1[i], i);
1910 	}
1911 
1912 	set_mas4_defaults();
1913 }
1914 #endif
1915 
1916 static void
1917 booke_pmap_init_qpages(void)
1918 {
1919 	struct pcpu *pc;
1920 	int i;
1921 
1922 	CPU_FOREACH(i) {
1923 		pc = pcpu_find(i);
1924 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1925 		if (pc->pc_qmap_addr == 0)
1926 			panic("pmap_init_qpages: unable to allocate KVA");
1927 	}
1928 }
1929 
1930 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1931 
1932 /*
1933  * Get the physical page address for the given pmap/virtual address.
1934  */
1935 static vm_paddr_t
1936 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1937 {
1938 	vm_paddr_t pa;
1939 
1940 	PMAP_LOCK(pmap);
1941 	pa = pte_vatopa(mmu, pmap, va);
1942 	PMAP_UNLOCK(pmap);
1943 
1944 	return (pa);
1945 }
1946 
1947 /*
1948  * Extract the physical page address associated with the given
1949  * kernel virtual address.
1950  */
1951 static vm_paddr_t
1952 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1953 {
1954 	tlb_entry_t e;
1955 	vm_paddr_t p = 0;
1956 	int i;
1957 
1958 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
1959 		p = pte_vatopa(mmu, kernel_pmap, va);
1960 
1961 	if (p == 0) {
1962 		/* Check TLB1 mappings */
1963 		for (i = 0; i < TLB1_ENTRIES; i++) {
1964 			tlb1_read_entry(&e, i);
1965 			if (!(e.mas1 & MAS1_VALID))
1966 				continue;
1967 			if (va >= e.virt && va < e.virt + e.size)
1968 				return (e.phys + (va - e.virt));
1969 		}
1970 	}
1971 
1972 	return (p);
1973 }
1974 
1975 /*
1976  * Initialize the pmap module.
1977  * Called by vm_init, to initialize any structures that the pmap
1978  * system needs to map virtual memory.
1979  */
1980 static void
1981 mmu_booke_init(mmu_t mmu)
1982 {
1983 	int shpgperproc = PMAP_SHPGPERPROC;
1984 
1985 	/*
1986 	 * Initialize the address space (zone) for the pv entries.  Set a
1987 	 * high water mark so that the system can recover from excessive
1988 	 * numbers of pv entries.
1989 	 */
1990 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1991 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1992 
1993 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1994 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1995 
1996 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1997 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1998 
1999 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2000 
2001 	/* Pre-fill pvzone with initial number of pv entries. */
2002 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2003 
2004 	/* Initialize ptbl allocation. */
2005 	ptbl_init();
2006 }
2007 
2008 /*
2009  * Map a list of wired pages into kernel virtual address space.  This is
2010  * intended for temporary mappings which do not need page modification or
2011  * references recorded.  Existing mappings in the region are overwritten.
2012  */
2013 static void
2014 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2015 {
2016 	vm_offset_t va;
2017 
2018 	va = sva;
2019 	while (count-- > 0) {
2020 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2021 		va += PAGE_SIZE;
2022 		m++;
2023 	}
2024 }
2025 
2026 /*
2027  * Remove page mappings from kernel virtual address space.  Intended for
2028  * temporary mappings entered by mmu_booke_qenter.
2029  */
2030 static void
2031 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2032 {
2033 	vm_offset_t va;
2034 
2035 	va = sva;
2036 	while (count-- > 0) {
2037 		mmu_booke_kremove(mmu, va);
2038 		va += PAGE_SIZE;
2039 	}
2040 }
2041 
2042 /*
2043  * Map a wired page into kernel virtual address space.
2044  */
2045 static void
2046 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2047 {
2048 
2049 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2050 }
2051 
2052 static void
2053 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2054 {
2055 	uint32_t flags;
2056 	pte_t *pte;
2057 
2058 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2059 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2060 
2061 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2062 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2063 	flags |= PTE_PS_4KB;
2064 
2065 	pte = pte_find(mmu, kernel_pmap, va);
2066 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2067 
2068 	mtx_lock_spin(&tlbivax_mutex);
2069 	tlb_miss_lock();
2070 
2071 	if (PTE_ISVALID(pte)) {
2072 
2073 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2074 
2075 		/* Flush entry from TLB0 */
2076 		tlb0_flush_entry(va);
2077 	}
2078 
2079 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2080 
2081 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2082 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2083 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2084 
2085 	/* Flush the real memory from the instruction cache. */
2086 	if ((flags & (PTE_I | PTE_G)) == 0)
2087 		__syncicache((void *)va, PAGE_SIZE);
2088 
2089 	tlb_miss_unlock();
2090 	mtx_unlock_spin(&tlbivax_mutex);
2091 }
2092 
2093 /*
2094  * Remove a page from kernel page table.
2095  */
2096 static void
2097 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2098 {
2099 	pte_t *pte;
2100 
2101 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
2102 
2103 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2104 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2105 	    ("mmu_booke_kremove: invalid va"));
2106 
2107 	pte = pte_find(mmu, kernel_pmap, va);
2108 
2109 	if (!PTE_ISVALID(pte)) {
2110 
2111 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2112 
2113 		return;
2114 	}
2115 
2116 	mtx_lock_spin(&tlbivax_mutex);
2117 	tlb_miss_lock();
2118 
2119 	/* Invalidate entry in TLB0, update PTE. */
2120 	tlb0_flush_entry(va);
2121 	*pte = 0;
2122 
2123 	tlb_miss_unlock();
2124 	mtx_unlock_spin(&tlbivax_mutex);
2125 }
2126 
2127 /*
2128  * Provide a kernel pointer corresponding to a given userland pointer.
2129  * The returned pointer is valid until the next time this function is
2130  * called in this thread. This is used internally in copyin/copyout.
2131  */
2132 int
2133 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2134     void **kaddr, size_t ulen, size_t *klen)
2135 {
2136 
2137 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2138 		return (EFAULT);
2139 
2140 	*kaddr = (void *)(uintptr_t)uaddr;
2141 	if (klen)
2142 		*klen = ulen;
2143 
2144 	return (0);
2145 }
2146 
2147 /*
2148  * Figure out where a given kernel pointer (usually in a fault) points
2149  * to from the VM's perspective, potentially remapping into userland's
2150  * address space.
2151  */
2152 static int
2153 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2154     vm_offset_t *decoded_addr)
2155 {
2156 
2157 	if (addr < VM_MAXUSER_ADDRESS)
2158 		*is_user = 1;
2159 	else
2160 		*is_user = 0;
2161 
2162 	*decoded_addr = addr;
2163 	return (0);
2164 }
2165 
2166 /*
2167  * Initialize pmap associated with process 0.
2168  */
2169 static void
2170 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2171 {
2172 
2173 	PMAP_LOCK_INIT(pmap);
2174 	mmu_booke_pinit(mmu, pmap);
2175 	PCPU_SET(curpmap, pmap);
2176 }
2177 
2178 /*
2179  * Initialize a preallocated and zeroed pmap structure,
2180  * such as one in a vmspace structure.
2181  */
2182 static void
2183 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2184 {
2185 	int i;
2186 
2187 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2188 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2189 
2190 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2191 
2192 	for (i = 0; i < MAXCPU; i++)
2193 		pmap->pm_tid[i] = TID_NONE;
2194 	CPU_ZERO(&kernel_pmap->pm_active);
2195 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2196 #ifdef __powerpc64__
2197 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2198 	TAILQ_INIT(&pmap->pm_pdir_list);
2199 #else
2200 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2201 #endif
2202 	TAILQ_INIT(&pmap->pm_ptbl_list);
2203 }
2204 
2205 /*
2206  * Release any resources held by the given physical map.
2207  * Called when a pmap initialized by mmu_booke_pinit is being released.
2208  * Should only be called if the map contains no valid mappings.
2209  */
2210 static void
2211 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2212 {
2213 
2214 	KASSERT(pmap->pm_stats.resident_count == 0,
2215 	    ("pmap_release: pmap resident count %ld != 0",
2216 	    pmap->pm_stats.resident_count));
2217 }
2218 
2219 /*
2220  * Insert the given physical page at the specified virtual address in the
2221  * target physical map with the protection requested. If specified the page
2222  * will be wired down.
2223  */
2224 static int
2225 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2226     vm_prot_t prot, u_int flags, int8_t psind)
2227 {
2228 	int error;
2229 
2230 	rw_wlock(&pvh_global_lock);
2231 	PMAP_LOCK(pmap);
2232 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2233 	PMAP_UNLOCK(pmap);
2234 	rw_wunlock(&pvh_global_lock);
2235 	return (error);
2236 }
2237 
2238 static int
2239 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2240     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2241 {
2242 	pte_t *pte;
2243 	vm_paddr_t pa;
2244 	uint32_t flags;
2245 	int error, su, sync;
2246 
2247 	pa = VM_PAGE_TO_PHYS(m);
2248 	su = (pmap == kernel_pmap);
2249 	sync = 0;
2250 
2251 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2252 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2253 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2254 	//		(u_int32_t)m, va, pa, prot, flags);
2255 
2256 	if (su) {
2257 		KASSERT(((va >= virtual_avail) &&
2258 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2259 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2260 	} else {
2261 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2262 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2263 	}
2264 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2265 		VM_OBJECT_ASSERT_LOCKED(m->object);
2266 
2267 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2268 
2269 	/*
2270 	 * If there is an existing mapping, and the physical address has not
2271 	 * changed, must be protection or wiring change.
