xref: /freebsd/sys/powerpc/booke/pmap.c (revision 60fde7ce5d7bf5d94290720ea53db5701ab406a8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - ptbl_bufs-1              : message buffer
63   *               ptbl_bufs - kernel_pdir-1            : kernel page tables
64   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
65   *             kernel_pp2d - .                        : kernel pointers to page directory
66   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
67   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
68   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
69   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
70   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
71   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
72   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
73   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
74   *                         - 0xffff_ffff_ffff_ffff    : device direct map
75   */
76 
77 #include <sys/cdefs.h>
78 __FBSDID("$FreeBSD$");
79 
80 #include "opt_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 "mmu_if.h"
127 
128 #define	SPARSE_MAPDEV
129 #ifdef  DEBUG
130 #define debugf(fmt, args...) printf(fmt, ##args)
131 #else
132 #define debugf(fmt, args...)
133 #endif
134 
135 #ifdef __powerpc64__
136 #define	PRI0ptrX	"016lx"
137 #else
138 #define	PRI0ptrX	"08x"
139 #endif
140 
141 #define TODO			panic("%s: not implemented", __func__);
142 
143 extern unsigned char _etext[];
144 extern unsigned char _end[];
145 
146 extern uint32_t *bootinfo;
147 
148 vm_paddr_t kernload;
149 vm_offset_t kernstart;
150 vm_size_t kernsize;
151 
152 /* Message buffer and tables. */
153 static vm_offset_t data_start;
154 static vm_size_t data_end;
155 
156 /* Phys/avail memory regions. */
157 static struct mem_region *availmem_regions;
158 static int availmem_regions_sz;
159 static struct mem_region *physmem_regions;
160 static int physmem_regions_sz;
161 
162 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
163 static vm_offset_t zero_page_va;
164 static struct mtx zero_page_mutex;
165 
166 static struct mtx tlbivax_mutex;
167 
168 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
169 static vm_offset_t copy_page_src_va;
170 static vm_offset_t copy_page_dst_va;
171 static struct mtx copy_page_mutex;
172 
173 /**************************************************************************/
174 /* PMAP */
175 /**************************************************************************/
176 
177 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
178     vm_prot_t, u_int flags, int8_t psind);
179 
180 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
181 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
182 #ifdef __powerpc64__
183 unsigned int kernel_pdirs;
184 #endif
185 
186 /*
187  * If user pmap is processed with mmu_booke_remove and the resident count
188  * drops to 0, there are no more pages to remove, so we need not continue.
189  */
190 #define PMAP_REMOVE_DONE(pmap) \
191 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
192 
193 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
194 extern int elf32_nxstack;
195 #endif
196 
197 /**************************************************************************/
198 /* TLB and TID handling */
199 /**************************************************************************/
200 
201 /* Translation ID busy table */
202 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
203 
204 /*
205  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
206  * core revisions and should be read from h/w registers during early config.
207  */
208 uint32_t tlb0_entries;
209 uint32_t tlb0_ways;
210 uint32_t tlb0_entries_per_way;
211 uint32_t tlb1_entries;
212 
213 #define TLB0_ENTRIES		(tlb0_entries)
214 #define TLB0_WAYS		(tlb0_ways)
215 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
216 
217 #define TLB1_ENTRIES (tlb1_entries)
218 
219 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
220 
221 static tlbtid_t tid_alloc(struct pmap *);
222 static void tid_flush(tlbtid_t tid);
223 
224 #ifdef __powerpc64__
225 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
226 #else
227 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
228 #endif
229 
230 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
231 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
232 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
233 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
234 
235 static vm_size_t tsize2size(unsigned int);
236 static unsigned int size2tsize(vm_size_t);
237 static unsigned int ilog2(unsigned int);
238 
239 static void set_mas4_defaults(void);
240 
241 static inline void tlb0_flush_entry(vm_offset_t);
242 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
243 
244 /**************************************************************************/
245 /* Page table management */
246 /**************************************************************************/
247 
248 static struct rwlock_padalign pvh_global_lock;
249 
250 /* Data for the pv entry allocation mechanism */
251 static uma_zone_t pvzone;
252 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
253 
254 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
255 
256 #ifndef PMAP_SHPGPERPROC
257 #define PMAP_SHPGPERPROC	200
258 #endif
259 
260 static void ptbl_init(void);
261 static struct ptbl_buf *ptbl_buf_alloc(void);
262 static void ptbl_buf_free(struct ptbl_buf *);
263 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
264 
265 #ifdef __powerpc64__
266 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
267 			 unsigned int, boolean_t);
268 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int);
269 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
270 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
271 #else
272 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
273 static void ptbl_free(mmu_t, pmap_t, unsigned int);
274 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
275 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
276 #endif
277 
278 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
279 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
280 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
281 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
282 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
283 
284 static pv_entry_t pv_alloc(void);
285 static void pv_free(pv_entry_t);
286 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
287 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
288 
289 static void booke_pmap_init_qpages(void);
290 
291 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
292 #ifdef __powerpc64__
293 #define PTBL_BUFS               (16UL * 16 * 16)
294 #else
295 #define PTBL_BUFS		(128 * 16)
296 #endif
297 
298 struct ptbl_buf {
299 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
300 	vm_offset_t kva;		/* va of mapping */
301 };
302 
303 /* ptbl free list and a lock used for access synchronization. */
304 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
305 static struct mtx ptbl_buf_freelist_lock;
306 
307 /* Base address of kva space allocated fot ptbl bufs. */
308 static vm_offset_t ptbl_buf_pool_vabase;
309 
310 /* Pointer to ptbl_buf structures. */
311 static struct ptbl_buf *ptbl_bufs;
312 
313 #ifdef SMP
314 extern tlb_entry_t __boot_tlb1[];
315 void pmap_bootstrap_ap(volatile uint32_t *);
316 #endif
317 
318 /*
319  * Kernel MMU interface
320  */
321 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
322 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
323     vm_size_t, vm_offset_t);
324 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
325 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
326     vm_offset_t, vm_page_t *, vm_offset_t, int);
327 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
328     vm_prot_t, u_int flags, int8_t psind);
329 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
330     vm_page_t, vm_prot_t);
331 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
332     vm_prot_t);
333 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
334 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
335     vm_prot_t);
336 static void		mmu_booke_init(mmu_t);
337 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
338 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
339 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
340 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
341 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
342     int);
343 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
344     vm_paddr_t *);
345 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
346     vm_object_t, vm_pindex_t, vm_size_t);
347 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
348 static void		mmu_booke_page_init(mmu_t, vm_page_t);
349 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
350 static void		mmu_booke_pinit(mmu_t, pmap_t);
351 static void		mmu_booke_pinit0(mmu_t, pmap_t);
352 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
353     vm_prot_t);
354 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
355 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
356 static void		mmu_booke_release(mmu_t, pmap_t);
357 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
358 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
359 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
360 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
361 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
362 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
363 static void		mmu_booke_activate(mmu_t, struct thread *);
364 static void		mmu_booke_deactivate(mmu_t, struct thread *);
365 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
366 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
367 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
368 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
369 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
370 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
371 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
372 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
373 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
374 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
375     vm_size_t);
376 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
377     void **);
378 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
379     void *);
380 static void		mmu_booke_scan_init(mmu_t);
381 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
382 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
383 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
384     vm_size_t sz, vm_memattr_t mode);
385 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
386     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
387 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
388     int *is_user, vm_offset_t *decoded_addr);
389 
390 
391 static mmu_method_t mmu_booke_methods[] = {
392 	/* pmap dispatcher interface */
393 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
394 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
395 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
396 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
397 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
398 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
399 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
400 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
401 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
402 	MMUMETHOD(mmu_init,		mmu_booke_init),
403 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
404 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
405 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
406 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
407 	MMUMETHOD(mmu_map,		mmu_booke_map),
408 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
409 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
410 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
411 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
412 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
413 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
414 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
415 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
416 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
417 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
418 	MMUMETHOD(mmu_release,		mmu_booke_release),
419 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
420 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
421 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
422 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
423 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
424 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
425 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
426 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
427 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
428 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
429 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
430 
431 	/* Internal interfaces */
432 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
433 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
434 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
435 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
436 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
437 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
438 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
439 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
440 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
441 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
442 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
443 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
444 
445 	/* dumpsys() support */
446 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
447 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
448 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
449 
450 	{ 0, 0 }
451 };
452 
453 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
454 
455 static __inline uint32_t
456 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
457 {
458 	uint32_t attrib;
459 	int i;
460 
461 	if (ma != VM_MEMATTR_DEFAULT) {
462 		switch (ma) {
463 		case VM_MEMATTR_UNCACHEABLE:
464 			return (MAS2_I | MAS2_G);
465 		case VM_MEMATTR_WRITE_COMBINING:
466 		case VM_MEMATTR_WRITE_BACK:
467 		case VM_MEMATTR_PREFETCHABLE:
468 			return (MAS2_I);
469 		case VM_MEMATTR_WRITE_THROUGH:
470 			return (MAS2_W | MAS2_M);
471 		case VM_MEMATTR_CACHEABLE:
472 			return (MAS2_M);
473 		}
474 	}
475 
476 	/*
477 	 * Assume the page is cache inhibited and access is guarded unless
478 	 * it's in our available memory array.
479 	 */
480 	attrib = _TLB_ENTRY_IO;
481 	for (i = 0; i < physmem_regions_sz; i++) {
482 		if ((pa >= physmem_regions[i].mr_start) &&
483 		    (pa < (physmem_regions[i].mr_start +
484 		     physmem_regions[i].mr_size))) {
485 			attrib = _TLB_ENTRY_MEM;
486 			break;
487 		}
488 	}
489 
490 	return (attrib);
491 }
492 
493 static inline void
494 tlb_miss_lock(void)
495 {
496 #ifdef SMP
497 	struct pcpu *pc;
498 
499 	if (!smp_started)
500 		return;
501 
502 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
503 		if (pc != pcpup) {
504 
505 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
506 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
507 
508 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
509 			    ("tlb_miss_lock: tried to lock self"));
510 
511 			tlb_lock(pc->pc_booke.tlb_lock);
512 
513 			CTR1(KTR_PMAP, "%s: locked", __func__);
514 		}
515 	}
516 #endif
517 }
518 
519 static inline void
520 tlb_miss_unlock(void)
521 {
522 #ifdef SMP
523 	struct pcpu *pc;
524 
525 	if (!smp_started)
526 		return;
527 
528 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
529 		if (pc != pcpup) {
530 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
531 			    __func__, pc->pc_cpuid);
532 
533 			tlb_unlock(pc->pc_booke.tlb_lock);
534 
535 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
536 		}
537 	}
538 #endif
539 }
540 
541 /* Return number of entries in TLB0. */
542 static __inline void
543 tlb0_get_tlbconf(void)
544 {
545 	uint32_t tlb0_cfg;
546 
547 	tlb0_cfg = mfspr(SPR_TLB0CFG);
548 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
549 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
550 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
551 }
552 
553 /* Return number of entries in TLB1. */
554 static __inline void
555 tlb1_get_tlbconf(void)
556 {
557 	uint32_t tlb1_cfg;
558 
559 	tlb1_cfg = mfspr(SPR_TLB1CFG);
560 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
561 }
562 
563 /**************************************************************************/
564 /* Page table related */
565 /**************************************************************************/
566 
567 #ifdef __powerpc64__
568 /* Initialize pool of kva ptbl buffers. */
569 static void
570 ptbl_init(void)
571 {
572 	int		i;
573 
574 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
575 	TAILQ_INIT(&ptbl_buf_freelist);
576 
577 	for (i = 0; i < PTBL_BUFS; i++) {
578 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase +
579 		    i * MAX(PTBL_PAGES,PDIR_PAGES) * PAGE_SIZE;
580 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
581 	}
582 }
583 
584 /* Get an sf_buf from the freelist. */
585 static struct ptbl_buf *
586 ptbl_buf_alloc(void)
587 {
588 	struct ptbl_buf *buf;
589 
590 	mtx_lock(&ptbl_buf_freelist_lock);
591 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
592 	if (buf != NULL)
593 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
594 	mtx_unlock(&ptbl_buf_freelist_lock);
595 
596 	return (buf);
597 }
598 
599 /* Return ptbl buff to free pool. */
600 static void
601 ptbl_buf_free(struct ptbl_buf *buf)
602 {
603 	mtx_lock(&ptbl_buf_freelist_lock);
604 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
605 	mtx_unlock(&ptbl_buf_freelist_lock);
606 }
607 
608 /*
609  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
610  */
611 static void
612 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t * ptbl)
613 {
614 	struct ptbl_buf *pbuf;
615 
616 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) {
617 		if (pbuf->kva == (vm_offset_t) ptbl) {
618 			/* Remove from pmap ptbl buf list. */
619 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
620 
621 			/* Free corresponding ptbl buf. */
622 			ptbl_buf_free(pbuf);
623 
624 			break;
625 		}
626 	}
627 }
628 
629 /* Get a pointer to a PTE in a page table. */
630 static __inline pte_t *
631 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
632 {
633 	pte_t         **pdir;
634 	pte_t          *ptbl;
635 
636 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
637 
638 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
639 	if (!pdir)
640 		return NULL;
641 	ptbl = pdir[PDIR_IDX(va)];
642 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
643 }
644 
645 /*
646  * Search the list of allocated pdir bufs and find on list of allocated pdirs
647  */
648 static void
649 ptbl_free_pmap_pdir(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
650 {
651 	struct ptbl_buf *pbuf;
652 
653 	TAILQ_FOREACH(pbuf, &pmap->pm_pdir_list, link) {
654 		if (pbuf->kva == (vm_offset_t) pdir) {
655 			/* Remove from pmap ptbl buf list. */
656 			TAILQ_REMOVE(&pmap->pm_pdir_list, pbuf, link);
657 
658 			/* Free corresponding pdir buf. */
659 			ptbl_buf_free(pbuf);
660 
661 			break;
662 		}
663 	}
664 }
665 /* Free pdir pages and invalidate pdir entry. */
666 static void
667 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
668 {
669 	pte_t         **pdir;
670 	vm_paddr_t	pa;
671 	vm_offset_t	va;
672 	vm_page_t	m;
673 	int		i;
674 
675 	pdir = pmap->pm_pp2d[pp2d_idx];
676 
677 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
678 
679 	pmap->pm_pp2d[pp2d_idx] = NULL;
680 
681 	for (i = 0; i < PDIR_PAGES; i++) {
682 		va = ((vm_offset_t) pdir + (i * PAGE_SIZE));
683 		pa = pte_vatopa(mmu, kernel_pmap, va);
684 		m = PHYS_TO_VM_PAGE(pa);
685 		vm_page_free_zero(m);
686 		vm_wire_sub(1);
687 		pmap_kremove(va);
688 	}
689 
690 	ptbl_free_pmap_pdir(mmu, pmap, pdir);
691 }
692 
693 /*
694  * Decrement pdir pages hold count and attempt to free pdir pages. Called
695  * when removing directory entry from pdir.
696  *
697  * Return 1 if pdir pages were freed.
698  */
699 static int
700 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
701 {
702 	pte_t         **pdir;
703 	vm_paddr_t	pa;
704 	vm_page_t	m;
705 	int		i;
706 
707 	KASSERT((pmap != kernel_pmap),
708 		("pdir_unhold: unholding kernel pdir!"));
709 
710 	pdir = pmap->pm_pp2d[pp2d_idx];
711 
712 	KASSERT(((vm_offset_t) pdir >= VM_MIN_KERNEL_ADDRESS),
713 	    ("pdir_unhold: non kva pdir"));
714 
715 	/* decrement hold count */
716 	for (i = 0; i < PDIR_PAGES; i++) {
717 		pa = pte_vatopa(mmu, kernel_pmap,
718 		    (vm_offset_t) pdir + (i * PAGE_SIZE));
719 		m = PHYS_TO_VM_PAGE(pa);
720 		m->wire_count--;
721 	}
722 
723 	/*
724 	 * Free pdir pages if there are no dir entries in this pdir.
