xref: /freebsd/sys/powerpc/booke/pmap.c (revision b13788e396c2b24f88697e7d4a74bab429ef4d0c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - kernel_pdir-1            : message buffer
63   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
64   *             kernel_pp2d - .                        : kernel pointers to page directory
65   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
66   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
67   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
68   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
69   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
70   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
71   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
72   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
73   *                         - 0xffff_ffff_ffff_ffff    : device direct map
74   */
75 
76 #include <sys/cdefs.h>
77 __FBSDID("$FreeBSD$");
78 
79 #include "opt_ddb.h"
80 #include "opt_kstack_pages.h"
81 
82 #include <sys/param.h>
83 #include <sys/conf.h>
84 #include <sys/malloc.h>
85 #include <sys/ktr.h>
86 #include <sys/proc.h>
87 #include <sys/user.h>
88 #include <sys/queue.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/kerneldump.h>
92 #include <sys/linker.h>
93 #include <sys/msgbuf.h>
94 #include <sys/lock.h>
95 #include <sys/mutex.h>
96 #include <sys/rwlock.h>
97 #include <sys/sched.h>
98 #include <sys/smp.h>
99 #include <sys/vmmeter.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_kern.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_param.h>
108 #include <vm/vm_map.h>
109 #include <vm/vm_pager.h>
110 #include <vm/vm_phys.h>
111 #include <vm/vm_pagequeue.h>
112 #include <vm/uma.h>
113 
114 #include <machine/_inttypes.h>
115 #include <machine/cpu.h>
116 #include <machine/pcb.h>
117 #include <machine/platform.h>
118 
119 #include <machine/tlb.h>
120 #include <machine/spr.h>
121 #include <machine/md_var.h>
122 #include <machine/mmuvar.h>
123 #include <machine/pmap.h>
124 #include <machine/pte.h>
125 
126 #include <ddb/ddb.h>
127 
128 #include "mmu_if.h"
129 
130 #define	SPARSE_MAPDEV
131 
132 /* Use power-of-two mappings in mmu_booke_mapdev(), to save entries. */
133 #define	POW2_MAPPINGS
134 
135 #ifdef  DEBUG
136 #define debugf(fmt, args...) printf(fmt, ##args)
137 #else
138 #define debugf(fmt, args...)
139 #endif
140 
141 #ifdef __powerpc64__
142 #define	PRI0ptrX	"016lx"
143 #else
144 #define	PRI0ptrX	"08x"
145 #endif
146 
147 #define TODO			panic("%s: not implemented", __func__);
148 
149 extern unsigned char _etext[];
150 extern unsigned char _end[];
151 
152 extern uint32_t *bootinfo;
153 
154 vm_paddr_t kernload;
155 vm_offset_t kernstart;
156 vm_size_t kernsize;
157 
158 /* Message buffer and tables. */
159 static vm_offset_t data_start;
160 static vm_size_t data_end;
161 
162 /* Phys/avail memory regions. */
163 static struct mem_region *availmem_regions;
164 static int availmem_regions_sz;
165 static struct mem_region *physmem_regions;
166 static int physmem_regions_sz;
167 
168 #ifndef __powerpc64__
169 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
170 static vm_offset_t zero_page_va;
171 static struct mtx zero_page_mutex;
172 
173 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
174 static vm_offset_t copy_page_src_va;
175 static vm_offset_t copy_page_dst_va;
176 static struct mtx copy_page_mutex;
177 #endif
178 
179 static struct mtx tlbivax_mutex;
180 
181 /**************************************************************************/
182 /* PMAP */
183 /**************************************************************************/
184 
185 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
186     vm_prot_t, u_int flags, int8_t psind);
187 
188 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
189 static uma_zone_t ptbl_root_zone;
190 
191 /*
192  * If user pmap is processed with mmu_booke_remove and the resident count
193  * drops to 0, there are no more pages to remove, so we need not continue.
194  */
195 #define PMAP_REMOVE_DONE(pmap) \
196 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
197 
198 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
199 extern int elf32_nxstack;
200 #endif
201 
202 /**************************************************************************/
203 /* TLB and TID handling */
204 /**************************************************************************/
205 
206 /* Translation ID busy table */
207 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
208 
209 /*
210  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
211  * core revisions and should be read from h/w registers during early config.
212  */
213 uint32_t tlb0_entries;
214 uint32_t tlb0_ways;
215 uint32_t tlb0_entries_per_way;
216 uint32_t tlb1_entries;
217 
218 #define TLB0_ENTRIES		(tlb0_entries)
219 #define TLB0_WAYS		(tlb0_ways)
220 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
221 
222 #define TLB1_ENTRIES (tlb1_entries)
223 
224 static tlbtid_t tid_alloc(struct pmap *);
225 
226 #ifdef DDB
227 #ifdef __powerpc64__
228 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
229 #else
230 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
231 #endif
232 #endif
233 
234 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
235 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
236 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
237 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t, int);
238 
239 static __inline uint32_t tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma);
240 
241 static vm_size_t tsize2size(unsigned int);
242 static unsigned int size2tsize(vm_size_t);
243 static unsigned long ilog2(unsigned long);
244 
245 static void set_mas4_defaults(void);
246 
247 static inline void tlb0_flush_entry(vm_offset_t);
248 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
249 
250 /**************************************************************************/
251 /* Page table management */
252 /**************************************************************************/
253 
254 static struct rwlock_padalign pvh_global_lock;
255 
256 /* Data for the pv entry allocation mechanism */
257 static uma_zone_t pvzone;
258 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
259 
260 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
261 
262 #ifndef PMAP_SHPGPERPROC
263 #define PMAP_SHPGPERPROC	200
264 #endif
265 
266 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
267 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
268 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
269 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
270 static void kernel_pte_alloc(vm_offset_t, vm_offset_t);
271 
272 static pv_entry_t pv_alloc(void);
273 static void pv_free(pv_entry_t);
274 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
275 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
276 
277 static void booke_pmap_init_qpages(void);
278 
279 static inline void tlb_miss_lock(void);
280 static inline void tlb_miss_unlock(void);
281 
282 #ifdef SMP
283 extern tlb_entry_t __boot_tlb1[];
284 void pmap_bootstrap_ap(volatile uint32_t *);
285 #endif
286 
287 /*
288  * Kernel MMU interface
289  */
290 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
291 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
292     vm_size_t, vm_offset_t);
293 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
294 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
295     vm_offset_t, vm_page_t *, vm_offset_t, int);
296 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
297     vm_prot_t, u_int flags, int8_t psind);
298 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
299     vm_page_t, vm_prot_t);
300 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
301     vm_prot_t);
302 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
303 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
304     vm_prot_t);
305 static void		mmu_booke_init(mmu_t);
306 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
307 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
308 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
309 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
310 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
311     int);
312 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
313     vm_paddr_t *);
314 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
315     vm_object_t, vm_pindex_t, vm_size_t);
316 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
317 static void		mmu_booke_page_init(mmu_t, vm_page_t);
318 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
319 static void		mmu_booke_pinit(mmu_t, pmap_t);
320 static void		mmu_booke_pinit0(mmu_t, pmap_t);
321 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
322     vm_prot_t);
323 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
324 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
325 static void		mmu_booke_release(mmu_t, pmap_t);
326 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
327 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
328 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
329 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
330 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
331 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
332 static void		mmu_booke_activate(mmu_t, struct thread *);
333 static void		mmu_booke_deactivate(mmu_t, struct thread *);
334 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
335 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
336 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
337 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
338 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
339 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
340 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
341 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
342 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
343 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
344     vm_size_t);
345 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
346     void **);
347 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
348     void *);
349 static void		mmu_booke_scan_init(mmu_t);
350 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
351 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
352 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
353     vm_size_t sz, vm_memattr_t mode);
354 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
355     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
356 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
357     int *is_user, vm_offset_t *decoded_addr);
358 static void		mmu_booke_page_array_startup(mmu_t , long);
359 static boolean_t mmu_booke_page_is_mapped(mmu_t mmu, vm_page_t m);
360 
361 
362 static mmu_method_t mmu_booke_methods[] = {
363 	/* pmap dispatcher interface */
364 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
365 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
366 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
367 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
368 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
369 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
370 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
371 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
372 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
373 	MMUMETHOD(mmu_init,		mmu_booke_init),
374 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
375 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
376 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
377 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
378 	MMUMETHOD(mmu_map,		mmu_booke_map),
379 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
380 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
381 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
382 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
383 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
384 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
385 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
386 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
387 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
388 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
389 	MMUMETHOD(mmu_release,		mmu_booke_release),
390 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
391 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
392 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
393 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
394 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
395 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
396 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
397 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
398 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
399 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
400 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
401 	MMUMETHOD(mmu_page_array_startup,	mmu_booke_page_array_startup),
402 	MMUMETHOD(mmu_page_is_mapped,	mmu_booke_page_is_mapped),
403 
404 	/* Internal interfaces */
405 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
406 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
407 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
408 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
409 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
410 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
411 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
412 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
413 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
414 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
415 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
416 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
417 
418 	/* dumpsys() support */
419 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
420 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
421 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
422 
423 	{ 0, 0 }
424 };
425 
426 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
427 
428 #ifdef __powerpc64__
429 #include "pmap_64.c"
430 #else
431 #include "pmap_32.c"
432 #endif
433 
434 static vm_offset_t tlb1_map_base = VM_MAPDEV_BASE;
435 
436 static __inline uint32_t
437 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
438 {
439 	uint32_t attrib;
440 	int i;
441 
442 	if (ma != VM_MEMATTR_DEFAULT) {
443 		switch (ma) {
444 		case VM_MEMATTR_UNCACHEABLE:
445 			return (MAS2_I | MAS2_G);
446 		case VM_MEMATTR_WRITE_COMBINING:
447 		case VM_MEMATTR_WRITE_BACK:
448 		case VM_MEMATTR_PREFETCHABLE:
449 			return (MAS2_I);
450 		case VM_MEMATTR_WRITE_THROUGH:
451 			return (MAS2_W | MAS2_M);
452 		case VM_MEMATTR_CACHEABLE:
453 			return (MAS2_M);
454 		}
455 	}
456 
457 	/*
458 	 * Assume the page is cache inhibited and access is guarded unless
459 	 * it's in our available memory array.