2272 	 */
2273 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2274 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2275 
2276 		/*
2277 		 * Before actually updating pte->flags we calculate and
2278 		 * prepare its new value in a helper var.
2279 		 */
2280 		flags = *pte;
2281 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2282 
2283 		/* Wiring change, just update stats. */
2284 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2285 			if (!PTE_ISWIRED(pte)) {
2286 				flags |= PTE_WIRED;
2287 				pmap->pm_stats.wired_count++;
2288 			}
2289 		} else {
2290 			if (PTE_ISWIRED(pte)) {
2291 				flags &= ~PTE_WIRED;
2292 				pmap->pm_stats.wired_count--;
2293 			}
2294 		}
2295 
2296 		if (prot & VM_PROT_WRITE) {
2297 			/* Add write permissions. */
2298 			flags |= PTE_SW;
2299 			if (!su)
2300 				flags |= PTE_UW;
2301 
2302 			if ((flags & PTE_MANAGED) != 0)
2303 				vm_page_aflag_set(m, PGA_WRITEABLE);
2304 		} else {
2305 			/* Handle modified pages, sense modify status. */
2306 
2307 			/*
2308 			 * The PTE_MODIFIED flag could be set by underlying
2309 			 * TLB misses since we last read it (above), possibly
2310 			 * other CPUs could update it so we check in the PTE
2311 			 * directly rather than rely on that saved local flags
2312 			 * copy.
2313 			 */
2314 			if (PTE_ISMODIFIED(pte))
2315 				vm_page_dirty(m);
2316 		}
2317 
2318 		if (prot & VM_PROT_EXECUTE) {
2319 			flags |= PTE_SX;
2320 			if (!su)
2321 				flags |= PTE_UX;
2322 
2323 			/*
2324 			 * Check existing flags for execute permissions: if we
2325 			 * are turning execute permissions on, icache should
2326 			 * be flushed.
2327 			 */
2328 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2329 				sync++;
2330 		}
2331 
2332 		flags &= ~PTE_REFERENCED;
2333 
2334 		/*
2335 		 * The new flags value is all calculated -- only now actually
2336 		 * update the PTE.
2337 		 */
2338 		mtx_lock_spin(&tlbivax_mutex);
2339 		tlb_miss_lock();
2340 
2341 		tlb0_flush_entry(va);
2342 		*pte &= ~PTE_FLAGS_MASK;
2343 		*pte |= flags;
2344 
2345 		tlb_miss_unlock();
2346 		mtx_unlock_spin(&tlbivax_mutex);
2347 
2348 	} else {
2349 		/*
2350 		 * If there is an existing mapping, but it's for a different
2351 		 * physical address, pte_enter() will delete the old mapping.
2352 		 */
2353 		//if ((pte != NULL) && PTE_ISVALID(pte))
2354 		//	debugf("mmu_booke_enter_locked: replace\n");
2355 		//else
2356 		//	debugf("mmu_booke_enter_locked: new\n");
2357 
2358 		/* Now set up the flags and install the new mapping. */
2359 		flags = (PTE_SR | PTE_VALID);
2360 		flags |= PTE_M;
2361 
2362 		if (!su)
2363 			flags |= PTE_UR;
2364 
2365 		if (prot & VM_PROT_WRITE) {
2366 			flags |= PTE_SW;
2367 			if (!su)
2368 				flags |= PTE_UW;
2369 
2370 			if ((m->oflags & VPO_UNMANAGED) == 0)
2371 				vm_page_aflag_set(m, PGA_WRITEABLE);
2372 		}
2373 
2374 		if (prot & VM_PROT_EXECUTE) {
2375 			flags |= PTE_SX;
2376 			if (!su)
2377 				flags |= PTE_UX;
2378 		}
2379 
2380 		/* If its wired update stats. */
2381 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2382 			flags |= PTE_WIRED;
2383 
2384 		error = pte_enter(mmu, pmap, m, va, flags,
2385 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2386 		if (error != 0)
2387 			return (KERN_RESOURCE_SHORTAGE);
2388 
2389 		if ((flags & PMAP_ENTER_WIRED) != 0)
2390 			pmap->pm_stats.wired_count++;
2391 
2392 		/* Flush the real memory from the instruction cache. */
2393 		if (prot & VM_PROT_EXECUTE)
2394 			sync++;
2395 	}
2396 
2397 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2398 		__syncicache((void *)va, PAGE_SIZE);
2399 		sync = 0;
2400 	}
2401 
2402 	return (KERN_SUCCESS);
2403 }
2404 
2405 /*
2406  * Maps a sequence of resident pages belonging to the same object.
2407  * The sequence begins with the given page m_start.  This page is
2408  * mapped at the given virtual address start.  Each subsequent page is
2409  * mapped at a virtual address that is offset from start by the same
2410  * amount as the page is offset from m_start within the object.  The
2411  * last page in the sequence is the page with the largest offset from
2412  * m_start that can be mapped at a virtual address less than the given
2413  * virtual address end.  Not every virtual page between start and end
2414  * is mapped; only those for which a resident page exists with the
2415  * corresponding offset from m_start are mapped.
2416  */
2417 static void
2418 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2419     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2420 {
2421 	vm_page_t m;
2422 	vm_pindex_t diff, psize;
2423 
2424 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2425 
2426 	psize = atop(end - start);
2427 	m = m_start;
2428 	rw_wlock(&pvh_global_lock);
2429 	PMAP_LOCK(pmap);
2430 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2431 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2432 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2433 		    PMAP_ENTER_NOSLEEP, 0);
2434 		m = TAILQ_NEXT(m, listq);
2435 	}
2436 	rw_wunlock(&pvh_global_lock);
2437 	PMAP_UNLOCK(pmap);
2438 }
2439 
2440 static void
2441 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2442     vm_prot_t prot)
2443 {
2444 
2445 	rw_wlock(&pvh_global_lock);
2446 	PMAP_LOCK(pmap);
2447 	mmu_booke_enter_locked(mmu, pmap, va, m,
2448 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2449 	    0);
2450 	rw_wunlock(&pvh_global_lock);
2451 	PMAP_UNLOCK(pmap);
2452 }
2453 
2454 /*
2455  * Remove the given range of addresses from the specified map.
2456  *
2457  * It is assumed that the start and end are properly rounded to the page size.
2458  */
2459 static void
2460 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2461 {
2462 	pte_t *pte;
2463 	uint8_t hold_flag;
2464 
2465 	int su = (pmap == kernel_pmap);
2466 
2467 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2468 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2469 
2470 	if (su) {
2471 		KASSERT(((va >= virtual_avail) &&
2472 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2473 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2474 	} else {
2475 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2476 		    ("mmu_booke_remove: user pmap, non user va"));
2477 	}
2478 
2479 	if (PMAP_REMOVE_DONE(pmap)) {
2480 		//debugf("mmu_booke_remove: e (empty)\n");
2481 		return;
2482 	}
2483 
2484 	hold_flag = PTBL_HOLD_FLAG(pmap);
2485 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2486 
2487 	rw_wlock(&pvh_global_lock);
2488 	PMAP_LOCK(pmap);
2489 	for (; va < endva; va += PAGE_SIZE) {
2490 		pte = pte_find(mmu, pmap, va);
2491 		if ((pte != NULL) && PTE_ISVALID(pte))
2492 			pte_remove(mmu, pmap, va, hold_flag);
2493 	}
2494 	PMAP_UNLOCK(pmap);
2495 	rw_wunlock(&pvh_global_lock);
2496 
2497 	//debugf("mmu_booke_remove: e\n");
2498 }
2499 
2500 /*
2501  * Remove physical page from all pmaps in which it resides.
2502  */
2503 static void
2504 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2505 {
2506 	pv_entry_t pv, pvn;
2507 	uint8_t hold_flag;
2508 
2509 	rw_wlock(&pvh_global_lock);
2510 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2511 		pvn = TAILQ_NEXT(pv, pv_link);
2512 
2513 		PMAP_LOCK(pv->pv_pmap);
2514 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2515 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2516 		PMAP_UNLOCK(pv->pv_pmap);
2517 	}
2518 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2519 	rw_wunlock(&pvh_global_lock);
2520 }
2521 
2522 /*
2523  * Map a range of physical addresses into kernel virtual address space.
