xref: /freebsd/sys/powerpc/booke/pmap.c (revision 0b37c1590418417c894529d371800dfac71ef887)
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 #ifndef __powerpc64__
165 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
166 static vm_offset_t zero_page_va;
167 static struct mtx zero_page_mutex;
168 
169 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
170 static vm_offset_t copy_page_src_va;
171 static vm_offset_t copy_page_dst_va;
172 static struct mtx copy_page_mutex;
173 #endif
174 
175 static struct mtx tlbivax_mutex;
176 
177 /**************************************************************************/
178 /* PMAP */
179 /**************************************************************************/
180 
181 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
182     vm_prot_t, u_int flags, int8_t psind);
183 
184 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
185 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
186 #ifdef __powerpc64__
187 unsigned int kernel_pdirs;
188 #endif
189 static uma_zone_t ptbl_root_zone;
190 
191 /*
192  * If user pmap is processed with mmu_booke_remove and the resident count
193  * drops to 0, there are no more pages to remove, so we need not continue.
194  */
195 #define PMAP_REMOVE_DONE(pmap) \
196 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
197 
198 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
199 extern int elf32_nxstack;
200 #endif
201 
202 /**************************************************************************/
203 /* TLB and TID handling */
204 /**************************************************************************/
205 
206 /* Translation ID busy table */
207 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
208 
209 /*
210  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
211  * core revisions and should be read from h/w registers during early config.
212  */
213 uint32_t tlb0_entries;
214 uint32_t tlb0_ways;
215 uint32_t tlb0_entries_per_way;
216 uint32_t tlb1_entries;
217 
218 #define TLB0_ENTRIES		(tlb0_entries)
219 #define TLB0_WAYS		(tlb0_ways)
220 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
221 
222 #define TLB1_ENTRIES (tlb1_entries)
223 
224 static vm_offset_t tlb1_map_base = (vm_offset_t)VM_MAXUSER_ADDRESS + PAGE_SIZE;
225 
226 static tlbtid_t tid_alloc(struct pmap *);
227 static void tid_flush(tlbtid_t tid);
228 
229 #ifdef DDB
230 #ifdef __powerpc64__
231 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
232 #else
233 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
234 #endif
235 #endif
236 
237 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
238 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
239 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
240 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t, int);
241 
242 static vm_size_t tsize2size(unsigned int);
243 static unsigned int size2tsize(vm_size_t);
244 static unsigned long ilog2(unsigned long);
245 
246 static void set_mas4_defaults(void);
247 
248 static inline void tlb0_flush_entry(vm_offset_t);
249 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
250 
251 /**************************************************************************/
252 /* Page table management */
253 /**************************************************************************/
254 
255 static struct rwlock_padalign pvh_global_lock;
256 
257 /* Data for the pv entry allocation mechanism */
258 static uma_zone_t pvzone;
259 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
260 
261 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
262 
263 #ifndef PMAP_SHPGPERPROC
264 #define PMAP_SHPGPERPROC	200
265 #endif
266 
267 #ifdef __powerpc64__
268 #define PMAP_ROOT_SIZE	(sizeof(pte_t***) * PP2D_NENTRIES)
269 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
270 			 unsigned int, boolean_t);
271 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int, vm_page_t);
272 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
273 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
274 #else
275 #define PMAP_ROOT_SIZE	(sizeof(pte_t**) * PDIR_NENTRIES)
276 static void ptbl_init(void);
277 static struct ptbl_buf *ptbl_buf_alloc(void);
278 static void ptbl_buf_free(struct ptbl_buf *);
279 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
280 
281 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
282 static void ptbl_free(mmu_t, pmap_t, unsigned int);
283 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
284 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
285 #endif
286 
287 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
288 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
289 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
290 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
291 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
292 
293 static pv_entry_t pv_alloc(void);
294 static void pv_free(pv_entry_t);
295 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
296 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
297 
298 static void booke_pmap_init_qpages(void);
299 
300 struct ptbl_buf {
301 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
302 	vm_offset_t kva;		/* va of mapping */
303 };
304 
305 #ifndef __powerpc64__
306 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
307 #define PTBL_BUFS		(128 * 16)
308 
309 /* ptbl free list and a lock used for access synchronization. */
310 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
311 static struct mtx ptbl_buf_freelist_lock;
312 
313 /* Base address of kva space allocated fot ptbl bufs. */
314 static vm_offset_t ptbl_buf_pool_vabase;
315 
316 /* Pointer to ptbl_buf structures. */
317 static struct ptbl_buf *ptbl_bufs;
318 #endif
319 
320 #ifdef SMP
321 extern tlb_entry_t __boot_tlb1[];
322 void pmap_bootstrap_ap(volatile uint32_t *);
323 #endif
324 
325 /*
326  * Kernel MMU interface
327  */
328 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
329 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
330     vm_size_t, vm_offset_t);
331 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
332 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
333     vm_offset_t, vm_page_t *, vm_offset_t, int);
334 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
335     vm_prot_t, u_int flags, int8_t psind);
336 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
337     vm_page_t, vm_prot_t);
338 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
339     vm_prot_t);
340 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
341 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
342     vm_prot_t);
343 static void		mmu_booke_init(mmu_t);
344 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
345 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
346 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
347 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
348 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
349     int);
350 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
351     vm_paddr_t *);
352 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
353     vm_object_t, vm_pindex_t, vm_size_t);
354 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
355 static void		mmu_booke_page_init(mmu_t, vm_page_t);
356 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
357 static void		mmu_booke_pinit(mmu_t, pmap_t);
358 static void		mmu_booke_pinit0(mmu_t, pmap_t);
359 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
360     vm_prot_t);
361 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
362 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
363 static void		mmu_booke_release(mmu_t, pmap_t);
364 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
365 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
366 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
367 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
368 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
369 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
370 static void		mmu_booke_activate(mmu_t, struct thread *);
371 static void		mmu_booke_deactivate(mmu_t, struct thread *);
372 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
373 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
374 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
375 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
376 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
377 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
378 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
379 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
380 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
381 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
382     vm_size_t);
383 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
384     void **);
385 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
386     void *);
387 static void		mmu_booke_scan_init(mmu_t);
388 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
389 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
390 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
391     vm_size_t sz, vm_memattr_t mode);
392 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
393     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
394 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
395     int *is_user, vm_offset_t *decoded_addr);
396 static void		mmu_booke_page_array_startup(mmu_t , long);
397 
398 
399 static mmu_method_t mmu_booke_methods[] = {
400 	/* pmap dispatcher interface */
401 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
402 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
403 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
404 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
405 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
406 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
407 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
408 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
409 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
410 	MMUMETHOD(mmu_init,		mmu_booke_init),
411 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
412 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
413 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
414 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
415 	MMUMETHOD(mmu_map,		mmu_booke_map),
416 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
417 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
418 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
419 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
420 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
421 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
422 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
423 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
424 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
425 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
426 	MMUMETHOD(mmu_release,		mmu_booke_release),
427 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
428 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
429 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
430 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
431 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
432 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
433 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
434 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
435 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
436 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
437 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
438 	MMUMETHOD(mmu_page_array_startup,	mmu_booke_page_array_startup),
439 
440 	/* Internal interfaces */
441 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
442 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
443 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
444 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
445 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
446 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
447 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
448 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
449 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
450 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
451 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
452 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
453 
454 	/* dumpsys() support */
455 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
456 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
457 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
458 
459 	{ 0, 0 }
460 };
461 
462 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
463 
464 static __inline uint32_t
465 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
466 {
467 	uint32_t attrib;
468 	int i;
469 
470 	if (ma != VM_MEMATTR_DEFAULT) {
471 		switch (ma) {
472 		case VM_MEMATTR_UNCACHEABLE:
473 			return (MAS2_I | MAS2_G);
474 		case VM_MEMATTR_WRITE_COMBINING:
475 		case VM_MEMATTR_WRITE_BACK:
476 		case VM_MEMATTR_PREFETCHABLE:
477 			return (MAS2_I);
478 		case VM_MEMATTR_WRITE_THROUGH:
479 			return (MAS2_W | MAS2_M);
480 		case VM_MEMATTR_CACHEABLE:
481 			return (MAS2_M);
482 		}
483 	}
484 
485 	/*
486 	 * Assume the page is cache inhibited and access is guarded unless
487 	 * it's in our available memory array.
488 	 */
489 	attrib = _TLB_ENTRY_IO;
490 	for (i = 0; i < physmem_regions_sz; i++) {
491 		if ((pa >= physmem_regions[i].mr_start) &&
492 		    (pa < (physmem_regions[i].mr_start +
493 		     physmem_regions[i].mr_size))) {
494 			attrib = _TLB_ENTRY_MEM;
495 			break;
496 		}
497 	}
498 
499 	return (attrib);
500 }
501 
502 static inline void
503 tlb_miss_lock(void)
504 {
505 #ifdef SMP
506 	struct pcpu *pc;
507 
508 	if (!smp_started)
509 		return;
510 
511 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
512 		if (pc != pcpup) {
513 
514 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
515 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
516 
517 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
518 			    ("tlb_miss_lock: tried to lock self"));
519 
520 			tlb_lock(pc->pc_booke.tlb_lock);
521 
522 			CTR1(KTR_PMAP, "%s: locked", __func__);
523 		}
524 	}
525 #endif
526 }
527 
528 static inline void
529 tlb_miss_unlock(void)
530 {
531 #ifdef SMP
532 	struct pcpu *pc;
533 
534 	if (!smp_started)
535 		return;
536 
537 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
538 		if (pc != pcpup) {
539 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
540 			    __func__, pc->pc_cpuid);
541 
542 			tlb_unlock(pc->pc_booke.tlb_lock);
543 
544 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
545 		}
546 	}
547 #endif
548 }
549 
550 /* Return number of entries in TLB0. */
551 static __inline void
552 tlb0_get_tlbconf(void)
553 {
554 	uint32_t tlb0_cfg;
555 
556 	tlb0_cfg = mfspr(SPR_TLB0CFG);
557 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
558 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
559 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
560 }
561 
562 /* Return number of entries in TLB1. */
563 static __inline void
564 tlb1_get_tlbconf(void)
565 {
566 	uint32_t tlb1_cfg;
567 
568 	tlb1_cfg = mfspr(SPR_TLB1CFG);
569 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
570 }
571 
572 /**************************************************************************/
573 /* Page table related */
574 /**************************************************************************/
575 
576 #ifdef __powerpc64__
577 /* Initialize pool of kva ptbl buffers. */
578 static void
579 ptbl_init(void)
580 {
581 }
582 
583 /* Get a pointer to a PTE in a page table. */
584 static __inline pte_t *
585 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
586 {
587 	pte_t         **pdir;
588 	pte_t          *ptbl;
589 
590 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
591 
592 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
593 	if (!pdir)
594 		return NULL;
595 	ptbl = pdir[PDIR_IDX(va)];
596 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
597 }
598 
599 /*
600  * allocate a page of pointers to page directories, do not preallocate the
601  * page tables
602  */
603 static pte_t  **
604 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
605 {
606 	vm_page_t	m;
607 	pte_t          **pdir;
608 	int		req;
609 
610 	req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
611 	while ((m = vm_page_alloc(NULL, pp2d_idx, req)) == NULL) {
612 		PMAP_UNLOCK(pmap);
613 		if (nosleep) {
614 			return (NULL);
615 		}
616 		vm_wait(NULL);
617 		PMAP_LOCK(pmap);
618 	}
619 
620 	/* Zero whole ptbl. */
621 	pdir = (pte_t **)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
622 	mmu_booke_zero_page(mmu, m);
623 
624 	return (pdir);
625 }
626 
627 /* Free pdir pages and invalidate pdir entry. */
628 static void
629 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, vm_page_t m)
630 {
631 	pte_t         **pdir;
632 
633 	pdir = pmap->pm_pp2d[pp2d_idx];
634 
635 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
636 
637 	pmap->pm_pp2d[pp2d_idx] = NULL;
638 
639 	vm_wire_sub(1);
640 	vm_page_free_zero(m);
641 }
642 
643 /*
644  * Decrement pdir pages hold count and attempt to free pdir pages. Called
645  * when removing directory entry from pdir.
646  *
647  * Return 1 if pdir pages were freed.
648  */
649 static int
650 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
651 {
652 	pte_t         **pdir;
653 	vm_paddr_t	pa;
654 	vm_page_t	m;
655 
656 	KASSERT((pmap != kernel_pmap),
657 		("pdir_unhold: unholding kernel pdir!"));
658 
659 	pdir = pmap->pm_pp2d[pp2d_idx];
660 
661 	/* decrement hold count */
662 	pa = DMAP_TO_PHYS((vm_offset_t) pdir);
663 	m = PHYS_TO_VM_PAGE(pa);
664 
665 	/*
666 	 * Free pdir page if there are no dir entries in this pdir.
667 	 */
668 	m->ref_count--;
669 	if (m->ref_count == 0) {
670 		pdir_free(mmu, pmap, pp2d_idx, m);
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 	m->ref_count++;
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 	mmu_booke_zero_page(mmu, m);
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, vm_page_t m)
728 {
729 	pte_t          *ptbl;
730 
731 	ptbl = pdir[pdir_idx];
732 
733 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
734 
735 	pdir[pdir_idx] = NULL;
736 
737 	vm_wire_sub(1);
738 	vm_page_free_zero(m);
739 }
740 
741 /*
742  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
743  * when removing pte entry from ptbl.
744  *
745  * Return 1 if ptbl pages were freed.
746  */
747 static int
748 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
749 {
750 	pte_t          *ptbl;
751 	vm_page_t	m;
752 	u_int		pp2d_idx;
753 	pte_t         **pdir;
754 	u_int		pdir_idx;
755 
756 	pp2d_idx = PP2D_IDX(va);
757 	pdir_idx = PDIR_IDX(va);
758 
759 	KASSERT((pmap != kernel_pmap),
760 		("ptbl_unhold: unholding kernel ptbl!"));
761 
762 	pdir = pmap->pm_pp2d[pp2d_idx];
763 	ptbl = pdir[pdir_idx];
764 
765 	/* decrement hold count */
766 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
767 
768 	/*
769 	 * Free ptbl pages if there are no pte entries in this ptbl.
770 	 * ref_count has the same value for all ptbl pages, so check the
771 	 * last page.
772 	 */
773 	m->ref_count--;
774 	if (m->ref_count == 0) {
775 		ptbl_free(mmu, pmap, pdir, pdir_idx, m);
776 		pdir_unhold(mmu, pmap, pp2d_idx);
777 		return (1);
778 	}
779 	return (0);
780 }
781 
782 /*
783  * Increment hold count for ptbl pages. This routine is used when new pte
784  * entry is being inserted into ptbl.