460 	 */
461 	attrib = _TLB_ENTRY_IO;
462 	for (i = 0; i < physmem_regions_sz; i++) {
463 		if ((pa >= physmem_regions[i].mr_start) &&
464 		    (pa < (physmem_regions[i].mr_start +
465 		     physmem_regions[i].mr_size))) {
466 			attrib = _TLB_ENTRY_MEM;
467 			break;
468 		}
469 	}
470 
471 	return (attrib);
472 }
473 
474 static inline void
475 tlb_miss_lock(void)
476 {
477 #ifdef SMP
478 	struct pcpu *pc;
479 
480 	if (!smp_started)
481 		return;
482 
483 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
484 		if (pc != pcpup) {
485 
486 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
487 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
488 
489 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
490 			    ("tlb_miss_lock: tried to lock self"));
491 
492 			tlb_lock(pc->pc_booke.tlb_lock);
493 
494 			CTR1(KTR_PMAP, "%s: locked", __func__);
495 		}
496 	}
497 #endif
498 }
499 
500 static inline void
501 tlb_miss_unlock(void)
502 {
503 #ifdef SMP
504 	struct pcpu *pc;
505 
506 	if (!smp_started)
507 		return;
508 
509 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
510 		if (pc != pcpup) {
511 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
512 			    __func__, pc->pc_cpuid);
513 
514 			tlb_unlock(pc->pc_booke.tlb_lock);
515 
516 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
517 		}
518 	}
519 #endif
520 }
521 
522 /* Return number of entries in TLB0. */
523 static __inline void
524 tlb0_get_tlbconf(void)
525 {
526 	uint32_t tlb0_cfg;
527 
528 	tlb0_cfg = mfspr(SPR_TLB0CFG);
529 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
530 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
531 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
532 }
533 
534 /* Return number of entries in TLB1. */
535 static __inline void
536 tlb1_get_tlbconf(void)
537 {
538 	uint32_t tlb1_cfg;
539 
540 	tlb1_cfg = mfspr(SPR_TLB1CFG);
541 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
542 }
543 
544 /**************************************************************************/
545 /* Page table related */
546 /**************************************************************************/
547 
548 /* Allocate pv_entry structure. */
549 pv_entry_t
550 pv_alloc(void)
551 {
552 	pv_entry_t pv;
553 
554 	pv_entry_count++;
555 	if (pv_entry_count > pv_entry_high_water)
556 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
557 	pv = uma_zalloc(pvzone, M_NOWAIT);
558 
559 	return (pv);
560 }
561 
562 /* Free pv_entry structure. */
563 static __inline void
564 pv_free(pv_entry_t pve)
565 {
566 
567 	pv_entry_count--;
568 	uma_zfree(pvzone, pve);
569 }
570 
571 
572 /* Allocate and initialize pv_entry structure. */
573 static void
574 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
575 {
576 	pv_entry_t pve;
577 
578 	//int su = (pmap == kernel_pmap);
579 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
580 	//	(u_int32_t)pmap, va, (u_int32_t)m);
581 
582 	pve = pv_alloc();
583 	if (pve == NULL)
584 		panic("pv_insert: no pv entries!");
585 
586 	pve->pv_pmap = pmap;
587 	pve->pv_va = va;
588 
589 	/* add to pv_list */
590 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
591 	rw_assert(&pvh_global_lock, RA_WLOCKED);
592 
593 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
594 
595 	//debugf("pv_insert: e\n");
596 }
597 
598 /* Destroy pv entry. */
599 static void
600 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
601 {
602 	pv_entry_t pve;
603 
604 	//int su = (pmap == kernel_pmap);
605 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
606 
607 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
608 	rw_assert(&pvh_global_lock, RA_WLOCKED);
609 
610 	/* find pv entry */
611 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
612 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
613 			/* remove from pv_list */
614 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
615 			if (TAILQ_EMPTY(&m->md.pv_list))
616 				vm_page_aflag_clear(m, PGA_WRITEABLE);
617 
618 			/* free pv entry struct */
619 			pv_free(pve);
620 			break;
621 		}
622 	}
623 
624 	//debugf("pv_remove: e\n");
625 }
626 
627 /**************************************************************************/
628 /* PMAP related */
629 /**************************************************************************/
630 
631 /*
632  * This is called during booke_init, before the system is really initialized.
633  */
634 static void
635 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
636 {
637 	vm_paddr_t phys_kernelend;
638 	struct mem_region *mp, *mp1;
639 	int cnt, i, j;
640 	vm_paddr_t s, e, sz;
641 	vm_paddr_t physsz, hwphyssz;
642 	u_int phys_avail_count;
643 	vm_size_t kstack0_sz;
644 	vm_paddr_t kstack0_phys;
645 	vm_offset_t kstack0;
646 	void *dpcpu;
647 
648 	debugf("mmu_booke_bootstrap: entered\n");
649 
650 	/* Set interesting system properties */
651 #ifdef __powerpc64__
652 	hw_direct_map = 1;
653 #else
654 	hw_direct_map = 0;
655 #endif
656 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
657 	elf32_nxstack = 1;
658 #endif
659 
660 	/* Initialize invalidation mutex */
661 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
662 
663 	/* Read TLB0 size and associativity. */
664 	tlb0_get_tlbconf();
665 
666 	/*
667 	 * Align kernel start and end address (kernel image).
668 	 * Note that kernel end does not necessarily relate to kernsize.
669 	 * kernsize is the size of the kernel that is actually mapped.
670 	 */
671 	data_start = round_page(kernelend);
672 	data_end = data_start;
673 
674 	/* Allocate the dynamic per-cpu area. */
675 	dpcpu = (void *)data_end;
676 	data_end += DPCPU_SIZE;
677 
678 	/* Allocate space for the message buffer. */
679 	msgbufp = (struct msgbuf *)data_end;
680 	data_end += msgbufsize;
681 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
682 	    (uintptr_t)msgbufp, data_end);
683 
684 	data_end = round_page(data_end);
685 	data_end = round_page(mmu_booke_alloc_kernel_pgtables(data_end));
686 
687 	/* Retrieve phys/avail mem regions */
688 	mem_regions(&physmem_regions, &physmem_regions_sz,
689 	    &availmem_regions, &availmem_regions_sz);
690 
691 	if (PHYS_AVAIL_ENTRIES < availmem_regions_sz)
692 		panic("mmu_booke_bootstrap: phys_avail too small");
693 
694 	data_end = round_page(data_end);
695 	vm_page_array = (vm_page_t)data_end;
696 	/*
697 	 * Get a rough idea (upper bound) on the size of the page array.  The
698 	 * vm_page_array will not handle any more pages than we have in the
699 	 * avail_regions array, and most likely much less.
700 	 */
701 	sz = 0;
702 	for (mp = availmem_regions; mp->mr_size; mp++) {
703 		sz += mp->mr_size;
704 	}
705 	sz = (round_page(sz) / (PAGE_SIZE + sizeof(struct vm_page)));
706 	data_end += round_page(sz * sizeof(struct vm_page));
707 
708 	/* Pre-round up to 1MB.  This wastes some space, but saves TLB entries */
709 	data_end = roundup2(data_end, 1 << 20);
710 
711 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
712 	debugf(" kernstart: %#zx\n", kernstart);
713 	debugf(" kernsize: %#zx\n", kernsize);
714 
715 	if (data_end - kernstart > kernsize) {
716 		kernsize += tlb1_mapin_region(kernstart + kernsize,
717 		    kernload + kernsize, (data_end - kernstart) - kernsize,
718 		    _TLB_ENTRY_MEM);
719 	}
720 	data_end = kernstart + kernsize;
721 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
722 
723 	/*
724 	 * Clear the structures - note we can only do it safely after the
725 	 * possible additional TLB1 translations are in place (above) so that
726 	 * all range up to the currently calculated 'data_end' is covered.
727 	 */
728 	bzero((void *)data_start, data_end - data_start);
729 	dpcpu_init(dpcpu, 0);
730 
731 	/*******************************************************/
732 	/* Set the start and end of kva. */
733 	/*******************************************************/
734 	virtual_avail = round_page(data_end);
735 	virtual_end = VM_MAX_KERNEL_ADDRESS;
736 
737 #ifndef __powerpc64__
738 	/* Allocate KVA space for page zero/copy operations. */
739 	zero_page_va = virtual_avail;
740 	virtual_avail += PAGE_SIZE;
741 	copy_page_src_va = virtual_avail;
742 	virtual_avail += PAGE_SIZE;
743 	copy_page_dst_va = virtual_avail;
744 	virtual_avail += PAGE_SIZE;
745 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
746 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
747 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
748 
749 	/* Initialize page zero/copy mutexes. */
750 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
751 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
752 
753 	/* Allocate KVA space for ptbl bufs. */
754 	ptbl_buf_pool_vabase = virtual_avail;
755 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
756 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
757 	    ptbl_buf_pool_vabase, virtual_avail);
758 #endif
759 
760 	/* Calculate corresponding physical addresses for the kernel region. */
761 	phys_kernelend = kernload + kernsize;
762 	debugf("kernel image and allocated data:\n");
763 	debugf(" kernload    = 0x%09jx\n", (uintmax_t)kernload);
764 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
765 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
766 
767 	/*
768 	 * Remove kernel physical address range from avail regions list. Page
769 	 * align all regions.  Non-page aligned memory isn't very interesting
770 	 * to us.  Also, sort the entries for ascending addresses.
771 	 */
772 
773 	sz = 0;
774 	cnt = availmem_regions_sz;
775 	debugf("processing avail regions:\n");
776 	for (mp = availmem_regions; mp->mr_size; mp++) {
777 		s = mp->mr_start;
778 		e = mp->mr_start + mp->mr_size;
779 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
780 		/* Check whether this region holds all of the kernel. */
781 		if (s < kernload && e > phys_kernelend) {
782 			availmem_regions[cnt].mr_start = phys_kernelend;
783 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
784 			e = kernload;
785 		}
786 		/* Look whether this regions starts within the kernel. */
787 		if (s >= kernload && s < phys_kernelend) {
788 			if (e <= phys_kernelend)
789 				goto empty;
790 			s = phys_kernelend;
791 		}
792 		/* Now look whether this region ends within the kernel. */
793 		if (e > kernload && e <= phys_kernelend) {
794 			if (s >= kernload)
795 				goto empty;
796 			e = kernload;
797 		}
798 		/* Now page align the start and size of the region. */
799 		s = round_page(s);
800 		e = trunc_page(e);
801 		if (e < s)
802 			e = s;
803 		sz = e - s;
804 		debugf("%09jx-%09jx = %jx\n",
805 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
806 
807 		/* Check whether some memory is left here. */
808 		if (sz == 0) {
809 		empty:
810 			memmove(mp, mp + 1,
811 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
812 			cnt--;
813 			mp--;
814 			continue;
815 		}
816 
817 		/* Do an insertion sort. */
818 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
819 			if (s < mp1->mr_start)
820 				break;
821 		if (mp1 < mp) {
822 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
823 			mp1->mr_start = s;
824 			mp1->mr_size = sz;
825 		} else {
826 			mp->mr_start = s;
827 			mp->mr_size = sz;
828 		}
829 	}
830 	availmem_regions_sz = cnt;
831 
832 	/*******************************************************/
833 	/* Steal physical memory for kernel stack from the end */
834 	/* of the first avail region                           */
835 	/*******************************************************/
836 	kstack0_sz = kstack_pages * PAGE_SIZE;
837 	kstack0_phys = availmem_regions[0].mr_start +
838 	    availmem_regions[0].mr_size;
839 	kstack0_phys -= kstack0_sz;
840 	availmem_regions[0].mr_size -= kstack0_sz;
841 
842 	/*******************************************************/
843 	/* Fill in phys_avail table, based on availmem_regions */
844 	/*******************************************************/
845 	phys_avail_count = 0;
846 	physsz = 0;
847 	hwphyssz = 0;
848 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
849 
850 	debugf("fill in phys_avail:\n");
851 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
852 
853 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
854 		    (uintmax_t)availmem_regions[i].mr_start,
855 		    (uintmax_t)availmem_regions[i].mr_start +
856 		        availmem_regions[i].mr_size,
857 		    (uintmax_t)availmem_regions[i].mr_size);
858 
859 		if (hwphyssz != 0 &&
860 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
861 			debugf(" hw.physmem adjust\n");
862 			if (physsz < hwphyssz) {
863 				phys_avail[j] = availmem_regions[i].mr_start;
864 				phys_avail[j + 1] =
865 				    availmem_regions[i].mr_start +
866 				    hwphyssz - physsz;
867 				physsz = hwphyssz;
868 				phys_avail_count++;
869 				dump_avail[j] = phys_avail[j];
870 				dump_avail[j + 1] = phys_avail[j + 1];
871 			}
872 			break;
873 		}
874 
875 		phys_avail[j] = availmem_regions[i].mr_start;
876 		phys_avail[j + 1] = availmem_regions[i].mr_start +
877 		    availmem_regions[i].mr_size;
878 		phys_avail_count++;
879 		physsz += availmem_regions[i].mr_size;
880 		dump_avail[j] = phys_avail[j];
881 		dump_avail[j + 1] = phys_avail[j + 1];
882 	}
883 	physmem = btoc(physsz);
884 
885 	/* Calculate the last available physical address. */
886 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
887 		;
888 	Maxmem = powerpc_btop(phys_avail[i + 1]);
889 
890 	debugf("Maxmem = 0x%08lx\n", Maxmem);
891 	debugf("phys_avail_count = %d\n", phys_avail_count);
892 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
893 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
894 
895 #ifdef __powerpc64__
896 	/*
897 	 * Map the physical memory contiguously in TLB1.