2524  */
2525 static vm_offset_t
2526 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2527     vm_paddr_t pa_end, int prot)
2528 {
2529 	vm_offset_t sva = *virt;
2530 	vm_offset_t va = sva;
2531 
2532 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2533 	//		sva, pa_start, pa_end);
2534 
2535 	while (pa_start < pa_end) {
2536 		mmu_booke_kenter(mmu, va, pa_start);
2537 		va += PAGE_SIZE;
2538 		pa_start += PAGE_SIZE;
2539 	}
2540 	*virt = va;
2541 
2542 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2543 	return (sva);
2544 }
2545 
2546 /*
2547  * The pmap must be activated before it's address space can be accessed in any
2548  * way.
2549  */
2550 static void
2551 mmu_booke_activate(mmu_t mmu, struct thread *td)
2552 {
2553 	pmap_t pmap;
2554 	u_int cpuid;
2555 
2556 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2557 
2558 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
2559 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2560 
2561 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2562 
2563 	sched_pin();
2564 
2565 	cpuid = PCPU_GET(cpuid);
2566 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2567 	PCPU_SET(curpmap, pmap);
2568 
2569 	if (pmap->pm_tid[cpuid] == TID_NONE)
2570 		tid_alloc(pmap);
2571 
2572 	/* Load PID0 register with pmap tid value. */
2573 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2574 	__asm __volatile("isync");
2575 
2576 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2577 
2578 	sched_unpin();
2579 
2580 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2581 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2582 }
2583 
2584 /*
2585  * Deactivate the specified process's address space.
2586  */
2587 static void
2588 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2589 {
2590 	pmap_t pmap;
2591 
2592 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2593 
2594 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
2595 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2596 
2597 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2598 
2599 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2600 	PCPU_SET(curpmap, NULL);
2601 }
2602 
2603 /*
2604  * Copy the range specified by src_addr/len
2605  * from the source map to the range dst_addr/len
2606  * in the destination map.
2607  *
2608  * This routine is only advisory and need not do anything.
2609  */
2610 static void
2611 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2612     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2613 {
2614 
2615 }
2616 
2617 /*
2618  * Set the physical protection on the specified range of this map as requested.
2619  */
2620 static void
2621 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2622     vm_prot_t prot)
2623 {
2624 	vm_offset_t va;
2625 	vm_page_t m;
2626 	pte_t *pte;
2627 
2628 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2629 		mmu_booke_remove(mmu, pmap, sva, eva);
2630 		return;
2631 	}
2632 
2633 	if (prot & VM_PROT_WRITE)
2634 		return;
2635 
2636 	PMAP_LOCK(pmap);
2637 	for (va = sva; va < eva; va += PAGE_SIZE) {
2638 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2639 			if (PTE_ISVALID(pte)) {
2640 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2641 
2642 				mtx_lock_spin(&tlbivax_mutex);
2643 				tlb_miss_lock();
2644 
2645 				/* Handle modified pages. */
2646 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2647 					vm_page_dirty(m);
2648 
2649 				tlb0_flush_entry(va);
2650 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2651 
2652 				tlb_miss_unlock();
2653 				mtx_unlock_spin(&tlbivax_mutex);
2654 			}
2655 		}
2656 	}
2657 	PMAP_UNLOCK(pmap);
2658 }
2659 
2660 /*
2661  * Clear the write and modified bits in each of the given page's mappings.
2662  */
2663 static void
2664 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2665 {
2666 	pv_entry_t pv;
2667 	pte_t *pte;
2668 
2669 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2670 	    ("mmu_booke_remove_write: page %p is not managed", m));
2671 
2672 	/*
2673 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2674 	 * set by another thread while the object is locked.  Thus,
2675 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2676 	 */
2677 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2678 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2679 		return;
2680 	rw_wlock(&pvh_global_lock);
2681 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2682 		PMAP_LOCK(pv->pv_pmap);
2683 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2684 			if (PTE_ISVALID(pte)) {
2685 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2686 
2687 				mtx_lock_spin(&tlbivax_mutex);
2688 				tlb_miss_lock();
2689 
2690 				/* Handle modified pages. */
2691 				if (PTE_ISMODIFIED(pte))
2692 					vm_page_dirty(m);
2693 
2694 				/* Flush mapping from TLB0. */
2695 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2696 
2697 				tlb_miss_unlock();
2698 				mtx_unlock_spin(&tlbivax_mutex);
2699 			}
2700 		}
2701 		PMAP_UNLOCK(pv->pv_pmap);
2702 	}
2703 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2704 	rw_wunlock(&pvh_global_lock);
2705 }
2706 
2707 static void
2708 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2709 {
2710 	pte_t *pte;
2711 	vm_paddr_t pa = 0;
2712 	int sync_sz, valid;
2713 #ifndef __powerpc64__
2714 	pmap_t pmap;
2715 	vm_page_t m;
2716 	vm_offset_t addr;
2717 	int active;
2718 #endif
2719 
2720 #ifndef __powerpc64__
2721 	rw_wlock(&pvh_global_lock);
2722 	pmap = PCPU_GET(curpmap);
2723 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2724 #endif
2725 	while (sz > 0) {
2726 		PMAP_LOCK(pm);
2727 		pte = pte_find(mmu, pm, va);
2728 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2729 		if (valid)
2730 			pa = PTE_PA(pte);
2731 		PMAP_UNLOCK(pm);
2732 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
2733 		sync_sz = min(sync_sz, sz);
2734 		if (valid) {
2735 #ifdef __powerpc64__
2736 			pa += (va & PAGE_MASK);
2737 			__syncicache((void *)PHYS_TO_DMAP(pa), sync_sz);
2738 #else
2739 			if (!active) {
2740 				/* Create a mapping in the active pmap. */
2741 				addr = 0;
2742 				m = PHYS_TO_VM_PAGE(pa);
2743 				PMAP_LOCK(pmap);
2744 				pte_enter(mmu, pmap, m, addr,
2745 				    PTE_SR | PTE_VALID, FALSE);
2746 				addr += (va & PAGE_MASK);
2747 				__syncicache((void *)addr, sync_sz);
2748 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2749 				PMAP_UNLOCK(pmap);
2750 			} else
2751 				__syncicache((void *)va, sync_sz);
2752 #endif
2753 		}
2754 		va += sync_sz;
2755 		sz -= sync_sz;
2756 	}
2757 #ifndef __powerpc64__
2758 	rw_wunlock(&pvh_global_lock);
2759 #endif
2760 }
2761 
2762 /*
2763  * Atomically extract and hold the physical page with the given
2764  * pmap and virtual address pair if that mapping permits the given
2765  * protection.
2766  */
2767 static vm_page_t
2768 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2769     vm_prot_t prot)
2770 {
2771 	pte_t *pte;
2772 	vm_page_t m;
2773 	uint32_t pte_wbit;
2774 	vm_paddr_t pa;
2775 
2776 	m = NULL;
2777 	pa = 0;
2778 	PMAP_LOCK(pmap);
2779 retry:
2780 	pte = pte_find(mmu, pmap, va);
2781 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2782 		if (pmap == kernel_pmap)
2783 			pte_wbit = PTE_SW;
2784 		else
2785 			pte_wbit = PTE_UW;
2786 
2787 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2788 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2789 				goto retry;
2790 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2791 			vm_page_wire(m);
2792 		}
2793 	}
2794 
2795 	PA_UNLOCK_COND(pa);
2796 	PMAP_UNLOCK(pmap);
2797 	return (m);
2798 }
2799 
2800 /*
2801  * Initialize a vm_page's machine-dependent fields.
2802  */
2803 static void
2804 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2805 {
2806 
2807 	m->md.pv_tracked = 0;
2808 	TAILQ_INIT(&m->md.pv_list);
2809 }
2810 
2811 /*
2812  * mmu_booke_zero_page_area zeros the specified hardware page by
2813  * mapping it into virtual memory and using bzero to clear
2814  * its contents.
2815  *
2816  * off and size must reside within a single page.
2817  */
2818 static void
2819 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2820 {
2821 	vm_offset_t va;
2822 
2823 	/* XXX KASSERT off and size are within a single page? */
2824 
2825 #ifdef __powerpc64__
2826 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2827 	bzero((caddr_t)va + off, size);
2828 #else
2829 	mtx_lock(&zero_page_mutex);
2830 	va = zero_page_va;
2831 
2832 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2833 	bzero((caddr_t)va + off, size);
2834 	mmu_booke_kremove(mmu, va);
2835 
2836 	mtx_unlock(&zero_page_mutex);
2837 #endif
2838 }
2839 
2840 /*
2841  * mmu_booke_zero_page zeros the specified hardware page.
2842  */
2843 static void
2844 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2845 {
2846 	vm_offset_t off, va;
2847 
2848 #ifdef __powerpc64__
2849 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2850 
2851 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2852 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2853 #else
2854 	va = zero_page_va;
2855 	mtx_lock(&zero_page_mutex);
2856 
2857 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2858 
2859 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2860 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2861 
2862 	mmu_booke_kremove(mmu, va);
2863 
2864 	mtx_unlock(&zero_page_mutex);
2865 #endif
2866 }
2867 
2868 /*
2869  * mmu_booke_copy_page copies the specified (machine independent) page by
2870  * mapping the page into virtual memory and using memcopy to copy the page,
2871  * one machine dependent page at a time.