725 	 * wire_count has the same value for all ptbl pages, so check the
726 	 * last page.
727 	 */
728 	if (m->wire_count == 0) {
729 		pdir_free(mmu, pmap, pp2d_idx);
730 		return (1);
731 	}
732 	return (0);
733 }
734 
735 /*
736  * Increment hold count for pdir pages. This routine is used when new ptlb
737  * entry is being inserted into pdir.
738  */
739 static void
740 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
741 {
742 	vm_paddr_t	pa;
743 	vm_page_t	m;
744 	int		i;
745 
746 	KASSERT((pmap != kernel_pmap),
747 		("pdir_hold: holding kernel pdir!"));
748 
749 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
750 
751 	for (i = 0; i < PDIR_PAGES; i++) {
752 		pa = pte_vatopa(mmu, kernel_pmap,
753 				(vm_offset_t) pdir + (i * PAGE_SIZE));
754 		m = PHYS_TO_VM_PAGE(pa);
755 		m->wire_count++;
756 	}
757 }
758 
759 /* Allocate page table. */
760 static pte_t   *
761 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
762     boolean_t nosleep)
763 {
764 	vm_page_t	mtbl  [PTBL_PAGES];
765 	vm_page_t	m;
766 	struct ptbl_buf *pbuf;
767 	unsigned int	pidx;
768 	pte_t          *ptbl;
769 	int		i, j;
770 	int		req;
771 
772 	KASSERT((pdir[pdir_idx] == NULL),
773 		("%s: valid ptbl entry exists!", __func__));
774 
775 	pbuf = ptbl_buf_alloc();
776 	if (pbuf == NULL)
777 		panic("%s: couldn't alloc kernel virtual memory", __func__);
778 
779 	ptbl = (pte_t *) pbuf->kva;
780 
781 	for (i = 0; i < PTBL_PAGES; i++) {
782 		pidx = (PTBL_PAGES * pdir_idx) + i;
783 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
784 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
785 			PMAP_UNLOCK(pmap);
786 			rw_wunlock(&pvh_global_lock);
787 			if (nosleep) {
788 				ptbl_free_pmap_ptbl(pmap, ptbl);
789 				for (j = 0; j < i; j++)
790 					vm_page_free(mtbl[j]);
791 				vm_wire_sub(i);
792 				return (NULL);
793 			}
794 			vm_wait(NULL);
795 			rw_wlock(&pvh_global_lock);
796 			PMAP_LOCK(pmap);
797 		}
798 		mtbl[i] = m;
799 	}
800 
801 	/* Mapin allocated pages into kernel_pmap. */
802 	mmu_booke_qenter(mmu, (vm_offset_t) ptbl, mtbl, PTBL_PAGES);
803 	/* Zero whole ptbl. */
804 	bzero((caddr_t) ptbl, PTBL_PAGES * PAGE_SIZE);
805 
806 	/* Add pbuf to the pmap ptbl bufs list. */
807 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
808 
809 	return (ptbl);
810 }
811 
812 /* Free ptbl pages and invalidate pdir entry. */
813 static void
814 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
815 {
816 	pte_t          *ptbl;
817 	vm_paddr_t	pa;
818 	vm_offset_t	va;
819 	vm_page_t	m;
820 	int		i;
821 
822 	ptbl = pdir[pdir_idx];
823 
824 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
825 
826 	pdir[pdir_idx] = NULL;
827 
828 	for (i = 0; i < PTBL_PAGES; i++) {
829 		va = ((vm_offset_t) ptbl + (i * PAGE_SIZE));
830 		pa = pte_vatopa(mmu, kernel_pmap, va);
831 		m = PHYS_TO_VM_PAGE(pa);
832 		vm_page_free_zero(m);
833 		vm_wire_sub(1);
834 		pmap_kremove(va);
835 	}
836 
837 	ptbl_free_pmap_ptbl(pmap, ptbl);
838 }
839 
840 /*
841  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
842  * when removing pte entry from ptbl.
843  *
844  * Return 1 if ptbl pages were freed.
845  */
846 static int
847 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
848 {
849 	pte_t          *ptbl;
850 	vm_paddr_t	pa;
851 	vm_page_t	m;
852 	u_int		pp2d_idx;
853 	pte_t         **pdir;
854 	u_int		pdir_idx;
855 	int		i;
856 
857 	pp2d_idx = PP2D_IDX(va);
858 	pdir_idx = PDIR_IDX(va);
859 
860 	KASSERT((pmap != kernel_pmap),
861 		("ptbl_unhold: unholding kernel ptbl!"));
862 
863 	pdir = pmap->pm_pp2d[pp2d_idx];
864 	ptbl = pdir[pdir_idx];
865 
866 	KASSERT(((vm_offset_t) ptbl >= VM_MIN_KERNEL_ADDRESS),
867 	    ("ptbl_unhold: non kva ptbl"));
868 
869 	/* decrement hold count */
870 	for (i = 0; i < PTBL_PAGES; i++) {
871 		pa = pte_vatopa(mmu, kernel_pmap,
872 		    (vm_offset_t) ptbl + (i * PAGE_SIZE));
873 		m = PHYS_TO_VM_PAGE(pa);
874 		m->wire_count--;
875 	}
876 
877 	/*
878 	 * Free ptbl pages if there are no pte entries in this ptbl.
879 	 * wire_count has the same value for all ptbl pages, so check the
880 	 * last page.
881 	 */
882 	if (m->wire_count == 0) {
883 		/* A pair of indirect entries might point to this ptbl page */
884 #if 0
885 		tlb_flush_entry(pmap, va & ~((2UL * PAGE_SIZE_1M) - 1),
886 				TLB_SIZE_1M, MAS6_SIND);
887 		tlb_flush_entry(pmap, (va & ~((2UL * PAGE_SIZE_1M) - 1)) | PAGE_SIZE_1M,
888 				TLB_SIZE_1M, MAS6_SIND);
889 #endif
890 		ptbl_free(mmu, pmap, pdir, pdir_idx);
891 		pdir_unhold(mmu, pmap, pp2d_idx);
892 		return (1);
893 	}
894 	return (0);
895 }
896 
897 /*
898  * Increment hold count for ptbl pages. This routine is used when new pte
899  * entry is being inserted into ptbl.
900  */
901 static void
902 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
903 {
904 	vm_paddr_t	pa;
905 	pte_t          *ptbl;
906 	vm_page_t	m;
907 	int		i;
908 
909 	KASSERT((pmap != kernel_pmap),
910 		("ptbl_hold: holding kernel ptbl!"));
911 
912 	ptbl = pdir[pdir_idx];
913 
914 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
915 
916 	for (i = 0; i < PTBL_PAGES; i++) {
917 		pa = pte_vatopa(mmu, kernel_pmap,
918 				(vm_offset_t) ptbl + (i * PAGE_SIZE));
919 		m = PHYS_TO_VM_PAGE(pa);
920 		m->wire_count++;
921 	}
922 }
923 #else
924 
925 /* Initialize pool of kva ptbl buffers. */
926 static void
927 ptbl_init(void)
928 {
929 	int i;
930 
931 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
932 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
933 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
934 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
935 
936 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
937 	TAILQ_INIT(&ptbl_buf_freelist);
938 
939 	for (i = 0; i < PTBL_BUFS; i++) {
940 		ptbl_bufs[i].kva =
941 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
942 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
943 	}
944 }
945 
946 /* Get a ptbl_buf from the freelist. */
947 static struct ptbl_buf *
948 ptbl_buf_alloc(void)
949 {
950 	struct ptbl_buf *buf;
951 
952 	mtx_lock(&ptbl_buf_freelist_lock);
953 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
954 	if (buf != NULL)
955 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
956 	mtx_unlock(&ptbl_buf_freelist_lock);
957 
958 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
959 
960 	return (buf);
961 }
962 
963 /* Return ptbl buff to free pool. */
964 static void
965 ptbl_buf_free(struct ptbl_buf *buf)
966 {
967 
968 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
969 
970 	mtx_lock(&ptbl_buf_freelist_lock);
971 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
972 	mtx_unlock(&ptbl_buf_freelist_lock);
973 }
974 
975 /*
976  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
977  */
978 static void
979 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
980 {
981 	struct ptbl_buf *pbuf;
982 
983 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
984 
985 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
986 
987 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
988 		if (pbuf->kva == (vm_offset_t)ptbl) {
989 			/* Remove from pmap ptbl buf list. */
990 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
991 
992 			/* Free corresponding ptbl buf. */
993 			ptbl_buf_free(pbuf);
994 			break;
995 		}
996 }
997 
998 /* Allocate page table. */
999 static pte_t *
1000 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
1001 {
1002 	vm_page_t mtbl[PTBL_PAGES];
1003 	vm_page_t m;
1004 	struct ptbl_buf *pbuf;
1005 	unsigned int pidx;
1006 	pte_t *ptbl;
1007 	int i, j;
1008 
1009 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1010 	    (pmap == kernel_pmap), pdir_idx);
1011 
1012 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1013 	    ("ptbl_alloc: invalid pdir_idx"));
1014 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
1015 	    ("pte_alloc: valid ptbl entry exists!"));
1016 
1017 	pbuf = ptbl_buf_alloc();
1018 	if (pbuf == NULL)
1019 		panic("pte_alloc: couldn't alloc kernel virtual memory");
1020 
1021 	ptbl = (pte_t *)pbuf->kva;
1022 
1023 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
1024 
1025 	for (i = 0; i < PTBL_PAGES; i++) {
1026 		pidx = (PTBL_PAGES * pdir_idx) + i;
1027 		while ((m = vm_page_alloc(NULL, pidx,
1028 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
1029 			PMAP_UNLOCK(pmap);
1030 			rw_wunlock(&pvh_global_lock);
1031 			if (nosleep) {
1032 				ptbl_free_pmap_ptbl(pmap, ptbl);
1033 				for (j = 0; j < i; j++)
1034 					vm_page_free(mtbl[j]);
1035 				vm_wire_sub(i);
1036 				return (NULL);
1037 			}
1038 			vm_wait(NULL);
1039 			rw_wlock(&pvh_global_lock);
1040 			PMAP_LOCK(pmap);
1041 		}
1042 		mtbl[i] = m;
1043 	}
1044 
1045 	/* Map allocated pages into kernel_pmap. */
1046 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
1047 
1048 	/* Zero whole ptbl. */
1049 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
1050 
1051 	/* Add pbuf to the pmap ptbl bufs list. */
1052 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
1053 
1054 	return (ptbl);
1055 }
1056 
1057 /* Free ptbl pages and invalidate pdir entry. */
1058 static void
1059 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1060 {
1061 	pte_t *ptbl;
1062 	vm_paddr_t pa;
1063 	vm_offset_t va;
1064 	vm_page_t m;
1065 	int i;
1066 
1067 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1068 	    (pmap == kernel_pmap), pdir_idx);
1069 
1070 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1071 	    ("ptbl_free: invalid pdir_idx"));
1072 
1073 	ptbl = pmap->pm_pdir[pdir_idx];
1074 
1075 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
1076 
1077 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
1078 
1079 	/*
1080 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
1081 	 * don't attempt to look up the page tables we are releasing.
1082 	 */
1083 	mtx_lock_spin(&tlbivax_mutex);
1084 	tlb_miss_lock();
1085 
1086 	pmap->pm_pdir[pdir_idx] = NULL;
1087 
1088 	tlb_miss_unlock();
1089 	mtx_unlock_spin(&tlbivax_mutex);
1090 
1091 	for (i = 0; i < PTBL_PAGES; i++) {
1092 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
1093 		pa = pte_vatopa(mmu, kernel_pmap, va);
1094 		m = PHYS_TO_VM_PAGE(pa);
1095 		vm_page_free_zero(m);
1096 		vm_wire_sub(1);
1097 		mmu_booke_kremove(mmu, va);
1098 	}
1099 
1100 	ptbl_free_pmap_ptbl(pmap, ptbl);
1101 }
1102 
1103 /*
1104  * Decrement ptbl pages hold count and attempt to free ptbl pages.
1105  * Called when removing pte entry from ptbl.
1106  *
1107  * Return 1 if ptbl pages were freed.
1108  */
1109 static int
1110 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1111 {
1112 	pte_t *ptbl;
1113 	vm_paddr_t pa;
1114 	vm_page_t m;
1115 	int i;
1116 
1117 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1118 	    (pmap == kernel_pmap), pdir_idx);
1119 
1120 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1121 	    ("ptbl_unhold: invalid pdir_idx"));
1122 	KASSERT((pmap != kernel_pmap),
1123 	    ("ptbl_unhold: unholding kernel ptbl!"));
1124 
1125 	ptbl = pmap->pm_pdir[pdir_idx];
1126 
1127 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1128 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1129 	    ("ptbl_unhold: non kva ptbl"));
1130 
1131 	/* decrement hold count */
1132 	for (i = 0; i < PTBL_PAGES; i++) {
1133 		pa = pte_vatopa(mmu, kernel_pmap,
1134 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1135 		m = PHYS_TO_VM_PAGE(pa);
1136 		m->wire_count--;
1137 	}
1138 
1139 	/*
1140 	 * Free ptbl pages if there are no pte etries in this ptbl.
1141 	 * wire_count has the same value for all ptbl pages, so check the last
1142 	 * page.
1143 	 */
1144 	if (m->wire_count == 0) {
1145 		ptbl_free(mmu, pmap, pdir_idx);
1146 
1147 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1148 		return (1);
1149 	}
1150 
1151 	return (0);
1152 }
1153 
1154 /*
1155  * Increment hold count for ptbl pages. This routine is used when a new pte
1156  * entry is being inserted into the ptbl.
1157  */
1158 static void
1159 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1160 {
1161 	vm_paddr_t pa;
1162 	pte_t *ptbl;
1163 	vm_page_t m;
1164 	int i;
1165 
1166 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1167 	    pdir_idx);
1168 
1169 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1170 	    ("ptbl_hold: invalid pdir_idx"));
1171 	KASSERT((pmap != kernel_pmap),
1172 	    ("ptbl_hold: holding kernel ptbl!"));
1173 
1174 	ptbl = pmap->pm_pdir[pdir_idx];
1175 
1176 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1177 
1178 	for (i = 0; i < PTBL_PAGES; i++) {
1179 		pa = pte_vatopa(mmu, kernel_pmap,
1180 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1181 		m = PHYS_TO_VM_PAGE(pa);
1182 		m->wire_count++;
1183 	}
1184 }
1185 #endif
1186 
1187 /* Allocate pv_entry structure. */
1188 pv_entry_t
1189 pv_alloc(void)
1190 {
1191 	pv_entry_t pv;
1192 
1193 	pv_entry_count++;
1194 	if (pv_entry_count > pv_entry_high_water)
1195 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1196 	pv = uma_zalloc(pvzone, M_NOWAIT);
1197 
1198 	return (pv);
1199 }
1200 
1201 /* Free pv_entry structure. */
1202 static __inline void
1203 pv_free(pv_entry_t pve)
1204 {
1205 
1206 	pv_entry_count--;
1207 	uma_zfree(pvzone, pve);
1208 }
1209 
1210 
1211 /* Allocate and initialize pv_entry structure. */
1212 static void
1213 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1214 {
1215 	pv_entry_t pve;
1216 
1217 	//int su = (pmap == kernel_pmap);
1218 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1219 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1220 
1221 	pve = pv_alloc();
1222 	if (pve == NULL)
1223 		panic("pv_insert: no pv entries!");
1224 
1225 	pve->pv_pmap = pmap;
1226 	pve->pv_va = va;
1227 
1228 	/* add to pv_list */
1229 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1230 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1231 
1232 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1233 
1234 	//debugf("pv_insert: e\n");
1235 }
1236 
1237 /* Destroy pv entry. */
1238 static void
1239 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1240 {
1241 	pv_entry_t pve;
1242 
1243 	//int su = (pmap == kernel_pmap);
1244 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1245 
1246 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1247 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1248 
1249 	/* find pv entry */
1250 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1251 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1252 			/* remove from pv_list */
1253 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1254 			if (TAILQ_EMPTY(&m->md.pv_list))
1255 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1256 
1257 			/* free pv entry struct */
1258 			pv_free(pve);
1259 			break;
1260 		}
1261 	}
1262 
1263 	//debugf("pv_remove: e\n");
1264 }
1265 
1266 #ifdef __powerpc64__
1267 /*
1268  * Clean pte entry, try to free page table page if requested.