898 	 * Round so it fits into a single mapping.
899 	 */
900 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
901 	    phys_avail[i + 1], _TLB_ENTRY_MEM);
902 #endif
903 
904 	/*******************************************************/
905 	/* Initialize (statically allocated) kernel pmap. */
906 	/*******************************************************/
907 	PMAP_LOCK_INIT(kernel_pmap);
908 
909 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
910 	kernel_pte_alloc(virtual_avail, kernstart);
911 	for (i = 0; i < MAXCPU; i++) {
912 		kernel_pmap->pm_tid[i] = TID_KERNEL;
913 
914 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
915 		tidbusy[i][TID_KERNEL] = kernel_pmap;
916 	}
917 
918 	/* Mark kernel_pmap active on all CPUs */
919 	CPU_FILL(&kernel_pmap->pm_active);
920 
921  	/*
922 	 * Initialize the global pv list lock.
923 	 */
924 	rw_init(&pvh_global_lock, "pmap pv global");
925 
926 	/*******************************************************/
927 	/* Final setup */
928 	/*******************************************************/
929 
930 	/* Enter kstack0 into kernel map, provide guard page */
931 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
932 	thread0.td_kstack = kstack0;
933 	thread0.td_kstack_pages = kstack_pages;
934 
935 	debugf("kstack_sz = 0x%08jx\n", (uintmax_t)kstack0_sz);
936 	debugf("kstack0_phys at 0x%09jx - 0x%09jx\n",
937 	    (uintmax_t)kstack0_phys, (uintmax_t)kstack0_phys + kstack0_sz);
938 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
939 	    kstack0, kstack0 + kstack0_sz);
940 
941 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
942 	for (i = 0; i < kstack_pages; i++) {
943 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
944 		kstack0 += PAGE_SIZE;
945 		kstack0_phys += PAGE_SIZE;
946 	}
947 
948 	pmap_bootstrapped = 1;
949 
950 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
951 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
952 
953 	debugf("mmu_booke_bootstrap: exit\n");
954 }
955 
956 #ifdef SMP
957 void
958 tlb1_ap_prep(void)
959 {
960 	tlb_entry_t *e, tmp;
961 	unsigned int i;
962 
963 	/* Prepare TLB1 image for AP processors */
964 	e = __boot_tlb1;
965 	for (i = 0; i < TLB1_ENTRIES; i++) {
966 		tlb1_read_entry(&tmp, i);
967 
968 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
969 			memcpy(e++, &tmp, sizeof(tmp));
970 	}
971 }
972 
973 void
974 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
975 {
976 	int i;
977 
978 	/*
979 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
980 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
981 	 * created by tlb1_ap_prep(), so use these values directly to
982 	 * (re)program AP's TLB1 hardware.
983 	 *
984 	 * Start at index 1 because index 0 has the kernel map.
985 	 */
986 	for (i = 1; i < TLB1_ENTRIES; i++) {
987 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
988 			tlb1_write_entry(&__boot_tlb1[i], i);
989 	}
990 
991 	set_mas4_defaults();
992 }
993 #endif
994 
995 static void
996 booke_pmap_init_qpages(void)
997 {
998 	struct pcpu *pc;
999 	int i;
1000 
1001 	CPU_FOREACH(i) {
1002 		pc = pcpu_find(i);
1003 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1004 		if (pc->pc_qmap_addr == 0)
1005 			panic("pmap_init_qpages: unable to allocate KVA");
1006 	}
1007 }
1008 
1009 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1010 
1011 /*
1012  * Get the physical page address for the given pmap/virtual address.
1013  */
1014 static vm_paddr_t
1015 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1016 {
1017 	vm_paddr_t pa;
1018 
1019 	PMAP_LOCK(pmap);
1020 	pa = pte_vatopa(mmu, pmap, va);
1021 	PMAP_UNLOCK(pmap);
1022 
1023 	return (pa);
1024 }
1025 
1026 /*
1027  * Extract the physical page address associated with the given
1028  * kernel virtual address.
1029  */
1030 static vm_paddr_t
1031 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1032 {
1033 	tlb_entry_t e;
1034 	vm_paddr_t p = 0;
1035 	int i;
1036 
1037 #ifdef __powerpc64__
1038 	if (va >= DMAP_BASE_ADDRESS && va <= DMAP_MAX_ADDRESS)
1039 		return (DMAP_TO_PHYS(va));
1040 #endif
1041 
1042 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
1043 		p = pte_vatopa(mmu, kernel_pmap, va);
1044 
1045 	if (p == 0) {
1046 		/* Check TLB1 mappings */
1047 		for (i = 0; i < TLB1_ENTRIES; i++) {
1048 			tlb1_read_entry(&e, i);
1049 			if (!(e.mas1 & MAS1_VALID))
1050 				continue;
1051 			if (va >= e.virt && va < e.virt + e.size)
1052 				return (e.phys + (va - e.virt));
1053 		}
1054 	}
1055 
1056 	return (p);
1057 }
1058 
1059 /*
1060  * Initialize the pmap module.
1061  * Called by vm_init, to initialize any structures that the pmap
1062  * system needs to map virtual memory.
1063  */
1064 static void
1065 mmu_booke_init(mmu_t mmu)
1066 {
1067 	int shpgperproc = PMAP_SHPGPERPROC;
1068 
1069 	/*
1070 	 * Initialize the address space (zone) for the pv entries.  Set a
1071 	 * high water mark so that the system can recover from excessive
1072 	 * numbers of pv entries.
1073 	 */
1074 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1075 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1076 
1077 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1078 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1079 
1080 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1081 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1082 
1083 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1084 
1085 	/* Pre-fill pvzone with initial number of pv entries. */
1086 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1087 
1088 	/* Create a UMA zone for page table roots. */
1089 	ptbl_root_zone = uma_zcreate("pmap root", PMAP_ROOT_SIZE,
1090 	    NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, UMA_ZONE_VM);
1091 
1092 	/* Initialize ptbl allocation. */
1093 	ptbl_init();
1094 }
1095 
1096 /*
1097  * Map a list of wired pages into kernel virtual address space.  This is
1098  * intended for temporary mappings which do not need page modification or
1099  * references recorded.  Existing mappings in the region are overwritten.
1100  */
1101 static void
1102 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1103 {
1104 	vm_offset_t va;
1105 
1106 	va = sva;
1107 	while (count-- > 0) {
1108 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1109 		va += PAGE_SIZE;
1110 		m++;
1111 	}
1112 }
1113 
1114 /*
1115  * Remove page mappings from kernel virtual address space.  Intended for
1116  * temporary mappings entered by mmu_booke_qenter.
1117  */
1118 static void
1119 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1120 {
1121 	vm_offset_t va;
1122 
1123 	va = sva;
1124 	while (count-- > 0) {
1125 		mmu_booke_kremove(mmu, va);
1126 		va += PAGE_SIZE;
1127 	}
1128 }
1129 
1130 /*
1131  * Map a wired page into kernel virtual address space.
1132  */
1133 static void
1134 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1135 {
1136 
1137 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1138 }
1139 
1140 static void
1141 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1142 {
1143 	uint32_t flags;
1144 	pte_t *pte;
1145 
1146 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1147 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1148 
1149 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1150 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1151 	flags |= PTE_PS_4KB;
1152 
1153 	pte = pte_find(mmu, kernel_pmap, va);
1154 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
1155 
1156 	mtx_lock_spin(&tlbivax_mutex);
1157 	tlb_miss_lock();
1158 
1159 	if (PTE_ISVALID(pte)) {
1160 
1161 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1162 
1163 		/* Flush entry from TLB0 */
1164 		tlb0_flush_entry(va);
1165 	}
1166 
1167 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1168 
1169 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1170 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1171 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1172 
1173 	/* Flush the real memory from the instruction cache. */
1174 	if ((flags & (PTE_I | PTE_G)) == 0)
1175 		__syncicache((void *)va, PAGE_SIZE);
1176 
1177 	tlb_miss_unlock();
1178 	mtx_unlock_spin(&tlbivax_mutex);
1179 }
1180 
1181 /*
1182  * Remove a page from kernel page table.
1183  */
1184 static void
1185 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1186 {
1187 	pte_t *pte;
1188 
1189 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
1190 
1191 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1192 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1193 	    ("mmu_booke_kremove: invalid va"));
1194 
1195 	pte = pte_find(mmu, kernel_pmap, va);
1196 
1197 	if (!PTE_ISVALID(pte)) {
1198 
1199 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1200 
1201 		return;
1202 	}
1203 
1204 	mtx_lock_spin(&tlbivax_mutex);
1205 	tlb_miss_lock();
1206 
1207 	/* Invalidate entry in TLB0, update PTE. */
1208 	tlb0_flush_entry(va);
1209 	*pte = 0;
1210 
1211 	tlb_miss_unlock();
1212 	mtx_unlock_spin(&tlbivax_mutex);
1213 }
1214 
1215 /*
1216  * Provide a kernel pointer corresponding to a given userland pointer.
1217  * The returned pointer is valid until the next time this function is
1218  * called in this thread. This is used internally in copyin/copyout.
1219  */
1220 int
1221 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
1222     void **kaddr, size_t ulen, size_t *klen)
1223 {
1224 
1225 	if (trunc_page((uintptr_t)uaddr + ulen) > VM_MAXUSER_ADDRESS)
1226 		return (EFAULT);
1227 
1228 	*kaddr = (void *)(uintptr_t)uaddr;
1229 	if (klen)
1230 		*klen = ulen;
1231 
1232 	return (0);
1233 }
1234 
1235 /*
1236  * Figure out where a given kernel pointer (usually in a fault) points
1237  * to from the VM's perspective, potentially remapping into userland's
1238  * address space.
1239  */
1240 static int
1241 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
1242     vm_offset_t *decoded_addr)
1243 {
1244 
1245 	if (trunc_page(addr) <= VM_MAXUSER_ADDRESS)
1246 		*is_user = 1;
1247 	else
1248 		*is_user = 0;
1249 
1250 	*decoded_addr = addr;
1251 	return (0);
1252 }
1253 
1254 static boolean_t
1255 mmu_booke_page_is_mapped(mmu_t mmu, vm_page_t m)
1256 {
1257 
1258 	return (!TAILQ_EMPTY(&(m)->md.pv_list));
1259 }
1260 
1261 /*
1262  * Initialize pmap associated with process 0.
1263  */
1264 static void
1265 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1266 {
1267 
1268 	PMAP_LOCK_INIT(pmap);
1269 	mmu_booke_pinit(mmu, pmap);
1270 	PCPU_SET(curpmap, pmap);
1271 }
1272 
1273 /*
1274  * Insert the given physical page at the specified virtual address in the
1275  * target physical map with the protection requested. If specified the page
1276  * will be wired down.