2872  */
2873 static void
2874 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2875 {
2876 	vm_offset_t sva, dva;
2877 
2878 #ifdef __powerpc64__
2879 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
2880 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
2881 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2882 #else
2883 	sva = copy_page_src_va;
2884 	dva = copy_page_dst_va;
2885 
2886 	mtx_lock(&copy_page_mutex);
2887 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2888 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2889 
2890 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2891 
2892 	mmu_booke_kremove(mmu, dva);
2893 	mmu_booke_kremove(mmu, sva);
2894 	mtx_unlock(&copy_page_mutex);
2895 #endif
2896 }
2897 
2898 static inline void
2899 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2900     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2901 {
2902 	void *a_cp, *b_cp;
2903 	vm_offset_t a_pg_offset, b_pg_offset;
2904 	int cnt;
2905 
2906 #ifdef __powerpc64__
2907 	vm_page_t pa, pb;
2908 
2909 	while (xfersize > 0) {
2910 		a_pg_offset = a_offset & PAGE_MASK;
2911 		pa = ma[a_offset >> PAGE_SHIFT];
2912 		b_pg_offset = b_offset & PAGE_MASK;
2913 		pb = mb[b_offset >> PAGE_SHIFT];
2914 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2915 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2916 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
2917 		    a_pg_offset);
2918 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
2919 		    b_pg_offset);
2920 		bcopy(a_cp, b_cp, cnt);
2921 		a_offset += cnt;
2922 		b_offset += cnt;
2923 		xfersize -= cnt;
2924 	}
2925 #else
2926 	mtx_lock(&copy_page_mutex);
2927 	while (xfersize > 0) {
2928 		a_pg_offset = a_offset & PAGE_MASK;
2929 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2930 		mmu_booke_kenter(mmu, copy_page_src_va,
2931 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2932 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2933 		b_pg_offset = b_offset & PAGE_MASK;
2934 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2935 		mmu_booke_kenter(mmu, copy_page_dst_va,
2936 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2937 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2938 		bcopy(a_cp, b_cp, cnt);
2939 		mmu_booke_kremove(mmu, copy_page_dst_va);
2940 		mmu_booke_kremove(mmu, copy_page_src_va);
2941 		a_offset += cnt;
2942 		b_offset += cnt;
2943 		xfersize -= cnt;
2944 	}
2945 	mtx_unlock(&copy_page_mutex);
2946 #endif
2947 }
2948 
2949 static vm_offset_t
2950 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2951 {
2952 #ifdef __powerpc64__
2953 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
2954 #else
2955 	vm_paddr_t paddr;
2956 	vm_offset_t qaddr;
2957 	uint32_t flags;
2958 	pte_t *pte;
2959 
2960 	paddr = VM_PAGE_TO_PHYS(m);
2961 
2962 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2963 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2964 	flags |= PTE_PS_4KB;
2965 
2966 	critical_enter();
2967 	qaddr = PCPU_GET(qmap_addr);
2968 
2969 	pte = pte_find(mmu, kernel_pmap, qaddr);
2970 
2971 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2972 
2973 	/*
2974 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2975  	 * not be present in other TLBs.  Is there a better instruction
2976 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2977 	 */
2978 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2979 	__asm __volatile("isync; msync");
2980 
2981 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
2982 
2983 	/* Flush the real memory from the instruction cache. */
2984 	if ((flags & (PTE_I | PTE_G)) == 0)
2985 		__syncicache((void *)qaddr, PAGE_SIZE);
2986 
2987 	return (qaddr);
2988 #endif
2989 }
2990 
2991 static void
2992 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
2993 {
2994 #ifndef __powerpc64__
2995 	pte_t *pte;
2996 
2997 	pte = pte_find(mmu, kernel_pmap, addr);
2998 
2999 	KASSERT(PCPU_GET(qmap_addr) == addr,
3000 	    ("mmu_booke_quick_remove_page: invalid address"));
3001 	KASSERT(*pte != 0,
3002 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3003 
3004 	*pte = 0;
3005 	critical_exit();
3006 #endif
3007 }
3008 
3009 /*
3010  * Return whether or not the specified physical page was modified
3011  * in any of physical maps.
3012  */
3013 static boolean_t
3014 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3015 {
3016 	pte_t *pte;
3017 	pv_entry_t pv;
3018 	boolean_t rv;
3019 
3020 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3021 	    ("mmu_booke_is_modified: page %p is not managed", m));
3022 	rv = FALSE;
3023 
3024 	/*
3025 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3026 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3027 	 * is clear, no PTEs can be modified.
3028 	 */
3029 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3030 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3031 		return (rv);
3032 	rw_wlock(&pvh_global_lock);
3033 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3034 		PMAP_LOCK(pv->pv_pmap);
3035 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3036 		    PTE_ISVALID(pte)) {
3037 			if (PTE_ISMODIFIED(pte))
3038 				rv = TRUE;
3039 		}
3040 		PMAP_UNLOCK(pv->pv_pmap);
3041 		if (rv)
3042 			break;
3043 	}
3044 	rw_wunlock(&pvh_global_lock);
3045 	return (rv);
3046 }
3047 
3048 /*
3049  * Return whether or not the specified virtual address is eligible
3050  * for prefault.
3051  */
3052 static boolean_t
3053 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3054 {
3055 
3056 	return (FALSE);
3057 }
3058 
3059 /*
3060  * Return whether or not the specified physical page was referenced
3061  * in any physical maps.
3062  */
3063 static boolean_t
3064 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3065 {
3066 	pte_t *pte;
3067 	pv_entry_t pv;
3068 	boolean_t rv;
3069 
3070 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3071 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3072 	rv = FALSE;
3073 	rw_wlock(&pvh_global_lock);
3074 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3075 		PMAP_LOCK(pv->pv_pmap);
3076 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3077 		    PTE_ISVALID(pte)) {
3078 			if (PTE_ISREFERENCED(pte))
3079 				rv = TRUE;
3080 		}
3081 		PMAP_UNLOCK(pv->pv_pmap);
3082 		if (rv)
3083 			break;
3084 	}
3085 	rw_wunlock(&pvh_global_lock);
3086 	return (rv);
3087 }
3088 
3089 /*
3090  * Clear the modify bits on the specified physical page.
3091  */
3092 static void
3093 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3094 {
3095 	pte_t *pte;
3096 	pv_entry_t pv;
3097 
3098 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3099 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3100 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3101 	KASSERT(!vm_page_xbusied(m),
3102 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3103 
3104 	/*
3105 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3106 	 * If the object containing the page is locked and the page is not
3107 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3108 	 */
3109 	if ((m->aflags & PGA_WRITEABLE) == 0)
3110 		return;
3111 	rw_wlock(&pvh_global_lock);
3112 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3113 		PMAP_LOCK(pv->pv_pmap);
3114 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3115 		    PTE_ISVALID(pte)) {
3116 			mtx_lock_spin(&tlbivax_mutex);
3117 			tlb_miss_lock();
3118 
3119 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3120 				tlb0_flush_entry(pv->pv_va);
3121 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3122 				    PTE_REFERENCED);
3123 			}
3124 
3125 			tlb_miss_unlock();
3126 			mtx_unlock_spin(&tlbivax_mutex);
3127 		}
3128 		PMAP_UNLOCK(pv->pv_pmap);
3129 	}
3130 	rw_wunlock(&pvh_global_lock);
3131 }
3132 
3133 /*
3134  * Return a count of reference bits for a page, clearing those bits.
3135  * It is not necessary for every reference bit to be cleared, but it
3136  * is necessary that 0 only be returned when there are truly no
3137  * reference bits set.
3138  *
3139  * As an optimization, update the page's dirty field if a modified bit is
3140  * found while counting reference bits.  This opportunistic update can be
3141  * performed at low cost and can eliminate the need for some future calls
3142  * to pmap_is_modified().  However, since this function stops after
3143  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3144  * dirty pages.  Those dirty pages will only be detected by a future call
3145  * to pmap_is_modified().