785  */
786 static void
787 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
788 {
789 	pte_t          *ptbl;
790 	vm_page_t	m;
791 
792 	KASSERT((pmap != kernel_pmap),
793 		("ptbl_hold: holding kernel ptbl!"));
794 
795 	ptbl = pdir[pdir_idx];
796 
797 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
798 
799 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
800 	m->ref_count++;
801 }
802 #else
803 
804 /* Initialize pool of kva ptbl buffers. */
805 static void
806 ptbl_init(void)
807 {
808 	int i;
809 
810 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
811 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
812 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
813 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
814 
815 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
816 	TAILQ_INIT(&ptbl_buf_freelist);
817 
818 	for (i = 0; i < PTBL_BUFS; i++) {
819 		ptbl_bufs[i].kva =
820 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
821 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
822 	}
823 }
824 
825 /* Get a ptbl_buf from the freelist. */
826 static struct ptbl_buf *
827 ptbl_buf_alloc(void)
828 {
829 	struct ptbl_buf *buf;
830 
831 	mtx_lock(&ptbl_buf_freelist_lock);
832 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
833 	if (buf != NULL)
834 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
835 	mtx_unlock(&ptbl_buf_freelist_lock);
836 
837 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
838 
839 	return (buf);
840 }
841 
842 /* Return ptbl buff to free pool. */
843 static void
844 ptbl_buf_free(struct ptbl_buf *buf)
845 {
846 
847 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
848 
849 	mtx_lock(&ptbl_buf_freelist_lock);
850 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
851 	mtx_unlock(&ptbl_buf_freelist_lock);
852 }
853 
854 /*
855  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
856  */
857 static void
858 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
859 {
860 	struct ptbl_buf *pbuf;
861 
862 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
863 
864 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
865 
866 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
867 		if (pbuf->kva == (vm_offset_t)ptbl) {
868 			/* Remove from pmap ptbl buf list. */
869 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
870 
871 			/* Free corresponding ptbl buf. */
872 			ptbl_buf_free(pbuf);
873 			break;
874 		}
875 }
876 
877 /* Allocate page table. */
878 static pte_t *
879 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
880 {
881 	vm_page_t mtbl[PTBL_PAGES];
882 	vm_page_t m;
883 	struct ptbl_buf *pbuf;
884 	unsigned int pidx;
885 	pte_t *ptbl;
886 	int i, j;
887 
888 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
889 	    (pmap == kernel_pmap), pdir_idx);
890 
891 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
892 	    ("ptbl_alloc: invalid pdir_idx"));
893 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
894 	    ("pte_alloc: valid ptbl entry exists!"));
895 
896 	pbuf = ptbl_buf_alloc();
897 	if (pbuf == NULL)
898 		panic("pte_alloc: couldn't alloc kernel virtual memory");
899 
900 	ptbl = (pte_t *)pbuf->kva;
901 
902 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
903 
904 	for (i = 0; i < PTBL_PAGES; i++) {
905 		pidx = (PTBL_PAGES * pdir_idx) + i;
906 		while ((m = vm_page_alloc(NULL, pidx,
907 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
908 			PMAP_UNLOCK(pmap);
909 			rw_wunlock(&pvh_global_lock);
910 			if (nosleep) {
911 				ptbl_free_pmap_ptbl(pmap, ptbl);
912 				for (j = 0; j < i; j++)
913 					vm_page_free(mtbl[j]);
914 				vm_wire_sub(i);
915 				return (NULL);
916 			}
917 			vm_wait(NULL);
918 			rw_wlock(&pvh_global_lock);
919 			PMAP_LOCK(pmap);
920 		}
921 		mtbl[i] = m;
922 	}
923 
924 	/* Map allocated pages into kernel_pmap. */
925 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
926 
927 	/* Zero whole ptbl. */
928 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
929 
930 	/* Add pbuf to the pmap ptbl bufs list. */
931 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
932 
933 	return (ptbl);
934 }
935 
936 /* Free ptbl pages and invalidate pdir entry. */
937 static void
938 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
939 {
940 	pte_t *ptbl;
941 	vm_paddr_t pa;
942 	vm_offset_t va;
943 	vm_page_t m;
944 	int i;
945 
946 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
947 	    (pmap == kernel_pmap), pdir_idx);
948 
949 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
950 	    ("ptbl_free: invalid pdir_idx"));
951 
952 	ptbl = pmap->pm_pdir[pdir_idx];
953 
954 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
955 
956 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
957 
958 	/*
959 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
960 	 * don't attempt to look up the page tables we are releasing.
961 	 */
962 	mtx_lock_spin(&tlbivax_mutex);
963 	tlb_miss_lock();
964 
965 	pmap->pm_pdir[pdir_idx] = NULL;
966 
967 	tlb_miss_unlock();
968 	mtx_unlock_spin(&tlbivax_mutex);
969 
970 	for (i = 0; i < PTBL_PAGES; i++) {
971 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
972 		pa = pte_vatopa(mmu, kernel_pmap, va);
973 		m = PHYS_TO_VM_PAGE(pa);
974 		vm_page_free_zero(m);
975 		vm_wire_sub(1);
976 		mmu_booke_kremove(mmu, va);
977 	}
978 
979 	ptbl_free_pmap_ptbl(pmap, ptbl);
980 }
981 
982 /*
983  * Decrement ptbl pages hold count and attempt to free ptbl pages.
984  * Called when removing pte entry from ptbl.
985  *
986  * Return 1 if ptbl pages were freed.
987  */
988 static int
989 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
990 {
991 	pte_t *ptbl;
992 	vm_paddr_t pa;
993 	vm_page_t m;
994 	int i;
995 
996 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
997 	    (pmap == kernel_pmap), pdir_idx);
998 
999 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1000 	    ("ptbl_unhold: invalid pdir_idx"));
1001 	KASSERT((pmap != kernel_pmap),
1002 	    ("ptbl_unhold: unholding kernel ptbl!"));
1003 
1004 	ptbl = pmap->pm_pdir[pdir_idx];
1005 
1006 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1007 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1008 	    ("ptbl_unhold: non kva ptbl"));
1009 
1010 	/* decrement hold count */
1011 	for (i = 0; i < PTBL_PAGES; i++) {
1012 		pa = pte_vatopa(mmu, kernel_pmap,
1013 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1014 		m = PHYS_TO_VM_PAGE(pa);
1015 		m->ref_count--;
1016 	}
1017 
1018 	/*
1019 	 * Free ptbl pages if there are no pte etries in this ptbl.
1020 	 * ref_count has the same value for all ptbl pages, so check the last
1021 	 * page.
1022 	 */
1023 	if (m->ref_count == 0) {
1024 		ptbl_free(mmu, pmap, pdir_idx);
1025 
1026 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1027 		return (1);
1028 	}
1029 
1030 	return (0);
1031 }
1032 
1033 /*
1034  * Increment hold count for ptbl pages. This routine is used when a new pte
1035  * entry is being inserted into the ptbl.
1036  */
1037 static void
1038 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1039 {
1040 	vm_paddr_t pa;
1041 	pte_t *ptbl;
1042 	vm_page_t m;
1043 	int i;
1044 
1045 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1046 	    pdir_idx);
1047 
1048 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1049 	    ("ptbl_hold: invalid pdir_idx"));
1050 	KASSERT((pmap != kernel_pmap),
1051 	    ("ptbl_hold: holding kernel ptbl!"));
1052 
1053 	ptbl = pmap->pm_pdir[pdir_idx];
1054 
1055 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1056 
1057 	for (i = 0; i < PTBL_PAGES; i++) {
1058 		pa = pte_vatopa(mmu, kernel_pmap,
1059 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1060 		m = PHYS_TO_VM_PAGE(pa);
1061 		m->ref_count++;
1062 	}
1063 }
1064 #endif
1065 
1066 /* Allocate pv_entry structure. */
1067 pv_entry_t
1068 pv_alloc(void)
1069 {
1070 	pv_entry_t pv;
1071 
1072 	pv_entry_count++;
1073 	if (pv_entry_count > pv_entry_high_water)
1074 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1075 	pv = uma_zalloc(pvzone, M_NOWAIT);
1076 
1077 	return (pv);
1078 }
1079 
1080 /* Free pv_entry structure. */
1081 static __inline void
1082 pv_free(pv_entry_t pve)
1083 {
1084 
1085 	pv_entry_count--;
1086 	uma_zfree(pvzone, pve);
1087 }
1088 
1089 
1090 /* Allocate and initialize pv_entry structure. */
1091 static void
1092 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1093 {
1094 	pv_entry_t pve;
1095 
1096 	//int su = (pmap == kernel_pmap);
1097 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1098 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1099 
1100 	pve = pv_alloc();
1101 	if (pve == NULL)
1102 		panic("pv_insert: no pv entries!");
1103 
1104 	pve->pv_pmap = pmap;
1105 	pve->pv_va = va;
1106 
1107 	/* add to pv_list */
1108 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1109 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1110 
1111 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1112 
1113 	//debugf("pv_insert: e\n");
1114 }
1115 
1116 /* Destroy pv entry. */
1117 static void
1118 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1119 {
1120 	pv_entry_t pve;
1121 
1122 	//int su = (pmap == kernel_pmap);
1123 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1124 
1125 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1126 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1127 
1128 	/* find pv entry */
1129 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1130 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1131 			/* remove from pv_list */
1132 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1133 			if (TAILQ_EMPTY(&m->md.pv_list))
1134 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1135 
1136 			/* free pv entry struct */
1137 			pv_free(pve);
1138 			break;
1139 		}
1140 	}
1141 
1142 	//debugf("pv_remove: e\n");
1143 }
1144 
1145 #ifdef __powerpc64__
1146 /*
1147  * Clean pte entry, try to free page table page if requested.
1148  *
1149  * Return 1 if ptbl pages were freed, otherwise return 0.
1150  */
1151 static int
1152 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1153 {
1154 	vm_page_t	m;
1155 	pte_t          *pte;
1156 
1157 	pte = pte_find(mmu, pmap, va);
1158 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1159 
1160 	if (!PTE_ISVALID(pte))
1161 		return (0);
1162 
1163 	/* Get vm_page_t for mapped pte. */
1164 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1165 
1166 	if (PTE_ISWIRED(pte))
1167 		pmap->pm_stats.wired_count--;
1168 
1169 	/* Handle managed entry. */
1170 	if (PTE_ISMANAGED(pte)) {
1171 
1172 		/* Handle modified pages. */
1173 		if (PTE_ISMODIFIED(pte))
1174 			vm_page_dirty(m);
1175 
1176 		/* Referenced pages. */
1177 		if (PTE_ISREFERENCED(pte))
1178 			vm_page_aflag_set(m, PGA_REFERENCED);
1179 
1180 		/* Remove pv_entry from pv_list. */
1181 		pv_remove(pmap, va, m);
1182 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
1183 		pv_remove(pmap, va, m);
1184 		if (TAILQ_EMPTY(&m->md.pv_list))
1185 			m->md.pv_tracked = false;
1186 	}
1187 	mtx_lock_spin(&tlbivax_mutex);
1188 	tlb_miss_lock();
1189 
1190 	tlb0_flush_entry(va);
1191 	*pte = 0;
1192 
1193 	tlb_miss_unlock();
1194 	mtx_unlock_spin(&tlbivax_mutex);
1195 
1196 	pmap->pm_stats.resident_count--;
1197 
1198 	if (flags & PTBL_UNHOLD) {
1199 		return (ptbl_unhold(mmu, pmap, va));
1200 	}
1201 	return (0);
1202 }
1203 
1204 /*
1205  * Insert PTE for a given page and virtual address.
1206  */
1207 static int
1208 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1209     boolean_t nosleep)
1210 {
1211 	unsigned int	pp2d_idx = PP2D_IDX(va);
1212 	unsigned int	pdir_idx = PDIR_IDX(va);
1213 	unsigned int	ptbl_idx = PTBL_IDX(va);
1214 	pte_t          *ptbl, *pte, pte_tmp;
1215 	pte_t         **pdir;
1216 
1217 	/* Get the page directory pointer. */
1218 	pdir = pmap->pm_pp2d[pp2d_idx];
1219 	if (pdir == NULL)
1220 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1221 
1222 	/* Get the page table pointer. */
1223 	ptbl = pdir[pdir_idx];
1224 
1225 	if (ptbl == NULL) {
1226 		/* Allocate page table pages. */
1227 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1228 		if (ptbl == NULL) {
1229 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1230 			return (ENOMEM);
1231 		}
1232 		pte = &ptbl[ptbl_idx];
1233 	} else {
1234 		/*
1235 		 * Check if there is valid mapping for requested va, if there
1236 		 * is, remove it.
1237 		 */
1238 		pte = &ptbl[ptbl_idx];
1239 		if (PTE_ISVALID(pte)) {
1240 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1241 		} else {
1242 			/*
1243 			 * pte is not used, increment hold count for ptbl
1244 			 * pages.
1245 			 */
1246 			if (pmap != kernel_pmap)
1247 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1248 		}
1249 	}
1250 
1251 	if (pdir[pdir_idx] == NULL) {
1252 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1253 			pdir_hold(mmu, pmap, pdir);
1254 		pdir[pdir_idx] = ptbl;
1255 	}
1256 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1257 		pmap->pm_pp2d[pp2d_idx] = pdir;
1258 
1259 	/*
1260 	 * Insert pv_entry into pv_list for mapped page if part of managed
1261 	 * memory.
1262 	 */
1263 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1264 		flags |= PTE_MANAGED;
1265 
1266 		/* Create and insert pv entry. */
1267 		pv_insert(pmap, va, m);
1268 	}
1269 
1270 	pmap->pm_stats.resident_count++;
1271 
1272 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1273 	pte_tmp |= (PTE_VALID | flags);
1274 
1275 	mtx_lock_spin(&tlbivax_mutex);
1276 	tlb_miss_lock();
1277 
1278 	tlb0_flush_entry(va);
1279 	*pte = pte_tmp;
1280 
1281 	tlb_miss_unlock();
1282 	mtx_unlock_spin(&tlbivax_mutex);
1283 
1284 	return (0);
1285 }
1286 
1287 /* Return the pa for the given pmap/va. */
1288 static	vm_paddr_t
1289 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1290 {
1291 	vm_paddr_t	pa = 0;
1292 	pte_t          *pte;
1293 
1294 	pte = pte_find(mmu, pmap, va);
1295 	if ((pte != NULL) && PTE_ISVALID(pte))
1296 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1297 	return (pa);
1298 }
1299 
1300 
1301 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1302 static void
1303 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1304 {
1305 	int		i, j;
1306 	vm_offset_t	va;
1307 	pte_t		*pte;
1308 
1309 	va = addr;
1310 	/* Initialize kernel pdir */
1311 	for (i = 0; i < kernel_pdirs; i++) {
1312 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1313 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1314 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1315 		    j < PDIR_NENTRIES; j++) {
1316 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1317 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE) +
1318 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE));
1319 		}
1320 	}
1321 
1322 	/*
1323 	 * Fill in PTEs covering kernel code and data. They are not required
1324 	 * for address translation, as this area is covered by static TLB1
1325 	 * entries, but for pte_vatopa() to work correctly with kernel area
1326 	 * addresses.
1327 	 */
1328 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1329 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1330 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1331 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1332 		    PTE_VALID | PTE_PS_4KB;
1333 	}
1334 }
1335 #else
1336 /*
1337  * Clean pte entry, try to free page table page if requested.
1338  *
1339  * Return 1 if ptbl pages were freed, otherwise return 0.
1340  */
1341 static int
1342 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1343 {
1344 	unsigned int pdir_idx = PDIR_IDX(va);
1345 	unsigned int ptbl_idx = PTBL_IDX(va);
1346 	vm_page_t m;
1347 	pte_t *ptbl;
1348 	pte_t *pte;
1349 
1350 	//int su = (pmap == kernel_pmap);
1351 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1352 	//		su, (u_int32_t)pmap, va, flags);
1353 
1354 	ptbl = pmap->pm_pdir[pdir_idx];
1355 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1356 
1357 	pte = &ptbl[ptbl_idx];
1358 
1359 	if (pte == NULL || !PTE_ISVALID(pte))
1360 		return (0);
1361 
1362 	if (PTE_ISWIRED(pte))
1363 		pmap->pm_stats.wired_count--;
1364 
1365 	/* Get vm_page_t for mapped pte. */
1366 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1367 
1368 	/* Handle managed entry. */
1369 	if (PTE_ISMANAGED(pte)) {
1370 
1371 		if (PTE_ISMODIFIED(pte))
1372 			vm_page_dirty(m);
1373 
1374 		if (PTE_ISREFERENCED(pte))
1375 			vm_page_aflag_set(m, PGA_REFERENCED);
1376 
1377 		pv_remove(pmap, va, m);
1378 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
1379 		/*
1380 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1381 		 * used.  This is needed by the NCSW support code for fast
1382 		 * VA<->PA translation.