1269  *
1270  * Return 1 if ptbl pages were freed, otherwise return 0.
1271  */
1272 static int
1273 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1274 {
1275 	vm_page_t	m;
1276 	pte_t          *pte;
1277 
1278 	pte = pte_find(mmu, pmap, va);
1279 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1280 
1281 	if (!PTE_ISVALID(pte))
1282 		return (0);
1283 
1284 	/* Get vm_page_t for mapped pte. */
1285 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1286 
1287 	if (PTE_ISWIRED(pte))
1288 		pmap->pm_stats.wired_count--;
1289 
1290 	/* Handle managed entry. */
1291 	if (PTE_ISMANAGED(pte)) {
1292 
1293 		/* Handle modified pages. */
1294 		if (PTE_ISMODIFIED(pte))
1295 			vm_page_dirty(m);
1296 
1297 		/* Referenced pages. */
1298 		if (PTE_ISREFERENCED(pte))
1299 			vm_page_aflag_set(m, PGA_REFERENCED);
1300 
1301 		/* Remove pv_entry from pv_list. */
1302 		pv_remove(pmap, va, m);
1303 	} else if (m->md.pv_tracked) {
1304 		pv_remove(pmap, va, m);
1305 		if (TAILQ_EMPTY(&m->md.pv_list))
1306 			m->md.pv_tracked = false;
1307 	}
1308 	mtx_lock_spin(&tlbivax_mutex);
1309 	tlb_miss_lock();
1310 
1311 	tlb0_flush_entry(va);
1312 	*pte = 0;
1313 
1314 	tlb_miss_unlock();
1315 	mtx_unlock_spin(&tlbivax_mutex);
1316 
1317 	pmap->pm_stats.resident_count--;
1318 
1319 	if (flags & PTBL_UNHOLD) {
1320 		return (ptbl_unhold(mmu, pmap, va));
1321 	}
1322 	return (0);
1323 }
1324 
1325 /*
1326  * allocate a page of pointers to page directories, do not preallocate the
1327  * page tables
1328  */
1329 static pte_t  **
1330 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
1331 {
1332 	vm_page_t	mtbl  [PDIR_PAGES];
1333 	vm_page_t	m;
1334 	struct ptbl_buf *pbuf;
1335 	pte_t         **pdir;
1336 	unsigned int	pidx;
1337 	int		i;
1338 	int		req;
1339 
1340 	pbuf = ptbl_buf_alloc();
1341 
1342 	if (pbuf == NULL)
1343 		panic("%s: couldn't alloc kernel virtual memory", __func__);
1344 
1345 	/* Allocate pdir pages, this will sleep! */
1346 	for (i = 0; i < PDIR_PAGES; i++) {
1347 		pidx = (PDIR_PAGES * pp2d_idx) + i;
1348 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
1349 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
1350 			PMAP_UNLOCK(pmap);
1351 			vm_wait(NULL);
1352 			PMAP_LOCK(pmap);
1353 		}
1354 		mtbl[i] = m;
1355 	}
1356 
1357 	/* Mapin allocated pages into kernel_pmap. */
1358 	pdir = (pte_t **) pbuf->kva;
1359 	pmap_qenter((vm_offset_t) pdir, mtbl, PDIR_PAGES);
1360 
1361 	/* Zero whole pdir. */
1362 	bzero((caddr_t) pdir, PDIR_PAGES * PAGE_SIZE);
1363 
1364 	/* Add pdir to the pmap pdir bufs list. */
1365 	TAILQ_INSERT_TAIL(&pmap->pm_pdir_list, pbuf, link);
1366 
1367 	return pdir;
1368 }
1369 
1370 /*
1371  * Insert PTE for a given page and virtual address.
1372  */
1373 static int
1374 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1375     boolean_t nosleep)
1376 {
1377 	unsigned int	pp2d_idx = PP2D_IDX(va);
1378 	unsigned int	pdir_idx = PDIR_IDX(va);
1379 	unsigned int	ptbl_idx = PTBL_IDX(va);
1380 	pte_t          *ptbl, *pte;
1381 	pte_t         **pdir;
1382 
1383 	/* Get the page directory pointer. */
1384 	pdir = pmap->pm_pp2d[pp2d_idx];
1385 	if (pdir == NULL)
1386 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1387 
1388 	/* Get the page table pointer. */
1389 	ptbl = pdir[pdir_idx];
1390 
1391 	if (ptbl == NULL) {
1392 		/* Allocate page table pages. */
1393 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1394 		if (ptbl == NULL) {
1395 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1396 			return (ENOMEM);
1397 		}
1398 	} else {
1399 		/*
1400 		 * Check if there is valid mapping for requested va, if there
1401 		 * is, remove it.
1402 		 */
1403 		pte = &pdir[pdir_idx][ptbl_idx];
1404 		if (PTE_ISVALID(pte)) {
1405 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1406 		} else {
1407 			/*
1408 			 * pte is not used, increment hold count for ptbl
1409 			 * pages.
1410 			 */
1411 			if (pmap != kernel_pmap)
1412 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1413 		}
1414 	}
1415 
1416 	if (pdir[pdir_idx] == NULL) {
1417 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1418 			pdir_hold(mmu, pmap, pdir);
1419 		pdir[pdir_idx] = ptbl;
1420 	}
1421 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1422 		pmap->pm_pp2d[pp2d_idx] = pdir;
1423 
1424 	/*
1425 	 * Insert pv_entry into pv_list for mapped page if part of managed
1426 	 * memory.
1427 	 */
1428 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1429 		flags |= PTE_MANAGED;
1430 
1431 		/* Create and insert pv entry. */
1432 		pv_insert(pmap, va, m);
1433 	}
1434 
1435 	mtx_lock_spin(&tlbivax_mutex);
1436 	tlb_miss_lock();
1437 
1438 	tlb0_flush_entry(va);
1439 	pmap->pm_stats.resident_count++;
1440 	pte = &pdir[pdir_idx][ptbl_idx];
1441 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1442 	*pte |= (PTE_VALID | flags);
1443 
1444 	tlb_miss_unlock();
1445 	mtx_unlock_spin(&tlbivax_mutex);
1446 
1447 	return (0);
1448 }
1449 
1450 /* Return the pa for the given pmap/va. */
1451 static	vm_paddr_t
1452 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1453 {
1454 	vm_paddr_t	pa = 0;
1455 	pte_t          *pte;
1456 
1457 	pte = pte_find(mmu, pmap, va);
1458 	if ((pte != NULL) && PTE_ISVALID(pte))
1459 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1460 	return (pa);
1461 }
1462 
1463 
1464 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1465 static void
1466 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1467 {
1468 	int		i, j;
1469 	vm_offset_t	va;
1470 	pte_t		*pte;
1471 
1472 	va = addr;
1473 	/* Initialize kernel pdir */
1474 	for (i = 0; i < kernel_pdirs; i++) {
1475 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1476 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1477 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1478 		    j < PDIR_NENTRIES; j++) {
1479 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1480 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE * PDIR_PAGES) +
1481 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE * PTBL_PAGES));
1482 		}
1483 	}
1484 
1485 	/*
1486 	 * Fill in PTEs covering kernel code and data. They are not required
1487 	 * for address translation, as this area is covered by static TLB1
1488 	 * entries, but for pte_vatopa() to work correctly with kernel area
1489 	 * addresses.
1490 	 */
1491 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1492 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1493 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1494 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1495 		    PTE_VALID | PTE_PS_4KB;
1496 	}
1497 }
1498 #else
1499 /*
1500  * Clean pte entry, try to free page table page if requested.
1501  *
1502  * Return 1 if ptbl pages were freed, otherwise return 0.
1503  */
1504 static int
1505 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1506 {
1507 	unsigned int pdir_idx = PDIR_IDX(va);
1508 	unsigned int ptbl_idx = PTBL_IDX(va);
1509 	vm_page_t m;
1510 	pte_t *ptbl;
1511 	pte_t *pte;
1512 
1513 	//int su = (pmap == kernel_pmap);
1514 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1515 	//		su, (u_int32_t)pmap, va, flags);
1516 
1517 	ptbl = pmap->pm_pdir[pdir_idx];
1518 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1519 
1520 	pte = &ptbl[ptbl_idx];
1521 
1522 	if (pte == NULL || !PTE_ISVALID(pte))
1523 		return (0);
1524 
1525 	if (PTE_ISWIRED(pte))
1526 		pmap->pm_stats.wired_count--;
1527 
1528 	/* Get vm_page_t for mapped pte. */
1529 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1530 
1531 	/* Handle managed entry. */
1532 	if (PTE_ISMANAGED(pte)) {
1533 
1534 		if (PTE_ISMODIFIED(pte))
1535 			vm_page_dirty(m);
1536 
1537 		if (PTE_ISREFERENCED(pte))
1538 			vm_page_aflag_set(m, PGA_REFERENCED);
1539 
1540 		pv_remove(pmap, va, m);
1541 	} else if (m->md.pv_tracked) {
1542 		/*
1543 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1544 		 * used.  This is needed by the NCSW support code for fast
1545 		 * VA<->PA translation.
1546 		 */
1547 		pv_remove(pmap, va, m);
1548 		if (TAILQ_EMPTY(&m->md.pv_list))
1549 			m->md.pv_tracked = false;
1550 	}
1551 
1552 	mtx_lock_spin(&tlbivax_mutex);
1553 	tlb_miss_lock();
1554 
1555 	tlb0_flush_entry(va);
1556 	*pte = 0;
1557 
1558 	tlb_miss_unlock();
1559 	mtx_unlock_spin(&tlbivax_mutex);
1560 
1561 	pmap->pm_stats.resident_count--;
1562 
1563 	if (flags & PTBL_UNHOLD) {
1564 		//debugf("pte_remove: e (unhold)\n");
1565 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1566 	}
1567 
1568 	//debugf("pte_remove: e\n");
1569 	return (0);
1570 }
1571 
1572 /*
1573  * Insert PTE for a given page and virtual address.
1574  */
1575 static int
1576 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1577     boolean_t nosleep)
1578 {
1579 	unsigned int pdir_idx = PDIR_IDX(va);
1580 	unsigned int ptbl_idx = PTBL_IDX(va);
1581 	pte_t *ptbl, *pte;
1582 
1583 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1584 	    pmap == kernel_pmap, pmap, va);
1585 
1586 	/* Get the page table pointer. */
1587 	ptbl = pmap->pm_pdir[pdir_idx];
1588 
1589 	if (ptbl == NULL) {
1590 		/* Allocate page table pages. */
1591 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1592 		if (ptbl == NULL) {
1593 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1594 			return (ENOMEM);
1595 		}
1596 	} else {
1597 		/*
1598 		 * Check if there is valid mapping for requested
1599 		 * va, if there is, remove it.
1600 		 */
1601 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1602 		if (PTE_ISVALID(pte)) {
1603 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1604 		} else {
1605 			/*
1606 			 * pte is not used, increment hold count
1607 			 * for ptbl pages.
1608 			 */
1609 			if (pmap != kernel_pmap)
1610 				ptbl_hold(mmu, pmap, pdir_idx);
1611 		}
1612 	}
1613 
1614 	/*
1615 	 * Insert pv_entry into pv_list for mapped page if part of managed
1616 	 * memory.
1617 	 */
1618 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1619 		flags |= PTE_MANAGED;
1620 
1621 		/* Create and insert pv entry. */
1622 		pv_insert(pmap, va, m);
1623 	}
1624 
1625 	pmap->pm_stats.resident_count++;
1626 
1627 	mtx_lock_spin(&tlbivax_mutex);
1628 	tlb_miss_lock();
1629 
1630 	tlb0_flush_entry(va);
1631 	if (pmap->pm_pdir[pdir_idx] == NULL) {
1632 		/*
1633 		 * If we just allocated a new page table, hook it in
1634 		 * the pdir.
1635 		 */
1636 		pmap->pm_pdir[pdir_idx] = ptbl;
1637 	}
1638 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
1639 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1640 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1641 
1642 	tlb_miss_unlock();
1643 	mtx_unlock_spin(&tlbivax_mutex);
1644 	return (0);
1645 }
1646 
1647 /* Return the pa for the given pmap/va. */
1648 static vm_paddr_t
1649 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1650 {
1651 	vm_paddr_t pa = 0;
1652 	pte_t *pte;
1653 
1654 	pte = pte_find(mmu, pmap, va);
1655 	if ((pte != NULL) && PTE_ISVALID(pte))
1656 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1657 	return (pa);
1658 }
1659 
1660 /* Get a pointer to a PTE in a page table. */
1661 static pte_t *
1662 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1663 {
1664 	unsigned int pdir_idx = PDIR_IDX(va);
1665 	unsigned int ptbl_idx = PTBL_IDX(va);
1666 
1667 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1668 
1669 	if (pmap->pm_pdir[pdir_idx])
1670 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1671 
1672 	return (NULL);
1673 }
1674 
1675 /* Set up kernel page tables. */
1676 static void
1677 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1678 {
1679 	int		i;
1680 	vm_offset_t	va;
1681 	pte_t		*pte;
1682 
1683 	/* Initialize kernel pdir */
1684 	for (i = 0; i < kernel_ptbls; i++)
1685 		kernel_pmap->pm_pdir[kptbl_min + i] =
1686 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1687 
1688 	/*
1689 	 * Fill in PTEs covering kernel code and data. They are not required
1690 	 * for address translation, as this area is covered by static TLB1
1691 	 * entries, but for pte_vatopa() to work correctly with kernel area
1692 	 * addresses.
1693 	 */
1694 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1695 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1696 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1697 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1698 		    PTE_VALID | PTE_PS_4KB;
1699 	}
1700 }
1701 #endif
1702 
1703 /**************************************************************************/
1704 /* PMAP related */
1705 /**************************************************************************/
1706 
1707 /*
1708  * This is called during booke_init, before the system is really initialized.
1709  */
1710 static void
1711 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1712 {
1713 	vm_paddr_t phys_kernelend;
1714 	struct mem_region *mp, *mp1;
1715 	int cnt, i, j;
1716 	vm_paddr_t s, e, sz;
1717 	vm_paddr_t physsz, hwphyssz;
1718 	u_int phys_avail_count;
1719 	vm_size_t kstack0_sz;
1720 	vm_offset_t kernel_pdir, kstack0;
1721 	vm_paddr_t kstack0_phys;
1722 	void *dpcpu;
1723 
1724 	debugf("mmu_booke_bootstrap: entered\n");
1725 
1726 	/* Set interesting system properties */
1727 	hw_direct_map = 0;
1728 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1729 	elf32_nxstack = 1;
1730 #endif
1731 
1732 	/* Initialize invalidation mutex */
1733 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1734 
1735 	/* Read TLB0 size and associativity. */
1736 	tlb0_get_tlbconf();
1737 
1738 	/*
1739 	 * Align kernel start and end address (kernel image).
1740 	 * Note that kernel end does not necessarily relate to kernsize.
1741 	 * kernsize is the size of the kernel that is actually mapped.