3146  */
3147 static int
3148 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3149 {
3150 	pte_t *pte;
3151 	pv_entry_t pv;
3152 	int count;
3153 
3154 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3155 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3156 	count = 0;
3157 	rw_wlock(&pvh_global_lock);
3158 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3159 		PMAP_LOCK(pv->pv_pmap);
3160 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3161 		    PTE_ISVALID(pte)) {
3162 			if (PTE_ISMODIFIED(pte))
3163 				vm_page_dirty(m);
3164 			if (PTE_ISREFERENCED(pte)) {
3165 				mtx_lock_spin(&tlbivax_mutex);
3166 				tlb_miss_lock();
3167 
3168 				tlb0_flush_entry(pv->pv_va);
3169 				*pte &= ~PTE_REFERENCED;
3170 
3171 				tlb_miss_unlock();
3172 				mtx_unlock_spin(&tlbivax_mutex);
3173 
3174 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3175 					PMAP_UNLOCK(pv->pv_pmap);
3176 					break;
3177 				}
3178 			}
3179 		}
3180 		PMAP_UNLOCK(pv->pv_pmap);
3181 	}
3182 	rw_wunlock(&pvh_global_lock);
3183 	return (count);
3184 }
3185 
3186 /*
3187  * Clear the wired attribute from the mappings for the specified range of
3188  * addresses in the given pmap.  Every valid mapping within that range must
3189  * have the wired attribute set.  In contrast, invalid mappings cannot have
3190  * the wired attribute set, so they are ignored.
3191  *
3192  * The wired attribute of the page table entry is not a hardware feature, so
3193  * there is no need to invalidate any TLB entries.
3194  */
3195 static void
3196 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3197 {
3198 	vm_offset_t va;
3199 	pte_t *pte;
3200 
3201 	PMAP_LOCK(pmap);
3202 	for (va = sva; va < eva; va += PAGE_SIZE) {
3203 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3204 		    PTE_ISVALID(pte)) {
3205 			if (!PTE_ISWIRED(pte))
3206 				panic("mmu_booke_unwire: pte %p isn't wired",
3207 				    pte);
3208 			*pte &= ~PTE_WIRED;
3209 			pmap->pm_stats.wired_count--;
3210 		}
3211 	}
3212 	PMAP_UNLOCK(pmap);
3213 
3214 }
3215 
3216 /*
3217  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3218  * page.  This count may be changed upwards or downwards in the future; it is
3219  * only necessary that true be returned for a small subset of pmaps for proper
3220  * page aging.
3221  */
3222 static boolean_t
3223 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3224 {
3225 	pv_entry_t pv;
3226 	int loops;
3227 	boolean_t rv;
3228 
3229 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3230 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3231 	loops = 0;
3232 	rv = FALSE;
3233 	rw_wlock(&pvh_global_lock);
3234 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3235 		if (pv->pv_pmap == pmap) {
3236 			rv = TRUE;
3237 			break;
3238 		}
3239 		if (++loops >= 16)
3240 			break;
3241 	}
3242 	rw_wunlock(&pvh_global_lock);
3243 	return (rv);
3244 }
3245 
3246 /*
3247  * Return the number of managed mappings to the given physical page that are
3248  * wired.
3249  */
3250 static int
3251 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3252 {
3253 	pv_entry_t pv;
3254 	pte_t *pte;
3255 	int count = 0;
3256 
3257 	if ((m->oflags & VPO_UNMANAGED) != 0)
3258 		return (count);
3259 	rw_wlock(&pvh_global_lock);
3260 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3261 		PMAP_LOCK(pv->pv_pmap);
3262 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3263 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3264 				count++;
3265 		PMAP_UNLOCK(pv->pv_pmap);
3266 	}
3267 	rw_wunlock(&pvh_global_lock);
3268 	return (count);
3269 }
3270 
3271 static int
3272 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3273 {
3274 	int i;
3275 	vm_offset_t va;
3276 
3277 	/*
3278 	 * This currently does not work for entries that
3279 	 * overlap TLB1 entries.
3280 	 */
3281 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3282 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3283 			return (0);
3284 	}
3285 
3286 	return (EFAULT);
3287 }
3288 
3289 void
3290 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3291 {
3292 	vm_paddr_t ppa;
3293 	vm_offset_t ofs;
3294 	vm_size_t gran;
3295 
3296 	/* Minidumps are based on virtual memory addresses. */
3297 	if (do_minidump) {
3298 		*va = (void *)(vm_offset_t)pa;
3299 		return;
3300 	}
3301 
3302 	/* Raw physical memory dumps don't have a virtual address. */
3303 	/* We always map a 256MB page at 256M. */
3304 	gran = 256 * 1024 * 1024;
3305 	ppa = rounddown2(pa, gran);
3306 	ofs = pa - ppa;
3307 	*va = (void *)gran;
3308 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3309 
3310 	if (sz > (gran - ofs))
3311 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3312 		    _TLB_ENTRY_IO);
3313 }
3314 
3315 void
3316 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3317 {
3318 	vm_paddr_t ppa;
3319 	vm_offset_t ofs;
3320 	vm_size_t gran;
3321 	tlb_entry_t e;
3322 	int i;
3323 
3324 	/* Minidumps are based on virtual memory addresses. */
3325 	/* Nothing to do... */
3326 	if (do_minidump)
3327 		return;
3328 
3329 	for (i = 0; i < TLB1_ENTRIES; i++) {
3330 		tlb1_read_entry(&e, i);
3331 		if (!(e.mas1 & MAS1_VALID))
3332 			break;
3333 	}
3334 
3335 	/* Raw physical memory dumps don't have a virtual address. */
3336 	i--;
3337 	e.mas1 = 0;
3338 	e.mas2 = 0;
3339 	e.mas3 = 0;
3340 	tlb1_write_entry(&e, i);
3341 
3342 	gran = 256 * 1024 * 1024;
3343 	ppa = rounddown2(pa, gran);
3344 	ofs = pa - ppa;
3345 	if (sz > (gran - ofs)) {
3346 		i--;
3347 		e.mas1 = 0;
3348 		e.mas2 = 0;
3349 		e.mas3 = 0;
3350 		tlb1_write_entry(&e, i);
3351 	}
3352 }
3353 
3354 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3355 
3356 void
3357 mmu_booke_scan_init(mmu_t mmu)
3358 {
3359 	vm_offset_t va;
3360 	pte_t *pte;
3361 	int i;
3362 
3363 	if (!do_minidump) {
3364 		/* Initialize phys. segments for dumpsys(). */
3365 		memset(&dump_map, 0, sizeof(dump_map));
3366 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3367 		    &availmem_regions_sz);
3368 		for (i = 0; i < physmem_regions_sz; i++) {
3369 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3370 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3371 		}
3372 		return;
3373 	}
3374 
3375 	/* Virtual segments for minidumps: */
3376 	memset(&dump_map, 0, sizeof(dump_map));
3377 
3378 	/* 1st: kernel .data and .bss. */
3379 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3380 	dump_map[0].pa_size =
3381 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3382 
3383 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3384 	dump_map[1].pa_start = data_start;
3385 	dump_map[1].pa_size = data_end - data_start;
3386 
3387 	/* 3rd: kernel VM. */
3388 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3389 	/* Find start of next chunk (from va). */
3390 	while (va < virtual_end) {
3391 		/* Don't dump the buffer cache. */
3392 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3393 			va = kmi.buffer_eva;
3394 			continue;
3395 		}
3396 		pte = pte_find(mmu, kernel_pmap, va);
3397 		if (pte != NULL && PTE_ISVALID(pte))
3398 			break;
3399 		va += PAGE_SIZE;
3400 	}
3401 	if (va < virtual_end) {
3402 		dump_map[2].pa_start = va;
3403 		va += PAGE_SIZE;
3404 		/* Find last page in chunk. */
3405 		while (va < virtual_end) {
3406 			/* Don't run into the buffer cache. */
3407 			if (va == kmi.buffer_sva)
3408 				break;
3409 			pte = pte_find(mmu, kernel_pmap, va);
3410 			if (pte == NULL || !PTE_ISVALID(pte))
3411 				break;
3412 			va += PAGE_SIZE;
3413 		}
3414 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3415 	}
3416 }
3417 
3418 /*
3419  * Map a set of physical memory pages into the kernel virtual address space.
3420  * Return a pointer to where it is mapped. This routine is intended to be used
3421  * for mapping device memory, NOT real memory.
3422  */
3423 static void *
3424 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3425 {
3426 
3427 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3428 }
3429 
3430 static void *
3431 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3432 {
3433 	tlb_entry_t e;
3434 	void *res;
3435 	uintptr_t va, tmpva;
3436 	vm_size_t sz;
3437 	int i;
3438 
3439 	/*
3440 	 * Check if this is premapped in TLB1. Note: this should probably also
3441 	 * check whether a sequence of TLB1 entries exist that match the
3442 	 * requirement, but now only checks the easy case.
3443 	 */
3444 	for (i = 0; i < TLB1_ENTRIES; i++) {
3445 		tlb1_read_entry(&e, i);
3446 		if (!(e.mas1 & MAS1_VALID))
3447 			continue;
3448 		if (pa >= e.phys &&
3449 		    (pa + size) <= (e.phys + e.size) &&
3450 		    (ma == VM_MEMATTR_DEFAULT ||
3451 		     tlb_calc_wimg(pa, ma) ==
3452 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3453 			return (void *)(e.virt +
3454 			    (vm_offset_t)(pa - e.phys));
3455 	}
3456 
3457 	size = roundup(size, PAGE_SIZE);
3458 
3459 	/*
3460 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3461 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3462 	 */
3463 #ifdef SPARSE_MAPDEV
3464 	/*
3465 	 * With a sparse mapdev, align to the largest starting region.  This
3466 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3467 	 * calculation could be very costly.