1277  */
1278 static int
1279 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1280     vm_prot_t prot, u_int flags, int8_t psind)
1281 {
1282 	int error;
1283 
1284 	rw_wlock(&pvh_global_lock);
1285 	PMAP_LOCK(pmap);
1286 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1287 	PMAP_UNLOCK(pmap);
1288 	rw_wunlock(&pvh_global_lock);
1289 	return (error);
1290 }
1291 
1292 static int
1293 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1294     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1295 {
1296 	pte_t *pte;
1297 	vm_paddr_t pa;
1298 	pte_t flags;
1299 	int error, su, sync;
1300 
1301 	pa = VM_PAGE_TO_PHYS(m);
1302 	su = (pmap == kernel_pmap);
1303 	sync = 0;
1304 
1305 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1306 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1307 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1308 	//		(u_int32_t)m, va, pa, prot, flags);
1309 
1310 	if (su) {
1311 		KASSERT(((va >= virtual_avail) &&
1312 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1313 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1314 	} else {
1315 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1316 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1317 	}
1318 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1319 		if ((pmap_flags & PMAP_ENTER_QUICK_LOCKED) == 0)
1320 			VM_PAGE_OBJECT_BUSY_ASSERT(m);
1321 		else
1322 			VM_OBJECT_ASSERT_LOCKED(m->object);
1323 	}
1324 
1325 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1326 
1327 	/*
1328 	 * If there is an existing mapping, and the physical address has not
1329 	 * changed, must be protection or wiring change.
1330 	 */
1331 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1332 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1333 
1334 		/*
1335 		 * Before actually updating pte->flags we calculate and
1336 		 * prepare its new value in a helper var.
1337 		 */
1338 		flags = *pte;
1339 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1340 
1341 		/* Wiring change, just update stats. */
1342 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1343 			if (!PTE_ISWIRED(pte)) {
1344 				flags |= PTE_WIRED;
1345 				pmap->pm_stats.wired_count++;
1346 			}
1347 		} else {
1348 			if (PTE_ISWIRED(pte)) {
1349 				flags &= ~PTE_WIRED;
1350 				pmap->pm_stats.wired_count--;
1351 			}
1352 		}
1353 
1354 		if (prot & VM_PROT_WRITE) {
1355 			/* Add write permissions. */
1356 			flags |= PTE_SW;
1357 			if (!su)
1358 				flags |= PTE_UW;
1359 
1360 			if ((flags & PTE_MANAGED) != 0)
1361 				vm_page_aflag_set(m, PGA_WRITEABLE);
1362 		} else {
1363 			/* Handle modified pages, sense modify status. */
1364 
1365 			/*
1366 			 * The PTE_MODIFIED flag could be set by underlying
1367 			 * TLB misses since we last read it (above), possibly
1368 			 * other CPUs could update it so we check in the PTE
1369 			 * directly rather than rely on that saved local flags
1370 			 * copy.
1371 			 */
1372 			if (PTE_ISMODIFIED(pte))
1373 				vm_page_dirty(m);
1374 		}
1375 
1376 		if (prot & VM_PROT_EXECUTE) {
1377 			flags |= PTE_SX;
1378 			if (!su)
1379 				flags |= PTE_UX;
1380 
1381 			/*
1382 			 * Check existing flags for execute permissions: if we
1383 			 * are turning execute permissions on, icache should
1384 			 * be flushed.
1385 			 */
1386 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1387 				sync++;
1388 		}
1389 
1390 		flags &= ~PTE_REFERENCED;
1391 
1392 		/*
1393 		 * The new flags value is all calculated -- only now actually
1394 		 * update the PTE.
1395 		 */
1396 		mtx_lock_spin(&tlbivax_mutex);
1397 		tlb_miss_lock();
1398 
1399 		tlb0_flush_entry(va);
1400 		*pte &= ~PTE_FLAGS_MASK;
1401 		*pte |= flags;
1402 
1403 		tlb_miss_unlock();
1404 		mtx_unlock_spin(&tlbivax_mutex);
1405 
1406 	} else {
1407 		/*
1408 		 * If there is an existing mapping, but it's for a different
1409 		 * physical address, pte_enter() will delete the old mapping.
1410 		 */
1411 		//if ((pte != NULL) && PTE_ISVALID(pte))
1412 		//	debugf("mmu_booke_enter_locked: replace\n");
1413 		//else
1414 		//	debugf("mmu_booke_enter_locked: new\n");
1415 
1416 		/* Now set up the flags and install the new mapping. */
1417 		flags = (PTE_SR | PTE_VALID);
1418 		flags |= PTE_M;
1419 
1420 		if (!su)
1421 			flags |= PTE_UR;
1422 
1423 		if (prot & VM_PROT_WRITE) {
1424 			flags |= PTE_SW;
1425 			if (!su)
1426 				flags |= PTE_UW;
1427 
1428 			if ((m->oflags & VPO_UNMANAGED) == 0)
1429 				vm_page_aflag_set(m, PGA_WRITEABLE);
1430 		}
1431 
1432 		if (prot & VM_PROT_EXECUTE) {
1433 			flags |= PTE_SX;
1434 			if (!su)
1435 				flags |= PTE_UX;
1436 		}
1437 
1438 		/* If its wired update stats. */
1439 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1440 			flags |= PTE_WIRED;
1441 
1442 		error = pte_enter(mmu, pmap, m, va, flags,
1443 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1444 		if (error != 0)
1445 			return (KERN_RESOURCE_SHORTAGE);
1446 
1447 		if ((flags & PMAP_ENTER_WIRED) != 0)
1448 			pmap->pm_stats.wired_count++;
1449 
1450 		/* Flush the real memory from the instruction cache. */
1451 		if (prot & VM_PROT_EXECUTE)
1452 			sync++;
1453 	}
1454 
1455 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1456 		__syncicache((void *)va, PAGE_SIZE);
1457 		sync = 0;
1458 	}
1459 
1460 	return (KERN_SUCCESS);
1461 }
1462 
1463 /*
1464  * Maps a sequence of resident pages belonging to the same object.
1465  * The sequence begins with the given page m_start.  This page is
1466  * mapped at the given virtual address start.  Each subsequent page is
1467  * mapped at a virtual address that is offset from start by the same
1468  * amount as the page is offset from m_start within the object.  The
1469  * last page in the sequence is the page with the largest offset from
1470  * m_start that can be mapped at a virtual address less than the given
1471  * virtual address end.  Not every virtual page between start and end
1472  * is mapped; only those for which a resident page exists with the
1473  * corresponding offset from m_start are mapped.
1474  */
1475 static void
1476 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1477     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1478 {
1479 	vm_page_t m;
1480 	vm_pindex_t diff, psize;
1481 
1482 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1483 
1484 	psize = atop(end - start);
1485 	m = m_start;
1486 	rw_wlock(&pvh_global_lock);
1487 	PMAP_LOCK(pmap);
1488 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1489 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1490 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1491 		    PMAP_ENTER_NOSLEEP | PMAP_ENTER_QUICK_LOCKED, 0);
1492 		m = TAILQ_NEXT(m, listq);
1493 	}
1494 	PMAP_UNLOCK(pmap);
1495 	rw_wunlock(&pvh_global_lock);
1496 }
1497 
1498 static void
1499 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1500     vm_prot_t prot)
1501 {
1502 
1503 	rw_wlock(&pvh_global_lock);
1504 	PMAP_LOCK(pmap);
1505 	mmu_booke_enter_locked(mmu, pmap, va, m,
1506 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP |
1507 	    PMAP_ENTER_QUICK_LOCKED, 0);
1508 	PMAP_UNLOCK(pmap);
1509 	rw_wunlock(&pvh_global_lock);
1510 }
1511 
1512 /*
1513  * Remove the given range of addresses from the specified map.
1514  *
1515  * It is assumed that the start and end are properly rounded to the page size.
1516  */
1517 static void
1518 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1519 {
1520 	pte_t *pte;
1521 	uint8_t hold_flag;
1522 
1523 	int su = (pmap == kernel_pmap);
1524 
1525 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1526 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1527 
1528 	if (su) {
1529 		KASSERT(((va >= virtual_avail) &&
1530 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1531 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1532 	} else {
1533 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1534 		    ("mmu_booke_remove: user pmap, non user va"));
1535 	}
1536 
1537 	if (PMAP_REMOVE_DONE(pmap)) {
1538 		//debugf("mmu_booke_remove: e (empty)\n");
1539 		return;
1540 	}
1541 
1542 	hold_flag = PTBL_HOLD_FLAG(pmap);
1543 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1544 
1545 	rw_wlock(&pvh_global_lock);
1546 	PMAP_LOCK(pmap);
1547 	for (; va < endva; va += PAGE_SIZE) {
1548 		pte = pte_find_next(mmu, pmap, &va);
1549 		if ((pte == NULL) || !PTE_ISVALID(pte))
1550 			break;
1551 		if (va >= endva)
1552 			break;
1553 		pte_remove(mmu, pmap, va, hold_flag);
1554 	}
1555 	PMAP_UNLOCK(pmap);
1556 	rw_wunlock(&pvh_global_lock);
1557 
1558 	//debugf("mmu_booke_remove: e\n");
1559 }
1560 
1561 /*
1562  * Remove physical page from all pmaps in which it resides.
1563  */
1564 static void
1565 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1566 {
1567 	pv_entry_t pv, pvn;
1568 	uint8_t hold_flag;
1569 
1570 	rw_wlock(&pvh_global_lock);
1571 	TAILQ_FOREACH_SAFE(pv, &m->md.pv_list, pv_link, pvn) {
1572 		PMAP_LOCK(pv->pv_pmap);
1573 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1574 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1575 		PMAP_UNLOCK(pv->pv_pmap);
1576 	}
1577 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1578 	rw_wunlock(&pvh_global_lock);
1579 }
1580 
1581 /*
1582  * Map a range of physical addresses into kernel virtual address space.
1583  */
1584 static vm_offset_t
1585 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1586     vm_paddr_t pa_end, int prot)
1587 {
1588 	vm_offset_t sva = *virt;
1589 	vm_offset_t va = sva;
1590 
1591 #ifdef __powerpc64__
1592 	/* XXX: Handle memory not starting at 0x0. */
1593 	if (pa_end < ctob(Maxmem))
1594 		return (PHYS_TO_DMAP(pa_start));
1595 #endif
1596 
1597 	while (pa_start < pa_end) {
1598 		mmu_booke_kenter(mmu, va, pa_start);
1599 		va += PAGE_SIZE;
1600 		pa_start += PAGE_SIZE;
1601 	}
1602 	*virt = va;
1603 
1604 	return (sva);
1605 }
1606 
1607 /*
1608  * The pmap must be activated before it's address space can be accessed in any
1609  * way.
1610  */
1611 static void
1612 mmu_booke_activate(mmu_t mmu, struct thread *td)
1613 {
1614 	pmap_t pmap;
1615 	u_int cpuid;
1616 
1617 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1618 
1619 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
1620 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1621 
1622 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1623 
1624 	sched_pin();
1625 
1626 	cpuid = PCPU_GET(cpuid);
1627 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1628 	PCPU_SET(curpmap, pmap);
1629 
1630 	if (pmap->pm_tid[cpuid] == TID_NONE)
1631 		tid_alloc(pmap);
1632 
1633 	/* Load PID0 register with pmap tid value. */
1634 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1635 	__asm __volatile("isync");
1636 
1637 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1638 
1639 	sched_unpin();
1640 
1641 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1642 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1643 }
1644 
1645 /*
1646  * Deactivate the specified process's address space.