1742 	 */
1743 	kernstart = trunc_page(start);
1744 	data_start = round_page(kernelend);
1745 	data_end = data_start;
1746 
1747 	/*
1748 	 * Addresses of preloaded modules (like file systems) use
1749 	 * physical addresses. Make sure we relocate those into
1750 	 * virtual addresses.
1751 	 */
1752 	preload_addr_relocate = kernstart - kernload;
1753 
1754 	/* Allocate the dynamic per-cpu area. */
1755 	dpcpu = (void *)data_end;
1756 	data_end += DPCPU_SIZE;
1757 
1758 	/* Allocate space for the message buffer. */
1759 	msgbufp = (struct msgbuf *)data_end;
1760 	data_end += msgbufsize;
1761 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1762 	    (uintptr_t)msgbufp, data_end);
1763 
1764 	data_end = round_page(data_end);
1765 
1766 	/* Allocate space for ptbl_bufs. */
1767 	ptbl_bufs = (struct ptbl_buf *)data_end;
1768 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1769 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1770 	    (uintptr_t)ptbl_bufs, data_end);
1771 
1772 	data_end = round_page(data_end);
1773 
1774 	/* Allocate PTE tables for kernel KVA. */
1775 	kernel_pdir = data_end;
1776 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1777 	    PDIR_SIZE);
1778 #ifdef __powerpc64__
1779 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1780 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1781 #endif
1782 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1783 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1784 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1785 	    kernel_pdir, data_end);
1786 
1787 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1788 	if (data_end - kernstart > kernsize) {
1789 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1790 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1791 	}
1792 	data_end = kernstart + kernsize;
1793 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1794 
1795 	/*
1796 	 * Clear the structures - note we can only do it safely after the
1797 	 * possible additional TLB1 translations are in place (above) so that
1798 	 * all range up to the currently calculated 'data_end' is covered.
1799 	 */
1800 	dpcpu_init(dpcpu, 0);
1801 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1802 #ifdef __powerpc64__
1803 	memset((void *)kernel_pdir, 0,
1804 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1805 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1806 #else
1807 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1808 #endif
1809 
1810 	/*******************************************************/
1811 	/* Set the start and end of kva. */
1812 	/*******************************************************/
1813 	virtual_avail = round_page(data_end);
1814 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1815 
1816 	/* Allocate KVA space for page zero/copy operations. */
1817 	zero_page_va = virtual_avail;
1818 	virtual_avail += PAGE_SIZE;
1819 	copy_page_src_va = virtual_avail;
1820 	virtual_avail += PAGE_SIZE;
1821 	copy_page_dst_va = virtual_avail;
1822 	virtual_avail += PAGE_SIZE;
1823 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1824 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1825 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1826 
1827 	/* Initialize page zero/copy mutexes. */
1828 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1829 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1830 
1831 	/* Allocate KVA space for ptbl bufs. */
1832 	ptbl_buf_pool_vabase = virtual_avail;
1833 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1834 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1835 	    ptbl_buf_pool_vabase, virtual_avail);
1836 
1837 	/* Calculate corresponding physical addresses for the kernel region. */
1838 	phys_kernelend = kernload + kernsize;
1839 	debugf("kernel image and allocated data:\n");
1840 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1841 	debugf(" kernstart   = 0x%08x\n", kernstart);
1842 	debugf(" kernsize    = 0x%08x\n", kernsize);
1843 
1844 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1845 		panic("mmu_booke_bootstrap: phys_avail too small");
1846 
1847 	/*
1848 	 * Remove kernel physical address range from avail regions list. Page
1849 	 * align all regions.  Non-page aligned memory isn't very interesting
1850 	 * to us.  Also, sort the entries for ascending addresses.
1851 	 */
1852 
1853 	/* Retrieve phys/avail mem regions */
1854 	mem_regions(&physmem_regions, &physmem_regions_sz,
1855 	    &availmem_regions, &availmem_regions_sz);
1856 	sz = 0;
1857 	cnt = availmem_regions_sz;
1858 	debugf("processing avail regions:\n");
1859 	for (mp = availmem_regions; mp->mr_size; mp++) {
1860 		s = mp->mr_start;
1861 		e = mp->mr_start + mp->mr_size;
1862 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1863 		/* Check whether this region holds all of the kernel. */
1864 		if (s < kernload && e > phys_kernelend) {
1865 			availmem_regions[cnt].mr_start = phys_kernelend;
1866 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1867 			e = kernload;
1868 		}
1869 		/* Look whether this regions starts within the kernel. */
1870 		if (s >= kernload && s < phys_kernelend) {
1871 			if (e <= phys_kernelend)
1872 				goto empty;
1873 			s = phys_kernelend;
1874 		}
1875 		/* Now look whether this region ends within the kernel. */
1876 		if (e > kernload && e <= phys_kernelend) {
1877 			if (s >= kernload)
1878 				goto empty;
1879 			e = kernload;
1880 		}
1881 		/* Now page align the start and size of the region. */
1882 		s = round_page(s);
1883 		e = trunc_page(e);
1884 		if (e < s)
1885 			e = s;
1886 		sz = e - s;
1887 		debugf("%09jx-%09jx = %jx\n",
1888 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1889 
1890 		/* Check whether some memory is left here. */
1891 		if (sz == 0) {
1892 		empty:
1893 			memmove(mp, mp + 1,
1894 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1895 			cnt--;
1896 			mp--;
1897 			continue;
1898 		}
1899 
1900 		/* Do an insertion sort. */
1901 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1902 			if (s < mp1->mr_start)
1903 				break;
1904 		if (mp1 < mp) {
1905 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1906 			mp1->mr_start = s;
1907 			mp1->mr_size = sz;
1908 		} else {
1909 			mp->mr_start = s;
1910 			mp->mr_size = sz;
1911 		}
1912 	}
1913 	availmem_regions_sz = cnt;
1914 
1915 	/*******************************************************/
1916 	/* Steal physical memory for kernel stack from the end */
1917 	/* of the first avail region                           */
1918 	/*******************************************************/
1919 	kstack0_sz = kstack_pages * PAGE_SIZE;
1920 	kstack0_phys = availmem_regions[0].mr_start +
1921 	    availmem_regions[0].mr_size;
1922 	kstack0_phys -= kstack0_sz;
1923 	availmem_regions[0].mr_size -= kstack0_sz;
1924 
1925 	/*******************************************************/
1926 	/* Fill in phys_avail table, based on availmem_regions */
1927 	/*******************************************************/
1928 	phys_avail_count = 0;
1929 	physsz = 0;
1930 	hwphyssz = 0;
1931 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1932 
1933 	debugf("fill in phys_avail:\n");
1934 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1935 
1936 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1937 		    (uintmax_t)availmem_regions[i].mr_start,
1938 		    (uintmax_t)availmem_regions[i].mr_start +
1939 		        availmem_regions[i].mr_size,
1940 		    (uintmax_t)availmem_regions[i].mr_size);
1941 
1942 		if (hwphyssz != 0 &&
1943 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1944 			debugf(" hw.physmem adjust\n");
1945 			if (physsz < hwphyssz) {
1946 				phys_avail[j] = availmem_regions[i].mr_start;
1947 				phys_avail[j + 1] =
1948 				    availmem_regions[i].mr_start +
1949 				    hwphyssz - physsz;
1950 				physsz = hwphyssz;
1951 				phys_avail_count++;
1952 			}
1953 			break;
1954 		}
1955 
1956 		phys_avail[j] = availmem_regions[i].mr_start;
1957 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1958 		    availmem_regions[i].mr_size;
1959 		phys_avail_count++;
1960 		physsz += availmem_regions[i].mr_size;
1961 	}
1962 	physmem = btoc(physsz);
1963 
1964 	/* Calculate the last available physical address. */
1965 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1966 		;
1967 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1968 
1969 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1970 	debugf("phys_avail_count = %d\n", phys_avail_count);
1971 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1972 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1973 
1974 	/*******************************************************/
1975 	/* Initialize (statically allocated) kernel pmap. */
1976 	/*******************************************************/
1977 	PMAP_LOCK_INIT(kernel_pmap);
1978 #ifndef __powerpc64__
1979 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1980 #endif
1981 
1982 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1983 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1984 	for (i = 0; i < MAXCPU; i++) {
1985 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1986 
1987 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1988 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1989 	}
1990 
1991 	/* Mark kernel_pmap active on all CPUs */
1992 	CPU_FILL(&kernel_pmap->pm_active);
1993 
1994  	/*
1995 	 * Initialize the global pv list lock.
1996 	 */
1997 	rw_init(&pvh_global_lock, "pmap pv global");
1998 
1999 	/*******************************************************/
2000 	/* Final setup */
2001 	/*******************************************************/
2002 
2003 	/* Enter kstack0 into kernel map, provide guard page */
2004 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
2005 	thread0.td_kstack = kstack0;
2006 	thread0.td_kstack_pages = kstack_pages;
2007 
2008 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
2009 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
2010 	    kstack0_phys, kstack0_phys + kstack0_sz);
2011 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
2012 	    kstack0, kstack0 + kstack0_sz);
2013 
2014 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
2015 	for (i = 0; i < kstack_pages; i++) {
2016 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
2017 		kstack0 += PAGE_SIZE;
2018 		kstack0_phys += PAGE_SIZE;
2019 	}
2020 
2021 	pmap_bootstrapped = 1;
2022 
2023 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
2024 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
2025 
2026 	debugf("mmu_booke_bootstrap: exit\n");
2027 }
2028 
2029 #ifdef SMP
2030  void
2031 tlb1_ap_prep(void)
2032 {
2033 	tlb_entry_t *e, tmp;
2034 	unsigned int i;
2035 
2036 	/* Prepare TLB1 image for AP processors */
2037 	e = __boot_tlb1;
2038 	for (i = 0; i < TLB1_ENTRIES; i++) {
2039 		tlb1_read_entry(&tmp, i);
2040 
2041 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
2042 			memcpy(e++, &tmp, sizeof(tmp));
2043 	}
2044 }
2045 
2046 void
2047 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
2048 {
2049 	int i;
2050 
2051 	/*
2052 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
2053 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
2054 	 * created by tlb1_ap_prep(), so use these values directly to
2055 	 * (re)program AP's TLB1 hardware.
2056 	 *
2057 	 * Start at index 1 because index 0 has the kernel map.
2058 	 */
2059 	for (i = 1; i < TLB1_ENTRIES; i++) {
2060 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
2061 			tlb1_write_entry(&__boot_tlb1[i], i);
2062 	}
2063 
2064 	set_mas4_defaults();
2065 }
2066 #endif
2067 
2068 static void
2069 booke_pmap_init_qpages(void)
2070 {
2071 	struct pcpu *pc;
2072 	int i;
2073 
2074 	CPU_FOREACH(i) {
2075 		pc = pcpu_find(i);
2076 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
2077 		if (pc->pc_qmap_addr == 0)
2078 			panic("pmap_init_qpages: unable to allocate KVA");
2079 	}
2080 }
2081 
2082 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
2083 
2084 /*
2085  * Get the physical page address for the given pmap/virtual address.
2086  */
2087 static vm_paddr_t
2088 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
2089 {
2090 	vm_paddr_t pa;
2091 
2092 	PMAP_LOCK(pmap);
2093 	pa = pte_vatopa(mmu, pmap, va);
2094 	PMAP_UNLOCK(pmap);
2095 
2096 	return (pa);
2097 }
2098 
2099 /*
2100  * Extract the physical page address associated with the given
2101  * kernel virtual address.
2102  */
2103 static vm_paddr_t
2104 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
2105 {
2106 	tlb_entry_t e;
2107 	vm_paddr_t p = 0;
2108 	int i;
2109 
2110 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
2111 		p = pte_vatopa(mmu, kernel_pmap, va);
2112 
2113 	if (p == 0) {
2114 		/* Check TLB1 mappings */
2115 		for (i = 0; i < TLB1_ENTRIES; i++) {
2116 			tlb1_read_entry(&e, i);
2117 			if (!(e.mas1 & MAS1_VALID))
2118 				continue;
2119 			if (va >= e.virt && va < e.virt + e.size)
2120 				return (e.phys + (va - e.virt));
2121 		}
2122 	}
2123 
2124 	return (p);
2125 }
2126 
2127 /*
2128  * Initialize the pmap module.
2129  * Called by vm_init, to initialize any structures that the pmap
2130  * system needs to map virtual memory.
2131  */
2132 static void
2133 mmu_booke_init(mmu_t mmu)
2134 {
2135 	int shpgperproc = PMAP_SHPGPERPROC;
2136 
2137 	/*
2138 	 * Initialize the address space (zone) for the pv entries.  Set a
2139 	 * high water mark so that the system can recover from excessive
2140 	 * numbers of pv entries.
2141 	 */
2142 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2143 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2144 
2145 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2146 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2147 
2148 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2149 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2150 
2151 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2152 
2153 	/* Pre-fill pvzone with initial number of pv entries. */
2154 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2155 
2156 	/* Initialize ptbl allocation. */
2157 	ptbl_init();
2158 }
2159 
2160 /*
2161  * Map a list of wired pages into kernel virtual address space.  This is
2162  * intended for temporary mappings which do not need page modification or
2163  * references recorded.  Existing mappings in the region are overwritten.
2164  */
2165 static void
2166 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2167 {
2168 	vm_offset_t va;
2169 
2170 	va = sva;
2171 	while (count-- > 0) {
2172 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2173 		va += PAGE_SIZE;
2174 		m++;
2175 	}
2176 }
2177 
2178 /*
2179  * Remove page mappings from kernel virtual address space.  Intended for
2180  * temporary mappings entered by mmu_booke_qenter.
2181  */
2182 static void
2183 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2184 {
2185 	vm_offset_t va;
2186 
2187 	va = sva;
2188 	while (count-- > 0) {
2189 		mmu_booke_kremove(mmu, va);
2190 		va += PAGE_SIZE;
2191 	}
2192 }
2193 
2194 /*
2195  * Map a wired page into kernel virtual address space.
2196  */
2197 static void
2198 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2199 {
2200 
2201 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2202 }
2203 
2204 static void
2205 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2206 {
2207 	uint32_t flags;
2208 	pte_t *pte;
2209 
2210 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2211 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2212 
2213 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2214 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2215 	flags |= PTE_PS_4KB;
2216 
2217 	pte = pte_find(mmu, kernel_pmap, va);
2218 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2219 
2220 	mtx_lock_spin(&tlbivax_mutex);
2221 	tlb_miss_lock();
2222 
2223 	if (PTE_ISVALID(pte)) {
2224 
2225 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2226 
2227 		/* Flush entry from TLB0 */
2228 		tlb0_flush_entry(va);
2229 	}
2230 
2231 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2232 
2233 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2234 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2235 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2236 
2237 	/* Flush the real memory from the instruction cache. */
2238 	if ((flags & (PTE_I | PTE_G)) == 0)
2239 		__syncicache((void *)va, PAGE_SIZE);
2240 
2241 	tlb_miss_unlock();
2242 	mtx_unlock_spin(&tlbivax_mutex);
2243 }
2244 
2245 /*
2246  * Remove a page from kernel page table.
2247  */
2248 static void
2249 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2250 {
2251 	pte_t *pte;
2252 
2253 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
2254 
2255 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2256 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2257 	    ("mmu_booke_kremove: invalid va"));
2258 
2259 	pte = pte_find(mmu, kernel_pmap, va);
2260 
2261 	if (!PTE_ISVALID(pte)) {
2262 
2263 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2264 
2265 		return;
2266 	}
2267 
2268 	mtx_lock_spin(&tlbivax_mutex);
2269 	tlb_miss_lock();
2270 
2271 	/* Invalidate entry in TLB0, update PTE. */
2272 	tlb0_flush_entry(va);
2273 	*pte = 0;
2274 
2275 	tlb_miss_unlock();
2276 	mtx_unlock_spin(&tlbivax_mutex);
2277 }
2278 
2279 /*
2280  * Provide a kernel pointer corresponding to a given userland pointer.