3468 	 * Align to the smaller of:
3469 	 * - first set bit in overlap of (pa & size mask)
3470 	 * - largest size envelope
3471 	 *
3472 	 * It's possible the device mapping may start at a PA that's not larger
3473 	 * than the size mask, so we need to offset in to maximize the TLB entry
3474 	 * range and minimize the number of used TLB entries.
3475 	 */
3476 	do {
3477 	    tmpva = tlb1_map_base;
3478 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3479 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3480 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3481 #ifdef __powerpc64__
3482 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3483 #else
3484 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3485 #endif
3486 #else
3487 #ifdef __powerpc64__
3488 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3489 #else
3490 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3491 #endif
3492 #endif
3493 	res = (void *)va;
3494 
3495 	do {
3496 		sz = 1 << (ilog2(size) & ~1);
3497 		/* Align size to PA */
3498 		if (pa % sz != 0) {
3499 			do {
3500 				sz >>= 2;
3501 			} while (pa % sz != 0);
3502 		}
3503 		/* Now align from there to VA */
3504 		if (va % sz != 0) {
3505 			do {
3506 				sz >>= 2;
3507 			} while (va % sz != 0);
3508 		}
3509 		if (bootverbose)
3510 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3511 			    va, (uintmax_t)pa, sz);
3512 		if (tlb1_set_entry(va, pa, sz,
3513 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3514 			return (NULL);
3515 		size -= sz;
3516 		pa += sz;
3517 		va += sz;
3518 	} while (size > 0);
3519 
3520 	return (res);
3521 }
3522 
3523 /*
3524  * 'Unmap' a range mapped by mmu_booke_mapdev().
3525  */
3526 static void
3527 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3528 {
3529 #ifdef SUPPORTS_SHRINKING_TLB1
3530 	vm_offset_t base, offset;
3531 
3532 	/*
3533 	 * Unmap only if this is inside kernel virtual space.
3534 	 */
3535 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3536 		base = trunc_page(va);
3537 		offset = va & PAGE_MASK;
3538 		size = roundup(offset + size, PAGE_SIZE);
3539 		kva_free(base, size);
3540 	}
3541 #endif
3542 }
3543 
3544 /*
3545  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3546  * specified pmap. This eliminates the blast of soft faults on process startup
3547  * and immediately after an mmap.
3548  */
3549 static void
3550 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3551     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3552 {
3553 
3554 	VM_OBJECT_ASSERT_WLOCKED(object);
3555 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3556 	    ("mmu_booke_object_init_pt: non-device object"));
3557 }
3558 
3559 /*
3560  * Perform the pmap work for mincore.
3561  */
3562 static int
3563 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3564     vm_paddr_t *locked_pa)
3565 {
3566 
3567 	/* XXX: this should be implemented at some point */
3568 	return (0);
3569 }
3570 
3571 static int
3572 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3573     vm_memattr_t mode)
3574 {
3575 	vm_offset_t va;
3576 	pte_t *pte;
3577 	int i, j;
3578 	tlb_entry_t e;
3579 
3580 	/* Check TLB1 mappings */
3581 	for (i = 0; i < TLB1_ENTRIES; i++) {
3582 		tlb1_read_entry(&e, i);
3583 		if (!(e.mas1 & MAS1_VALID))
3584 			continue;
3585 		if (addr >= e.virt && addr < e.virt + e.size)
3586 			break;
3587 	}
3588 	if (i < TLB1_ENTRIES) {
3589 		/* Only allow full mappings to be modified for now. */
3590 		/* Validate the range. */
3591 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3592 			tlb1_read_entry(&e, j);
3593 			if (va != e.virt || (sz - (va - addr) < e.size))
3594 				return (EINVAL);
3595 		}
3596 		for (va = addr; va < addr + sz; va += e.size, i++) {
3597 			tlb1_read_entry(&e, i);
3598 			e.mas2 &= ~MAS2_WIMGE_MASK;
3599 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3600 
3601 			/*
3602 			 * Write it out to the TLB.  Should really re-sync with other
3603 			 * cores.
3604 			 */
3605 			tlb1_write_entry(&e, i);
3606 		}
3607 		return (0);
3608 	}
3609 
3610 	/* Not in TLB1, try through pmap */
3611 	/* First validate the range. */
3612 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3613 		pte = pte_find(mmu, kernel_pmap, va);
3614 		if (pte == NULL || !PTE_ISVALID(pte))
3615 			return (EINVAL);
3616 	}
3617 
3618 	mtx_lock_spin(&tlbivax_mutex);
3619 	tlb_miss_lock();
3620 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3621 		pte = pte_find(mmu, kernel_pmap, va);
3622 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3623 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3624 		tlb0_flush_entry(va);
3625 	}
3626 	tlb_miss_unlock();
3627 	mtx_unlock_spin(&tlbivax_mutex);
3628 
3629 	return (0);
3630 }
3631 
3632 /**************************************************************************/
3633 /* TID handling */
3634 /**************************************************************************/
3635 
3636 /*
3637  * Allocate a TID. If necessary, steal one from someone else.
3638  * The new TID is flushed from the TLB before returning.
3639  */
3640 static tlbtid_t
3641 tid_alloc(pmap_t pmap)
3642 {
3643 	tlbtid_t tid;
3644 	int thiscpu;
3645 
3646 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3647 
3648 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3649 
3650 	thiscpu = PCPU_GET(cpuid);
3651 
3652 	tid = PCPU_GET(booke.tid_next);
3653 	if (tid > TID_MAX)
3654 		tid = TID_MIN;
3655 	PCPU_SET(booke.tid_next, tid + 1);
3656 
3657 	/* If we are stealing TID then clear the relevant pmap's field */
3658 	if (tidbusy[thiscpu][tid] != NULL) {
3659 
3660 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3661 
3662 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3663 
3664 		/* Flush all entries from TLB0 matching this TID. */
3665 		tid_flush(tid);
3666 	}
3667 
3668 	tidbusy[thiscpu][tid] = pmap;
3669 	pmap->pm_tid[thiscpu] = tid;
3670 	__asm __volatile("msync; isync");
3671 
3672 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3673 	    PCPU_GET(booke.tid_next));
3674 
3675 	return (tid);
3676 }
3677 
3678 /**************************************************************************/
3679 /* TLB0 handling */
3680 /**************************************************************************/
3681 
3682 /* Convert TLB0 va and way number to tlb0[] table index. */
3683 static inline unsigned int
3684 tlb0_tableidx(vm_offset_t va, unsigned int way)
3685 {
3686 	unsigned int idx;
3687 
3688 	idx = (way * TLB0_ENTRIES_PER_WAY);
3689 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3690 	return (idx);
3691 }
3692 
3693 /*
3694  * Invalidate TLB0 entry.
3695  */
3696 static inline void
3697 tlb0_flush_entry(vm_offset_t va)
3698 {
3699 
3700 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3701 
3702 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3703 
3704 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3705 	__asm __volatile("isync; msync");
3706 	__asm __volatile("tlbsync; msync");
3707 
3708 	CTR1(KTR_PMAP, "%s: e", __func__);
3709 }
3710 
3711 
3712 /**************************************************************************/
3713 /* TLB1 handling */
3714 /**************************************************************************/
3715 
3716 /*
3717  * TLB1 mapping notes:
3718  *
3719  * TLB1[0]	Kernel text and data.
3720  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3721  *		windows, other devices mappings.
3722  */
3723 
3724  /*
3725  * Read an entry from given TLB1 slot.