1383 		 */
1384 		pv_remove(pmap, va, m);
1385 		if (TAILQ_EMPTY(&m->md.pv_list))
1386 			m->md.pv_tracked = false;
1387 	}
1388 
1389 	mtx_lock_spin(&tlbivax_mutex);
1390 	tlb_miss_lock();
1391 
1392 	tlb0_flush_entry(va);
1393 	*pte = 0;
1394 
1395 	tlb_miss_unlock();
1396 	mtx_unlock_spin(&tlbivax_mutex);
1397 
1398 	pmap->pm_stats.resident_count--;
1399 
1400 	if (flags & PTBL_UNHOLD) {
1401 		//debugf("pte_remove: e (unhold)\n");
1402 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1403 	}
1404 
1405 	//debugf("pte_remove: e\n");
1406 	return (0);
1407 }
1408 
1409 /*
1410  * Insert PTE for a given page and virtual address.
1411  */
1412 static int
1413 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1414     boolean_t nosleep)
1415 {
1416 	unsigned int pdir_idx = PDIR_IDX(va);
1417 	unsigned int ptbl_idx = PTBL_IDX(va);
1418 	pte_t *ptbl, *pte, pte_tmp;
1419 
1420 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1421 	    pmap == kernel_pmap, pmap, va);
1422 
1423 	/* Get the page table pointer. */
1424 	ptbl = pmap->pm_pdir[pdir_idx];
1425 
1426 	if (ptbl == NULL) {
1427 		/* Allocate page table pages. */
1428 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1429 		if (ptbl == NULL) {
1430 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1431 			return (ENOMEM);
1432 		}
1433 		pmap->pm_pdir[pdir_idx] = ptbl;
1434 		pte = &ptbl[ptbl_idx];
1435 	} else {
1436 		/*
1437 		 * Check if there is valid mapping for requested
1438 		 * va, if there is, remove it.
1439 		 */
1440 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1441 		if (PTE_ISVALID(pte)) {
1442 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1443 		} else {
1444 			/*
1445 			 * pte is not used, increment hold count
1446 			 * for ptbl pages.
1447 			 */
1448 			if (pmap != kernel_pmap)
1449 				ptbl_hold(mmu, pmap, pdir_idx);
1450 		}
1451 	}
1452 
1453 	/*
1454 	 * Insert pv_entry into pv_list for mapped page if part of managed
1455 	 * memory.
1456 	 */
1457 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1458 		flags |= PTE_MANAGED;
1459 
1460 		/* Create and insert pv entry. */
1461 		pv_insert(pmap, va, m);
1462 	}
1463 
1464 	pmap->pm_stats.resident_count++;
1465 
1466 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1467 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1468 
1469 	mtx_lock_spin(&tlbivax_mutex);
1470 	tlb_miss_lock();
1471 
1472 	tlb0_flush_entry(va);
1473 	*pte = pte_tmp;
1474 
1475 	tlb_miss_unlock();
1476 	mtx_unlock_spin(&tlbivax_mutex);
1477 	return (0);
1478 }
1479 
1480 /* Return the pa for the given pmap/va. */
1481 static vm_paddr_t
1482 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1483 {
1484 	vm_paddr_t pa = 0;
1485 	pte_t *pte;
1486 
1487 	pte = pte_find(mmu, pmap, va);
1488 	if ((pte != NULL) && PTE_ISVALID(pte))
1489 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1490 	return (pa);
1491 }
1492 
1493 /* Get a pointer to a PTE in a page table. */
1494 static pte_t *
1495 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1496 {
1497 	unsigned int pdir_idx = PDIR_IDX(va);
1498 	unsigned int ptbl_idx = PTBL_IDX(va);
1499 
1500 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1501 
1502 	if (pmap->pm_pdir[pdir_idx])
1503 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1504 
1505 	return (NULL);
1506 }
1507 
1508 /* Set up kernel page tables. */
1509 static void
1510 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1511 {
1512 	int		i;
1513 	vm_offset_t	va;
1514 	pte_t		*pte;
1515 
1516 	/* Initialize kernel pdir */
1517 	for (i = 0; i < kernel_ptbls; i++)
1518 		kernel_pmap->pm_pdir[kptbl_min + i] =
1519 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1520 
1521 	/*
1522 	 * Fill in PTEs covering kernel code and data. They are not required
1523 	 * for address translation, as this area is covered by static TLB1
1524 	 * entries, but for pte_vatopa() to work correctly with kernel area
1525 	 * addresses.
1526 	 */
1527 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1528 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1529 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1530 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1531 		    PTE_VALID | PTE_PS_4KB;
1532 	}
1533 }
1534 #endif
1535 
1536 /**************************************************************************/
1537 /* PMAP related */
1538 /**************************************************************************/
1539 
1540 /*
1541  * This is called during booke_init, before the system is really initialized.
1542  */
1543 static void
1544 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1545 {
1546 	vm_paddr_t phys_kernelend;
1547 	struct mem_region *mp, *mp1;
1548 	int cnt, i, j;
1549 	vm_paddr_t s, e, sz;
1550 	vm_paddr_t physsz, hwphyssz;
1551 	u_int phys_avail_count;
1552 	vm_size_t kstack0_sz;
1553 	vm_offset_t kernel_pdir, kstack0;
1554 	vm_paddr_t kstack0_phys;
1555 	void *dpcpu;
1556 	vm_offset_t kernel_ptbl_root;
1557 
1558 	debugf("mmu_booke_bootstrap: entered\n");
1559 
1560 	/* Set interesting system properties */
1561 #ifdef __powerpc64__
1562 	hw_direct_map = 1;
1563 #else
1564 	hw_direct_map = 0;
1565 #endif
1566 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1567 	elf32_nxstack = 1;
1568 #endif
1569 
1570 	/* Initialize invalidation mutex */
1571 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1572 
1573 	/* Read TLB0 size and associativity. */
1574 	tlb0_get_tlbconf();
1575 
1576 	/*
1577 	 * Align kernel start and end address (kernel image).
1578 	 * Note that kernel end does not necessarily relate to kernsize.
1579 	 * kernsize is the size of the kernel that is actually mapped.
1580 	 */
1581 	data_start = round_page(kernelend);
1582 	data_end = data_start;
1583 
1584 	/* Allocate the dynamic per-cpu area. */
1585 	dpcpu = (void *)data_end;
1586 	data_end += DPCPU_SIZE;
1587 
1588 	/* Allocate space for the message buffer. */
1589 	msgbufp = (struct msgbuf *)data_end;
1590 	data_end += msgbufsize;
1591 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1592 	    (uintptr_t)msgbufp, data_end);
1593 
1594 	data_end = round_page(data_end);
1595 
1596 #ifdef __powerpc64__
1597 	kernel_ptbl_root = data_end;
1598 	data_end += PP2D_NENTRIES * sizeof(pte_t**);
1599 #else
1600 	/* Allocate space for ptbl_bufs. */
1601 	ptbl_bufs = (struct ptbl_buf *)data_end;
1602 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1603 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1604 	    (uintptr_t)ptbl_bufs, data_end);
1605 
1606 	data_end = round_page(data_end);
1607 	kernel_ptbl_root = data_end;
1608 	data_end += PDIR_NENTRIES * sizeof(pte_t*);
1609 #endif
1610 
1611 	/* Allocate PTE tables for kernel KVA. */
1612 	kernel_pdir = data_end;
1613 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1614 	    PDIR_SIZE);
1615 #ifdef __powerpc64__
1616 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1617 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1618 #endif
1619 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1620 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1621 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1622 	    kernel_pdir, data_end);
1623 
1624 	/* Retrieve phys/avail mem regions */
1625 	mem_regions(&physmem_regions, &physmem_regions_sz,
1626 	    &availmem_regions, &availmem_regions_sz);
1627 
1628 	if (PHYS_AVAIL_ENTRIES < availmem_regions_sz)
1629 		panic("mmu_booke_bootstrap: phys_avail too small");
1630 
1631 	data_end = round_page(data_end);
1632 	vm_page_array = (vm_page_t)data_end;
1633 	/*
1634 	 * Get a rough idea (upper bound) on the size of the page array.  The
1635 	 * vm_page_array will not handle any more pages than we have in the
1636 	 * avail_regions array, and most likely much less.
1637 	 */
1638 	sz = 0;
1639 	for (mp = availmem_regions; mp->mr_size; mp++) {
1640 		sz += mp->mr_size;
1641 	}
1642 	sz = (round_page(sz) / (PAGE_SIZE + sizeof(struct vm_page)));
1643 	data_end += round_page(sz * sizeof(struct vm_page));
1644 
1645 	/* Pre-round up to 1MB.  This wastes some space, but saves TLB entries */
1646 	data_end = roundup2(data_end, 1 << 20);
1647 
1648 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1649 	debugf(" kernstart: %#zx\n", kernstart);
1650 	debugf(" kernsize: %#zx\n", kernsize);
1651 
1652 	if (data_end - kernstart > kernsize) {
1653 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1654 		    kernload + kernsize, (data_end - kernstart) - kernsize,
1655 		    _TLB_ENTRY_MEM);
1656 	}
1657 	data_end = kernstart + kernsize;
1658 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1659 
1660 	/*
1661 	 * Clear the structures - note we can only do it safely after the
1662 	 * possible additional TLB1 translations are in place (above) so that
1663 	 * all range up to the currently calculated 'data_end' is covered.
1664 	 */
1665 	dpcpu_init(dpcpu, 0);
1666 #ifdef __powerpc64__
1667 	memset((void *)kernel_pdir, 0,
1668 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1669 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1670 #else
1671 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1672 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1673 #endif
1674 
1675 	/*******************************************************/
1676 	/* Set the start and end of kva. */
1677 	/*******************************************************/
1678 	virtual_avail = round_page(data_end);
1679 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1680 
1681 #ifndef __powerpc64__
1682 	/* Allocate KVA space for page zero/copy operations. */
1683 	zero_page_va = virtual_avail;
1684 	virtual_avail += PAGE_SIZE;
1685 	copy_page_src_va = virtual_avail;
1686 	virtual_avail += PAGE_SIZE;
1687 	copy_page_dst_va = virtual_avail;
1688 	virtual_avail += PAGE_SIZE;
1689 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
1690 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
1691 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
1692 
1693 	/* Initialize page zero/copy mutexes. */
1694 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1695 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1696 
1697 	/* Allocate KVA space for ptbl bufs. */
1698 	ptbl_buf_pool_vabase = virtual_avail;
1699 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1700 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1701 	    ptbl_buf_pool_vabase, virtual_avail);
1702 #endif
1703 
1704 	/* Calculate corresponding physical addresses for the kernel region. */
1705 	phys_kernelend = kernload + kernsize;
1706 	debugf("kernel image and allocated data:\n");
1707 	debugf(" kernload    = 0x%09jx\n", (uintmax_t)kernload);
1708 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
1709 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
1710 
1711 	/*
1712 	 * Remove kernel physical address range from avail regions list. Page
1713 	 * align all regions.  Non-page aligned memory isn't very interesting
1714 	 * to us.  Also, sort the entries for ascending addresses.
1715 	 */
1716 
1717 	sz = 0;
1718 	cnt = availmem_regions_sz;
1719 	debugf("processing avail regions:\n");
1720 	for (mp = availmem_regions; mp->mr_size; mp++) {
1721 		s = mp->mr_start;
1722 		e = mp->mr_start + mp->mr_size;
1723 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1724 		/* Check whether this region holds all of the kernel. */
1725 		if (s < kernload && e > phys_kernelend) {
1726 			availmem_regions[cnt].mr_start = phys_kernelend;
1727 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1728 			e = kernload;
1729 		}
1730 		/* Look whether this regions starts within the kernel. */
1731 		if (s >= kernload && s < phys_kernelend) {
1732 			if (e <= phys_kernelend)
1733 				goto empty;
1734 			s = phys_kernelend;
1735 		}
1736 		/* Now look whether this region ends within the kernel. */
1737 		if (e > kernload && e <= phys_kernelend) {
1738 			if (s >= kernload)
1739 				goto empty;
1740 			e = kernload;
1741 		}
1742 		/* Now page align the start and size of the region. */
1743 		s = round_page(s);
1744 		e = trunc_page(e);
1745 		if (e < s)
1746 			e = s;
1747 		sz = e - s;
1748 		debugf("%09jx-%09jx = %jx\n",
1749 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1750 
1751 		/* Check whether some memory is left here. */
1752 		if (sz == 0) {
1753 		empty:
1754 			memmove(mp, mp + 1,
1755 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1756 			cnt--;
1757 			mp--;
1758 			continue;
1759 		}
1760 
1761 		/* Do an insertion sort. */
1762 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1763 			if (s < mp1->mr_start)
1764 				break;
1765 		if (mp1 < mp) {
1766 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1767 			mp1->mr_start = s;
1768 			mp1->mr_size = sz;
1769 		} else {
1770 			mp->mr_start = s;
1771 			mp->mr_size = sz;
1772 		}
1773 	}
1774 	availmem_regions_sz = cnt;
1775 
1776 	/*******************************************************/
1777 	/* Steal physical memory for kernel stack from the end */
1778 	/* of the first avail region                           */
1779 	/*******************************************************/
1780 	kstack0_sz = kstack_pages * PAGE_SIZE;
1781 	kstack0_phys = availmem_regions[0].mr_start +
1782 	    availmem_regions[0].mr_size;
1783 	kstack0_phys -= kstack0_sz;
1784 	availmem_regions[0].mr_size -= kstack0_sz;
1785 
1786 	/*******************************************************/
1787 	/* Fill in phys_avail table, based on availmem_regions */
1788 	/*******************************************************/
1789 	phys_avail_count = 0;
1790 	physsz = 0;
1791 	hwphyssz = 0;
1792 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1793 
1794 	debugf("fill in phys_avail:\n");
1795 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1796 
1797 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1798 		    (uintmax_t)availmem_regions[i].mr_start,
1799 		    (uintmax_t)availmem_regions[i].mr_start +
1800 		        availmem_regions[i].mr_size,
1801 		    (uintmax_t)availmem_regions[i].mr_size);
1802 
1803 		if (hwphyssz != 0 &&
1804 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1805 			debugf(" hw.physmem adjust\n");
1806 			if (physsz < hwphyssz) {
1807 				phys_avail[j] = availmem_regions[i].mr_start;
1808 				phys_avail[j + 1] =
1809 				    availmem_regions[i].mr_start +
1810 				    hwphyssz - physsz;
1811 				physsz = hwphyssz;
1812 				phys_avail_count++;
1813 				dump_avail[j] = phys_avail[j];
1814 				dump_avail[j + 1] = phys_avail[j + 1];
1815 			}
1816 			break;
1817 		}
1818 
1819 		phys_avail[j] = availmem_regions[i].mr_start;
1820 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1821 		    availmem_regions[i].mr_size;
1822 		phys_avail_count++;
1823 		physsz += availmem_regions[i].mr_size;
1824 		dump_avail[j] = phys_avail[j];
1825 		dump_avail[j + 1] = phys_avail[j + 1];
1826 	}
1827 	physmem = btoc(physsz);
1828 
1829 	/* Calculate the last available physical address. */
1830 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1831 		;
1832 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1833 
1834 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1835 	debugf("phys_avail_count = %d\n", phys_avail_count);
1836 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1837 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1838 
1839 #ifdef __powerpc64__
1840 	/*
1841 	 * Map the physical memory contiguously in TLB1.