2281  * The returned pointer is valid until the next time this function is
2282  * called in this thread. This is used internally in copyin/copyout.
2283  */
2284 int
2285 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2286     void **kaddr, size_t ulen, size_t *klen)
2287 {
2288 
2289 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2290 		return (EFAULT);
2291 
2292 	*kaddr = (void *)(uintptr_t)uaddr;
2293 	if (klen)
2294 		*klen = ulen;
2295 
2296 	return (0);
2297 }
2298 
2299 /*
2300  * Figure out where a given kernel pointer (usually in a fault) points
2301  * to from the VM's perspective, potentially remapping into userland's
2302  * address space.
2303  */
2304 static int
2305 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2306     vm_offset_t *decoded_addr)
2307 {
2308 
2309 	if (addr < VM_MAXUSER_ADDRESS)
2310 		*is_user = 1;
2311 	else
2312 		*is_user = 0;
2313 
2314 	*decoded_addr = addr;
2315 	return (0);
2316 }
2317 
2318 /*
2319  * Initialize pmap associated with process 0.
2320  */
2321 static void
2322 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2323 {
2324 
2325 	PMAP_LOCK_INIT(pmap);
2326 	mmu_booke_pinit(mmu, pmap);
2327 	PCPU_SET(curpmap, pmap);
2328 }
2329 
2330 /*
2331  * Initialize a preallocated and zeroed pmap structure,
2332  * such as one in a vmspace structure.
2333  */
2334 static void
2335 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2336 {
2337 	int i;
2338 
2339 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2340 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2341 
2342 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2343 
2344 	for (i = 0; i < MAXCPU; i++)
2345 		pmap->pm_tid[i] = TID_NONE;
2346 	CPU_ZERO(&kernel_pmap->pm_active);
2347 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2348 #ifdef __powerpc64__
2349 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2350 	TAILQ_INIT(&pmap->pm_pdir_list);
2351 #else
2352 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2353 #endif
2354 	TAILQ_INIT(&pmap->pm_ptbl_list);
2355 }
2356 
2357 /*
2358  * Release any resources held by the given physical map.
2359  * Called when a pmap initialized by mmu_booke_pinit is being released.
2360  * Should only be called if the map contains no valid mappings.
2361  */
2362 static void
2363 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2364 {
2365 
2366 	KASSERT(pmap->pm_stats.resident_count == 0,
2367 	    ("pmap_release: pmap resident count %ld != 0",
2368 	    pmap->pm_stats.resident_count));
2369 }
2370 
2371 /*
2372  * Insert the given physical page at the specified virtual address in the
2373  * target physical map with the protection requested. If specified the page
2374  * will be wired down.
2375  */
2376 static int
2377 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2378     vm_prot_t prot, u_int flags, int8_t psind)
2379 {
2380 	int error;
2381 
2382 	rw_wlock(&pvh_global_lock);
2383 	PMAP_LOCK(pmap);
2384 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2385 	PMAP_UNLOCK(pmap);
2386 	rw_wunlock(&pvh_global_lock);
2387 	return (error);
2388 }
2389 
2390 static int
2391 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2392     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2393 {
2394 	pte_t *pte;
2395 	vm_paddr_t pa;
2396 	uint32_t flags;
2397 	int error, su, sync;
2398 
2399 	pa = VM_PAGE_TO_PHYS(m);
2400 	su = (pmap == kernel_pmap);
2401 	sync = 0;
2402 
2403 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2404 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2405 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2406 	//		(u_int32_t)m, va, pa, prot, flags);
2407 
2408 	if (su) {
2409 		KASSERT(((va >= virtual_avail) &&
2410 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2411 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2412 	} else {
2413 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2414 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2415 	}
2416 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2417 		VM_OBJECT_ASSERT_LOCKED(m->object);
2418 
2419 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2420 
2421 	/*
2422 	 * If there is an existing mapping, and the physical address has not
2423 	 * changed, must be protection or wiring change.
2424 	 */
2425 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2426 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2427 
2428 		/*
2429 		 * Before actually updating pte->flags we calculate and
2430 		 * prepare its new value in a helper var.
2431 		 */
2432 		flags = *pte;
2433 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2434 
2435 		/* Wiring change, just update stats. */
2436 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2437 			if (!PTE_ISWIRED(pte)) {
2438 				flags |= PTE_WIRED;
2439 				pmap->pm_stats.wired_count++;
2440 			}
2441 		} else {
2442 			if (PTE_ISWIRED(pte)) {
2443 				flags &= ~PTE_WIRED;
2444 				pmap->pm_stats.wired_count--;
2445 			}
2446 		}
2447 
2448 		if (prot & VM_PROT_WRITE) {
2449 			/* Add write permissions. */
2450 			flags |= PTE_SW;
2451 			if (!su)
2452 				flags |= PTE_UW;
2453 
2454 			if ((flags & PTE_MANAGED) != 0)
2455 				vm_page_aflag_set(m, PGA_WRITEABLE);
2456 		} else {
2457 			/* Handle modified pages, sense modify status. */
2458 
2459 			/*
2460 			 * The PTE_MODIFIED flag could be set by underlying
2461 			 * TLB misses since we last read it (above), possibly
2462 			 * other CPUs could update it so we check in the PTE
2463 			 * directly rather than rely on that saved local flags
2464 			 * copy.
2465 			 */
2466 			if (PTE_ISMODIFIED(pte))
2467 				vm_page_dirty(m);
2468 		}
2469 
2470 		if (prot & VM_PROT_EXECUTE) {
2471 			flags |= PTE_SX;
2472 			if (!su)
2473 				flags |= PTE_UX;
2474 
2475 			/*
2476 			 * Check existing flags for execute permissions: if we
2477 			 * are turning execute permissions on, icache should
2478 			 * be flushed.
2479 			 */
2480 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2481 				sync++;
2482 		}
2483 
2484 		flags &= ~PTE_REFERENCED;
2485 
2486 		/*
2487 		 * The new flags value is all calculated -- only now actually
2488 		 * update the PTE.
2489 		 */
2490 		mtx_lock_spin(&tlbivax_mutex);
2491 		tlb_miss_lock();
2492 
2493 		tlb0_flush_entry(va);
2494 		*pte &= ~PTE_FLAGS_MASK;
2495 		*pte |= flags;
2496 
2497 		tlb_miss_unlock();
2498 		mtx_unlock_spin(&tlbivax_mutex);
2499 
2500 	} else {
2501 		/*
2502 		 * If there is an existing mapping, but it's for a different
2503 		 * physical address, pte_enter() will delete the old mapping.
2504 		 */
2505 		//if ((pte != NULL) && PTE_ISVALID(pte))
2506 		//	debugf("mmu_booke_enter_locked: replace\n");
2507 		//else
2508 		//	debugf("mmu_booke_enter_locked: new\n");
2509 
2510 		/* Now set up the flags and install the new mapping. */
2511 		flags = (PTE_SR | PTE_VALID);
2512 		flags |= PTE_M;
2513 
2514 		if (!su)
2515 			flags |= PTE_UR;
2516 
2517 		if (prot & VM_PROT_WRITE) {
2518 			flags |= PTE_SW;
2519 			if (!su)
2520 				flags |= PTE_UW;
2521 
2522 			if ((m->oflags & VPO_UNMANAGED) == 0)
2523 				vm_page_aflag_set(m, PGA_WRITEABLE);
2524 		}
2525 
2526 		if (prot & VM_PROT_EXECUTE) {
2527 			flags |= PTE_SX;
2528 			if (!su)
2529 				flags |= PTE_UX;
2530 		}
2531 
2532 		/* If its wired update stats. */
2533 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2534 			flags |= PTE_WIRED;
2535 
2536 		error = pte_enter(mmu, pmap, m, va, flags,
2537 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2538 		if (error != 0)
2539 			return (KERN_RESOURCE_SHORTAGE);
2540 
2541 		if ((flags & PMAP_ENTER_WIRED) != 0)
2542 			pmap->pm_stats.wired_count++;
2543 
2544 		/* Flush the real memory from the instruction cache. */
2545 		if (prot & VM_PROT_EXECUTE)
2546 			sync++;
2547 	}
2548 
2549 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2550 		__syncicache((void *)va, PAGE_SIZE);
2551 		sync = 0;
2552 	}
2553 
2554 	return (KERN_SUCCESS);
2555 }
2556 
2557 /*
2558  * Maps a sequence of resident pages belonging to the same object.
2559  * The sequence begins with the given page m_start.  This page is
2560  * mapped at the given virtual address start.  Each subsequent page is
2561  * mapped at a virtual address that is offset from start by the same
2562  * amount as the page is offset from m_start within the object.  The
2563  * last page in the sequence is the page with the largest offset from
2564  * m_start that can be mapped at a virtual address less than the given
2565  * virtual address end.  Not every virtual page between start and end
2566  * is mapped; only those for which a resident page exists with the
2567  * corresponding offset from m_start are mapped.
2568  */
2569 static void
2570 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2571     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2572 {
2573 	vm_page_t m;
2574 	vm_pindex_t diff, psize;
2575 
2576 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2577 
2578 	psize = atop(end - start);
2579 	m = m_start;
2580 	rw_wlock(&pvh_global_lock);
2581 	PMAP_LOCK(pmap);
2582 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2583 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2584 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2585 		    PMAP_ENTER_NOSLEEP, 0);
2586 		m = TAILQ_NEXT(m, listq);
2587 	}
2588 	rw_wunlock(&pvh_global_lock);
2589 	PMAP_UNLOCK(pmap);
2590 }
2591 
2592 static void
2593 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2594     vm_prot_t prot)
2595 {
2596 
2597 	rw_wlock(&pvh_global_lock);
2598 	PMAP_LOCK(pmap);
2599 	mmu_booke_enter_locked(mmu, pmap, va, m,
2600 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2601 	    0);
2602 	rw_wunlock(&pvh_global_lock);
2603 	PMAP_UNLOCK(pmap);
2604 }
2605 
2606 /*
2607  * Remove the given range of addresses from the specified map.
2608  *
2609  * It is assumed that the start and end are properly rounded to the page size.
2610  */
2611 static void
2612 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2613 {
2614 	pte_t *pte;
2615 	uint8_t hold_flag;
2616 
2617 	int su = (pmap == kernel_pmap);
2618 
2619 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2620 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2621 
2622 	if (su) {
2623 		KASSERT(((va >= virtual_avail) &&
2624 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2625 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2626 	} else {
2627 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2628 		    ("mmu_booke_remove: user pmap, non user va"));
2629 	}
2630 
2631 	if (PMAP_REMOVE_DONE(pmap)) {
2632 		//debugf("mmu_booke_remove: e (empty)\n");
2633 		return;
2634 	}
2635 
2636 	hold_flag = PTBL_HOLD_FLAG(pmap);
2637 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2638 
2639 	rw_wlock(&pvh_global_lock);
2640 	PMAP_LOCK(pmap);
2641 	for (; va < endva; va += PAGE_SIZE) {
2642 		pte = pte_find(mmu, pmap, va);
2643 		if ((pte != NULL) && PTE_ISVALID(pte))
2644 			pte_remove(mmu, pmap, va, hold_flag);
2645 	}
2646 	PMAP_UNLOCK(pmap);
2647 	rw_wunlock(&pvh_global_lock);
2648 
2649 	//debugf("mmu_booke_remove: e\n");
2650 }
2651 
2652 /*
2653  * Remove physical page from all pmaps in which it resides.
2654  */
2655 static void
2656 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2657 {
2658 	pv_entry_t pv, pvn;
2659 	uint8_t hold_flag;
2660 
2661 	rw_wlock(&pvh_global_lock);
2662 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2663 		pvn = TAILQ_NEXT(pv, pv_link);
2664 
2665 		PMAP_LOCK(pv->pv_pmap);
2666 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2667 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2668 		PMAP_UNLOCK(pv->pv_pmap);
2669 	}
2670 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2671 	rw_wunlock(&pvh_global_lock);
2672 }
2673 
2674 /*
2675  * Map a range of physical addresses into kernel virtual address space.
2676  */
2677 static vm_offset_t
2678 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2679     vm_paddr_t pa_end, int prot)
2680 {
2681 	vm_offset_t sva = *virt;
2682 	vm_offset_t va = sva;
2683 
2684 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2685 	//		sva, pa_start, pa_end);
2686 
2687 	while (pa_start < pa_end) {
2688 		mmu_booke_kenter(mmu, va, pa_start);
2689 		va += PAGE_SIZE;
2690 		pa_start += PAGE_SIZE;
2691 	}
2692 	*virt = va;
2693 
2694 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2695 	return (sva);
2696 }
2697 
2698 /*
2699  * The pmap must be activated before it's address space can be accessed in any
2700  * way.
2701  */
2702 static void
2703 mmu_booke_activate(mmu_t mmu, struct thread *td)
2704 {
2705 	pmap_t pmap;
2706 	u_int cpuid;
2707 
2708 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2709 
2710 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
2711 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2712 
2713 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2714 
2715 	sched_pin();
2716 
2717 	cpuid = PCPU_GET(cpuid);
2718 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2719 	PCPU_SET(curpmap, pmap);
2720 
2721 	if (pmap->pm_tid[cpuid] == TID_NONE)
2722 		tid_alloc(pmap);
2723 
2724 	/* Load PID0 register with pmap tid value. */
2725 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2726 	__asm __volatile("isync");
2727 
2728 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2729 
2730 	sched_unpin();
2731 
2732 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2733 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2734 }
2735 
2736 /*
2737  * Deactivate the specified process's address space.
2738  */
2739 static void
2740 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2741 {
2742 	pmap_t pmap;
2743 
2744 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2745 
2746 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2747 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2748 
2749 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2750 
2751 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2752 	PCPU_SET(curpmap, NULL);
2753 }
2754 
2755 /*
2756  * Copy the range specified by src_addr/len
2757  * from the source map to the range dst_addr/len
2758  * in the destination map.
2759  *
2760  * This routine is only advisory and need not do anything.
2761  */
2762 static void
2763 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2764     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2765 {
2766 
2767 }
2768 
2769 /*
2770  * Set the physical protection on the specified range of this map as requested.
2771  */
2772 static void
2773 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2774     vm_prot_t prot)
2775 {
2776 	vm_offset_t va;
2777 	vm_page_t m;
2778 	pte_t *pte;
2779 
2780 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2781 		mmu_booke_remove(mmu, pmap, sva, eva);
2782 		return;
2783 	}
2784 
2785 	if (prot & VM_PROT_WRITE)
2786 		return;
2787 
2788 	PMAP_LOCK(pmap);
2789 	for (va = sva; va < eva; va += PAGE_SIZE) {
2790 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2791 			if (PTE_ISVALID(pte)) {
2792 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2793 
2794 				mtx_lock_spin(&tlbivax_mutex);
2795 				tlb_miss_lock();
2796 
2797 				/* Handle modified pages. */
2798 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2799 					vm_page_dirty(m);
2800 
2801 				tlb0_flush_entry(va);
2802 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2803 
2804 				tlb_miss_unlock();
2805 				mtx_unlock_spin(&tlbivax_mutex);
2806 			}
2807 		}
2808 	}
2809 	PMAP_UNLOCK(pmap);
2810 }
2811 
2812 /*
2813  * Clear the write and modified bits in each of the given page's mappings.