3726  */
3727 void
3728 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3729 {
3730 	register_t msr;
3731 	uint32_t mas0;
3732 
3733 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3734 
3735 	msr = mfmsr();
3736 	__asm __volatile("wrteei 0");
3737 
3738 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3739 	mtspr(SPR_MAS0, mas0);
3740 	__asm __volatile("isync; tlbre");
3741 
3742 	entry->mas1 = mfspr(SPR_MAS1);
3743 	entry->mas2 = mfspr(SPR_MAS2);
3744 	entry->mas3 = mfspr(SPR_MAS3);
3745 
3746 	switch ((mfpvr() >> 16) & 0xFFFF) {
3747 	case FSL_E500v2:
3748 	case FSL_E500mc:
3749 	case FSL_E5500:
3750 	case FSL_E6500:
3751 		entry->mas7 = mfspr(SPR_MAS7);
3752 		break;
3753 	default:
3754 		entry->mas7 = 0;
3755 		break;
3756 	}
3757 	__asm __volatile("wrtee %0" :: "r"(msr));
3758 
3759 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3760 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3761 	    (entry->mas3 & MAS3_RPN);
3762 	entry->size =
3763 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3764 }
3765 
3766 struct tlbwrite_args {
3767 	tlb_entry_t *e;
3768 	unsigned int idx;
3769 };
3770 
3771 static void
3772 tlb1_write_entry_int(void *arg)
3773 {
3774 	struct tlbwrite_args *args = arg;
3775 	uint32_t mas0;
3776 
3777 	/* Select entry */
3778 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3779 
3780 	mtspr(SPR_MAS0, mas0);
3781 	mtspr(SPR_MAS1, args->e->mas1);
3782 	mtspr(SPR_MAS2, args->e->mas2);
3783 	mtspr(SPR_MAS3, args->e->mas3);
3784 	switch ((mfpvr() >> 16) & 0xFFFF) {
3785 	case FSL_E500mc:
3786 	case FSL_E5500:
3787 	case FSL_E6500:
3788 		mtspr(SPR_MAS8, 0);
3789 		/* FALLTHROUGH */
3790 	case FSL_E500v2:
3791 		mtspr(SPR_MAS7, args->e->mas7);
3792 		break;
3793 	default:
3794 		break;
3795 	}
3796 
3797 	__asm __volatile("isync; tlbwe; isync; msync");
3798 
3799 }
3800 
3801 static void
3802 tlb1_write_entry_sync(void *arg)
3803 {
3804 	/* Empty synchronization point for smp_rendezvous(). */
3805 }
3806 
3807 /*
3808  * Write given entry to TLB1 hardware.
3809  */
3810 static void
3811 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3812 {
3813 	struct tlbwrite_args args;
3814 
3815 	args.e = e;
3816 	args.idx = idx;
3817 
3818 #ifdef SMP
3819 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3820 		mb();
3821 		smp_rendezvous(tlb1_write_entry_sync,
3822 		    tlb1_write_entry_int,
3823 		    tlb1_write_entry_sync, &args);
3824 	} else
3825 #endif
3826 	{
3827 		register_t msr;
3828 
3829 		msr = mfmsr();
3830 		__asm __volatile("wrteei 0");
3831 		tlb1_write_entry_int(&args);
3832 		__asm __volatile("wrtee %0" :: "r"(msr));
3833 	}
3834 }
3835 
3836 /*
3837  * Return the largest uint value log such that 2^log <= num.
3838  */
3839 static unsigned int
3840 ilog2(unsigned long num)
3841 {
3842 	long lz;
3843 
3844 #ifdef __powerpc64__
3845 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
3846 	return (63 - lz);
3847 #else
3848 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3849 	return (31 - lz);
3850 #endif
3851 }
3852 
3853 /*
3854  * Convert TLB TSIZE value to mapped region size.
3855  */
3856 static vm_size_t
3857 tsize2size(unsigned int tsize)
3858 {
3859 
3860 	/*
3861 	 * size = 4^tsize KB
3862 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3863 	 */
3864 
3865 	return ((1 << (2 * tsize)) * 1024);
3866 }
3867 
3868 /*
3869  * Convert region size (must be power of 4) to TLB TSIZE value.
3870  */
3871 static unsigned int
3872 size2tsize(vm_size_t size)
3873 {
3874 
3875 	return (ilog2(size) / 2 - 5);
3876 }
3877 
3878 /*
3879  * Register permanent kernel mapping in TLB1.
3880  *
3881  * Entries are created starting from index 0 (current free entry is
3882  * kept in tlb1_idx) and are not supposed to be invalidated.
3883  */
3884 int
3885 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3886     uint32_t flags)
3887 {
3888 	tlb_entry_t e;
3889 	uint32_t ts, tid;
3890 	int tsize, index;
3891 
3892 	for (index = 0; index < TLB1_ENTRIES; index++) {
3893 		tlb1_read_entry(&e, index);
3894 		if ((e.mas1 & MAS1_VALID) == 0)
3895 			break;
3896 		/* Check if we're just updating the flags, and update them. */
3897 		if (e.phys == pa && e.virt == va && e.size == size) {
3898 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
3899 			tlb1_write_entry(&e, index);
3900 			return (0);
3901 		}
3902 	}
3903 	if (index >= TLB1_ENTRIES) {
3904 		printf("tlb1_set_entry: TLB1 full!\n");
3905 		return (-1);
3906 	}
3907 
3908 	/* Convert size to TSIZE */
3909 	tsize = size2tsize(size);
3910 
3911 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3912 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3913 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3914 
3915 	e.phys = pa;
3916 	e.virt = va;
3917 	e.size = size;
3918 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3919 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3920 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
3921 
3922 	/* Set supervisor RWX permission bits */
3923 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3924 	e.mas7 = (pa >> 32) & MAS7_RPN;
3925 
3926 	tlb1_write_entry(&e, index);
3927 
3928 	/*
3929 	 * XXX in general TLB1 updates should be propagated between CPUs,
3930 	 * since current design assumes to have the same TLB1 set-up on all
3931 	 * cores.
3932 	 */
3933 	return (0);
3934 }
3935 
3936 /*
3937  * Map in contiguous RAM region into the TLB1 using maximum of
3938  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3939  *
3940  * If necessary round up last entry size and return total size
3941  * used by all allocated entries.
3942  */
3943 vm_size_t
3944 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3945 {
3946 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3947 	vm_size_t mapped, pgsz, base, mask;
3948 	int idx, nents;
3949 
3950 	/* Round up to the next 1M */
3951 	size = roundup2(size, 1 << 20);
3952 
3953 	mapped = 0;
3954 	idx = 0;
3955 	base = va;
3956 	pgsz = 64*1024*1024;
3957 	while (mapped < size) {
3958 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3959 			while (pgsz > (size - mapped))
3960 				pgsz >>= 2;
3961 			pgs[idx++] = pgsz;
3962 			mapped += pgsz;
3963 		}
3964 
3965 		/* We under-map. Correct for this. */
3966 		if (mapped < size) {
3967 			while (pgs[idx - 1] == pgsz) {
3968 				idx--;
3969 				mapped -= pgsz;
3970 			}
3971 			/* XXX We may increase beyond out starting point. */
3972 			pgsz <<= 2;
3973 			pgs[idx++] = pgsz;
3974 			mapped += pgsz;
3975 		}
3976 	}
3977 
3978 	nents = idx;
3979 	mask = pgs[0] - 1;
3980 	/* Align address to the boundary */
3981 	if (va & mask) {
3982 		va = (va + mask) & ~mask;
3983 		pa = (pa + mask) & ~mask;
3984 	}
3985 
3986 	for (idx = 0; idx < nents; idx++) {
3987 		pgsz = pgs[idx];
3988 		debugf("%u: %llx -> %jx, size=%jx\n", idx, pa,
3989 		    (uintmax_t)va, (uintmax_t)pgsz);
3990 		tlb1_set_entry(va, pa, pgsz,
3991 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
3992 		pa += pgsz;
3993 		va += pgsz;
3994 	}
3995 
3996 	mapped = (va - base);
3997 	if (bootverbose)
3998 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
3999 		    mapped, mapped - size);
4000 	return (mapped);
4001 }
4002 
4003 /*
4004  * TLB1 initialization routine, to be called after the very first
4005  * assembler level setup done in locore.S.
4006  */
4007 void
4008 tlb1_init()
4009 {
4010 	uint32_t mas0, mas1, mas2, mas3, mas7;
4011 	uint32_t tsz;
4012 
4013 	tlb1_get_tlbconf();
4014 
4015 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4016 	mtspr(SPR_MAS0, mas0);
4017 	__asm __volatile("isync; tlbre");
4018 
4019 	mas1 = mfspr(SPR_MAS1);
4020 	mas2 = mfspr(SPR_MAS2);
4021 	mas3 = mfspr(SPR_MAS3);
4022 	mas7 = mfspr(SPR_MAS7);
4023 
4024 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4025 	    (mas3 & MAS3_RPN);
4026 
4027 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4028 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4029 
4030 	/* Setup TLB miss defaults */
4031 	set_mas4_defaults();
4032 }
4033 
4034 /*
4035  * pmap_early_io_unmap() should be used in short conjunction with
4036  * pmap_early_io_map(), as in the following snippet:
4037  *
4038  * x = pmap_early_io_map(...);
4039  * <do something with x>
4040  * pmap_early_io_unmap(x, size);
4041  *
4042  * And avoiding more allocations between.