1647  */
1648 static void
1649 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1650 {
1651 	pmap_t pmap;
1652 
1653 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1654 
1655 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
1656 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1657 
1658 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1659 
1660 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1661 	PCPU_SET(curpmap, NULL);
1662 }
1663 
1664 /*
1665  * Copy the range specified by src_addr/len
1666  * from the source map to the range dst_addr/len
1667  * in the destination map.
1668  *
1669  * This routine is only advisory and need not do anything.
1670  */
1671 static void
1672 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1673     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1674 {
1675 
1676 }
1677 
1678 /*
1679  * Set the physical protection on the specified range of this map as requested.
1680  */
1681 static void
1682 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1683     vm_prot_t prot)
1684 {
1685 	vm_offset_t va;
1686 	vm_page_t m;
1687 	pte_t *pte;
1688 
1689 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1690 		mmu_booke_remove(mmu, pmap, sva, eva);
1691 		return;
1692 	}
1693 
1694 	if (prot & VM_PROT_WRITE)
1695 		return;
1696 
1697 	PMAP_LOCK(pmap);
1698 	for (va = sva; va < eva; va += PAGE_SIZE) {
1699 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1700 			if (PTE_ISVALID(pte)) {
1701 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1702 
1703 				mtx_lock_spin(&tlbivax_mutex);
1704 				tlb_miss_lock();
1705 
1706 				/* Handle modified pages. */
1707 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1708 					vm_page_dirty(m);
1709 
1710 				tlb0_flush_entry(va);
1711 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1712 
1713 				tlb_miss_unlock();
1714 				mtx_unlock_spin(&tlbivax_mutex);
1715 			}
1716 		}
1717 	}
1718 	PMAP_UNLOCK(pmap);
1719 }
1720 
1721 /*
1722  * Clear the write and modified bits in each of the given page's mappings.
1723  */
1724 static void
1725 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1726 {
1727 	pv_entry_t pv;
1728 	pte_t *pte;
1729 
1730 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1731 	    ("mmu_booke_remove_write: page %p is not managed", m));
1732 	vm_page_assert_busied(m);
1733 
1734 	if (!pmap_page_is_write_mapped(m))
1735 	        return;
1736 	rw_wlock(&pvh_global_lock);
1737 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1738 		PMAP_LOCK(pv->pv_pmap);
1739 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1740 			if (PTE_ISVALID(pte)) {
1741 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1742 
1743 				mtx_lock_spin(&tlbivax_mutex);
1744 				tlb_miss_lock();
1745 
1746 				/* Handle modified pages. */
1747 				if (PTE_ISMODIFIED(pte))
1748 					vm_page_dirty(m);
1749 
1750 				/* Flush mapping from TLB0. */
1751 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1752 
1753 				tlb_miss_unlock();
1754 				mtx_unlock_spin(&tlbivax_mutex);
1755 			}
1756 		}
1757 		PMAP_UNLOCK(pv->pv_pmap);
1758 	}
1759 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1760 	rw_wunlock(&pvh_global_lock);
1761 }
1762 
1763 /*
1764  * Atomically extract and hold the physical page with the given
1765  * pmap and virtual address pair if that mapping permits the given
1766  * protection.
1767  */
1768 static vm_page_t
1769 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
1770     vm_prot_t prot)
1771 {
1772 	pte_t *pte;
1773 	vm_page_t m;
1774 	uint32_t pte_wbit;
1775 
1776 	m = NULL;
1777 	PMAP_LOCK(pmap);
1778 	pte = pte_find(mmu, pmap, va);
1779 	if ((pte != NULL) && PTE_ISVALID(pte)) {
1780 		if (pmap == kernel_pmap)
1781 			pte_wbit = PTE_SW;
1782 		else
1783 			pte_wbit = PTE_UW;
1784 
1785 		if ((*pte & pte_wbit) != 0 || (prot & VM_PROT_WRITE) == 0) {
1786 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1787 			if (!vm_page_wire_mapped(m))
1788 				m = NULL;
1789 		}
1790 	}
1791 	PMAP_UNLOCK(pmap);
1792 	return (m);
1793 }
1794 
1795 /*
1796  * Initialize a vm_page's machine-dependent fields.
1797  */
1798 static void
1799 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
1800 {
1801 
1802 	m->md.pv_tracked = 0;
1803 	TAILQ_INIT(&m->md.pv_list);
1804 }
1805 
1806 /*
1807  * Return whether or not the specified physical page was modified
1808  * in any of physical maps.
1809  */
1810 static boolean_t
1811 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
1812 {
1813 	pte_t *pte;
1814 	pv_entry_t pv;
1815 	boolean_t rv;
1816 
1817 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1818 	    ("mmu_booke_is_modified: page %p is not managed", m));
1819 	rv = FALSE;
1820 
1821 	/*
1822 	 * If the page is not busied then this check is racy.
1823 	 */
1824 	if (!pmap_page_is_write_mapped(m))
1825 		return (FALSE);
1826 
1827 	rw_wlock(&pvh_global_lock);
1828 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1829 		PMAP_LOCK(pv->pv_pmap);
1830 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
1831 		    PTE_ISVALID(pte)) {
1832 			if (PTE_ISMODIFIED(pte))
1833 				rv = TRUE;
1834 		}
1835 		PMAP_UNLOCK(pv->pv_pmap);
1836 		if (rv)
1837 			break;
1838 	}
1839 	rw_wunlock(&pvh_global_lock);
1840 	return (rv);
1841 }
1842 
1843 /*
1844  * Return whether or not the specified virtual address is eligible
1845  * for prefault.
1846  */
1847 static boolean_t
1848 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
1849 {
1850 
1851 	return (FALSE);
1852 }
1853 
1854 /*
1855  * Return whether or not the specified physical page was referenced
1856  * in any physical maps.
1857  */
1858 static boolean_t
1859 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
1860 {
1861 	pte_t *pte;
1862 	pv_entry_t pv;
1863 	boolean_t rv;
1864 
1865 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1866 	    ("mmu_booke_is_referenced: page %p is not managed", m));
1867 	rv = FALSE;
1868 	rw_wlock(&pvh_global_lock);
1869 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1870 		PMAP_LOCK(pv->pv_pmap);
1871 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
1872 		    PTE_ISVALID(pte)) {
1873 			if (PTE_ISREFERENCED(pte))
1874 				rv = TRUE;
1875 		}
1876 		PMAP_UNLOCK(pv->pv_pmap);
1877 		if (rv)
1878 			break;
1879 	}
1880 	rw_wunlock(&pvh_global_lock);
1881 	return (rv);
1882 }
1883 
1884 /*
1885  * Clear the modify bits on the specified physical page.
1886  */
1887 static void
1888 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
1889 {
1890 	pte_t *pte;
1891 	pv_entry_t pv;
1892 
1893 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1894 	    ("mmu_booke_clear_modify: page %p is not managed", m));
1895 	vm_page_assert_busied(m);
1896 
1897 	if (!pmap_page_is_write_mapped(m))
1898 	        return;
1899 
1900 	rw_wlock(&pvh_global_lock);
1901 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1902 		PMAP_LOCK(pv->pv_pmap);
1903 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
1904 		    PTE_ISVALID(pte)) {
1905 			mtx_lock_spin(&tlbivax_mutex);
1906 			tlb_miss_lock();
1907 
1908 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
1909 				tlb0_flush_entry(pv->pv_va);
1910 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
1911 				    PTE_REFERENCED);
1912 			}
1913 
1914 			tlb_miss_unlock();
1915 			mtx_unlock_spin(&tlbivax_mutex);
1916 		}
1917 		PMAP_UNLOCK(pv->pv_pmap);
1918 	}
1919 	rw_wunlock(&pvh_global_lock);
1920 }
1921 
1922 /*
1923  * Return a count of reference bits for a page, clearing those bits.
1924  * It is not necessary for every reference bit to be cleared, but it
1925  * is necessary that 0 only be returned when there are truly no
1926  * reference bits set.
1927  *
1928  * As an optimization, update the page's dirty field if a modified bit is
1929  * found while counting reference bits.  This opportunistic update can be
1930  * performed at low cost and can eliminate the need for some future calls
1931  * to pmap_is_modified().  However, since this function stops after
1932  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
1933  * dirty pages.  Those dirty pages will only be detected by a future call
1934  * to pmap_is_modified().
1935  */
1936 static int
1937 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
1938 {
1939 	pte_t *pte;
1940 	pv_entry_t pv;
1941 	int count;
1942 
1943 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1944 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
1945 	count = 0;
1946 	rw_wlock(&pvh_global_lock);
1947 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1948 		PMAP_LOCK(pv->pv_pmap);
1949 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
1950 		    PTE_ISVALID(pte)) {
1951 			if (PTE_ISMODIFIED(pte))
1952 				vm_page_dirty(m);
1953 			if (PTE_ISREFERENCED(pte)) {
1954 				mtx_lock_spin(&tlbivax_mutex);
1955 				tlb_miss_lock();
1956 
1957 				tlb0_flush_entry(pv->pv_va);
1958 				*pte &= ~PTE_REFERENCED;
1959 
1960 				tlb_miss_unlock();
1961 				mtx_unlock_spin(&tlbivax_mutex);
1962 
1963 				if (++count >= PMAP_TS_REFERENCED_MAX) {
1964 					PMAP_UNLOCK(pv->pv_pmap);
1965 					break;
1966 				}
1967 			}
1968 		}
1969 		PMAP_UNLOCK(pv->pv_pmap);
1970 	}
1971 	rw_wunlock(&pvh_global_lock);
1972 	return (count);
1973 }
1974 
1975 /*
1976  * Clear the wired attribute from the mappings for the specified range of
1977  * addresses in the given pmap.  Every valid mapping within that range must
1978  * have the wired attribute set.  In contrast, invalid mappings cannot have
1979  * the wired attribute set, so they are ignored.
1980  *
1981  * The wired attribute of the page table entry is not a hardware feature, so
1982  * there is no need to invalidate any TLB entries.
1983  */
1984 static void
1985 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
1986 {
1987 	vm_offset_t va;
1988 	pte_t *pte;
1989 
1990 	PMAP_LOCK(pmap);
1991 	for (va = sva; va < eva; va += PAGE_SIZE) {
1992 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
1993 		    PTE_ISVALID(pte)) {
1994 			if (!PTE_ISWIRED(pte))
1995 				panic("mmu_booke_unwire: pte %p isn't wired",
1996 				    pte);
1997 			*pte &= ~PTE_WIRED;
1998 			pmap->pm_stats.wired_count--;
1999 		}
2000 	}
2001 	PMAP_UNLOCK(pmap);
2002 
2003 }
2004 
2005 /*
2006  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2007  * page.  This count may be changed upwards or downwards in the future; it is
2008  * only necessary that true be returned for a small subset of pmaps for proper
2009  * page aging.
2010  */
2011 static boolean_t
2012 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2013 {
2014 	pv_entry_t pv;
2015 	int loops;
2016 	boolean_t rv;
2017 
2018 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2019 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2020 	loops = 0;
2021 	rv = FALSE;
2022 	rw_wlock(&pvh_global_lock);
2023 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2024 		if (pv->pv_pmap == pmap) {
2025 			rv = TRUE;
2026 			break;
2027 		}
2028 		if (++loops >= 16)
2029 			break;
2030 	}
2031 	rw_wunlock(&pvh_global_lock);
2032 	return (rv);
2033 }
2034 
2035 /*
2036  * Return the number of managed mappings to the given physical page that are
2037  * wired.