2814  */
2815 static void
2816 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2817 {
2818 	pv_entry_t pv;
2819 	pte_t *pte;
2820 
2821 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2822 	    ("mmu_booke_remove_write: page %p is not managed", m));
2823 
2824 	/*
2825 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2826 	 * set by another thread while the object is locked.  Thus,
2827 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2828 	 */
2829 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2830 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2831 		return;
2832 	rw_wlock(&pvh_global_lock);
2833 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2834 		PMAP_LOCK(pv->pv_pmap);
2835 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2836 			if (PTE_ISVALID(pte)) {
2837 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2838 
2839 				mtx_lock_spin(&tlbivax_mutex);
2840 				tlb_miss_lock();
2841 
2842 				/* Handle modified pages. */
2843 				if (PTE_ISMODIFIED(pte))
2844 					vm_page_dirty(m);
2845 
2846 				/* Flush mapping from TLB0. */
2847 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2848 
2849 				tlb_miss_unlock();
2850 				mtx_unlock_spin(&tlbivax_mutex);
2851 			}
2852 		}
2853 		PMAP_UNLOCK(pv->pv_pmap);
2854 	}
2855 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2856 	rw_wunlock(&pvh_global_lock);
2857 }
2858 
2859 static void
2860 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2861 {
2862 	pte_t *pte;
2863 	pmap_t pmap;
2864 	vm_page_t m;
2865 	vm_offset_t addr;
2866 	vm_paddr_t pa = 0;
2867 	int active, valid;
2868 
2869 	va = trunc_page(va);
2870 	sz = round_page(sz);
2871 
2872 	rw_wlock(&pvh_global_lock);
2873 	pmap = PCPU_GET(curpmap);
2874 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2875 	while (sz > 0) {
2876 		PMAP_LOCK(pm);
2877 		pte = pte_find(mmu, pm, va);
2878 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2879 		if (valid)
2880 			pa = PTE_PA(pte);
2881 		PMAP_UNLOCK(pm);
2882 		if (valid) {
2883 			if (!active) {
2884 				/* Create a mapping in the active pmap. */
2885 				addr = 0;
2886 				m = PHYS_TO_VM_PAGE(pa);
2887 				PMAP_LOCK(pmap);
2888 				pte_enter(mmu, pmap, m, addr,
2889 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2890 				__syncicache((void *)addr, PAGE_SIZE);
2891 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2892 				PMAP_UNLOCK(pmap);
2893 			} else
2894 				__syncicache((void *)va, PAGE_SIZE);
2895 		}
2896 		va += PAGE_SIZE;
2897 		sz -= PAGE_SIZE;
2898 	}
2899 	rw_wunlock(&pvh_global_lock);
2900 }
2901 
2902 /*
2903  * Atomically extract and hold the physical page with the given
2904  * pmap and virtual address pair if that mapping permits the given
2905  * protection.
2906  */
2907 static vm_page_t
2908 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2909     vm_prot_t prot)
2910 {
2911 	pte_t *pte;
2912 	vm_page_t m;
2913 	uint32_t pte_wbit;
2914 	vm_paddr_t pa;
2915 
2916 	m = NULL;
2917 	pa = 0;
2918 	PMAP_LOCK(pmap);
2919 retry:
2920 	pte = pte_find(mmu, pmap, va);
2921 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2922 		if (pmap == kernel_pmap)
2923 			pte_wbit = PTE_SW;
2924 		else
2925 			pte_wbit = PTE_UW;
2926 
2927 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2928 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2929 				goto retry;
2930 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2931 			vm_page_hold(m);
2932 		}
2933 	}
2934 
2935 	PA_UNLOCK_COND(pa);
2936 	PMAP_UNLOCK(pmap);
2937 	return (m);
2938 }
2939 
2940 /*
2941  * Initialize a vm_page's machine-dependent fields.
2942  */
2943 static void
2944 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2945 {
2946 
2947 	m->md.pv_tracked = 0;
2948 	TAILQ_INIT(&m->md.pv_list);
2949 }
2950 
2951 /*
2952  * mmu_booke_zero_page_area zeros the specified hardware page by
2953  * mapping it into virtual memory and using bzero to clear
2954  * its contents.
2955  *
2956  * off and size must reside within a single page.
2957  */
2958 static void
2959 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2960 {
2961 	vm_offset_t va;
2962 
2963 	/* XXX KASSERT off and size are within a single page? */
2964 
2965 	mtx_lock(&zero_page_mutex);
2966 	va = zero_page_va;
2967 
2968 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2969 	bzero((caddr_t)va + off, size);
2970 	mmu_booke_kremove(mmu, va);
2971 
2972 	mtx_unlock(&zero_page_mutex);
2973 }
2974 
2975 /*
2976  * mmu_booke_zero_page zeros the specified hardware page.
2977  */
2978 static void
2979 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2980 {
2981 	vm_offset_t off, va;
2982 
2983 	mtx_lock(&zero_page_mutex);
2984 	va = zero_page_va;
2985 
2986 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2987 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2988 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2989 	mmu_booke_kremove(mmu, va);
2990 
2991 	mtx_unlock(&zero_page_mutex);
2992 }
2993 
2994 /*
2995  * mmu_booke_copy_page copies the specified (machine independent) page by
2996  * mapping the page into virtual memory and using memcopy to copy the page,
2997  * one machine dependent page at a time.
2998  */
2999 static void
3000 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
3001 {
3002 	vm_offset_t sva, dva;
3003 
3004 	sva = copy_page_src_va;
3005 	dva = copy_page_dst_va;
3006 
3007 	mtx_lock(&copy_page_mutex);
3008 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
3009 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
3010 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3011 	mmu_booke_kremove(mmu, dva);
3012 	mmu_booke_kremove(mmu, sva);
3013 	mtx_unlock(&copy_page_mutex);
3014 }
3015 
3016 static inline void
3017 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
3018     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
3019 {
3020 	void *a_cp, *b_cp;
3021 	vm_offset_t a_pg_offset, b_pg_offset;
3022 	int cnt;
3023 
3024 	mtx_lock(&copy_page_mutex);
3025 	while (xfersize > 0) {
3026 		a_pg_offset = a_offset & PAGE_MASK;
3027 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3028 		mmu_booke_kenter(mmu, copy_page_src_va,
3029 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
3030 		a_cp = (char *)copy_page_src_va + a_pg_offset;
3031 		b_pg_offset = b_offset & PAGE_MASK;
3032 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3033 		mmu_booke_kenter(mmu, copy_page_dst_va,
3034 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
3035 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
3036 		bcopy(a_cp, b_cp, cnt);
3037 		mmu_booke_kremove(mmu, copy_page_dst_va);
3038 		mmu_booke_kremove(mmu, copy_page_src_va);
3039 		a_offset += cnt;
3040 		b_offset += cnt;
3041 		xfersize -= cnt;
3042 	}
3043 	mtx_unlock(&copy_page_mutex);
3044 }
3045 
3046 static vm_offset_t
3047 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
3048 {
3049 	vm_paddr_t paddr;
3050 	vm_offset_t qaddr;
3051 	uint32_t flags;
3052 	pte_t *pte;
3053 
3054 	paddr = VM_PAGE_TO_PHYS(m);
3055 
3056 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3057 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3058 	flags |= PTE_PS_4KB;
3059 
3060 	critical_enter();
3061 	qaddr = PCPU_GET(qmap_addr);
3062 
3063 	pte = pte_find(mmu, kernel_pmap, qaddr);
3064 
3065 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3066 
3067 	/*
3068 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3069  	 * not be present in other TLBs.  Is there a better instruction
3070 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3071 	 */
3072 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3073 	__asm __volatile("isync; msync");
3074 
3075 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3076 
3077 	/* Flush the real memory from the instruction cache. */
3078 	if ((flags & (PTE_I | PTE_G)) == 0)
3079 		__syncicache((void *)qaddr, PAGE_SIZE);
3080 
3081 	return (qaddr);
3082 }
3083 
3084 static void
3085 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3086 {
3087 	pte_t *pte;
3088 
3089 	pte = pte_find(mmu, kernel_pmap, addr);
3090 
3091 	KASSERT(PCPU_GET(qmap_addr) == addr,
3092 	    ("mmu_booke_quick_remove_page: invalid address"));
3093 	KASSERT(*pte != 0,
3094 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3095 
3096 	*pte = 0;
3097 	critical_exit();
3098 }
3099 
3100 /*
3101  * Return whether or not the specified physical page was modified
3102  * in any of physical maps.
3103  */
3104 static boolean_t
3105 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3106 {
3107 	pte_t *pte;
3108 	pv_entry_t pv;
3109 	boolean_t rv;
3110 
3111 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3112 	    ("mmu_booke_is_modified: page %p is not managed", m));
3113 	rv = FALSE;
3114 
3115 	/*
3116 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3117 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3118 	 * is clear, no PTEs can be modified.
3119 	 */
3120 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3121 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3122 		return (rv);
3123 	rw_wlock(&pvh_global_lock);
3124 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3125 		PMAP_LOCK(pv->pv_pmap);
3126 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3127 		    PTE_ISVALID(pte)) {
3128 			if (PTE_ISMODIFIED(pte))
3129 				rv = TRUE;
3130 		}
3131 		PMAP_UNLOCK(pv->pv_pmap);
3132 		if (rv)
3133 			break;
3134 	}
3135 	rw_wunlock(&pvh_global_lock);
3136 	return (rv);
3137 }
3138 
3139 /*
3140  * Return whether or not the specified virtual address is eligible
3141  * for prefault.
3142  */
3143 static boolean_t
3144 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3145 {
3146 
3147 	return (FALSE);
3148 }
3149 
3150 /*
3151  * Return whether or not the specified physical page was referenced
3152  * in any physical maps.
3153  */
3154 static boolean_t
3155 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3156 {
3157 	pte_t *pte;
3158 	pv_entry_t pv;
3159 	boolean_t rv;
3160 
3161 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3162 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3163 	rv = FALSE;
3164 	rw_wlock(&pvh_global_lock);
3165 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3166 		PMAP_LOCK(pv->pv_pmap);
3167 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3168 		    PTE_ISVALID(pte)) {
3169 			if (PTE_ISREFERENCED(pte))
3170 				rv = TRUE;
3171 		}
3172 		PMAP_UNLOCK(pv->pv_pmap);
3173 		if (rv)
3174 			break;
3175 	}
3176 	rw_wunlock(&pvh_global_lock);
3177 	return (rv);
3178 }
3179 
3180 /*
3181  * Clear the modify bits on the specified physical page.
3182  */
3183 static void
3184 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3185 {
3186 	pte_t *pte;
3187 	pv_entry_t pv;
3188 
3189 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3190 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3191 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3192 	KASSERT(!vm_page_xbusied(m),
3193 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3194 
3195 	/*
3196 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3197 	 * If the object containing the page is locked and the page is not
3198 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3199 	 */
3200 	if ((m->aflags & PGA_WRITEABLE) == 0)
3201 		return;
3202 	rw_wlock(&pvh_global_lock);
3203 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3204 		PMAP_LOCK(pv->pv_pmap);
3205 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3206 		    PTE_ISVALID(pte)) {
3207 			mtx_lock_spin(&tlbivax_mutex);
3208 			tlb_miss_lock();
3209 
3210 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3211 				tlb0_flush_entry(pv->pv_va);
3212 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3213 				    PTE_REFERENCED);
3214 			}
3215 
3216 			tlb_miss_unlock();
3217 			mtx_unlock_spin(&tlbivax_mutex);
3218 		}
3219 		PMAP_UNLOCK(pv->pv_pmap);
3220 	}
3221 	rw_wunlock(&pvh_global_lock);
3222 }
3223 
3224 /*
3225  * Return a count of reference bits for a page, clearing those bits.
3226  * It is not necessary for every reference bit to be cleared, but it
3227  * is necessary that 0 only be returned when there are truly no
3228  * reference bits set.
3229  *
3230  * As an optimization, update the page's dirty field if a modified bit is
3231  * found while counting reference bits.  This opportunistic update can be
3232  * performed at low cost and can eliminate the need for some future calls
3233  * to pmap_is_modified().  However, since this function stops after
3234  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3235  * dirty pages.  Those dirty pages will only be detected by a future call
3236  * to pmap_is_modified().
3237  */
3238 static int
3239 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3240 {
3241 	pte_t *pte;
3242 	pv_entry_t pv;
3243 	int count;
3244 
3245 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3246 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3247 	count = 0;
3248 	rw_wlock(&pvh_global_lock);
3249 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3250 		PMAP_LOCK(pv->pv_pmap);
3251 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3252 		    PTE_ISVALID(pte)) {
3253 			if (PTE_ISMODIFIED(pte))
3254 				vm_page_dirty(m);
3255 			if (PTE_ISREFERENCED(pte)) {
3256 				mtx_lock_spin(&tlbivax_mutex);
3257 				tlb_miss_lock();
3258 
3259 				tlb0_flush_entry(pv->pv_va);
3260 				*pte &= ~PTE_REFERENCED;
3261 
3262 				tlb_miss_unlock();
3263 				mtx_unlock_spin(&tlbivax_mutex);
3264 
3265 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3266 					PMAP_UNLOCK(pv->pv_pmap);
3267 					break;
3268 				}
3269 			}
3270 		}
3271 		PMAP_UNLOCK(pv->pv_pmap);
3272 	}
3273 	rw_wunlock(&pvh_global_lock);
3274 	return (count);
3275 }
3276 
3277 /*
3278  * Clear the wired attribute from the mappings for the specified range of
3279  * addresses in the given pmap.  Every valid mapping within that range must
3280  * have the wired attribute set.  In contrast, invalid mappings cannot have
3281  * the wired attribute set, so they are ignored.
3282  *
3283  * The wired attribute of the page table entry is not a hardware feature, so
3284  * there is no need to invalidate any TLB entries.
3285  */
3286 static void
3287 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3288 {
3289 	vm_offset_t va;
3290 	pte_t *pte;
3291 
3292 	PMAP_LOCK(pmap);
3293 	for (va = sva; va < eva; va += PAGE_SIZE) {
3294 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3295 		    PTE_ISVALID(pte)) {
3296 			if (!PTE_ISWIRED(pte))
3297 				panic("mmu_booke_unwire: pte %p isn't wired",
3298 				    pte);
3299 			*pte &= ~PTE_WIRED;
3300 			pmap->pm_stats.wired_count--;
3301 		}
3302 	}
3303 	PMAP_UNLOCK(pmap);
3304 
3305 }
3306 
3307 /*
3308  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3309  * page.  This count may be changed upwards or downwards in the future; it is
3310  * only necessary that true be returned for a small subset of pmaps for proper
3311  * page aging.
3312  */
3313 static boolean_t
3314 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3315 {
3316 	pv_entry_t pv;
3317 	int loops;
3318 	boolean_t rv;
3319 
3320 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3321 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3322 	loops = 0;
3323 	rv = FALSE;
3324 	rw_wlock(&pvh_global_lock);
3325 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3326 		if (pv->pv_pmap == pmap) {
3327 			rv = TRUE;
3328 			break;
3329 		}
3330 		if (++loops >= 16)
3331 			break;
3332 	}
3333 	rw_wunlock(&pvh_global_lock);
3334 	return (rv);
3335 }
3336 
3337 /*
3338  * Return the number of managed mappings to the given physical page that are
3339  * wired.
3340  */
3341 static int
3342 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3343 {
3344 	pv_entry_t pv;
3345 	pte_t *pte;
3346 	int count = 0;
3347 
3348 	if ((m->oflags & VPO_UNMANAGED) != 0)
3349 		return (count);
3350 	rw_wlock(&pvh_global_lock);
3351 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3352 		PMAP_LOCK(pv->pv_pmap);
3353 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3354 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3355 				count++;
3356 		PMAP_UNLOCK(pv->pv_pmap);
3357 	}
3358 	rw_wunlock(&pvh_global_lock);
3359 	return (count);
3360 }
3361 
3362 static int
3363 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3364 {
3365 	int i;
3366 	vm_offset_t va;
3367 
3368 	/*
3369 	 * This currently does not work for entries that
3370 	 * overlap TLB1 entries.