4043  */
4044 void
4045 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4046 {
4047 	int i;
4048 	tlb_entry_t e;
4049 	vm_size_t isize;
4050 
4051 	size = roundup(size, PAGE_SIZE);
4052 	isize = size;
4053 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4054 		tlb1_read_entry(&e, i);
4055 		if (!(e.mas1 & MAS1_VALID))
4056 			continue;
4057 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4058 			size -= e.size;
4059 			e.mas1 &= ~MAS1_VALID;
4060 			tlb1_write_entry(&e, i);
4061 		}
4062 	}
4063 	if (tlb1_map_base == va + isize)
4064 		tlb1_map_base -= isize;
4065 }
4066 
4067 vm_offset_t
4068 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4069 {
4070 	vm_paddr_t pa_base;
4071 	vm_offset_t va, sz;
4072 	int i;
4073 	tlb_entry_t e;
4074 
4075 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4076 
4077 	for (i = 0; i < TLB1_ENTRIES; i++) {
4078 		tlb1_read_entry(&e, i);
4079 		if (!(e.mas1 & MAS1_VALID))
4080 			continue;
4081 		if (pa >= e.phys && (pa + size) <=
4082 		    (e.phys + e.size))
4083 			return (e.virt + (pa - e.phys));
4084 	}
4085 
4086 	pa_base = rounddown(pa, PAGE_SIZE);
4087 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4088 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4089 	va = tlb1_map_base + (pa - pa_base);
4090 
4091 	do {
4092 		sz = 1 << (ilog2(size) & ~1);
4093 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4094 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4095 		size -= sz;
4096 		pa_base += sz;
4097 		tlb1_map_base += sz;
4098 	} while (size > 0);
4099 
4100 	return (va);
4101 }
4102 
4103 void
4104 pmap_track_page(pmap_t pmap, vm_offset_t va)
4105 {
4106 	vm_paddr_t pa;
4107 	vm_page_t page;
4108 	struct pv_entry *pve;
4109 
4110 	va = trunc_page(va);
4111 	pa = pmap_kextract(va);
4112 	page = PHYS_TO_VM_PAGE(pa);
4113 
4114 	rw_wlock(&pvh_global_lock);
4115 	PMAP_LOCK(pmap);
4116 
4117 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4118 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4119 			goto out;
4120 		}
4121 	}
4122 	page->md.pv_tracked = true;
4123 	pv_insert(pmap, va, page);
4124 out:
4125 	PMAP_UNLOCK(pmap);
4126 	rw_wunlock(&pvh_global_lock);
4127 }
4128 
4129 
4130 /*
4131  * Setup MAS4 defaults.
4132  * These values are loaded to MAS0-2 on a TLB miss.
4133  */
4134 static void
4135 set_mas4_defaults(void)
4136 {
4137 	uint32_t mas4;
4138 
4139 	/* Defaults: TLB0, PID0, TSIZED=4K */
4140 	mas4 = MAS4_TLBSELD0;
4141 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4142 #ifdef SMP
4143 	mas4 |= MAS4_MD;
4144 #endif
4145 	mtspr(SPR_MAS4, mas4);
4146 	__asm __volatile("isync");
4147 }
4148 
4149 
4150 /*
4151  * Return 0 if the physical IO range is encompassed by one of the
4152  * the TLB1 entries, otherwise return related error code.
4153  */
4154 static int
4155 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4156 {
4157 	uint32_t prot;
4158 	vm_paddr_t pa_start;
4159 	vm_paddr_t pa_end;
4160 	unsigned int entry_tsize;
4161 	vm_size_t entry_size;
4162 	tlb_entry_t e;
4163 
4164 	*va = (vm_offset_t)NULL;
4165 
4166 	tlb1_read_entry(&e, i);
4167 	/* Skip invalid entries */
4168 	if (!(e.mas1 & MAS1_VALID))
4169 		return (EINVAL);
4170 
4171 	/*
4172 	 * The entry must be cache-inhibited, guarded, and r/w
4173 	 * so it can function as an i/o page
4174 	 */
4175 	prot = e.mas2 & (MAS2_I | MAS2_G);
4176 	if (prot != (MAS2_I | MAS2_G))
4177 		return (EPERM);
4178 
4179 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4180 	if (prot != (MAS3_SR | MAS3_SW))
4181 		return (EPERM);
4182 
4183 	/* The address should be within the entry range. */
4184 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4185 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4186 
4187 	entry_size = tsize2size(entry_tsize);
4188 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4189 	    (e.mas3 & MAS3_RPN);
4190 	pa_end = pa_start + entry_size;
4191 
4192 	if ((pa < pa_start) || ((pa + size) > pa_end))
4193 		return (ERANGE);
4194 
4195 	/* Return virtual address of this mapping. */
4196 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4197 	return (0);
4198 }
4199 
4200 /*
4201  * Invalidate all TLB0 entries which match the given TID. Note this is
4202  * dedicated for cases when invalidations should NOT be propagated to other
4203  * CPUs.
4204  */
4205 static void
4206 tid_flush(tlbtid_t tid)
4207 {
4208 	register_t msr;
4209 	uint32_t mas0, mas1, mas2;
4210 	int entry, way;
4211 
4212 
4213 	/* Don't evict kernel translations */
4214 	if (tid == TID_KERNEL)
4215 		return;
4216 
4217 	msr = mfmsr();
4218 	__asm __volatile("wrteei 0");
4219 
4220 	/*
4221 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
4222 	 * it for PID invalidation.
4223 	 */
4224 	switch ((mfpvr() >> 16) & 0xffff) {
4225 	case FSL_E500mc:
4226 	case FSL_E5500:
4227 	case FSL_E6500:
4228 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
4229 		/* tlbilxpid */
4230 		__asm __volatile("isync; .long 0x7c000024; isync; msync");
4231 		__asm __volatile("wrtee %0" :: "r"(msr));
4232 		return;
4233 	}
4234 
4235 	for (way = 0; way < TLB0_WAYS; way++)
4236 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4237 
4238 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4239 			mtspr(SPR_MAS0, mas0);
4240 
4241 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4242 			mtspr(SPR_MAS2, mas2);
4243 
4244 			__asm __volatile("isync; tlbre");
4245 
4246 			mas1 = mfspr(SPR_MAS1);
4247 
4248 			if (!(mas1 & MAS1_VALID))
4249 				continue;
4250 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4251 				continue;
4252 			mas1 &= ~MAS1_VALID;
4253 			mtspr(SPR_MAS1, mas1);
4254 			__asm __volatile("isync; tlbwe; isync; msync");
4255 		}
4256 	__asm __volatile("wrtee %0" :: "r"(msr));
4257 }
4258 
4259 #ifdef DDB
4260 /* Print out contents of the MAS registers for each TLB0 entry */
4261 static void
4262 #ifdef __powerpc64__
4263 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
4264 #else
4265 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
4266 #endif
4267     uint32_t mas7)
4268 {
4269 	int as;
4270 	char desc[3];
4271 	tlbtid_t tid;
4272 	vm_size_t size;
4273 	unsigned int tsize;
4274 
4275 	desc[2] = '\0';
4276 	if (mas1 & MAS1_VALID)
4277 		desc[0] = 'V';
4278 	else
4279 		desc[0] = ' ';
4280 
4281 	if (mas1 & MAS1_IPROT)
4282 		desc[1] = 'P';
4283 	else
4284 		desc[1] = ' ';
4285 
4286 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
4287 	tid = MAS1_GETTID(mas1);
4288 
4289 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4290 	size = 0;
4291 	if (tsize)
4292 		size = tsize2size(tsize);
4293 
4294 	printf("%3d: (%s) [AS=%d] "
4295 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
4296 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
4297 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
4298 }
4299 
4300 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
4301 {
4302 	uint32_t mas0, mas1, mas3, mas7;
4303 #ifdef __powerpc64__
4304 	uint64_t mas2;
4305 #else
4306 	uint32_t mas2;
4307 #endif
4308 	int entryidx, way, idx;
4309 
4310 	printf("TLB0 entries:\n");
4311 	for (way = 0; way < TLB0_WAYS; way ++)
4312 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
4313 
4314 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4315 			mtspr(SPR_MAS0, mas0);
4316 
4317 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
4318 			mtspr(SPR_MAS2, mas2);
4319 
4320 			__asm __volatile("isync; tlbre");
4321 
4322 			mas1 = mfspr(SPR_MAS1);
4323 			mas2 = mfspr(SPR_MAS2);
4324 			mas3 = mfspr(SPR_MAS3);
4325 			mas7 = mfspr(SPR_MAS7);
4326 
4327 			idx = tlb0_tableidx(mas2, way);
4328 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
4329 		}
4330 }
4331 
4332 /*
4333  * Print out contents of the MAS registers for each TLB1 entry
4334  */
4335 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
4336 {
4337 	uint32_t mas0, mas1, mas3, mas7;
4338 #ifdef __powerpc64__
4339 	uint64_t mas2;
4340 #else
4341 	uint32_t mas2;
4342 #endif
4343 	int i;
4344 
4345 	printf("TLB1 entries:\n");
4346 	for (i = 0; i < TLB1_ENTRIES; i++) {
4347 
4348 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4349 		mtspr(SPR_MAS0, mas0);
4350 
4351 		__asm __volatile("isync; tlbre");
4352 
4353 		mas1 = mfspr(SPR_MAS1);
4354 		mas2 = mfspr(SPR_MAS2);
4355 		mas3 = mfspr(SPR_MAS3);
4356 		mas7 = mfspr(SPR_MAS7);
4357 
4358 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4359 	}
4360 }
4361 #endif
4362