2038  */
2039 static int
2040 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2041 {
2042 	pv_entry_t pv;
2043 	pte_t *pte;
2044 	int count = 0;
2045 
2046 	if ((m->oflags & VPO_UNMANAGED) != 0)
2047 		return (count);
2048 	rw_wlock(&pvh_global_lock);
2049 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2050 		PMAP_LOCK(pv->pv_pmap);
2051 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2052 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2053 				count++;
2054 		PMAP_UNLOCK(pv->pv_pmap);
2055 	}
2056 	rw_wunlock(&pvh_global_lock);
2057 	return (count);
2058 }
2059 
2060 static int
2061 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2062 {
2063 	int i;
2064 	vm_offset_t va;
2065 
2066 	/*
2067 	 * This currently does not work for entries that
2068 	 * overlap TLB1 entries.
2069 	 */
2070 	for (i = 0; i < TLB1_ENTRIES; i ++) {
2071 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2072 			return (0);
2073 	}
2074 
2075 	return (EFAULT);
2076 }
2077 
2078 void
2079 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2080 {
2081 	vm_paddr_t ppa;
2082 	vm_offset_t ofs;
2083 	vm_size_t gran;
2084 
2085 	/* Minidumps are based on virtual memory addresses. */
2086 	if (do_minidump) {
2087 		*va = (void *)(vm_offset_t)pa;
2088 		return;
2089 	}
2090 
2091 	/* Raw physical memory dumps don't have a virtual address. */
2092 	/* We always map a 256MB page at 256M. */
2093 	gran = 256 * 1024 * 1024;
2094 	ppa = rounddown2(pa, gran);
2095 	ofs = pa - ppa;
2096 	*va = (void *)gran;
2097 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2098 
2099 	if (sz > (gran - ofs))
2100 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2101 		    _TLB_ENTRY_IO);
2102 }
2103 
2104 void
2105 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2106 {
2107 	vm_paddr_t ppa;
2108 	vm_offset_t ofs;
2109 	vm_size_t gran;
2110 	tlb_entry_t e;
2111 	int i;
2112 
2113 	/* Minidumps are based on virtual memory addresses. */
2114 	/* Nothing to do... */
2115 	if (do_minidump)
2116 		return;
2117 
2118 	for (i = 0; i < TLB1_ENTRIES; i++) {
2119 		tlb1_read_entry(&e, i);
2120 		if (!(e.mas1 & MAS1_VALID))
2121 			break;
2122 	}
2123 
2124 	/* Raw physical memory dumps don't have a virtual address. */
2125 	i--;
2126 	e.mas1 = 0;
2127 	e.mas2 = 0;
2128 	e.mas3 = 0;
2129 	tlb1_write_entry(&e, i);
2130 
2131 	gran = 256 * 1024 * 1024;
2132 	ppa = rounddown2(pa, gran);
2133 	ofs = pa - ppa;
2134 	if (sz > (gran - ofs)) {
2135 		i--;
2136 		e.mas1 = 0;
2137 		e.mas2 = 0;
2138 		e.mas3 = 0;
2139 		tlb1_write_entry(&e, i);
2140 	}
2141 }
2142 
2143 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2144 
2145 void
2146 mmu_booke_scan_init(mmu_t mmu)
2147 {
2148 	vm_offset_t va;
2149 	pte_t *pte;
2150 	int i;
2151 
2152 	if (!do_minidump) {
2153 		/* Initialize phys. segments for dumpsys(). */
2154 		memset(&dump_map, 0, sizeof(dump_map));
2155 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2156 		    &availmem_regions_sz);
2157 		for (i = 0; i < physmem_regions_sz; i++) {
2158 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2159 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2160 		}
2161 		return;
2162 	}
2163 
2164 	/* Virtual segments for minidumps: */
2165 	memset(&dump_map, 0, sizeof(dump_map));
2166 
2167 	/* 1st: kernel .data and .bss. */
2168 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2169 	dump_map[0].pa_size =
2170 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2171 
2172 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2173 	dump_map[1].pa_start = data_start;
2174 	dump_map[1].pa_size = data_end - data_start;
2175 
2176 	/* 3rd: kernel VM. */
2177 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2178 	/* Find start of next chunk (from va). */
2179 	while (va < virtual_end) {
2180 		/* Don't dump the buffer cache. */
2181 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2182 			va = kmi.buffer_eva;
2183 			continue;
2184 		}
2185 		pte = pte_find(mmu, kernel_pmap, va);
2186 		if (pte != NULL && PTE_ISVALID(pte))
2187 			break;
2188 		va += PAGE_SIZE;
2189 	}
2190 	if (va < virtual_end) {
2191 		dump_map[2].pa_start = va;
2192 		va += PAGE_SIZE;
2193 		/* Find last page in chunk. */
2194 		while (va < virtual_end) {
2195 			/* Don't run into the buffer cache. */
2196 			if (va == kmi.buffer_sva)
2197 				break;
2198 			pte = pte_find(mmu, kernel_pmap, va);
2199 			if (pte == NULL || !PTE_ISVALID(pte))
2200 				break;
2201 			va += PAGE_SIZE;
2202 		}
2203 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2204 	}
2205 }
2206 
2207 /*
2208  * Map a set of physical memory pages into the kernel virtual address space.
2209  * Return a pointer to where it is mapped. This routine is intended to be used
2210  * for mapping device memory, NOT real memory.
2211  */
2212 static void *
2213 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2214 {
2215 
2216 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2217 }
2218 
2219 static int
2220 tlb1_find_pa(vm_paddr_t pa, tlb_entry_t *e)
2221 {
2222 	int i;
2223 
2224 	for (i = 0; i < TLB1_ENTRIES; i++) {
2225 		tlb1_read_entry(e, i);
2226 		if ((e->mas1 & MAS1_VALID) == 0)
2227 			continue;
2228 		if (e->phys == pa)
2229 			return (i);
2230 	}
2231 	return (-1);
2232 }
2233 
2234 static void *
2235 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2236 {
2237 	tlb_entry_t e;
2238 	vm_paddr_t tmppa;
2239 #ifndef __powerpc64__
2240 	uintptr_t tmpva;
2241 #endif
2242 	uintptr_t va, retva;
2243 	vm_size_t sz;
2244 	int i;
2245 	int wimge;
2246 
2247 	/*
2248 	 * Check if this is premapped in TLB1.
2249 	 */
2250 	sz = size;
2251 	tmppa = pa;
2252 	va = ~0;
2253 	wimge = tlb_calc_wimg(pa, ma);
2254 	for (i = 0; i < TLB1_ENTRIES; i++) {
2255 		tlb1_read_entry(&e, i);
2256 		if (!(e.mas1 & MAS1_VALID))
2257 			continue;
2258 		if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
2259 			continue;
2260 		if (tmppa >= e.phys && tmppa < e.phys + e.size) {
2261 			va = e.virt + (pa - e.phys);
2262 			tmppa = e.phys + e.size;
2263 			sz -= MIN(sz, e.size - (pa - e.phys));
2264 			while (sz > 0 && (i = tlb1_find_pa(tmppa, &e)) != -1) {
2265 				if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
2266 					break;
2267 				sz -= MIN(sz, e.size);
2268 				tmppa = e.phys + e.size;
2269 			}
2270 			if (sz != 0)
2271 				break;
2272 			return ((void *)va);
2273 		}
2274 	}
2275 
2276 	size = roundup(size, PAGE_SIZE);
2277 
2278 #ifdef __powerpc64__
2279 	KASSERT(pa < VM_MAPDEV_PA_MAX,
2280 	    ("Unsupported physical address! %lx", pa));
2281 	va = VM_MAPDEV_BASE + pa;
2282 	retva = va;
2283 #ifdef POW2_MAPPINGS
2284 	/*
2285 	 * Align the mapping to a power of 2 size, taking into account that we
2286 	 * may need to increase the size multiple times to satisfy the size and
2287 	 * alignment requirements.
2288 	 *
2289 	 * This works in the general case because it's very rare (near never?)
2290 	 * to have different access properties (WIMG) within a single
2291 	 * power-of-two region.  If a design does call for that, POW2_MAPPINGS
2292 	 * can be undefined, and exact mappings will be used instead.
2293 	 */
2294 	sz = size;
2295 	size = roundup2(size, 1 << ilog2(size));
2296 	while (rounddown2(va, size) + size < va + sz)
2297 		size <<= 1;
2298 	va = rounddown2(va, size);
2299 	pa = rounddown2(pa, size);
2300 #endif
2301 #else
2302 	/*
2303 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
2304 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
2305 	 */
2306 #ifdef SPARSE_MAPDEV
2307 	/*
2308 	 * With a sparse mapdev, align to the largest starting region.  This
2309 	 * could feasibly be optimized for a 'best-fit' alignment, but that
2310 	 * calculation could be very costly.
2311 	 * Align to the smaller of:
2312 	 * - first set bit in overlap of (pa & size mask)
2313 	 * - largest size envelope
2314 	 *
2315 	 * It's possible the device mapping may start at a PA that's not larger
2316 	 * than the size mask, so we need to offset in to maximize the TLB entry
2317 	 * range and minimize the number of used TLB entries.
2318 	 */
2319 	do {
2320 	    tmpva = tlb1_map_base;
2321 	    sz = ffsl((~((1 << flsl(size-1)) - 1)) & pa);
2322 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
2323 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
2324 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
2325 #endif
2326 	va = atomic_fetchadd_int(&tlb1_map_base, size);
2327 	retva = va;
2328 #endif
2329 
2330 	if (tlb1_mapin_region(va, pa, size, tlb_calc_wimg(pa, ma)) != size)
2331 		return (NULL);
2332 
2333 	return ((void *)retva);
2334 }
2335 
2336 /*
2337  * 'Unmap' a range mapped by mmu_booke_mapdev().
2338  */
2339 static void
2340 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2341 {
2342 #ifdef SUPPORTS_SHRINKING_TLB1
2343 	vm_offset_t base, offset;
2344 
2345 	/*
2346 	 * Unmap only if this is inside kernel virtual space.
2347 	 */
2348 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2349 		base = trunc_page(va);
2350 		offset = va & PAGE_MASK;
2351 		size = roundup(offset + size, PAGE_SIZE);
2352 		kva_free(base, size);
2353 	}
2354 #endif
2355 }
2356 
2357 /*
2358  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2359  * specified pmap. This eliminates the blast of soft faults on process startup
2360  * and immediately after an mmap.
2361  */
2362 static void
2363 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2364     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2365 {
2366 
2367 	VM_OBJECT_ASSERT_WLOCKED(object);
2368 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2369 	    ("mmu_booke_object_init_pt: non-device object"));
2370 }
2371 
2372 /*
2373  * Perform the pmap work for mincore.