3371 	 */
3372 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3373 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3374 			return (0);
3375 	}
3376 
3377 	return (EFAULT);
3378 }
3379 
3380 void
3381 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3382 {
3383 	vm_paddr_t ppa;
3384 	vm_offset_t ofs;
3385 	vm_size_t gran;
3386 
3387 	/* Minidumps are based on virtual memory addresses. */
3388 	if (do_minidump) {
3389 		*va = (void *)(vm_offset_t)pa;
3390 		return;
3391 	}
3392 
3393 	/* Raw physical memory dumps don't have a virtual address. */
3394 	/* We always map a 256MB page at 256M. */
3395 	gran = 256 * 1024 * 1024;
3396 	ppa = rounddown2(pa, gran);
3397 	ofs = pa - ppa;
3398 	*va = (void *)gran;
3399 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3400 
3401 	if (sz > (gran - ofs))
3402 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3403 		    _TLB_ENTRY_IO);
3404 }
3405 
3406 void
3407 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3408 {
3409 	vm_paddr_t ppa;
3410 	vm_offset_t ofs;
3411 	vm_size_t gran;
3412 	tlb_entry_t e;
3413 	int i;
3414 
3415 	/* Minidumps are based on virtual memory addresses. */
3416 	/* Nothing to do... */
3417 	if (do_minidump)
3418 		return;
3419 
3420 	for (i = 0; i < TLB1_ENTRIES; i++) {
3421 		tlb1_read_entry(&e, i);
3422 		if (!(e.mas1 & MAS1_VALID))
3423 			break;
3424 	}
3425 
3426 	/* Raw physical memory dumps don't have a virtual address. */
3427 	i--;
3428 	e.mas1 = 0;
3429 	e.mas2 = 0;
3430 	e.mas3 = 0;
3431 	tlb1_write_entry(&e, i);
3432 
3433 	gran = 256 * 1024 * 1024;
3434 	ppa = rounddown2(pa, gran);
3435 	ofs = pa - ppa;
3436 	if (sz > (gran - ofs)) {
3437 		i--;
3438 		e.mas1 = 0;
3439 		e.mas2 = 0;
3440 		e.mas3 = 0;
3441 		tlb1_write_entry(&e, i);
3442 	}
3443 }
3444 
3445 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3446 
3447 void
3448 mmu_booke_scan_init(mmu_t mmu)
3449 {
3450 	vm_offset_t va;
3451 	pte_t *pte;
3452 	int i;
3453 
3454 	if (!do_minidump) {
3455 		/* Initialize phys. segments for dumpsys(). */
3456 		memset(&dump_map, 0, sizeof(dump_map));
3457 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3458 		    &availmem_regions_sz);
3459 		for (i = 0; i < physmem_regions_sz; i++) {
3460 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3461 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3462 		}
3463 		return;
3464 	}
3465 
3466 	/* Virtual segments for minidumps: */
3467 	memset(&dump_map, 0, sizeof(dump_map));
3468 
3469 	/* 1st: kernel .data and .bss. */
3470 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3471 	dump_map[0].pa_size =
3472 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3473 
3474 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3475 	dump_map[1].pa_start = data_start;
3476 	dump_map[1].pa_size = data_end - data_start;
3477 
3478 	/* 3rd: kernel VM. */
3479 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3480 	/* Find start of next chunk (from va). */
3481 	while (va < virtual_end) {
3482 		/* Don't dump the buffer cache. */
3483 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3484 			va = kmi.buffer_eva;
3485 			continue;
3486 		}
3487 		pte = pte_find(mmu, kernel_pmap, va);
3488 		if (pte != NULL && PTE_ISVALID(pte))
3489 			break;
3490 		va += PAGE_SIZE;
3491 	}
3492 	if (va < virtual_end) {
3493 		dump_map[2].pa_start = va;
3494 		va += PAGE_SIZE;
3495 		/* Find last page in chunk. */
3496 		while (va < virtual_end) {
3497 			/* Don't run into the buffer cache. */
3498 			if (va == kmi.buffer_sva)
3499 				break;
3500 			pte = pte_find(mmu, kernel_pmap, va);
3501 			if (pte == NULL || !PTE_ISVALID(pte))
3502 				break;
3503 			va += PAGE_SIZE;
3504 		}
3505 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3506 	}
3507 }
3508 
3509 /*
3510  * Map a set of physical memory pages into the kernel virtual address space.
3511  * Return a pointer to where it is mapped. This routine is intended to be used
3512  * for mapping device memory, NOT real memory.
3513  */
3514 static void *
3515 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3516 {
3517 
3518 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3519 }
3520 
3521 static void *
3522 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3523 {
3524 	tlb_entry_t e;
3525 	void *res;
3526 	uintptr_t va, tmpva;
3527 	vm_size_t sz;
3528 	int i;
3529 
3530 	/*
3531 	 * Check if this is premapped in TLB1. Note: this should probably also
3532 	 * check whether a sequence of TLB1 entries exist that match the
3533 	 * requirement, but now only checks the easy case.
3534 	 */
3535 	for (i = 0; i < TLB1_ENTRIES; i++) {
3536 		tlb1_read_entry(&e, i);
3537 		if (!(e.mas1 & MAS1_VALID))
3538 			continue;
3539 		if (pa >= e.phys &&
3540 		    (pa + size) <= (e.phys + e.size) &&
3541 		    (ma == VM_MEMATTR_DEFAULT ||
3542 		     tlb_calc_wimg(pa, ma) ==
3543 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3544 			return (void *)(e.virt +
3545 			    (vm_offset_t)(pa - e.phys));
3546 	}
3547 
3548 	size = roundup(size, PAGE_SIZE);
3549 
3550 	/*
3551 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3552 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3553 	 */
3554 #ifdef SPARSE_MAPDEV
3555 	/*
3556 	 * With a sparse mapdev, align to the largest starting region.  This
3557 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3558 	 * calculation could be very costly.
3559 	 * Align to the smaller of:
3560 	 * - first set bit in overlap of (pa & size mask)
3561 	 * - largest size envelope
3562 	 *
3563 	 * It's possible the device mapping may start at a PA that's not larger
3564 	 * than the size mask, so we need to offset in to maximize the TLB entry
3565 	 * range and minimize the number of used TLB entries.
3566 	 */
3567 	do {
3568 	    tmpva = tlb1_map_base;
3569 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3570 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3571 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3572 #ifdef __powerpc64__
3573 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3574 #else
3575 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3576 #endif
3577 #else
3578 #ifdef __powerpc64__
3579 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3580 #else
3581 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3582 #endif
3583 #endif
3584 	res = (void *)va;
3585 
3586 	do {
3587 		sz = 1 << (ilog2(size) & ~1);
3588 		/* Align size to PA */
3589 		if (pa % sz != 0) {
3590 			do {
3591 				sz >>= 2;
3592 			} while (pa % sz != 0);
3593 		}
3594 		/* Now align from there to VA */
3595 		if (va % sz != 0) {
3596 			do {
3597 				sz >>= 2;
3598 			} while (va % sz != 0);
3599 		}
3600 		if (bootverbose)
3601 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3602 			    va, (uintmax_t)pa, sz);
3603 		if (tlb1_set_entry(va, pa, sz,
3604 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3605 			return (NULL);
3606 		size -= sz;
3607 		pa += sz;
3608 		va += sz;
3609 	} while (size > 0);
3610 
3611 	return (res);
3612 }
3613 
3614 /*
3615  * 'Unmap' a range mapped by mmu_booke_mapdev().
3616  */
3617 static void
3618 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3619 {
3620 #ifdef SUPPORTS_SHRINKING_TLB1
3621 	vm_offset_t base, offset;
3622 
3623 	/*
3624 	 * Unmap only if this is inside kernel virtual space.
3625 	 */
3626 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3627 		base = trunc_page(va);
3628 		offset = va & PAGE_MASK;
3629 		size = roundup(offset + size, PAGE_SIZE);
3630 		kva_free(base, size);
3631 	}
3632 #endif
3633 }
3634 
3635 /*
3636  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3637  * specified pmap. This eliminates the blast of soft faults on process startup
3638  * and immediately after an mmap.
3639  */
3640 static void
3641 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3642     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3643 {
3644 
3645 	VM_OBJECT_ASSERT_WLOCKED(object);
3646 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3647 	    ("mmu_booke_object_init_pt: non-device object"));
3648 }
3649 
3650 /*
3651  * Perform the pmap work for mincore.
3652  */
3653 static int
3654 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3655     vm_paddr_t *locked_pa)
3656 {
3657 
3658 	/* XXX: this should be implemented at some point */
3659 	return (0);
3660 }
3661 
3662 static int
3663 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3664     vm_memattr_t mode)
3665 {
3666 	vm_offset_t va;
3667 	pte_t *pte;
3668 	int i, j;
3669 	tlb_entry_t e;
3670 
3671 	/* Check TLB1 mappings */
3672 	for (i = 0; i < TLB1_ENTRIES; i++) {
3673 		tlb1_read_entry(&e, i);
3674 		if (!(e.mas1 & MAS1_VALID))
3675 			continue;
3676 		if (addr >= e.virt && addr < e.virt + e.size)
3677 			break;
3678 	}
3679 	if (i < TLB1_ENTRIES) {
3680 		/* Only allow full mappings to be modified for now. */
3681 		/* Validate the range. */
3682 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3683 			tlb1_read_entry(&e, j);
3684 			if (va != e.virt || (sz - (va - addr) < e.size))
3685 				return (EINVAL);
3686 		}
3687 		for (va = addr; va < addr + sz; va += e.size, i++) {
3688 			tlb1_read_entry(&e, i);
3689 			e.mas2 &= ~MAS2_WIMGE_MASK;
3690 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3691 
3692 			/*
3693 			 * Write it out to the TLB.  Should really re-sync with other
3694 			 * cores.
3695 			 */
3696 			tlb1_write_entry(&e, i);
3697 		}
3698 		return (0);
3699 	}
3700 
3701 	/* Not in TLB1, try through pmap */
3702 	/* First validate the range. */
3703 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3704 		pte = pte_find(mmu, kernel_pmap, va);
3705 		if (pte == NULL || !PTE_ISVALID(pte))
3706 			return (EINVAL);
3707 	}
3708 
3709 	mtx_lock_spin(&tlbivax_mutex);
3710 	tlb_miss_lock();
3711 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3712 		pte = pte_find(mmu, kernel_pmap, va);
3713 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3714 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3715 		tlb0_flush_entry(va);
3716 	}
3717 	tlb_miss_unlock();
3718 	mtx_unlock_spin(&tlbivax_mutex);
3719 
3720 	return (0);
3721 }
3722 
3723 /**************************************************************************/
3724 /* TID handling */
3725 /**************************************************************************/
3726 
3727 /*
3728  * Allocate a TID. If necessary, steal one from someone else.
3729  * The new TID is flushed from the TLB before returning.
3730  */
3731 static tlbtid_t
3732 tid_alloc(pmap_t pmap)
3733 {
3734 	tlbtid_t tid;
3735 	int thiscpu;
3736 
3737 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3738 
3739 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3740 
3741 	thiscpu = PCPU_GET(cpuid);
3742 
3743 	tid = PCPU_GET(booke.tid_next);
3744 	if (tid > TID_MAX)
3745 		tid = TID_MIN;
3746 	PCPU_SET(booke.tid_next, tid + 1);
3747 
3748 	/* If we are stealing TID then clear the relevant pmap's field */
3749 	if (tidbusy[thiscpu][tid] != NULL) {
3750 
3751 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3752 
3753 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3754 
3755 		/* Flush all entries from TLB0 matching this TID. */
3756 		tid_flush(tid);
3757 	}
3758 
3759 	tidbusy[thiscpu][tid] = pmap;
3760 	pmap->pm_tid[thiscpu] = tid;
3761 	__asm __volatile("msync; isync");
3762 
3763 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3764 	    PCPU_GET(booke.tid_next));
3765 
3766 	return (tid);
3767 }
3768 
3769 /**************************************************************************/
3770 /* TLB0 handling */
3771 /**************************************************************************/
3772 
3773 static void
3774 #ifdef __powerpc64__
3775 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
3776 #else
3777 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3778 #endif
3779     uint32_t mas7)
3780 {
3781 	int as;
3782 	char desc[3];
3783 	tlbtid_t tid;
3784 	vm_size_t size;
3785 	unsigned int tsize;
3786 
3787 	desc[2] = '\0';
3788 	if (mas1 & MAS1_VALID)
3789 		desc[0] = 'V';
3790 	else
3791 		desc[0] = ' ';
3792 
3793 	if (mas1 & MAS1_IPROT)
3794 		desc[1] = 'P';
3795 	else
3796 		desc[1] = ' ';
3797 
3798 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3799 	tid = MAS1_GETTID(mas1);
3800 
3801 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3802 	size = 0;
3803 	if (tsize)
3804 		size = tsize2size(tsize);
3805 
3806 	debugf("%3d: (%s) [AS=%d] "
3807 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3808 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3809 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3810 }
3811 
3812 /* Convert TLB0 va and way number to tlb0[] table index. */
3813 static inline unsigned int
3814 tlb0_tableidx(vm_offset_t va, unsigned int way)
3815 {
3816 	unsigned int idx;
3817 
3818 	idx = (way * TLB0_ENTRIES_PER_WAY);
3819 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3820 	return (idx);
3821 }
3822 
3823 /*
3824  * Invalidate TLB0 entry.
3825  */
3826 static inline void
3827 tlb0_flush_entry(vm_offset_t va)
3828 {
3829 
3830 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3831 
3832 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3833 
3834 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3835 	__asm __volatile("isync; msync");
3836 	__asm __volatile("tlbsync; msync");
3837 
3838 	CTR1(KTR_PMAP, "%s: e", __func__);
3839 }
3840 
3841 /* Print out contents of the MAS registers for each TLB0 entry */
3842 void
3843 tlb0_print_tlbentries(void)
3844 {
3845 	uint32_t mas0, mas1, mas3, mas7;
3846 #ifdef __powerpc64__
3847 	uint64_t mas2;
3848 #else
3849 	uint32_t mas2;
3850 #endif
3851 	int entryidx, way, idx;
3852 
3853 	debugf("TLB0 entries:\n");
3854 	for (way = 0; way < TLB0_WAYS; way ++)
3855 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3856 
3857 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3858 			mtspr(SPR_MAS0, mas0);
3859 			__asm __volatile("isync");
3860 
3861 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3862 			mtspr(SPR_MAS2, mas2);
3863 
3864 			__asm __volatile("isync; tlbre");
3865 
3866 			mas1 = mfspr(SPR_MAS1);
3867 			mas2 = mfspr(SPR_MAS2);
3868 			mas3 = mfspr(SPR_MAS3);
3869 			mas7 = mfspr(SPR_MAS7);
3870 
3871 			idx = tlb0_tableidx(mas2, way);
3872 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3873 		}
3874 }
3875 
3876 /**************************************************************************/
3877 /* TLB1 handling */
3878 /**************************************************************************/
3879 
3880 /*
3881  * TLB1 mapping notes:
3882  *
3883  * TLB1[0]	Kernel text and data.
3884  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3885  *		windows, other devices mappings.
3886  */
3887 
3888  /*
3889  * Read an entry from given TLB1 slot.