1842 	 * Round so it fits into a single mapping.
1843 	 */
1844 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
1845 	    phys_avail[i + 1], _TLB_ENTRY_MEM);
1846 #endif
1847 
1848 	/*******************************************************/
1849 	/* Initialize (statically allocated) kernel pmap. */
1850 	/*******************************************************/
1851 	PMAP_LOCK_INIT(kernel_pmap);
1852 #ifdef __powerpc64__
1853 	kernel_pmap->pm_pp2d = (pte_t ***)kernel_ptbl_root;
1854 #else
1855 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1856 	kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
1857 #endif
1858 
1859 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1860 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1861 	for (i = 0; i < MAXCPU; i++) {
1862 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1863 
1864 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1865 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1866 	}
1867 
1868 	/* Mark kernel_pmap active on all CPUs */
1869 	CPU_FILL(&kernel_pmap->pm_active);
1870 
1871  	/*
1872 	 * Initialize the global pv list lock.
1873 	 */
1874 	rw_init(&pvh_global_lock, "pmap pv global");
1875 
1876 	/*******************************************************/
1877 	/* Final setup */
1878 	/*******************************************************/
1879 
1880 	/* Enter kstack0 into kernel map, provide guard page */
1881 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1882 	thread0.td_kstack = kstack0;
1883 	thread0.td_kstack_pages = kstack_pages;
1884 
1885 	debugf("kstack_sz = 0x%08jx\n", (uintmax_t)kstack0_sz);
1886 	debugf("kstack0_phys at 0x%09jx - 0x%09jx\n",
1887 	    (uintmax_t)kstack0_phys, (uintmax_t)kstack0_phys + kstack0_sz);
1888 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
1889 	    kstack0, kstack0 + kstack0_sz);
1890 
1891 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1892 	for (i = 0; i < kstack_pages; i++) {
1893 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1894 		kstack0 += PAGE_SIZE;
1895 		kstack0_phys += PAGE_SIZE;
1896 	}
1897 
1898 	pmap_bootstrapped = 1;
1899 
1900 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
1901 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
1902 
1903 	debugf("mmu_booke_bootstrap: exit\n");
1904 }
1905 
1906 #ifdef SMP
1907 void
1908 tlb1_ap_prep(void)
1909 {
1910 	tlb_entry_t *e, tmp;
1911 	unsigned int i;
1912 
1913 	/* Prepare TLB1 image for AP processors */
1914 	e = __boot_tlb1;
1915 	for (i = 0; i < TLB1_ENTRIES; i++) {
1916 		tlb1_read_entry(&tmp, i);
1917 
1918 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1919 			memcpy(e++, &tmp, sizeof(tmp));
1920 	}
1921 }
1922 
1923 void
1924 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1925 {
1926 	int i;
1927 
1928 	/*
1929 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1930 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1931 	 * created by tlb1_ap_prep(), so use these values directly to
1932 	 * (re)program AP's TLB1 hardware.
1933 	 *
1934 	 * Start at index 1 because index 0 has the kernel map.
1935 	 */
1936 	for (i = 1; i < TLB1_ENTRIES; i++) {
1937 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1938 			tlb1_write_entry(&__boot_tlb1[i], i);
1939 	}
1940 
1941 	set_mas4_defaults();
1942 }
1943 #endif
1944 
1945 static void
1946 booke_pmap_init_qpages(void)
1947 {
1948 	struct pcpu *pc;
1949 	int i;
1950 
1951 	CPU_FOREACH(i) {
1952 		pc = pcpu_find(i);
1953 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1954 		if (pc->pc_qmap_addr == 0)
1955 			panic("pmap_init_qpages: unable to allocate KVA");
1956 	}
1957 }
1958 
1959 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1960 
1961 /*
1962  * Get the physical page address for the given pmap/virtual address.
1963  */
1964 static vm_paddr_t
1965 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1966 {
1967 	vm_paddr_t pa;
1968 
1969 	PMAP_LOCK(pmap);
1970 	pa = pte_vatopa(mmu, pmap, va);
1971 	PMAP_UNLOCK(pmap);
1972 
1973 	return (pa);
1974 }
1975 
1976 /*
1977  * Extract the physical page address associated with the given
1978  * kernel virtual address.
1979  */
1980 static vm_paddr_t
1981 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1982 {
1983 	tlb_entry_t e;
1984 	vm_paddr_t p = 0;
1985 	int i;
1986 
1987 #ifdef __powerpc64__
1988 	if (va >= DMAP_BASE_ADDRESS && va <= DMAP_MAX_ADDRESS)
1989 		return (DMAP_TO_PHYS(va));
1990 #endif
1991 
1992 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
1993 		p = pte_vatopa(mmu, kernel_pmap, va);
1994 
1995 	if (p == 0) {
1996 		/* Check TLB1 mappings */
1997 		for (i = 0; i < TLB1_ENTRIES; i++) {
1998 			tlb1_read_entry(&e, i);
1999 			if (!(e.mas1 & MAS1_VALID))
2000 				continue;
2001 			if (va >= e.virt && va < e.virt + e.size)
2002 				return (e.phys + (va - e.virt));
2003 		}
2004 	}
2005 
2006 	return (p);
2007 }
2008 
2009 /*
2010  * Initialize the pmap module.
2011  * Called by vm_init, to initialize any structures that the pmap
2012  * system needs to map virtual memory.
2013  */
2014 static void
2015 mmu_booke_init(mmu_t mmu)
2016 {
2017 	int shpgperproc = PMAP_SHPGPERPROC;
2018 
2019 	/*
2020 	 * Initialize the address space (zone) for the pv entries.  Set a
2021 	 * high water mark so that the system can recover from excessive
2022 	 * numbers of pv entries.
2023 	 */
2024 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2025 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2026 
2027 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2028 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2029 
2030 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2031 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2032 
2033 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2034 
2035 	/* Pre-fill pvzone with initial number of pv entries. */
2036 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2037 
2038 	/* Create a UMA zone for page table roots. */
2039 	ptbl_root_zone = uma_zcreate("pmap root", PMAP_ROOT_SIZE,
2040 	    NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, UMA_ZONE_VM);
2041 
2042 	/* Initialize ptbl allocation. */
2043 	ptbl_init();
2044 }
2045 
2046 /*
2047  * Map a list of wired pages into kernel virtual address space.  This is
2048  * intended for temporary mappings which do not need page modification or
2049  * references recorded.  Existing mappings in the region are overwritten.
2050  */
2051 static void
2052 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2053 {
2054 	vm_offset_t va;
2055 
2056 	va = sva;
2057 	while (count-- > 0) {
2058 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2059 		va += PAGE_SIZE;
2060 		m++;
2061 	}
2062 }
2063 
2064 /*
2065  * Remove page mappings from kernel virtual address space.  Intended for
2066  * temporary mappings entered by mmu_booke_qenter.
2067  */
2068 static void
2069 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2070 {
2071 	vm_offset_t va;
2072 
2073 	va = sva;
2074 	while (count-- > 0) {
2075 		mmu_booke_kremove(mmu, va);
2076 		va += PAGE_SIZE;
2077 	}
2078 }
2079 
2080 /*
2081  * Map a wired page into kernel virtual address space.
2082  */
2083 static void
2084 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2085 {
2086 
2087 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2088 }
2089 
2090 static void
2091 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2092 {
2093 	uint32_t flags;
2094 	pte_t *pte;
2095 
2096 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2097 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2098 
2099 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2100 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2101 	flags |= PTE_PS_4KB;
2102 
2103 	pte = pte_find(mmu, kernel_pmap, va);
2104 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2105 
2106 	mtx_lock_spin(&tlbivax_mutex);
2107 	tlb_miss_lock();
2108 
2109 	if (PTE_ISVALID(pte)) {
2110 
2111 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2112 
2113 		/* Flush entry from TLB0 */
2114 		tlb0_flush_entry(va);
2115 	}
2116 
2117 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2118 
2119 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2120 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2121 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2122 
2123 	/* Flush the real memory from the instruction cache. */
2124 	if ((flags & (PTE_I | PTE_G)) == 0)
2125 		__syncicache((void *)va, PAGE_SIZE);
2126 
2127 	tlb_miss_unlock();
2128 	mtx_unlock_spin(&tlbivax_mutex);
2129 }
2130 
2131 /*
2132  * Remove a page from kernel page table.
2133  */
2134 static void
2135 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2136 {
2137 	pte_t *pte;
2138 
2139 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
2140 
2141 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2142 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2143 	    ("mmu_booke_kremove: invalid va"));
2144 
2145 	pte = pte_find(mmu, kernel_pmap, va);
2146 
2147 	if (!PTE_ISVALID(pte)) {
2148 
2149 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2150 
2151 		return;
2152 	}
2153 
2154 	mtx_lock_spin(&tlbivax_mutex);
2155 	tlb_miss_lock();
2156 
2157 	/* Invalidate entry in TLB0, update PTE. */
2158 	tlb0_flush_entry(va);
2159 	*pte = 0;
2160 
2161 	tlb_miss_unlock();
2162 	mtx_unlock_spin(&tlbivax_mutex);
2163 }
2164 
2165 /*
2166  * Provide a kernel pointer corresponding to a given userland pointer.
2167  * The returned pointer is valid until the next time this function is
2168  * called in this thread. This is used internally in copyin/copyout.
2169  */
2170 int
2171 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2172     void **kaddr, size_t ulen, size_t *klen)
2173 {
2174 
2175 	if (trunc_page((uintptr_t)uaddr + ulen) > VM_MAXUSER_ADDRESS)
2176 		return (EFAULT);
2177 
2178 	*kaddr = (void *)(uintptr_t)uaddr;
2179 	if (klen)
2180 		*klen = ulen;
2181 
2182 	return (0);
2183 }
2184 
2185 /*
2186  * Figure out where a given kernel pointer (usually in a fault) points
2187  * to from the VM's perspective, potentially remapping into userland's
2188  * address space.
2189  */
2190 static int
2191 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2192     vm_offset_t *decoded_addr)
2193 {
2194 
2195 	if (trunc_page(addr) <= VM_MAXUSER_ADDRESS)
2196 		*is_user = 1;
2197 	else
2198 		*is_user = 0;
2199 
2200 	*decoded_addr = addr;
2201 	return (0);
2202 }
2203 
2204 /*
2205  * Initialize pmap associated with process 0.
2206  */
2207 static void
2208 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2209 {
2210 
2211 	PMAP_LOCK_INIT(pmap);
2212 	mmu_booke_pinit(mmu, pmap);
2213 	PCPU_SET(curpmap, pmap);
2214 }
2215 
2216 /*
2217  * Initialize a preallocated and zeroed pmap structure,
2218  * such as one in a vmspace structure.
2219  */
2220 static void
2221 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2222 {
2223 	int i;
2224 
2225 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2226 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2227 
2228 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2229 
2230 	for (i = 0; i < MAXCPU; i++)
2231 		pmap->pm_tid[i] = TID_NONE;
2232 	CPU_ZERO(&kernel_pmap->pm_active);
2233 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2234 #ifdef __powerpc64__
2235 	pmap->pm_pp2d = uma_zalloc(ptbl_root_zone, M_WAITOK);
2236 	bzero(pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2237 #else
2238 	pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
2239 	bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2240 	TAILQ_INIT(&pmap->pm_ptbl_list);
2241 #endif
2242 }
2243 
2244 /*
2245  * Release any resources held by the given physical map.
2246  * Called when a pmap initialized by mmu_booke_pinit is being released.
2247  * Should only be called if the map contains no valid mappings.
2248  */
2249 static void
2250 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2251 {
2252 
2253 	KASSERT(pmap->pm_stats.resident_count == 0,
2254 	    ("pmap_release: pmap resident count %ld != 0",
2255 	    pmap->pm_stats.resident_count));
2256 #ifdef __powerpc64__
2257 	uma_zfree(ptbl_root_zone, pmap->pm_pp2d);
2258 #else
2259 	uma_zfree(ptbl_root_zone, pmap->pm_pdir);
2260 #endif
2261 }
2262 
2263 /*
2264  * Insert the given physical page at the specified virtual address in the
2265  * target physical map with the protection requested. If specified the page
2266  * will be wired down.
2267  */
2268 static int
2269 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2270     vm_prot_t prot, u_int flags, int8_t psind)
2271 {
2272 	int error;
2273 
2274 	rw_wlock(&pvh_global_lock);
2275 	PMAP_LOCK(pmap);
2276 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2277 	PMAP_UNLOCK(pmap);
2278 	rw_wunlock(&pvh_global_lock);
2279 	return (error);
2280 }
2281 
2282 static int
2283 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2284     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2285 {
2286 	pte_t *pte;
2287 	vm_paddr_t pa;
2288 	uint32_t flags;
2289 	int error, su, sync;
2290 
2291 	pa = VM_PAGE_TO_PHYS(m);
2292 	su = (pmap == kernel_pmap);
2293 	sync = 0;
2294 
2295 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2296 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2297 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2298 	//		(u_int32_t)m, va, pa, prot, flags);
2299 
2300 	if (su) {
2301 		KASSERT(((va >= virtual_avail) &&
2302 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2303 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2304 	} else {
2305 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2306 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2307 	}
2308 	if ((m->oflags & VPO_UNMANAGED) == 0) {
2309 		if ((pmap_flags & PMAP_ENTER_QUICK_LOCKED) == 0)
2310 			VM_PAGE_OBJECT_BUSY_ASSERT(m);
2311 		else
2312 			VM_OBJECT_ASSERT_LOCKED(m->object);
2313 	}
2314 
2315 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2316 
2317 	/*
2318 	 * If there is an existing mapping, and the physical address has not
2319 	 * changed, must be protection or wiring change.
2320 	 */
2321 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2322 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2323 
2324 		/*
2325 		 * Before actually updating pte->flags we calculate and
2326 		 * prepare its new value in a helper var.
2327 		 */
2328 		flags = *pte;
2329 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2330 
2331 		/* Wiring change, just update stats. */
2332 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2333 			if (!PTE_ISWIRED(pte)) {
2334 				flags |= PTE_WIRED;
2335 				pmap->pm_stats.wired_count++;
2336 			}
2337 		} else {
2338 			if (PTE_ISWIRED(pte)) {
2339 				flags &= ~PTE_WIRED;
2340 				pmap->pm_stats.wired_count--;
2341 			}
2342 		}
2343 
2344 		if (prot & VM_PROT_WRITE) {
2345 			/* Add write permissions. */
2346 			flags |= PTE_SW;
2347 			if (!su)
2348 				flags |= PTE_UW;
2349 
2350 			if ((flags & PTE_MANAGED) != 0)
2351 				vm_page_aflag_set(m, PGA_WRITEABLE);
2352 		} else {
2353 			/* Handle modified pages, sense modify status. */
2354 
2355 			/*
2356 			 * The PTE_MODIFIED flag could be set by underlying
2357 			 * TLB misses since we last read it (above), possibly
2358 			 * other CPUs could update it so we check in the PTE
2359 			 * directly rather than rely on that saved local flags
2360 			 * copy.