2374  */
2375 static int
2376 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2377     vm_paddr_t *pap)
2378 {
2379 
2380 	/* XXX: this should be implemented at some point */
2381 	return (0);
2382 }
2383 
2384 static int
2385 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
2386     vm_memattr_t mode)
2387 {
2388 	vm_offset_t va;
2389 	pte_t *pte;
2390 	int i, j;
2391 	tlb_entry_t e;
2392 
2393 	addr = trunc_page(addr);
2394 
2395 	/* Only allow changes to mapped kernel addresses.  This includes:
2396 	 * - KVA
2397 	 * - DMAP (powerpc64)
2398 	 * - Device mappings
2399 	 */
2400 	if (addr <= VM_MAXUSER_ADDRESS ||
2401 #ifdef __powerpc64__
2402 	    (addr >= tlb1_map_base && addr < DMAP_BASE_ADDRESS) ||
2403 	    (addr > DMAP_MAX_ADDRESS && addr < VM_MIN_KERNEL_ADDRESS) ||
2404 #else
2405 	    (addr >= tlb1_map_base && addr < VM_MIN_KERNEL_ADDRESS) ||
2406 #endif
2407 	    (addr > VM_MAX_KERNEL_ADDRESS))
2408 		return (EINVAL);
2409 
2410 	/* Check TLB1 mappings */
2411 	for (i = 0; i < TLB1_ENTRIES; i++) {
2412 		tlb1_read_entry(&e, i);
2413 		if (!(e.mas1 & MAS1_VALID))
2414 			continue;
2415 		if (addr >= e.virt && addr < e.virt + e.size)
2416 			break;
2417 	}
2418 	if (i < TLB1_ENTRIES) {
2419 		/* Only allow full mappings to be modified for now. */
2420 		/* Validate the range. */
2421 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
2422 			tlb1_read_entry(&e, j);
2423 			if (va != e.virt || (sz - (va - addr) < e.size))
2424 				return (EINVAL);
2425 		}
2426 		for (va = addr; va < addr + sz; va += e.size, i++) {
2427 			tlb1_read_entry(&e, i);
2428 			e.mas2 &= ~MAS2_WIMGE_MASK;
2429 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
2430 
2431 			/*
2432 			 * Write it out to the TLB.  Should really re-sync with other
2433 			 * cores.
2434 			 */
2435 			tlb1_write_entry(&e, i);
2436 		}
2437 		return (0);
2438 	}
2439 
2440 	/* Not in TLB1, try through pmap */
2441 	/* First validate the range. */
2442 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2443 		pte = pte_find(mmu, kernel_pmap, va);
2444 		if (pte == NULL || !PTE_ISVALID(pte))
2445 			return (EINVAL);
2446 	}
2447 
2448 	mtx_lock_spin(&tlbivax_mutex);
2449 	tlb_miss_lock();
2450 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2451 		pte = pte_find(mmu, kernel_pmap, va);
2452 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2453 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
2454 		tlb0_flush_entry(va);
2455 	}
2456 	tlb_miss_unlock();
2457 	mtx_unlock_spin(&tlbivax_mutex);
2458 
2459 	return (0);
2460 }
2461 
2462 static void
2463 mmu_booke_page_array_startup(mmu_t mmu, long pages)
2464 {
2465 	vm_page_array_size = pages;
2466 }
2467 
2468 /**************************************************************************/
2469 /* TID handling */
2470 /**************************************************************************/
2471 
2472 /*
2473  * Allocate a TID. If necessary, steal one from someone else.
2474  * The new TID is flushed from the TLB before returning.
2475  */
2476 static tlbtid_t
2477 tid_alloc(pmap_t pmap)
2478 {
2479 	tlbtid_t tid;
2480 	int thiscpu;
2481 
2482 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2483 
2484 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2485 
2486 	thiscpu = PCPU_GET(cpuid);
2487 
2488 	tid = PCPU_GET(booke.tid_next);
2489 	if (tid > TID_MAX)
2490 		tid = TID_MIN;
2491 	PCPU_SET(booke.tid_next, tid + 1);
2492 
2493 	/* If we are stealing TID then clear the relevant pmap's field */
2494 	if (tidbusy[thiscpu][tid] != NULL) {
2495 
2496 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2497 
2498 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2499 
2500 		/* Flush all entries from TLB0 matching this TID. */
2501 		tid_flush(tid);
2502 	}
2503 
2504 	tidbusy[thiscpu][tid] = pmap;
2505 	pmap->pm_tid[thiscpu] = tid;
2506 	__asm __volatile("msync; isync");
2507 
2508 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2509 	    PCPU_GET(booke.tid_next));
2510 
2511 	return (tid);
2512 }
2513 
2514 /**************************************************************************/
2515 /* TLB0 handling */
2516 /**************************************************************************/
2517 
2518 /* Convert TLB0 va and way number to tlb0[] table index. */
2519 static inline unsigned int
2520 tlb0_tableidx(vm_offset_t va, unsigned int way)
2521 {
2522 	unsigned int idx;
2523 
2524 	idx = (way * TLB0_ENTRIES_PER_WAY);
2525 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2526 	return (idx);
2527 }
2528 
2529 /*
2530  * Invalidate TLB0 entry.
2531  */
2532 static inline void
2533 tlb0_flush_entry(vm_offset_t va)
2534 {
2535 
2536 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2537 
2538 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2539 
2540 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2541 	__asm __volatile("isync; msync");
2542 	__asm __volatile("tlbsync; msync");
2543 
2544 	CTR1(KTR_PMAP, "%s: e", __func__);
2545 }
2546 
2547 
2548 /**************************************************************************/
2549 /* TLB1 handling */
2550 /**************************************************************************/
2551 
2552 /*
2553  * TLB1 mapping notes:
2554  *
2555  * TLB1[0]	Kernel text and data.
2556  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2557  *		windows, other devices mappings.
2558  */
2559 
2560  /*
2561  * Read an entry from given TLB1 slot.
2562  */
2563 void
2564 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
2565 {
2566 	register_t msr;
2567 	uint32_t mas0;
2568 
2569 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
2570 
2571 	msr = mfmsr();
2572 	__asm __volatile("wrteei 0");
2573 
2574 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
2575 	mtspr(SPR_MAS0, mas0);
2576 	__asm __volatile("isync; tlbre");
2577 
2578 	entry->mas1 = mfspr(SPR_MAS1);
2579 	entry->mas2 = mfspr(SPR_MAS2);
2580 	entry->mas3 = mfspr(SPR_MAS3);
2581 
2582 	switch ((mfpvr() >> 16) & 0xFFFF) {
2583 	case FSL_E500v2:
2584 	case FSL_E500mc:
2585 	case FSL_E5500:
2586 	case FSL_E6500:
2587 		entry->mas7 = mfspr(SPR_MAS7);
2588 		break;
2589 	default:
2590 		entry->mas7 = 0;
2591 		break;
2592 	}
2593 	__asm __volatile("wrtee %0" :: "r"(msr));
2594 
2595 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
2596 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
2597 	    (entry->mas3 & MAS3_RPN);
2598 	entry->size =
2599 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
2600 }
2601 
2602 struct tlbwrite_args {
2603 	tlb_entry_t *e;
2604 	unsigned int idx;
2605 };
2606 
2607 static uint32_t
2608 tlb1_find_free(void)
2609 {
2610 	tlb_entry_t e;
2611 	int i;
2612 
2613 	for (i = 0; i < TLB1_ENTRIES; i++) {
2614 		tlb1_read_entry(&e, i);
2615 		if ((e.mas1 & MAS1_VALID) == 0)
2616 			return (i);
2617 	}
2618 	return (-1);
2619 }
2620 
2621 static void
2622 tlb1_purge_va_range(vm_offset_t va, vm_size_t size)
2623 {
2624 	tlb_entry_t e;
2625 	int i;
2626 
2627 	for (i = 0; i < TLB1_ENTRIES; i++) {
2628 		tlb1_read_entry(&e, i);
2629 		if ((e.mas1 & MAS1_VALID) == 0)
2630 			continue;
2631 		if ((e.mas2 & MAS2_EPN_MASK) >= va &&
2632 		    (e.mas2 & MAS2_EPN_MASK) < va + size) {
2633 			mtspr(SPR_MAS1, e.mas1 & ~MAS1_VALID);
2634 			__asm __volatile("isync; tlbwe; isync; msync");
2635 		}
2636 	}
2637 }
2638 
2639 static void
2640 tlb1_write_entry_int(void *arg)
2641 {
2642 	struct tlbwrite_args *args = arg;
2643 	uint32_t idx, mas0;
2644 
2645 	idx = args->idx;
2646 	if (idx == -1) {
2647 		tlb1_purge_va_range(args->e->virt, args->e->size);
2648 		idx = tlb1_find_free();
2649 		if (idx == -1)
2650 			panic("No free TLB1 entries!\n");
2651 	}
2652 	/* Select entry */
2653 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2654 
2655 	mtspr(SPR_MAS0, mas0);
2656 	mtspr(SPR_MAS1, args->e->mas1);
2657 	mtspr(SPR_MAS2, args->e->mas2);
2658 	mtspr(SPR_MAS3, args->e->mas3);
2659 	switch ((mfpvr() >> 16) & 0xFFFF) {
2660 	case FSL_E500mc:
2661 	case FSL_E5500:
2662 	case FSL_E6500:
2663 		mtspr(SPR_MAS8, 0);
2664 		/* FALLTHROUGH */
2665 	case FSL_E500v2:
2666 		mtspr(SPR_MAS7, args->e->mas7);
2667 		break;
2668 	default:
2669 		break;
2670 	}
2671 
2672 	__asm __volatile("isync; tlbwe; isync; msync");
2673 
2674 }
2675 
2676 static void
2677 tlb1_write_entry_sync(void *arg)
2678 {
2679 	/* Empty synchronization point for smp_rendezvous(). */
2680 }
2681 
2682 /*
2683  * Write given entry to TLB1 hardware.
2684  */
2685 static void
2686 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
2687 {
2688 	struct tlbwrite_args args;
2689 
2690 	args.e = e;
2691 	args.idx = idx;
2692 
2693 #ifdef SMP
2694 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
2695 		mb();
2696 		smp_rendezvous(tlb1_write_entry_sync,
2697 		    tlb1_write_entry_int,
2698 		    tlb1_write_entry_sync, &args);
2699 	} else
2700 #endif
2701 	{
2702 		register_t msr;
2703 
2704 		msr = mfmsr();
2705 		__asm __volatile("wrteei 0");
2706 		tlb1_write_entry_int(&args);
2707 		__asm __volatile("wrtee %0" :: "r"(msr));
2708 	}
2709 }
2710 
2711 /*
2712  * Convert TLB TSIZE value to mapped region size.
2713  */
2714 static vm_size_t
2715 tsize2size(unsigned int tsize)
2716 {
2717 
2718 	/*
2719 	 * size = 4^tsize KB
2720 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2721 	 */
2722 
2723 	return ((1 << (2 * tsize)) * 1024);
2724 }
2725 
2726 /*
2727  * Convert region size (must be power of 4) to TLB TSIZE value.
2728  */
2729 static unsigned int
2730 size2tsize(vm_size_t size)
2731 {
2732 
2733 	return (ilog2(size) / 2 - 5);
2734 }
2735 
2736 /*
2737  * Register permanent kernel mapping in TLB1.
2738  *
2739  * Entries are created starting from index 0 (current free entry is
2740  * kept in tlb1_idx) and are not supposed to be invalidated.
2741  */
2742 int
2743 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
2744     uint32_t flags)
2745 {
2746 	tlb_entry_t e;
2747 	uint32_t ts, tid;
2748 	int tsize, index;
2749 
2750 	/* First try to update an existing entry. */
2751 	for (index = 0; index < TLB1_ENTRIES; index++) {
2752 		tlb1_read_entry(&e, index);
2753 		/* Check if we're just updating the flags, and update them. */
2754 		if (e.phys == pa && e.virt == va && e.size == size) {
2755 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
2756 			tlb1_write_entry(&e, index);
2757 			return (0);
2758 		}
2759 	}
2760 
2761 	/* Convert size to TSIZE */
2762 	tsize = size2tsize(size);
2763 
2764 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2765 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2766 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2767 
2768 	e.phys = pa;
2769 	e.virt = va;
2770 	e.size = size;
2771 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2772 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2773 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
2774 
2775 	/* Set supervisor RWX permission bits */
2776 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2777 	e.mas7 = (pa >> 32) & MAS7_RPN;
2778 
2779 	tlb1_write_entry(&e, -1);
2780 
2781 	return (0);
2782 }
2783 
2784 /*
2785  * Map in contiguous RAM region into the TLB1.