3890  */
3891 void
3892 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3893 {
3894 	register_t msr;
3895 	uint32_t mas0;
3896 
3897 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3898 
3899 	msr = mfmsr();
3900 	__asm __volatile("wrteei 0");
3901 
3902 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3903 	mtspr(SPR_MAS0, mas0);
3904 	__asm __volatile("isync; tlbre");
3905 
3906 	entry->mas1 = mfspr(SPR_MAS1);
3907 	entry->mas2 = mfspr(SPR_MAS2);
3908 	entry->mas3 = mfspr(SPR_MAS3);
3909 
3910 	switch ((mfpvr() >> 16) & 0xFFFF) {
3911 	case FSL_E500v2:
3912 	case FSL_E500mc:
3913 	case FSL_E5500:
3914 	case FSL_E6500:
3915 		entry->mas7 = mfspr(SPR_MAS7);
3916 		break;
3917 	default:
3918 		entry->mas7 = 0;
3919 		break;
3920 	}
3921 	mtmsr(msr);
3922 
3923 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3924 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3925 	    (entry->mas3 & MAS3_RPN);
3926 	entry->size =
3927 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3928 }
3929 
3930 struct tlbwrite_args {
3931 	tlb_entry_t *e;
3932 	unsigned int idx;
3933 };
3934 
3935 static void
3936 tlb1_write_entry_int(void *arg)
3937 {
3938 	struct tlbwrite_args *args = arg;
3939 	uint32_t mas0;
3940 
3941 	/* Select entry */
3942 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3943 
3944 	mtspr(SPR_MAS0, mas0);
3945 	__asm __volatile("isync");
3946 	mtspr(SPR_MAS1, args->e->mas1);
3947 	__asm __volatile("isync");
3948 	mtspr(SPR_MAS2, args->e->mas2);
3949 	__asm __volatile("isync");
3950 	mtspr(SPR_MAS3, args->e->mas3);
3951 	__asm __volatile("isync");
3952 	switch ((mfpvr() >> 16) & 0xFFFF) {
3953 	case FSL_E500mc:
3954 	case FSL_E5500:
3955 	case FSL_E6500:
3956 		mtspr(SPR_MAS8, 0);
3957 		__asm __volatile("isync");
3958 		/* FALLTHROUGH */
3959 	case FSL_E500v2:
3960 		mtspr(SPR_MAS7, args->e->mas7);
3961 		__asm __volatile("isync");
3962 		break;
3963 	default:
3964 		break;
3965 	}
3966 
3967 	__asm __volatile("tlbwe; isync; msync");
3968 
3969 }
3970 
3971 static void
3972 tlb1_write_entry_sync(void *arg)
3973 {
3974 	/* Empty synchronization point for smp_rendezvous(). */
3975 }
3976 
3977 /*
3978  * Write given entry to TLB1 hardware.
3979  */
3980 static void
3981 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3982 {
3983 	struct tlbwrite_args args;
3984 
3985 	args.e = e;
3986 	args.idx = idx;
3987 
3988 #ifdef SMP
3989 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3990 		mb();
3991 		smp_rendezvous(tlb1_write_entry_sync,
3992 		    tlb1_write_entry_int,
3993 		    tlb1_write_entry_sync, &args);
3994 	} else
3995 #endif
3996 	{
3997 		register_t msr;
3998 
3999 		msr = mfmsr();
4000 		__asm __volatile("wrteei 0");
4001 		tlb1_write_entry_int(&args);
4002 		mtmsr(msr);
4003 	}
4004 }
4005 
4006 /*
4007  * Return the largest uint value log such that 2^log <= num.
4008  */
4009 static unsigned int
4010 ilog2(unsigned int num)
4011 {
4012 	int lz;
4013 
4014 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
4015 	return (31 - lz);
4016 }
4017 
4018 /*
4019  * Convert TLB TSIZE value to mapped region size.
4020  */
4021 static vm_size_t
4022 tsize2size(unsigned int tsize)
4023 {
4024 
4025 	/*
4026 	 * size = 4^tsize KB
4027 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
4028 	 */
4029 
4030 	return ((1 << (2 * tsize)) * 1024);
4031 }
4032 
4033 /*
4034  * Convert region size (must be power of 4) to TLB TSIZE value.
4035  */
4036 static unsigned int
4037 size2tsize(vm_size_t size)
4038 {
4039 
4040 	return (ilog2(size) / 2 - 5);
4041 }
4042 
4043 /*
4044  * Register permanent kernel mapping in TLB1.
4045  *
4046  * Entries are created starting from index 0 (current free entry is
4047  * kept in tlb1_idx) and are not supposed to be invalidated.
4048  */
4049 int
4050 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
4051     uint32_t flags)
4052 {
4053 	tlb_entry_t e;
4054 	uint32_t ts, tid;
4055 	int tsize, index;
4056 
4057 	for (index = 0; index < TLB1_ENTRIES; index++) {
4058 		tlb1_read_entry(&e, index);
4059 		if ((e.mas1 & MAS1_VALID) == 0)
4060 			break;
4061 		/* Check if we're just updating the flags, and update them. */
4062 		if (e.phys == pa && e.virt == va && e.size == size) {
4063 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
4064 			tlb1_write_entry(&e, index);
4065 			return (0);
4066 		}
4067 	}
4068 	if (index >= TLB1_ENTRIES) {
4069 		printf("tlb1_set_entry: TLB1 full!\n");
4070 		return (-1);
4071 	}
4072 
4073 	/* Convert size to TSIZE */
4074 	tsize = size2tsize(size);
4075 
4076 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
4077 	/* XXX TS is hard coded to 0 for now as we only use single address space */
4078 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
4079 
4080 	e.phys = pa;
4081 	e.virt = va;
4082 	e.size = size;
4083 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
4084 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
4085 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
4086 
4087 	/* Set supervisor RWX permission bits */
4088 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
4089 	e.mas7 = (pa >> 32) & MAS7_RPN;
4090 
4091 	tlb1_write_entry(&e, index);
4092 
4093 	/*
4094 	 * XXX in general TLB1 updates should be propagated between CPUs,
4095 	 * since current design assumes to have the same TLB1 set-up on all
4096 	 * cores.
4097 	 */
4098 	return (0);
4099 }
4100 
4101 /*
4102  * Map in contiguous RAM region into the TLB1 using maximum of
4103  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
4104  *
4105  * If necessary round up last entry size and return total size
4106  * used by all allocated entries.
4107  */
4108 vm_size_t
4109 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
4110 {
4111 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
4112 	vm_size_t mapped, pgsz, base, mask;
4113 	int idx, nents;
4114 
4115 	/* Round up to the next 1M */
4116 	size = roundup2(size, 1 << 20);
4117 
4118 	mapped = 0;
4119 	idx = 0;
4120 	base = va;
4121 	pgsz = 64*1024*1024;
4122 	while (mapped < size) {
4123 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
4124 			while (pgsz > (size - mapped))
4125 				pgsz >>= 2;
4126 			pgs[idx++] = pgsz;
4127 			mapped += pgsz;
4128 		}
4129 
4130 		/* We under-map. Correct for this. */
4131 		if (mapped < size) {
4132 			while (pgs[idx - 1] == pgsz) {
4133 				idx--;
4134 				mapped -= pgsz;
4135 			}
4136 			/* XXX We may increase beyond out starting point. */
4137 			pgsz <<= 2;
4138 			pgs[idx++] = pgsz;
4139 			mapped += pgsz;
4140 		}
4141 	}
4142 
4143 	nents = idx;
4144 	mask = pgs[0] - 1;
4145 	/* Align address to the boundary */
4146 	if (va & mask) {
4147 		va = (va + mask) & ~mask;
4148 		pa = (pa + mask) & ~mask;
4149 	}
4150 
4151 	for (idx = 0; idx < nents; idx++) {
4152 		pgsz = pgs[idx];
4153 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
4154 		tlb1_set_entry(va, pa, pgsz,
4155 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4156 		pa += pgsz;
4157 		va += pgsz;
4158 	}
4159 
4160 	mapped = (va - base);
4161 	printf("mapped size 0x%"PRI0ptrX" (wasted space 0x%"PRIxPTR")\n",
4162 	    mapped, mapped - size);
4163 	return (mapped);
4164 }
4165 
4166 /*
4167  * TLB1 initialization routine, to be called after the very first
4168  * assembler level setup done in locore.S.
4169  */
4170 void
4171 tlb1_init()
4172 {
4173 	uint32_t mas0, mas1, mas2, mas3, mas7;
4174 	uint32_t tsz;
4175 
4176 	tlb1_get_tlbconf();
4177 
4178 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4179 	mtspr(SPR_MAS0, mas0);
4180 	__asm __volatile("isync; tlbre");
4181 
4182 	mas1 = mfspr(SPR_MAS1);
4183 	mas2 = mfspr(SPR_MAS2);
4184 	mas3 = mfspr(SPR_MAS3);
4185 	mas7 = mfspr(SPR_MAS7);
4186 
4187 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4188 	    (mas3 & MAS3_RPN);
4189 
4190 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4191 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4192 
4193 	/* Setup TLB miss defaults */
4194 	set_mas4_defaults();
4195 }
4196 
4197 /*
4198  * pmap_early_io_unmap() should be used in short conjunction with
4199  * pmap_early_io_map(), as in the following snippet:
4200  *
4201  * x = pmap_early_io_map(...);
4202  * <do something with x>
4203  * pmap_early_io_unmap(x, size);
4204  *
4205  * And avoiding more allocations between.
4206  */
4207 void
4208 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4209 {
4210 	int i;
4211 	tlb_entry_t e;
4212 	vm_size_t isize;
4213 
4214 	size = roundup(size, PAGE_SIZE);
4215 	isize = size;
4216 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4217 		tlb1_read_entry(&e, i);
4218 		if (!(e.mas1 & MAS1_VALID))
4219 			continue;
4220 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4221 			size -= e.size;
4222 			e.mas1 &= ~MAS1_VALID;
4223 			tlb1_write_entry(&e, i);
4224 		}
4225 	}
4226 	if (tlb1_map_base == va + isize)
4227 		tlb1_map_base -= isize;
4228 }
4229 
4230 vm_offset_t
4231 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4232 {
4233 	vm_paddr_t pa_base;
4234 	vm_offset_t va, sz;
4235 	int i;
4236 	tlb_entry_t e;
4237 
4238 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4239 
4240 	for (i = 0; i < TLB1_ENTRIES; i++) {
4241 		tlb1_read_entry(&e, i);
4242 		if (!(e.mas1 & MAS1_VALID))
4243 			continue;
4244 		if (pa >= e.phys && (pa + size) <=
4245 		    (e.phys + e.size))
4246 			return (e.virt + (pa - e.phys));
4247 	}
4248 
4249 	pa_base = rounddown(pa, PAGE_SIZE);
4250 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4251 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4252 	va = tlb1_map_base + (pa - pa_base);
4253 
4254 	do {
4255 		sz = 1 << (ilog2(size) & ~1);
4256 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4257 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4258 		size -= sz;
4259 		pa_base += sz;
4260 		tlb1_map_base += sz;
4261 	} while (size > 0);
4262 
4263 	return (va);
4264 }
4265 
4266 void
4267 pmap_track_page(pmap_t pmap, vm_offset_t va)
4268 {
4269 	vm_paddr_t pa;
4270 	vm_page_t page;
4271 	struct pv_entry *pve;
4272 
4273 	va = trunc_page(va);
4274 	pa = pmap_kextract(va);
4275 	page = PHYS_TO_VM_PAGE(pa);
4276 
4277 	rw_wlock(&pvh_global_lock);
4278 	PMAP_LOCK(pmap);
4279 
4280 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4281 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4282 			goto out;
4283 		}
4284 	}
4285 	page->md.pv_tracked = true;
4286 	pv_insert(pmap, va, page);
4287 out:
4288 	PMAP_UNLOCK(pmap);
4289 	rw_wunlock(&pvh_global_lock);
4290 }
4291 
4292 
4293 /*
4294  * Setup MAS4 defaults.
4295  * These values are loaded to MAS0-2 on a TLB miss.
4296  */
4297 static void
4298 set_mas4_defaults(void)
4299 {
4300 	uint32_t mas4;
4301 
4302 	/* Defaults: TLB0, PID0, TSIZED=4K */
4303 	mas4 = MAS4_TLBSELD0;
4304 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4305 #ifdef SMP
4306 	mas4 |= MAS4_MD;
4307 #endif
4308 	mtspr(SPR_MAS4, mas4);
4309 	__asm __volatile("isync");
4310 }
4311 
4312 /*
4313  * Print out contents of the MAS registers for each TLB1 entry
4314  */
4315 void
4316 tlb1_print_tlbentries(void)
4317 {
4318 	uint32_t mas0, mas1, mas3, mas7;
4319 #ifdef __powerpc64__
4320 	uint64_t mas2;
4321 #else
4322 	uint32_t mas2;
4323 #endif
4324 	int i;
4325 
4326 	debugf("TLB1 entries:\n");
4327 	for (i = 0; i < TLB1_ENTRIES; i++) {
4328 
4329 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4330 		mtspr(SPR_MAS0, mas0);
4331 
4332 		__asm __volatile("isync; tlbre");
4333 
4334 		mas1 = mfspr(SPR_MAS1);
4335 		mas2 = mfspr(SPR_MAS2);
4336 		mas3 = mfspr(SPR_MAS3);
4337 		mas7 = mfspr(SPR_MAS7);
4338 
4339 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4340 	}
4341 }
4342 
4343 /*
4344  * Return 0 if the physical IO range is encompassed by one of the
4345  * the TLB1 entries, otherwise return related error code.
4346  */
4347 static int
4348 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4349 {
4350 	uint32_t prot;
4351 	vm_paddr_t pa_start;
4352 	vm_paddr_t pa_end;
4353 	unsigned int entry_tsize;
4354 	vm_size_t entry_size;
4355 	tlb_entry_t e;
4356 
4357 	*va = (vm_offset_t)NULL;
4358 
4359 	tlb1_read_entry(&e, i);
4360 	/* Skip invalid entries */
4361 	if (!(e.mas1 & MAS1_VALID))
4362 		return (EINVAL);
4363 
4364 	/*
4365 	 * The entry must be cache-inhibited, guarded, and r/w
4366 	 * so it can function as an i/o page
4367 	 */
4368 	prot = e.mas2 & (MAS2_I | MAS2_G);
4369 	if (prot != (MAS2_I | MAS2_G))
4370 		return (EPERM);
4371 
4372 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4373 	if (prot != (MAS3_SR | MAS3_SW))
4374 		return (EPERM);
4375 
4376 	/* The address should be within the entry range. */
4377 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4378 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4379 
4380 	entry_size = tsize2size(entry_tsize);
4381 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4382 	    (e.mas3 & MAS3_RPN);
4383 	pa_end = pa_start + entry_size;
4384 
4385 	if ((pa < pa_start) || ((pa + size) > pa_end))
4386 		return (ERANGE);
4387 
4388 	/* Return virtual address of this mapping. */
4389 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4390 	return (0);
4391 }
4392 
4393 /*
4394  * Invalidate all TLB0 entries which match the given TID. Note this is
4395  * dedicated for cases when invalidations should NOT be propagated to other
4396  * CPUs.
4397  */
4398 static void
4399 tid_flush(tlbtid_t tid)
4400 {
4401 	register_t msr;
4402 	uint32_t mas0, mas1, mas2;
4403 	int entry, way;
4404 
4405 
4406 	/* Don't evict kernel translations */
4407 	if (tid == TID_KERNEL)
4408 		return;
4409 
4410 	msr = mfmsr();
4411 	__asm __volatile("wrteei 0");
4412 
4413 	for (way = 0; way < TLB0_WAYS; way++)
4414 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4415 
4416 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4417 			mtspr(SPR_MAS0, mas0);
4418 			__asm __volatile("isync");
4419 
4420 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4421 			mtspr(SPR_MAS2, mas2);
4422 
4423 			__asm __volatile("isync; tlbre");
4424 
4425 			mas1 = mfspr(SPR_MAS1);
4426 
4427 			if (!(mas1 & MAS1_VALID))
4428 				continue;
4429 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4430 				continue;
4431 			mas1 &= ~MAS1_VALID;
4432 			mtspr(SPR_MAS1, mas1);
4433 			__asm __volatile("isync; tlbwe; isync; msync");
4434 		}
4435 	mtmsr(msr);
4436 }
4437