2361 			 */
2362 			if (PTE_ISMODIFIED(pte))
2363 				vm_page_dirty(m);
2364 		}
2365 
2366 		if (prot & VM_PROT_EXECUTE) {
2367 			flags |= PTE_SX;
2368 			if (!su)
2369 				flags |= PTE_UX;
2370 
2371 			/*
2372 			 * Check existing flags for execute permissions: if we
2373 			 * are turning execute permissions on, icache should
2374 			 * be flushed.
2375 			 */
2376 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2377 				sync++;
2378 		}
2379 
2380 		flags &= ~PTE_REFERENCED;
2381 
2382 		/*
2383 		 * The new flags value is all calculated -- only now actually
2384 		 * update the PTE.
2385 		 */
2386 		mtx_lock_spin(&tlbivax_mutex);
2387 		tlb_miss_lock();
2388 
2389 		tlb0_flush_entry(va);
2390 		*pte &= ~PTE_FLAGS_MASK;
2391 		*pte |= flags;
2392 
2393 		tlb_miss_unlock();
2394 		mtx_unlock_spin(&tlbivax_mutex);
2395 
2396 	} else {
2397 		/*
2398 		 * If there is an existing mapping, but it's for a different
2399 		 * physical address, pte_enter() will delete the old mapping.
2400 		 */
2401 		//if ((pte != NULL) && PTE_ISVALID(pte))
2402 		//	debugf("mmu_booke_enter_locked: replace\n");
2403 		//else
2404 		//	debugf("mmu_booke_enter_locked: new\n");
2405 
2406 		/* Now set up the flags and install the new mapping. */
2407 		flags = (PTE_SR | PTE_VALID);
2408 		flags |= PTE_M;
2409 
2410 		if (!su)
2411 			flags |= PTE_UR;
2412 
2413 		if (prot & VM_PROT_WRITE) {
2414 			flags |= PTE_SW;
2415 			if (!su)
2416 				flags |= PTE_UW;
2417 
2418 			if ((m->oflags & VPO_UNMANAGED) == 0)
2419 				vm_page_aflag_set(m, PGA_WRITEABLE);
2420 		}
2421 
2422 		if (prot & VM_PROT_EXECUTE) {
2423 			flags |= PTE_SX;
2424 			if (!su)
2425 				flags |= PTE_UX;
2426 		}
2427 
2428 		/* If its wired update stats. */
2429 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2430 			flags |= PTE_WIRED;
2431 
2432 		error = pte_enter(mmu, pmap, m, va, flags,
2433 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2434 		if (error != 0)
2435 			return (KERN_RESOURCE_SHORTAGE);
2436 
2437 		if ((flags & PMAP_ENTER_WIRED) != 0)
2438 			pmap->pm_stats.wired_count++;
2439 
2440 		/* Flush the real memory from the instruction cache. */
2441 		if (prot & VM_PROT_EXECUTE)
2442 			sync++;
2443 	}
2444 
2445 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2446 		__syncicache((void *)va, PAGE_SIZE);
2447 		sync = 0;
2448 	}
2449 
2450 	return (KERN_SUCCESS);
2451 }
2452 
2453 /*
2454  * Maps a sequence of resident pages belonging to the same object.
2455  * The sequence begins with the given page m_start.  This page is
2456  * mapped at the given virtual address start.  Each subsequent page is
2457  * mapped at a virtual address that is offset from start by the same
2458  * amount as the page is offset from m_start within the object.  The
2459  * last page in the sequence is the page with the largest offset from
2460  * m_start that can be mapped at a virtual address less than the given
2461  * virtual address end.  Not every virtual page between start and end
2462  * is mapped; only those for which a resident page exists with the
2463  * corresponding offset from m_start are mapped.
2464  */
2465 static void
2466 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2467     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2468 {
2469 	vm_page_t m;
2470 	vm_pindex_t diff, psize;
2471 
2472 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2473 
2474 	psize = atop(end - start);
2475 	m = m_start;
2476 	rw_wlock(&pvh_global_lock);
2477 	PMAP_LOCK(pmap);
2478 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2479 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2480 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2481 		    PMAP_ENTER_NOSLEEP | PMAP_ENTER_QUICK_LOCKED, 0);
2482 		m = TAILQ_NEXT(m, listq);
2483 	}
2484 	rw_wunlock(&pvh_global_lock);
2485 	PMAP_UNLOCK(pmap);
2486 }
2487 
2488 static void
2489 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2490     vm_prot_t prot)
2491 {
2492 
2493 	rw_wlock(&pvh_global_lock);
2494 	PMAP_LOCK(pmap);
2495 	mmu_booke_enter_locked(mmu, pmap, va, m,
2496 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP |
2497 	    PMAP_ENTER_QUICK_LOCKED, 0);
2498 	rw_wunlock(&pvh_global_lock);
2499 	PMAP_UNLOCK(pmap);
2500 }
2501 
2502 /*
2503  * Remove the given range of addresses from the specified map.
2504  *
2505  * It is assumed that the start and end are properly rounded to the page size.
2506  */
2507 static void
2508 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2509 {
2510 	pte_t *pte;
2511 	uint8_t hold_flag;
2512 
2513 	int su = (pmap == kernel_pmap);
2514 
2515 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2516 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2517 
2518 	if (su) {
2519 		KASSERT(((va >= virtual_avail) &&
2520 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2521 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2522 	} else {
2523 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2524 		    ("mmu_booke_remove: user pmap, non user va"));
2525 	}
2526 
2527 	if (PMAP_REMOVE_DONE(pmap)) {
2528 		//debugf("mmu_booke_remove: e (empty)\n");
2529 		return;
2530 	}
2531 
2532 	hold_flag = PTBL_HOLD_FLAG(pmap);
2533 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2534 
2535 	rw_wlock(&pvh_global_lock);
2536 	PMAP_LOCK(pmap);
2537 	for (; va < endva; va += PAGE_SIZE) {
2538 		pte = pte_find(mmu, pmap, va);
2539 		if ((pte != NULL) && PTE_ISVALID(pte))
2540 			pte_remove(mmu, pmap, va, hold_flag);
2541 	}
2542 	PMAP_UNLOCK(pmap);
2543 	rw_wunlock(&pvh_global_lock);
2544 
2545 	//debugf("mmu_booke_remove: e\n");
2546 }
2547 
2548 /*
2549  * Remove physical page from all pmaps in which it resides.
2550  */
2551 static void
2552 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2553 {
2554 	pv_entry_t pv, pvn;
2555 	uint8_t hold_flag;
2556 
2557 	rw_wlock(&pvh_global_lock);
2558 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2559 		pvn = TAILQ_NEXT(pv, pv_link);
2560 
2561 		PMAP_LOCK(pv->pv_pmap);
2562 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2563 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2564 		PMAP_UNLOCK(pv->pv_pmap);
2565 	}
2566 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2567 	rw_wunlock(&pvh_global_lock);
2568 }
2569 
2570 /*
2571  * Map a range of physical addresses into kernel virtual address space.
2572  */
2573 static vm_offset_t
2574 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2575     vm_paddr_t pa_end, int prot)
2576 {
2577 	vm_offset_t sva = *virt;
2578 	vm_offset_t va = sva;
2579 
2580 #ifdef __powerpc64__
2581 	/* XXX: Handle memory not starting at 0x0. */
2582 	if (pa_end < ctob(Maxmem))
2583 		return (PHYS_TO_DMAP(pa_start));
2584 #endif
2585 
2586 	while (pa_start < pa_end) {
2587 		mmu_booke_kenter(mmu, va, pa_start);
2588 		va += PAGE_SIZE;
2589 		pa_start += PAGE_SIZE;
2590 	}
2591 	*virt = va;
2592 
2593 	return (sva);
2594 }
2595 
2596 /*
2597  * The pmap must be activated before it's address space can be accessed in any
2598  * way.
2599  */
2600 static void
2601 mmu_booke_activate(mmu_t mmu, struct thread *td)
2602 {
2603 	pmap_t pmap;
2604 	u_int cpuid;
2605 
2606 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2607 
2608 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
2609 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2610 
2611 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2612 
2613 	sched_pin();
2614 
2615 	cpuid = PCPU_GET(cpuid);
2616 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2617 	PCPU_SET(curpmap, pmap);
2618 
2619 	if (pmap->pm_tid[cpuid] == TID_NONE)
2620 		tid_alloc(pmap);
2621 
2622 	/* Load PID0 register with pmap tid value. */
2623 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2624 	__asm __volatile("isync");
2625 
2626 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2627 
2628 	sched_unpin();
2629 
2630 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2631 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2632 }
2633 
2634 /*
2635  * Deactivate the specified process's address space.
2636  */
2637 static void
2638 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2639 {
2640 	pmap_t pmap;
2641 
2642 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2643 
2644 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
2645 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2646 
2647 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2648 
2649 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2650 	PCPU_SET(curpmap, NULL);
2651 }
2652 
2653 /*
2654  * Copy the range specified by src_addr/len
2655  * from the source map to the range dst_addr/len
2656  * in the destination map.
2657  *
2658  * This routine is only advisory and need not do anything.
2659  */
2660 static void
2661 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2662     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2663 {
2664 
2665 }
2666 
2667 /*
2668  * Set the physical protection on the specified range of this map as requested.
2669  */
2670 static void
2671 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2672     vm_prot_t prot)
2673 {
2674 	vm_offset_t va;
2675 	vm_page_t m;
2676 	pte_t *pte;
2677 
2678 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2679 		mmu_booke_remove(mmu, pmap, sva, eva);
2680 		return;
2681 	}
2682 
2683 	if (prot & VM_PROT_WRITE)
2684 		return;
2685 
2686 	PMAP_LOCK(pmap);
2687 	for (va = sva; va < eva; va += PAGE_SIZE) {
2688 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2689 			if (PTE_ISVALID(pte)) {
2690 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2691 
2692 				mtx_lock_spin(&tlbivax_mutex);
2693 				tlb_miss_lock();
2694 
2695 				/* Handle modified pages. */
2696 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2697 					vm_page_dirty(m);
2698 
2699 				tlb0_flush_entry(va);
2700 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2701 
2702 				tlb_miss_unlock();
2703 				mtx_unlock_spin(&tlbivax_mutex);
2704 			}
2705 		}
2706 	}
2707 	PMAP_UNLOCK(pmap);
2708 }
2709 
2710 /*
2711  * Clear the write and modified bits in each of the given page's mappings.
2712  */
2713 static void
2714 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2715 {
2716 	pv_entry_t pv;
2717 	pte_t *pte;
2718 
2719 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2720 	    ("mmu_booke_remove_write: page %p is not managed", m));
2721 	vm_page_assert_busied(m);
2722 
2723 	if (!pmap_page_is_write_mapped(m))
2724 	        return;
2725 	rw_wlock(&pvh_global_lock);
2726 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2727 		PMAP_LOCK(pv->pv_pmap);
2728 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2729 			if (PTE_ISVALID(pte)) {
2730 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2731 
2732 				mtx_lock_spin(&tlbivax_mutex);
2733 				tlb_miss_lock();
2734 
2735 				/* Handle modified pages. */
2736 				if (PTE_ISMODIFIED(pte))
2737 					vm_page_dirty(m);
2738 
2739 				/* Flush mapping from TLB0. */
2740 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2741 
2742 				tlb_miss_unlock();
2743 				mtx_unlock_spin(&tlbivax_mutex);
2744 			}
2745 		}
2746 		PMAP_UNLOCK(pv->pv_pmap);
2747 	}
2748 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2749 	rw_wunlock(&pvh_global_lock);
2750 }
2751 
2752 static void
2753 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2754 {
2755 	pte_t *pte;
2756 	vm_paddr_t pa = 0;
2757 	int sync_sz, valid;
2758 #ifndef __powerpc64__
2759 	pmap_t pmap;
2760 	vm_page_t m;
2761 	vm_offset_t addr;
2762 	int active;
2763 #endif
2764 
2765 #ifndef __powerpc64__
2766 	rw_wlock(&pvh_global_lock);
2767 	pmap = PCPU_GET(curpmap);
2768 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2769 #endif
2770 	while (sz > 0) {
2771 		PMAP_LOCK(pm);
2772 		pte = pte_find(mmu, pm, va);
2773 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2774 		if (valid)
2775 			pa = PTE_PA(pte);
2776 		PMAP_UNLOCK(pm);
2777 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
2778 		sync_sz = min(sync_sz, sz);
2779 		if (valid) {
2780 #ifdef __powerpc64__
2781 			pa += (va & PAGE_MASK);
2782 			__syncicache((void *)PHYS_TO_DMAP(pa), sync_sz);
2783 #else
2784 			if (!active) {
2785 				/* Create a mapping in the active pmap. */
2786 				addr = 0;
2787 				m = PHYS_TO_VM_PAGE(pa);
2788 				PMAP_LOCK(pmap);
2789 				pte_enter(mmu, pmap, m, addr,
2790 				    PTE_SR | PTE_VALID, FALSE);
2791 				addr += (va & PAGE_MASK);
2792 				__syncicache((void *)addr, sync_sz);
2793 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2794 				PMAP_UNLOCK(pmap);
2795 			} else
2796 				__syncicache((void *)va, sync_sz);
2797 #endif
2798 		}
2799 		va += sync_sz;
2800 		sz -= sync_sz;
2801 	}
2802 #ifndef __powerpc64__
2803 	rw_wunlock(&pvh_global_lock);
2804 #endif
2805 }
2806 
2807 /*
2808  * Atomically extract and hold the physical page with the given
2809  * pmap and virtual address pair if that mapping permits the given
2810  * protection.
2811  */
2812 static vm_page_t
2813 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2814     vm_prot_t prot)
2815 {
2816 	pte_t *pte;
2817 	vm_page_t m;
2818 	uint32_t pte_wbit;
2819 
2820 	m = NULL;
2821 	PMAP_LOCK(pmap);
2822 	pte = pte_find(mmu, pmap, va);
2823 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2824 		if (pmap == kernel_pmap)
2825 			pte_wbit = PTE_SW;
2826 		else
2827 			pte_wbit = PTE_UW;
2828 
2829 		if ((*pte & pte_wbit) != 0 || (prot & VM_PROT_WRITE) == 0) {
2830 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2831 			if (!vm_page_wire_mapped(m))
2832 				m = NULL;
2833 		}
2834 	}
2835 	PMAP_UNLOCK(pmap);
2836 	return (m);
2837 }
2838 
2839 /*
2840  * Initialize a vm_page's machine-dependent fields.
2841  */
2842 static void
2843 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2844 {
2845 
2846 	m->md.pv_tracked = 0;
2847 	TAILQ_INIT(&m->md.pv_list);
2848 }
2849 
2850 /*
2851  * mmu_booke_zero_page_area zeros the specified hardware page by
2852  * mapping it into virtual memory and using bzero to clear
2853  * its contents.
2854  *
2855  * off and size must reside within a single page.
2856  */
2857 static void
2858 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2859 {
2860 	vm_offset_t va;
2861 
2862 	/* XXX KASSERT off and size are within a single page? */
2863 
2864 #ifdef __powerpc64__
2865 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2866 	bzero((caddr_t)va + off, size);
2867 #else
2868 	mtx_lock(&zero_page_mutex);
2869 	va = zero_page_va;
2870 
2871 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2872 	bzero((caddr_t)va + off, size);
2873 	mmu_booke_kremove(mmu, va);
2874 
2875 	mtx_unlock(&zero_page_mutex);
2876 #endif
2877 }
2878 
2879 /*
2880  * mmu_booke_zero_page zeros the specified hardware page.