2786  */
2787 static vm_size_t
2788 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size, int wimge)
2789 {
2790 	vm_offset_t base;
2791 	vm_size_t mapped, sz, ssize;
2792 
2793 	mapped = 0;
2794 	base = va;
2795 	ssize = size;
2796 
2797 	while (size > 0) {
2798 		sz = 1UL << (ilog2(size) & ~1);
2799 		/* Align size to PA */
2800 		if (pa % sz != 0) {
2801 			do {
2802 				sz >>= 2;
2803 			} while (pa % sz != 0);
2804 		}
2805 		/* Now align from there to VA */
2806 		if (va % sz != 0) {
2807 			do {
2808 				sz >>= 2;
2809 			} while (va % sz != 0);
2810 		}
2811 #ifdef __powerpc64__
2812 		/*
2813 		 * Clamp TLB1 entries to 4G.
2814 		 *
2815 		 * While the e6500 supports up to 1TB mappings, the e5500
2816 		 * only supports up to 4G mappings. (0b1011)
2817 		 *
2818 		 * If any e6500 machines capable of supporting a very
2819 		 * large amount of memory appear in the future, we can
2820 		 * revisit this.
2821 		 *
2822 		 * For now, though, since we have plenty of space in TLB1,
2823 		 * always avoid creating entries larger than 4GB.
2824 		 */
2825 		sz = MIN(sz, 1UL << 32);
2826 #endif
2827 		if (bootverbose)
2828 			printf("Wiring VA=%p to PA=%jx (size=%lx)\n",
2829 			    (void *)va, (uintmax_t)pa, (long)sz);
2830 		if (tlb1_set_entry(va, pa, sz,
2831 		    _TLB_ENTRY_SHARED | wimge) < 0)
2832 			return (mapped);
2833 		size -= sz;
2834 		pa += sz;
2835 		va += sz;
2836 	}
2837 
2838 	mapped = (va - base);
2839 	if (bootverbose)
2840 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
2841 		    mapped, mapped - ssize);
2842 
2843 	return (mapped);
2844 }
2845 
2846 /*
2847  * TLB1 initialization routine, to be called after the very first
2848  * assembler level setup done in locore.S.
2849  */
2850 void
2851 tlb1_init()
2852 {
2853 	vm_offset_t mas2;
2854 	uint32_t mas0, mas1, mas3, mas7;
2855 	uint32_t tsz;
2856 
2857 	tlb1_get_tlbconf();
2858 
2859 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
2860 	mtspr(SPR_MAS0, mas0);
2861 	__asm __volatile("isync; tlbre");
2862 
2863 	mas1 = mfspr(SPR_MAS1);
2864 	mas2 = mfspr(SPR_MAS2);
2865 	mas3 = mfspr(SPR_MAS3);
2866 	mas7 = mfspr(SPR_MAS7);
2867 
2868 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
2869 	    (mas3 & MAS3_RPN);
2870 
2871 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2872 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
2873 	kernstart = trunc_page(mas2);
2874 
2875 	/* Setup TLB miss defaults */
2876 	set_mas4_defaults();
2877 }
2878 
2879 /*
2880  * pmap_early_io_unmap() should be used in short conjunction with
2881  * pmap_early_io_map(), as in the following snippet:
2882  *
2883  * x = pmap_early_io_map(...);
2884  * <do something with x>
2885  * pmap_early_io_unmap(x, size);
2886  *
2887  * And avoiding more allocations between.
2888  */
2889 void
2890 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
2891 {
2892 	int i;
2893 	tlb_entry_t e;
2894 	vm_size_t isize;
2895 
2896 	size = roundup(size, PAGE_SIZE);
2897 	isize = size;
2898 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
2899 		tlb1_read_entry(&e, i);
2900 		if (!(e.mas1 & MAS1_VALID))
2901 			continue;
2902 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
2903 			size -= e.size;
2904 			e.mas1 &= ~MAS1_VALID;
2905 			tlb1_write_entry(&e, i);
2906 		}
2907 	}
2908 	if (tlb1_map_base == va + isize)
2909 		tlb1_map_base -= isize;
2910 }
2911 
2912 vm_offset_t
2913 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
2914 {
2915 	vm_paddr_t pa_base;
2916 	vm_offset_t va, sz;
2917 	int i;
2918 	tlb_entry_t e;
2919 
2920 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
2921 
2922 	for (i = 0; i < TLB1_ENTRIES; i++) {
2923 		tlb1_read_entry(&e, i);
2924 		if (!(e.mas1 & MAS1_VALID))
2925 			continue;
2926 		if (pa >= e.phys && (pa + size) <=
2927 		    (e.phys + e.size))
2928 			return (e.virt + (pa - e.phys));
2929 	}
2930 
2931 	pa_base = rounddown(pa, PAGE_SIZE);
2932 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
2933 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
2934 	va = tlb1_map_base + (pa - pa_base);
2935 
2936 	do {
2937 		sz = 1 << (ilog2(size) & ~1);
2938 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
2939 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
2940 		size -= sz;
2941 		pa_base += sz;
2942 		tlb1_map_base += sz;
2943 	} while (size > 0);
2944 
2945 	return (va);
2946 }
2947 
2948 void
2949 pmap_track_page(pmap_t pmap, vm_offset_t va)
2950 {
2951 	vm_paddr_t pa;
2952 	vm_page_t page;
2953 	struct pv_entry *pve;
2954 
2955 	va = trunc_page(va);
2956 	pa = pmap_kextract(va);
2957 	page = PHYS_TO_VM_PAGE(pa);
2958 
2959 	rw_wlock(&pvh_global_lock);
2960 	PMAP_LOCK(pmap);
2961 
2962 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
2963 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
2964 			goto out;
2965 		}
2966 	}
2967 	page->md.pv_tracked = true;
2968 	pv_insert(pmap, va, page);
2969 out:
2970 	PMAP_UNLOCK(pmap);
2971 	rw_wunlock(&pvh_global_lock);
2972 }
2973 
2974 
2975 /*
2976  * Setup MAS4 defaults.
2977  * These values are loaded to MAS0-2 on a TLB miss.
2978  */
2979 static void
2980 set_mas4_defaults(void)
2981 {
2982 	uint32_t mas4;
2983 
2984 	/* Defaults: TLB0, PID0, TSIZED=4K */
2985 	mas4 = MAS4_TLBSELD0;
2986 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
2987 #ifdef SMP
2988 	mas4 |= MAS4_MD;
2989 #endif
2990 	mtspr(SPR_MAS4, mas4);
2991 	__asm __volatile("isync");
2992 }
2993 
2994 
2995 /*
2996  * Return 0 if the physical IO range is encompassed by one of the
2997  * the TLB1 entries, otherwise return related error code.
2998  */
2999 static int
3000 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3001 {
3002 	uint32_t prot;
3003 	vm_paddr_t pa_start;
3004 	vm_paddr_t pa_end;
3005 	unsigned int entry_tsize;
3006 	vm_size_t entry_size;
3007 	tlb_entry_t e;
3008 
3009 	*va = (vm_offset_t)NULL;
3010 
3011 	tlb1_read_entry(&e, i);
3012 	/* Skip invalid entries */
3013 	if (!(e.mas1 & MAS1_VALID))
3014 		return (EINVAL);
3015 
3016 	/*
3017 	 * The entry must be cache-inhibited, guarded, and r/w
3018 	 * so it can function as an i/o page
3019 	 */
3020 	prot = e.mas2 & (MAS2_I | MAS2_G);
3021 	if (prot != (MAS2_I | MAS2_G))
3022 		return (EPERM);
3023 
3024 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
3025 	if (prot != (MAS3_SR | MAS3_SW))
3026 		return (EPERM);
3027 
3028 	/* The address should be within the entry range. */
3029 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3030 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3031 
3032 	entry_size = tsize2size(entry_tsize);
3033 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
3034 	    (e.mas3 & MAS3_RPN);
3035 	pa_end = pa_start + entry_size;
3036 
3037 	if ((pa < pa_start) || ((pa + size) > pa_end))
3038 		return (ERANGE);
3039 
3040 	/* Return virtual address of this mapping. */
3041 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3042 	return (0);
3043 }
3044 
3045 #ifdef DDB
3046 /* Print out contents of the MAS registers for each TLB0 entry */
3047 static void
3048 #ifdef __powerpc64__
3049 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
3050 #else
3051 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3052 #endif
3053     uint32_t mas7)
3054 {
3055 	int as;
3056 	char desc[3];
3057 	tlbtid_t tid;
3058 	vm_size_t size;
3059 	unsigned int tsize;
3060 
3061 	desc[2] = '\0';
3062 	if (mas1 & MAS1_VALID)
3063 		desc[0] = 'V';
3064 	else
3065 		desc[0] = ' ';
3066 
3067 	if (mas1 & MAS1_IPROT)
3068 		desc[1] = 'P';
3069 	else
3070 		desc[1] = ' ';
3071 
3072 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3073 	tid = MAS1_GETTID(mas1);
3074 
3075 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3076 	size = 0;
3077 	if (tsize)
3078 		size = tsize2size(tsize);
3079 
3080 	printf("%3d: (%s) [AS=%d] "
3081 	    "sz = 0x%jx tsz = %d tid = %d mas1 = 0x%08x "
3082 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3083 	    i, desc, as, (uintmax_t)size, tsize, tid, mas1, mas2, mas3, mas7);
3084 }
3085 
3086 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
3087 {
3088 	uint32_t mas0, mas1, mas3, mas7;
3089 #ifdef __powerpc64__
3090 	uint64_t mas2;
3091 #else
3092 	uint32_t mas2;
3093 #endif
3094 	int entryidx, way, idx;
3095 
3096 	printf("TLB0 entries:\n");
3097 	for (way = 0; way < TLB0_WAYS; way ++)
3098 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3099 
3100 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3101 			mtspr(SPR_MAS0, mas0);
3102 
3103 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3104 			mtspr(SPR_MAS2, mas2);
3105 
3106 			__asm __volatile("isync; tlbre");
3107 
3108 			mas1 = mfspr(SPR_MAS1);
3109 			mas2 = mfspr(SPR_MAS2);
3110 			mas3 = mfspr(SPR_MAS3);
3111 			mas7 = mfspr(SPR_MAS7);
3112 
3113 			idx = tlb0_tableidx(mas2, way);
3114 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3115 		}
3116 }
3117 
3118 /*
3119  * Print out contents of the MAS registers for each TLB1 entry
3120  */
3121 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
3122 {
3123 	uint32_t mas0, mas1, mas3, mas7;
3124 #ifdef __powerpc64__
3125 	uint64_t mas2;
3126 #else
3127 	uint32_t mas2;
3128 #endif
3129 	int i;
3130 
3131 	printf("TLB1 entries:\n");
3132 	for (i = 0; i < TLB1_ENTRIES; i++) {
3133 
3134 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3135 		mtspr(SPR_MAS0, mas0);
3136 
3137 		__asm __volatile("isync; tlbre");
3138 
3139 		mas1 = mfspr(SPR_MAS1);
3140 		mas2 = mfspr(SPR_MAS2);
3141 		mas3 = mfspr(SPR_MAS3);
3142 		mas7 = mfspr(SPR_MAS7);
3143 
3144 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3145 	}
3146 }
3147 #endif
3148