2881  */
2882 static void
2883 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2884 {
2885 	vm_offset_t off, va;
2886 
2887 #ifdef __powerpc64__
2888 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2889 
2890 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2891 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2892 #else
2893 	va = zero_page_va;
2894 	mtx_lock(&zero_page_mutex);
2895 
2896 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2897 
2898 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2899 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2900 
2901 	mmu_booke_kremove(mmu, va);
2902 
2903 	mtx_unlock(&zero_page_mutex);
2904 #endif
2905 }
2906 
2907 /*
2908  * mmu_booke_copy_page copies the specified (machine independent) page by
2909  * mapping the page into virtual memory and using memcopy to copy the page,
2910  * one machine dependent page at a time.
2911  */
2912 static void
2913 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2914 {
2915 	vm_offset_t sva, dva;
2916 
2917 #ifdef __powerpc64__
2918 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
2919 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
2920 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2921 #else
2922 	sva = copy_page_src_va;
2923 	dva = copy_page_dst_va;
2924 
2925 	mtx_lock(&copy_page_mutex);
2926 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2927 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2928 
2929 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2930 
2931 	mmu_booke_kremove(mmu, dva);
2932 	mmu_booke_kremove(mmu, sva);
2933 	mtx_unlock(&copy_page_mutex);
2934 #endif
2935 }
2936 
2937 static inline void
2938 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2939     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2940 {
2941 	void *a_cp, *b_cp;
2942 	vm_offset_t a_pg_offset, b_pg_offset;
2943 	int cnt;
2944 
2945 #ifdef __powerpc64__
2946 	vm_page_t pa, pb;
2947 
2948 	while (xfersize > 0) {
2949 		a_pg_offset = a_offset & PAGE_MASK;
2950 		pa = ma[a_offset >> PAGE_SHIFT];
2951 		b_pg_offset = b_offset & PAGE_MASK;
2952 		pb = mb[b_offset >> PAGE_SHIFT];
2953 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2954 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2955 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
2956 		    a_pg_offset);
2957 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
2958 		    b_pg_offset);
2959 		bcopy(a_cp, b_cp, cnt);
2960 		a_offset += cnt;
2961 		b_offset += cnt;
2962 		xfersize -= cnt;
2963 	}
2964 #else
2965 	mtx_lock(&copy_page_mutex);
2966 	while (xfersize > 0) {
2967 		a_pg_offset = a_offset & PAGE_MASK;
2968 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2969 		mmu_booke_kenter(mmu, copy_page_src_va,
2970 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2971 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2972 		b_pg_offset = b_offset & PAGE_MASK;
2973 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2974 		mmu_booke_kenter(mmu, copy_page_dst_va,
2975 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2976 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2977 		bcopy(a_cp, b_cp, cnt);
2978 		mmu_booke_kremove(mmu, copy_page_dst_va);
2979 		mmu_booke_kremove(mmu, copy_page_src_va);
2980 		a_offset += cnt;
2981 		b_offset += cnt;
2982 		xfersize -= cnt;
2983 	}
2984 	mtx_unlock(&copy_page_mutex);
2985 #endif
2986 }
2987 
2988 static vm_offset_t
2989 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2990 {
2991 #ifdef __powerpc64__
2992 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
2993 #else
2994 	vm_paddr_t paddr;
2995 	vm_offset_t qaddr;
2996 	uint32_t flags;
2997 	pte_t *pte;
2998 
2999 	paddr = VM_PAGE_TO_PHYS(m);
3000 
3001 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3002 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3003 	flags |= PTE_PS_4KB;
3004 
3005 	critical_enter();
3006 	qaddr = PCPU_GET(qmap_addr);
3007 
3008 	pte = pte_find(mmu, kernel_pmap, qaddr);
3009 
3010 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3011 
3012 	/*
3013 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3014  	 * not be present in other TLBs.  Is there a better instruction
3015 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3016 	 */
3017 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3018 	__asm __volatile("isync; msync");
3019 
3020 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3021 
3022 	/* Flush the real memory from the instruction cache. */
3023 	if ((flags & (PTE_I | PTE_G)) == 0)
3024 		__syncicache((void *)qaddr, PAGE_SIZE);
3025 
3026 	return (qaddr);
3027 #endif
3028 }
3029 
3030 static void
3031 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3032 {
3033 #ifndef __powerpc64__
3034 	pte_t *pte;
3035 
3036 	pte = pte_find(mmu, kernel_pmap, addr);
3037 
3038 	KASSERT(PCPU_GET(qmap_addr) == addr,
3039 	    ("mmu_booke_quick_remove_page: invalid address"));
3040 	KASSERT(*pte != 0,
3041 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3042 
3043 	*pte = 0;
3044 	critical_exit();
3045 #endif
3046 }
3047 
3048 /*
3049  * Return whether or not the specified physical page was modified
3050  * in any of physical maps.
3051  */
3052 static boolean_t
3053 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3054 {
3055 	pte_t *pte;
3056 	pv_entry_t pv;
3057 	boolean_t rv;
3058 
3059 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3060 	    ("mmu_booke_is_modified: page %p is not managed", m));
3061 	rv = FALSE;
3062 
3063 	/*
3064 	 * If the page is not busied then this check is racy.
3065 	 */
3066 	if (!pmap_page_is_write_mapped(m))
3067 		return (FALSE);
3068 
3069 	rw_wlock(&pvh_global_lock);
3070 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3071 		PMAP_LOCK(pv->pv_pmap);
3072 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3073 		    PTE_ISVALID(pte)) {
3074 			if (PTE_ISMODIFIED(pte))
3075 				rv = TRUE;
3076 		}
3077 		PMAP_UNLOCK(pv->pv_pmap);
3078 		if (rv)
3079 			break;
3080 	}
3081 	rw_wunlock(&pvh_global_lock);
3082 	return (rv);
3083 }
3084 
3085 /*
3086  * Return whether or not the specified virtual address is eligible
3087  * for prefault.
3088  */
3089 static boolean_t
3090 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3091 {
3092 
3093 	return (FALSE);
3094 }
3095 
3096 /*
3097  * Return whether or not the specified physical page was referenced
3098  * in any physical maps.
3099  */
3100 static boolean_t
3101 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3102 {
3103 	pte_t *pte;
3104 	pv_entry_t pv;
3105 	boolean_t rv;
3106 
3107 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3108 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3109 	rv = FALSE;
3110 	rw_wlock(&pvh_global_lock);
3111 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3112 		PMAP_LOCK(pv->pv_pmap);
3113 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3114 		    PTE_ISVALID(pte)) {
3115 			if (PTE_ISREFERENCED(pte))
3116 				rv = TRUE;
3117 		}
3118 		PMAP_UNLOCK(pv->pv_pmap);
3119 		if (rv)
3120 			break;
3121 	}
3122 	rw_wunlock(&pvh_global_lock);
3123 	return (rv);
3124 }
3125 
3126 /*
3127  * Clear the modify bits on the specified physical page.
3128  */
3129 static void
3130 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3131 {
3132 	pte_t *pte;
3133 	pv_entry_t pv;
3134 
3135 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3136 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3137 	vm_page_assert_busied(m);
3138 
3139 	if (!pmap_page_is_write_mapped(m))
3140 	        return;
3141 
3142 	rw_wlock(&pvh_global_lock);
3143 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3144 		PMAP_LOCK(pv->pv_pmap);
3145 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3146 		    PTE_ISVALID(pte)) {
3147 			mtx_lock_spin(&tlbivax_mutex);
3148 			tlb_miss_lock();
3149 
3150 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3151 				tlb0_flush_entry(pv->pv_va);
3152 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3153 				    PTE_REFERENCED);
3154 			}
3155 
3156 			tlb_miss_unlock();
3157 			mtx_unlock_spin(&tlbivax_mutex);
3158 		}
3159 		PMAP_UNLOCK(pv->pv_pmap);
3160 	}
3161 	rw_wunlock(&pvh_global_lock);
3162 }
3163 
3164 /*
3165  * Return a count of reference bits for a page, clearing those bits.
3166  * It is not necessary for every reference bit to be cleared, but it
3167  * is necessary that 0 only be returned when there are truly no
3168  * reference bits set.
3169  *
3170  * As an optimization, update the page's dirty field if a modified bit is
3171  * found while counting reference bits.  This opportunistic update can be
3172  * performed at low cost and can eliminate the need for some future calls
3173  * to pmap_is_modified().  However, since this function stops after
3174  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3175  * dirty pages.  Those dirty pages will only be detected by a future call
3176  * to pmap_is_modified().
3177  */
3178 static int
3179 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3180 {
3181 	pte_t *pte;
3182 	pv_entry_t pv;
3183 	int count;
3184 
3185 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3186 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3187 	count = 0;
3188 	rw_wlock(&pvh_global_lock);
3189 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3190 		PMAP_LOCK(pv->pv_pmap);
3191 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3192 		    PTE_ISVALID(pte)) {
3193 			if (PTE_ISMODIFIED(pte))
3194 				vm_page_dirty(m);
3195 			if (PTE_ISREFERENCED(pte)) {
3196 				mtx_lock_spin(&tlbivax_mutex);
3197 				tlb_miss_lock();
3198 
3199 				tlb0_flush_entry(pv->pv_va);
3200 				*pte &= ~PTE_REFERENCED;
3201 
3202 				tlb_miss_unlock();
3203 				mtx_unlock_spin(&tlbivax_mutex);
3204 
3205 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3206 					PMAP_UNLOCK(pv->pv_pmap);
3207 					break;
3208 				}
3209 			}
3210 		}
3211 		PMAP_UNLOCK(pv->pv_pmap);
3212 	}
3213 	rw_wunlock(&pvh_global_lock);
3214 	return (count);
3215 }
3216 
3217 /*
3218  * Clear the wired attribute from the mappings for the specified range of
3219  * addresses in the given pmap.  Every valid mapping within that range must
3220  * have the wired attribute set.  In contrast, invalid mappings cannot have
3221  * the wired attribute set, so they are ignored.
3222  *
3223  * The wired attribute of the page table entry is not a hardware feature, so
3224  * there is no need to invalidate any TLB entries.
3225  */
3226 static void
3227 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3228 {
3229 	vm_offset_t va;
3230 	pte_t *pte;
3231 
3232 	PMAP_LOCK(pmap);
3233 	for (va = sva; va < eva; va += PAGE_SIZE) {
3234 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3235 		    PTE_ISVALID(pte)) {
3236 			if (!PTE_ISWIRED(pte))
3237 				panic("mmu_booke_unwire: pte %p isn't wired",
3238 				    pte);
3239 			*pte &= ~PTE_WIRED;
3240 			pmap->pm_stats.wired_count--;
3241 		}
3242 	}
3243 	PMAP_UNLOCK(pmap);
3244 
3245 }
3246 
3247 /*
3248  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3249  * page.  This count may be changed upwards or downwards in the future; it is
3250  * only necessary that true be returned for a small subset of pmaps for proper
3251  * page aging.
3252  */
3253 static boolean_t
3254 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3255 {
3256 	pv_entry_t pv;
3257 	int loops;
3258 	boolean_t rv;
3259 
3260 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3261 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3262 	loops = 0;
3263 	rv = FALSE;
3264 	rw_wlock(&pvh_global_lock);
3265 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3266 		if (pv->pv_pmap == pmap) {
3267 			rv = TRUE;
3268 			break;
3269 		}
3270 		if (++loops >= 16)
3271 			break;
3272 	}
3273 	rw_wunlock(&pvh_global_lock);
3274 	return (rv);
3275 }
3276 
3277 /*
3278  * Return the number of managed mappings to the given physical page that are
3279  * wired.
3280  */
3281 static int
3282 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3283 {
3284 	pv_entry_t pv;
3285 	pte_t *pte;
3286 	int count = 0;
3287 
3288 	if ((m->oflags & VPO_UNMANAGED) != 0)
3289 		return (count);
3290 	rw_wlock(&pvh_global_lock);
3291 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3292 		PMAP_LOCK(pv->pv_pmap);
3293 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3294 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3295 				count++;
3296 		PMAP_UNLOCK(pv->pv_pmap);
3297 	}
3298 	rw_wunlock(&pvh_global_lock);
3299 	return (count);
3300 }
3301 
3302 static int
3303 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3304 {
3305 	int i;
3306 	vm_offset_t va;
3307 
3308 	/*
3309 	 * This currently does not work for entries that
3310 	 * overlap TLB1 entries.
3311 	 */
3312 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3313 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3314 			return (0);
3315 	}
3316 
3317 	return (EFAULT);
3318 }
3319 
3320 void
3321 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3322 {
3323 	vm_paddr_t ppa;
3324 	vm_offset_t ofs;
3325 	vm_size_t gran;
3326 
3327 	/* Minidumps are based on virtual memory addresses. */
3328 	if (do_minidump) {
3329 		*va = (void *)(vm_offset_t)pa;
3330 		return;
3331 	}
3332 
3333 	/* Raw physical memory dumps don't have a virtual address. */
3334 	/* We always map a 256MB page at 256M. */
3335 	gran = 256 * 1024 * 1024;
3336 	ppa = rounddown2(pa, gran);
3337 	ofs = pa - ppa;
3338 	*va = (void *)gran;
3339 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3340 
3341 	if (sz > (gran - ofs))
3342 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3343 		    _TLB_ENTRY_IO);
3344 }
3345 
3346 void
3347 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3348 {
3349 	vm_paddr_t ppa;
3350 	vm_offset_t ofs;
3351 	vm_size_t gran;
3352 	tlb_entry_t e;
3353 	int i;
3354 
3355 	/* Minidumps are based on virtual memory addresses. */
3356 	/* Nothing to do... */
3357 	if (do_minidump)
3358 		return;
3359 
3360 	for (i = 0; i < TLB1_ENTRIES; i++) {
3361 		tlb1_read_entry(&e, i);
3362 		if (!(e.mas1 & MAS1_VALID))
3363 			break;
3364 	}
3365 
3366 	/* Raw physical memory dumps don't have a virtual address. */
3367 	i--;
3368 	e.mas1 = 0;
3369 	e.mas2 = 0;
3370 	e.mas3 = 0;
3371 	tlb1_write_entry(&e, i);
3372 
3373 	gran = 256 * 1024 * 1024;
3374 	ppa = rounddown2(pa, gran);
3375 	ofs = pa - ppa;
3376 	if (sz > (gran - ofs)) {
3377 		i--;
3378 		e.mas1 = 0;
3379 		e.mas2 = 0;
3380 		e.mas3 = 0;
3381 		tlb1_write_entry(&e, i);
3382 	}
3383 }
3384 
3385 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3386 
3387 void
3388 mmu_booke_scan_init(mmu_t mmu)
3389 {
3390 	vm_offset_t va;
3391 	pte_t *pte;
3392 	int i;
3393 
3394 	if (!do_minidump) {
3395 		/* Initialize phys. segments for dumpsys(). */
3396 		memset(&dump_map, 0, sizeof(dump_map));
3397 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3398 		    &availmem_regions_sz);
3399 		for (i = 0; i < physmem_regions_sz; i++) {
3400 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3401 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3402 		}
3403 		return;
3404 	}
3405 
3406 	/* Virtual segments for minidumps: */
3407 	memset(&dump_map, 0, sizeof(dump_map));
3408 
3409 	/* 1st: kernel .data and .bss. */
3410 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3411 	dump_map[0].pa_size =
3412 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3413 
3414 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3415 	dump_map[1].pa_start = data_start;
3416 	dump_map[1].pa_size = data_end - data_start;
3417 
3418 	/* 3rd: kernel VM. */
3419 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3420 	/* Find start of next chunk (from va). */
3421 	while (va < virtual_end) {
3422 		/* Don't dump the buffer cache. */
3423 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3424 			va = kmi.buffer_eva;
3425 			continue;
3426 		}
3427 		pte = pte_find(mmu, kernel_pmap, va);
3428 		if (pte != NULL && PTE_ISVALID(pte))
3429 			break;
3430 		va += PAGE_SIZE;
3431 	}
3432 	if (va < virtual_end) {
3433 		dump_map[2].pa_start = va;
3434 		va += PAGE_SIZE;
3435 		/* Find last page in chunk. */
3436 		while (va < virtual_end) {
3437 			/* Don't run into the buffer cache. */
3438 			if (va == kmi.buffer_sva)
3439 				break;
3440 			pte = pte_find(mmu, kernel_pmap, va);
3441 			if (pte == NULL || !PTE_ISVALID(pte))
3442 				break;
3443 			va += PAGE_SIZE;
3444 		}
3445 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3446 	}
3447 }
3448 
3449 /*
3450  * Map a set of physical memory pages into the kernel virtual address space.
3451  * Return a pointer to where it is mapped. This routine is intended to be used
3452  * for mapping device memory, NOT real memory.
3453  */
3454 static void *
3455 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3456 {
3457 
3458 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3459 }
3460 
3461 static int
3462 tlb1_find_pa(vm_paddr_t pa, tlb_entry_t *e)
3463 {
3464 	int i;
3465 
3466 	for (i = 0; i < TLB1_ENTRIES; i++) {
3467 		tlb1_read_entry(e, i);
3468 		if ((e->mas1 & MAS1_VALID) == 0)
3469 			return (i);
3470 	}
3471 	return (-1);
3472 }
3473 
3474 static void *
3475 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3476 {
3477 	tlb_entry_t e;
3478 	vm_paddr_t tmppa;
3479 	void *res;
3480 	uintptr_t va, tmpva;
3481 	vm_size_t sz;
3482 	int i;
3483 	int wimge;
3484 
3485 	/*
3486 	 * Check if this is premapped in TLB1.
3487 	 */
3488 	sz = size;
3489 	tmppa = pa;
3490 	va = ~0;
3491 	wimge = tlb_calc_wimg(pa, ma);
3492 	for (i = 0; i < TLB1_ENTRIES; i++) {
3493 		tlb1_read_entry(&e, i);
3494 		if (!(e.mas1 & MAS1_VALID))
3495 			continue;
3496 		if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
3497 			continue;
3498 		if (tmppa >= e.phys && tmppa < e.phys + e.size) {
3499 			va = e.virt + (pa - e.phys);
3500 			tmppa = e.phys + e.size;
3501 			sz -= MIN(sz, e.size);
3502 			while (sz > 0 && (i = tlb1_find_pa(tmppa, &e)) != -1) {
3503 				if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
3504 					break;
3505 				sz -= MIN(sz, e.size);
3506 				tmppa = e.phys + e.size;
3507 			}
3508 			if (sz != 0)
3509 				break;
3510 			return ((void *)va);
3511 		}
3512 	}
3513 
3514 	size = roundup(size, PAGE_SIZE);
3515 
3516 	/*
3517 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3518 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3519 	 */
3520 #ifdef SPARSE_MAPDEV
3521 	/*
3522 	 * With a sparse mapdev, align to the largest starting region.  This
3523 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3524 	 * calculation could be very costly.
3525 	 * Align to the smaller of:
3526 	 * - first set bit in overlap of (pa & size mask)
3527 	 * - largest size envelope
3528 	 *
3529 	 * It's possible the device mapping may start at a PA that's not larger
3530 	 * than the size mask, so we need to offset in to maximize the TLB entry
3531 	 * range and minimize the number of used TLB entries.
3532 	 */
3533 	do {
3534 	    tmpva = tlb1_map_base;
3535 	    sz = ffsl((~((1 << flsl(size-1)) - 1)) & pa);
3536 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3537 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3538 #ifdef __powerpc64__
3539 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3540 #else
3541 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3542 #endif
3543 #else
3544 #ifdef __powerpc64__
3545 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3546 #else
3547 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3548 #endif
3549 #endif
3550 	res = (void *)va;
3551 
3552 	if (tlb1_mapin_region(va, pa, size, tlb_calc_wimg(pa, ma)) != size)
3553 		return (NULL);
3554 
3555 	return (res);
3556 }
3557 
3558 /*
3559  * 'Unmap' a range mapped by mmu_booke_mapdev().
3560  */
3561 static void
3562 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3563 {
3564 #ifdef SUPPORTS_SHRINKING_TLB1
3565 	vm_offset_t base, offset;
3566 
3567 	/*
3568 	 * Unmap only if this is inside kernel virtual space.
3569 	 */
3570 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3571 		base = trunc_page(va);
3572 		offset = va & PAGE_MASK;
3573 		size = roundup(offset + size, PAGE_SIZE);
3574 		kva_free(base, size);
3575 	}
3576 #endif
3577 }
3578 
3579 /*
3580  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3581  * specified pmap. This eliminates the blast of soft faults on process startup
3582  * and immediately after an mmap.
3583  */
3584 static void
3585 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3586     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3587 {
3588 
3589 	VM_OBJECT_ASSERT_WLOCKED(object);
3590 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3591 	    ("mmu_booke_object_init_pt: non-device object"));
3592 }
3593 
3594 /*
3595  * Perform the pmap work for mincore.
3596  */
3597 static int
3598 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3599     vm_paddr_t *pap)
3600 {
3601 
3602 	/* XXX: this should be implemented at some point */
3603 	return (0);
3604 }
3605 
3606 static int
3607 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3608     vm_memattr_t mode)
3609 {
3610 	vm_offset_t va;
3611 	pte_t *pte;
3612 	int i, j;
3613 	tlb_entry_t e;
3614 
3615 	addr = trunc_page(addr);
3616 
3617 	/* Only allow changes to mapped kernel addresses.  This includes:
3618 	 * - KVA
3619 	 * - DMAP (powerpc64)
3620 	 * - Device mappings
3621 	 */
3622 	if (addr <= VM_MAXUSER_ADDRESS ||
3623 #ifdef __powerpc64__
3624 	    (addr >= tlb1_map_base && addr < DMAP_BASE_ADDRESS) ||
3625 	    (addr > DMAP_MAX_ADDRESS && addr < VM_MIN_KERNEL_ADDRESS) ||
3626 #else
3627 	    (addr >= tlb1_map_base && addr < VM_MIN_KERNEL_ADDRESS) ||
3628 #endif
3629 	    (addr > VM_MAX_KERNEL_ADDRESS))
3630 		return (EINVAL);
3631 
3632 	/* Check TLB1 mappings */
3633 	for (i = 0; i < TLB1_ENTRIES; i++) {
3634 		tlb1_read_entry(&e, i);
3635 		if (!(e.mas1 & MAS1_VALID))
3636 			continue;
3637 		if (addr >= e.virt && addr < e.virt + e.size)
3638 			break;
3639 	}
3640 	if (i < TLB1_ENTRIES) {
3641 		/* Only allow full mappings to be modified for now. */
3642 		/* Validate the range. */
3643 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3644 			tlb1_read_entry(&e, j);
3645 			if (va != e.virt || (sz - (va - addr) < e.size))
3646 				return (EINVAL);
3647 		}
3648 		for (va = addr; va < addr + sz; va += e.size, i++) {
3649 			tlb1_read_entry(&e, i);
3650 			e.mas2 &= ~MAS2_WIMGE_MASK;
3651 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3652 
3653 			/*
3654 			 * Write it out to the TLB.  Should really re-sync with other
3655 			 * cores.
3656 			 */
3657 			tlb1_write_entry(&e, i);
3658 		}
3659 		return (0);
3660 	}
3661 
3662 	/* Not in TLB1, try through pmap */
3663 	/* First validate the range. */
3664 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3665 		pte = pte_find(mmu, kernel_pmap, va);
3666 		if (pte == NULL || !PTE_ISVALID(pte))
3667 			return (EINVAL);
3668 	}
3669 
3670 	mtx_lock_spin(&tlbivax_mutex);
3671 	tlb_miss_lock();
3672 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3673 		pte = pte_find(mmu, kernel_pmap, va);
3674 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3675 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3676 		tlb0_flush_entry(va);
3677 	}
3678 	tlb_miss_unlock();
3679 	mtx_unlock_spin(&tlbivax_mutex);
3680 
3681 	return (0);
3682 }
3683 
3684 static void
3685 mmu_booke_page_array_startup(mmu_t mmu, long pages)
3686 {
3687 	vm_page_array_size = pages;
3688 }
3689 
3690 /**************************************************************************/
3691 /* TID handling */
3692 /**************************************************************************/
3693 
3694 /*
3695  * Allocate a TID. If necessary, steal one from someone else.
3696  * The new TID is flushed from the TLB before returning.
3697  */
3698 static tlbtid_t
3699 tid_alloc(pmap_t pmap)
3700 {
3701 	tlbtid_t tid;
3702 	int thiscpu;
3703 
3704 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3705 
3706 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3707 
3708 	thiscpu = PCPU_GET(cpuid);
3709 
3710 	tid = PCPU_GET(booke.tid_next);
3711 	if (tid > TID_MAX)
3712 		tid = TID_MIN;
3713 	PCPU_SET(booke.tid_next, tid + 1);
3714 
3715 	/* If we are stealing TID then clear the relevant pmap's field */
3716 	if (tidbusy[thiscpu][tid] != NULL) {
3717 
3718 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3719 
3720 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3721 
3722 		/* Flush all entries from TLB0 matching this TID. */
3723 		tid_flush(tid);
3724 	}
3725 
3726 	tidbusy[thiscpu][tid] = pmap;
3727 	pmap->pm_tid[thiscpu] = tid;
3728 	__asm __volatile("msync; isync");
3729 
3730 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3731 	    PCPU_GET(booke.tid_next));
3732 
3733 	return (tid);
3734 }
3735 
3736 /**************************************************************************/
3737 /* TLB0 handling */
3738 /**************************************************************************/
3739 
3740 /* Convert TLB0 va and way number to tlb0[] table index. */
3741 static inline unsigned int
3742 tlb0_tableidx(vm_offset_t va, unsigned int way)
3743 {
3744 	unsigned int idx;
3745 
3746 	idx = (way * TLB0_ENTRIES_PER_WAY);
3747 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3748 	return (idx);
3749 }
3750 
3751 /*
3752  * Invalidate TLB0 entry.
3753  */
3754 static inline void
3755 tlb0_flush_entry(vm_offset_t va)
3756 {
3757 
3758 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3759 
3760 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3761 
3762 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3763 	__asm __volatile("isync; msync");
3764 	__asm __volatile("tlbsync; msync");
3765 
3766 	CTR1(KTR_PMAP, "%s: e", __func__);
3767 }
3768 
3769 
3770 /**************************************************************************/
3771 /* TLB1 handling */
3772 /**************************************************************************/
3773 
3774 /*
3775  * TLB1 mapping notes:
3776  *
3777  * TLB1[0]	Kernel text and data.
3778  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3779  *		windows, other devices mappings.
3780  */
3781 
3782  /*
3783  * Read an entry from given TLB1 slot.
3784  */
3785 void
3786 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3787 {
3788 	register_t msr;
3789 	uint32_t mas0;
3790 
3791 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3792 
3793 	msr = mfmsr();
3794 	__asm __volatile("wrteei 0");
3795 
3796 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3797 	mtspr(SPR_MAS0, mas0);
3798 	__asm __volatile("isync; tlbre");
3799 
3800 	entry->mas1 = mfspr(SPR_MAS1);
3801 	entry->mas2 = mfspr(SPR_MAS2);
3802 	entry->mas3 = mfspr(SPR_MAS3);
3803 
3804 	switch ((mfpvr() >> 16) & 0xFFFF) {
3805 	case FSL_E500v2:
3806 	case FSL_E500mc:
3807 	case FSL_E5500:
3808 	case FSL_E6500:
3809 		entry->mas7 = mfspr(SPR_MAS7);
3810 		break;
3811 	default:
3812 		entry->mas7 = 0;
3813 		break;
3814 	}
3815 	__asm __volatile("wrtee %0" :: "r"(msr));
3816 
3817 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3818 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3819 	    (entry->mas3 & MAS3_RPN);
3820 	entry->size =
3821 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3822 }
3823 
3824 struct tlbwrite_args {
3825 	tlb_entry_t *e;
3826 	unsigned int idx;
3827 };
3828 
3829 static uint32_t
3830 tlb1_find_free(void)
3831 {
3832 	tlb_entry_t e;
3833 	int i;
3834 
3835 	for (i = 0; i < TLB1_ENTRIES; i++) {
3836 		tlb1_read_entry(&e, i);
3837 		if ((e.mas1 & MAS1_VALID) == 0)
3838 			return (i);
3839 	}
3840 	return (-1);
3841 }
3842 
3843 static void
3844 tlb1_write_entry_int(void *arg)
3845 {
3846 	struct tlbwrite_args *args = arg;
3847 	uint32_t idx, mas0;
3848 
3849 	idx = args->idx;
3850 	if (idx == -1) {
3851 		idx = tlb1_find_free();
3852 		if (idx == -1)
3853 			panic("No free TLB1 entries!\n");
3854 	}
3855 	/* Select entry */
3856 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
3857 
3858 	mtspr(SPR_MAS0, mas0);
3859 	mtspr(SPR_MAS1, args->e->mas1);
3860 	mtspr(SPR_MAS2, args->e->mas2);
3861 	mtspr(SPR_MAS3, args->e->mas3);
3862 	switch ((mfpvr() >> 16) & 0xFFFF) {
3863 	case FSL_E500mc:
3864 	case FSL_E5500:
3865 	case FSL_E6500:
3866 		mtspr(SPR_MAS8, 0);
3867 		/* FALLTHROUGH */
3868 	case FSL_E500v2:
3869 		mtspr(SPR_MAS7, args->e->mas7);
3870 		break;
3871 	default:
3872 		break;
3873 	}
3874 
3875 	__asm __volatile("isync; tlbwe; isync; msync");
3876 
3877 }
3878 
3879 static void
3880 tlb1_write_entry_sync(void *arg)
3881 {
3882 	/* Empty synchronization point for smp_rendezvous(). */
3883 }
3884 
3885 /*
3886  * Write given entry to TLB1 hardware.
3887  */
3888 static void
3889 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3890 {
3891 	struct tlbwrite_args args;
3892 
3893 	args.e = e;
3894 	args.idx = idx;
3895 
3896 #ifdef SMP
3897 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3898 		mb();
3899 		smp_rendezvous(tlb1_write_entry_sync,
3900 		    tlb1_write_entry_int,
3901 		    tlb1_write_entry_sync, &args);
3902 	} else
3903 #endif
3904 	{
3905 		register_t msr;
3906 
3907 		msr = mfmsr();
3908 		__asm __volatile("wrteei 0");
3909 		tlb1_write_entry_int(&args);
3910 		__asm __volatile("wrtee %0" :: "r"(msr));
3911 	}
3912 }
3913 
3914 /*
3915  * Return the largest uint value log such that 2^log <= num.
3916  */
3917 static unsigned long
3918 ilog2(unsigned long num)
3919 {
3920 	long lz;
3921 
3922 #ifdef __powerpc64__
3923 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
3924 	return (63 - lz);
3925 #else
3926 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3927 	return (31 - lz);
3928 #endif
3929 }
3930 
3931 /*
3932  * Convert TLB TSIZE value to mapped region size.
3933  */
3934 static vm_size_t
3935 tsize2size(unsigned int tsize)
3936 {
3937 
3938 	/*
3939 	 * size = 4^tsize KB
3940 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3941 	 */
3942 
3943 	return ((1 << (2 * tsize)) * 1024);
3944 }
3945 
3946 /*
3947  * Convert region size (must be power of 4) to TLB TSIZE value.
3948  */
3949 static unsigned int
3950 size2tsize(vm_size_t size)
3951 {
3952 
3953 	return (ilog2(size) / 2 - 5);
3954 }
3955 
3956 /*
3957  * Register permanent kernel mapping in TLB1.
3958  *
3959  * Entries are created starting from index 0 (current free entry is
3960  * kept in tlb1_idx) and are not supposed to be invalidated.
3961  */
3962 int
3963 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3964     uint32_t flags)
3965 {
3966 	tlb_entry_t e;
3967 	uint32_t ts, tid;
3968 	int tsize, index;
3969 
3970 	/* First try to update an existing entry. */
3971 	for (index = 0; index < TLB1_ENTRIES; index++) {
3972 		tlb1_read_entry(&e, index);
3973 		/* Check if we're just updating the flags, and update them. */
3974 		if (e.phys == pa && e.virt == va && e.size == size) {
3975 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
3976 			tlb1_write_entry(&e, index);
3977 			return (0);
3978 		}
3979 	}
3980 
3981 	/* Convert size to TSIZE */
3982 	tsize = size2tsize(size);
3983 
3984 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3985 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3986 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3987 
3988 	e.phys = pa;
3989 	e.virt = va;
3990 	e.size = size;
3991 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3992 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3993 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
3994 
3995 	/* Set supervisor RWX permission bits */
3996 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3997 	e.mas7 = (pa >> 32) & MAS7_RPN;
3998 
3999 	tlb1_write_entry(&e, -1);
4000 
4001 	return (0);
4002 }
4003 
4004 /*
4005  * Map in contiguous RAM region into the TLB1.
4006  */
4007 static vm_size_t
4008 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size, int wimge)
4009 {
4010 	vm_offset_t base;
4011 	vm_size_t mapped, sz, ssize;
4012 
4013 	mapped = 0;
4014 	base = va;
4015 	ssize = size;
4016 
4017 	while (size > 0) {
4018 		sz = 1UL << (ilog2(size) & ~1);
4019 		/* Align size to PA */
4020 		if (pa % sz != 0) {
4021 			do {
4022 				sz >>= 2;
4023 			} while (pa % sz != 0);
4024 		}
4025 		/* Now align from there to VA */
4026 		if (va % sz != 0) {
4027 			do {
4028 				sz >>= 2;
4029 			} while (va % sz != 0);
4030 		}
4031 #ifdef __powerpc64__
4032 		/*
4033 		 * Clamp TLB1 entries to 4G.
4034 		 *
4035 		 * While the e6500 supports up to 1TB mappings, the e5500
4036 		 * only supports up to 4G mappings. (0b1011)
4037 		 *
4038 		 * If any e6500 machines capable of supporting a very
4039 		 * large amount of memory appear in the future, we can
4040 		 * revisit this.
4041 		 *
4042 		 * For now, though, since we have plenty of space in TLB1,
4043 		 * always avoid creating entries larger than 4GB.
4044 		 */
4045 		sz = MIN(sz, 1UL << 32);
4046 #endif
4047 		if (bootverbose)
4048 			printf("Wiring VA=%p to PA=%jx (size=%lx)\n",
4049 			    (void *)va, (uintmax_t)pa, (long)sz);
4050 		if (tlb1_set_entry(va, pa, sz,
4051 		    _TLB_ENTRY_SHARED | wimge) < 0)
4052 			return (mapped);
4053 		size -= sz;
4054 		pa += sz;
4055 		va += sz;
4056 	}
4057 
4058 	mapped = (va - base);
4059 	if (bootverbose)
4060 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
4061 		    mapped, mapped - ssize);
4062 
4063 	return (mapped);
4064 }
4065 
4066 /*
4067  * TLB1 initialization routine, to be called after the very first
4068  * assembler level setup done in locore.S.
4069  */
4070 void
4071 tlb1_init()
4072 {
4073 	vm_offset_t mas2;
4074 	uint32_t mas0, mas1, mas3, mas7;
4075 	uint32_t tsz;
4076 
4077 	tlb1_get_tlbconf();
4078 
4079 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4080 	mtspr(SPR_MAS0, mas0);
4081 	__asm __volatile("isync; tlbre");
4082 
4083 	mas1 = mfspr(SPR_MAS1);
4084 	mas2 = mfspr(SPR_MAS2);
4085 	mas3 = mfspr(SPR_MAS3);
4086 	mas7 = mfspr(SPR_MAS7);
4087 
4088 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4089 	    (mas3 & MAS3_RPN);
4090 
4091 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4092 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4093 	kernstart = trunc_page(mas2);
4094 
4095 	/* Setup TLB miss defaults */
4096 	set_mas4_defaults();
4097 }
4098 
4099 /*
4100  * pmap_early_io_unmap() should be used in short conjunction with
4101  * pmap_early_io_map(), as in the following snippet:
4102  *
4103  * x = pmap_early_io_map(...);
4104  * <do something with x>
4105  * pmap_early_io_unmap(x, size);
4106  *
4107  * And avoiding more allocations between.
4108  */
4109 void
4110 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4111 {
4112 	int i;
4113 	tlb_entry_t e;
4114 	vm_size_t isize;
4115 
4116 	size = roundup(size, PAGE_SIZE);
4117 	isize = size;
4118 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4119 		tlb1_read_entry(&e, i);
4120 		if (!(e.mas1 & MAS1_VALID))
4121 			continue;
4122 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4123 			size -= e.size;
4124 			e.mas1 &= ~MAS1_VALID;
4125 			tlb1_write_entry(&e, i);
4126 		}
4127 	}
4128 	if (tlb1_map_base == va + isize)
4129 		tlb1_map_base -= isize;
4130 }
4131 
4132 vm_offset_t
4133 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4134 {
4135 	vm_paddr_t pa_base;
4136 	vm_offset_t va, sz;
4137 	int i;
4138 	tlb_entry_t e;
4139 
4140 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4141 
4142 	for (i = 0; i < TLB1_ENTRIES; i++) {
4143 		tlb1_read_entry(&e, i);
4144 		if (!(e.mas1 & MAS1_VALID))
4145 			continue;
4146 		if (pa >= e.phys && (pa + size) <=
4147 		    (e.phys + e.size))
4148 			return (e.virt + (pa - e.phys));
4149 	}
4150 
4151 	pa_base = rounddown(pa, PAGE_SIZE);
4152 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4153 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4154 	va = tlb1_map_base + (pa - pa_base);
4155 
4156 	do {
4157 		sz = 1 << (ilog2(size) & ~1);
4158 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4159 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4160 		size -= sz;
4161 		pa_base += sz;
4162 		tlb1_map_base += sz;
4163 	} while (size > 0);
4164 
4165 	return (va);
4166 }
4167 
4168 void
4169 pmap_track_page(pmap_t pmap, vm_offset_t va)
4170 {
4171 	vm_paddr_t pa;
4172 	vm_page_t page;
4173 	struct pv_entry *pve;
4174 
4175 	va = trunc_page(va);
4176 	pa = pmap_kextract(va);
4177 	page = PHYS_TO_VM_PAGE(pa);
4178 
4179 	rw_wlock(&pvh_global_lock);
4180 	PMAP_LOCK(pmap);
4181 
4182 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4183 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4184 			goto out;
4185 		}
4186 	}
4187 	page->md.pv_tracked = true;
4188 	pv_insert(pmap, va, page);
4189 out:
4190 	PMAP_UNLOCK(pmap);
4191 	rw_wunlock(&pvh_global_lock);
4192 }
4193 
4194 
4195 /*
4196  * Setup MAS4 defaults.
4197  * These values are loaded to MAS0-2 on a TLB miss.
4198  */
4199 static void
4200 set_mas4_defaults(void)
4201 {
4202 	uint32_t mas4;
4203 
4204 	/* Defaults: TLB0, PID0, TSIZED=4K */
4205 	mas4 = MAS4_TLBSELD0;
4206 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4207 #ifdef SMP
4208 	mas4 |= MAS4_MD;
4209 #endif
4210 	mtspr(SPR_MAS4, mas4);
4211 	__asm __volatile("isync");
4212 }
4213 
4214 
4215 /*
4216  * Return 0 if the physical IO range is encompassed by one of the
4217  * the TLB1 entries, otherwise return related error code.
4218  */
4219 static int
4220 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4221 {
4222 	uint32_t prot;
4223 	vm_paddr_t pa_start;
4224 	vm_paddr_t pa_end;
4225 	unsigned int entry_tsize;
4226 	vm_size_t entry_size;
4227 	tlb_entry_t e;
4228 
4229 	*va = (vm_offset_t)NULL;
4230 
4231 	tlb1_read_entry(&e, i);
4232 	/* Skip invalid entries */
4233 	if (!(e.mas1 & MAS1_VALID))
4234 		return (EINVAL);
4235 
4236 	/*
4237 	 * The entry must be cache-inhibited, guarded, and r/w
4238 	 * so it can function as an i/o page
4239 	 */
4240 	prot = e.mas2 & (MAS2_I | MAS2_G);
4241 	if (prot != (MAS2_I | MAS2_G))
4242 		return (EPERM);
4243 
4244 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4245 	if (prot != (MAS3_SR | MAS3_SW))
4246 		return (EPERM);
4247 
4248 	/* The address should be within the entry range. */
4249 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4250 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4251 
4252 	entry_size = tsize2size(entry_tsize);
4253 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4254 	    (e.mas3 & MAS3_RPN);
4255 	pa_end = pa_start + entry_size;
4256 
4257 	if ((pa < pa_start) || ((pa + size) > pa_end))
4258 		return (ERANGE);
4259 
4260 	/* Return virtual address of this mapping. */
4261 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4262 	return (0);
4263 }
4264 
4265 /*
4266  * Invalidate all TLB0 entries which match the given TID. Note this is
4267  * dedicated for cases when invalidations should NOT be propagated to other
4268  * CPUs.
4269  */
4270 static void
4271 tid_flush(tlbtid_t tid)
4272 {
4273 	register_t msr;
4274 	uint32_t mas0, mas1, mas2;
4275 	int entry, way;
4276 
4277 
4278 	/* Don't evict kernel translations */
4279 	if (tid == TID_KERNEL)
4280 		return;
4281 
4282 	msr = mfmsr();
4283 	__asm __volatile("wrteei 0");
4284 
4285 	/*
4286 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
4287 	 * it for PID invalidation.
4288 	 */
4289 	switch ((mfpvr() >> 16) & 0xffff) {
4290 	case FSL_E500mc:
4291 	case FSL_E5500:
4292 	case FSL_E6500:
4293 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
4294 		/* tlbilxpid */
4295 		__asm __volatile("isync; .long 0x7c200024; isync; msync");
4296 		__asm __volatile("wrtee %0" :: "r"(msr));
4297 		return;
4298 	}
4299 
4300 	for (way = 0; way < TLB0_WAYS; way++)
4301 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4302 
4303 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4304 			mtspr(SPR_MAS0, mas0);
4305 
4306 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4307 			mtspr(SPR_MAS2, mas2);
4308 
4309 			__asm __volatile("isync; tlbre");
4310 
4311 			mas1 = mfspr(SPR_MAS1);
4312 
4313 			if (!(mas1 & MAS1_VALID))
4314 				continue;
4315 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4316 				continue;
4317 			mas1 &= ~MAS1_VALID;
4318 			mtspr(SPR_MAS1, mas1);
4319 			__asm __volatile("isync; tlbwe; isync; msync");
4320 		}
4321 	__asm __volatile("wrtee %0" :: "r"(msr));
4322 }
4323 
4324 #ifdef DDB
4325 /* Print out contents of the MAS registers for each TLB0 entry */
4326 static void
4327 #ifdef __powerpc64__
4328 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
4329 #else
4330 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
4331 #endif
4332     uint32_t mas7)
4333 {
4334 	int as;
4335 	char desc[3];
4336 	tlbtid_t tid;
4337 	vm_size_t size;
4338 	unsigned int tsize;
4339 
4340 	desc[2] = '\0';
4341 	if (mas1 & MAS1_VALID)
4342 		desc[0] = 'V';
4343 	else
4344 		desc[0] = ' ';
4345 
4346 	if (mas1 & MAS1_IPROT)
4347 		desc[1] = 'P';
4348 	else
4349 		desc[1] = ' ';
4350 
4351 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
4352 	tid = MAS1_GETTID(mas1);
4353 
4354 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4355 	size = 0;
4356 	if (tsize)
4357 		size = tsize2size(tsize);
4358 
4359 	printf("%3d: (%s) [AS=%d] "
4360 	    "sz = 0x%jx tsz = %d tid = %d mas1 = 0x%08x "
4361 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
4362 	    i, desc, as, (uintmax_t)size, tsize, tid, mas1, mas2, mas3, mas7);
4363 }
4364 
4365 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
4366 {
4367 	uint32_t mas0, mas1, mas3, mas7;
4368 #ifdef __powerpc64__
4369 	uint64_t mas2;
4370 #else
4371 	uint32_t mas2;
4372 #endif
4373 	int entryidx, way, idx;
4374 
4375 	printf("TLB0 entries:\n");
4376 	for (way = 0; way < TLB0_WAYS; way ++)
4377 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
4378 
4379 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4380 			mtspr(SPR_MAS0, mas0);
4381 
4382 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
4383 			mtspr(SPR_MAS2, mas2);
4384 
4385 			__asm __volatile("isync; tlbre");
4386 
4387 			mas1 = mfspr(SPR_MAS1);
4388 			mas2 = mfspr(SPR_MAS2);
4389 			mas3 = mfspr(SPR_MAS3);
4390 			mas7 = mfspr(SPR_MAS7);
4391 
4392 			idx = tlb0_tableidx(mas2, way);
4393 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
4394 		}
4395 }
4396 
4397 /*
4398  * Print out contents of the MAS registers for each TLB1 entry
4399  */
4400 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
4401 {
4402 	uint32_t mas0, mas1, mas3, mas7;
4403 #ifdef __powerpc64__
4404 	uint64_t mas2;
4405 #else
4406 	uint32_t mas2;
4407 #endif
4408 	int i;
4409 
4410 	printf("TLB1 entries:\n");
4411 	for (i = 0; i < TLB1_ENTRIES; i++) {
4412 
4413 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4414 		mtspr(SPR_MAS0, mas0);
4415 
4416 		__asm __volatile("isync; tlbre");
4417 
4418 		mas1 = mfspr(SPR_MAS1);
4419 		mas2 = mfspr(SPR_MAS2);
4420 		mas3 = mfspr(SPR_MAS3);
4421 		mas7 = mfspr(SPR_MAS7);
4422 
4423 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4424 	}
4425 }
4426 #endif
4427