xref: /freebsd/sys/powerpc/booke/pmap.c (revision 83129c0b650734f4ca6c6f9c214aa3110d2f129d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - kernel_pdir-1            : message buffer
63   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
64   *             kernel_pp2d - .                        : kernel pointers to page directory
65   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
66   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
67   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
68   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
69   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
70   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
71   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
72   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
73   *                         - 0xffff_ffff_ffff_ffff    : device direct map
74   */
75 
76 #include <sys/cdefs.h>
77 __FBSDID("$FreeBSD$");
78 
79 #include "opt_ddb.h"
80 #include "opt_kstack_pages.h"
81 
82 #include <sys/param.h>
83 #include <sys/conf.h>
84 #include <sys/malloc.h>
85 #include <sys/ktr.h>
86 #include <sys/proc.h>
87 #include <sys/user.h>
88 #include <sys/queue.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/kerneldump.h>
92 #include <sys/linker.h>
93 #include <sys/msgbuf.h>
94 #include <sys/lock.h>
95 #include <sys/mutex.h>
96 #include <sys/rwlock.h>
97 #include <sys/sched.h>
98 #include <sys/smp.h>
99 #include <sys/vmmeter.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_kern.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_param.h>
108 #include <vm/vm_map.h>
109 #include <vm/vm_pager.h>
110 #include <vm/vm_phys.h>
111 #include <vm/vm_pagequeue.h>
112 #include <vm/uma.h>
113 
114 #include <machine/_inttypes.h>
115 #include <machine/cpu.h>
116 #include <machine/pcb.h>
117 #include <machine/platform.h>
118 
119 #include <machine/tlb.h>
120 #include <machine/spr.h>
121 #include <machine/md_var.h>
122 #include <machine/mmuvar.h>
123 #include <machine/pmap.h>
124 #include <machine/pte.h>
125 
126 #include <ddb/ddb.h>
127 
128 #include "mmu_if.h"
129 
130 #define	SPARSE_MAPDEV
131 #ifdef  DEBUG
132 #define debugf(fmt, args...) printf(fmt, ##args)
133 #else
134 #define debugf(fmt, args...)
135 #endif
136 
137 #ifdef __powerpc64__
138 #define	PRI0ptrX	"016lx"
139 #else
140 #define	PRI0ptrX	"08x"
141 #endif
142 
143 #define TODO			panic("%s: not implemented", __func__);
144 
145 extern unsigned char _etext[];
146 extern unsigned char _end[];
147 
148 extern uint32_t *bootinfo;
149 
150 vm_paddr_t kernload;
151 vm_offset_t kernstart;
152 vm_size_t kernsize;
153 
154 /* Message buffer and tables. */
155 static vm_offset_t data_start;
156 static vm_size_t data_end;
157 
158 /* Phys/avail memory regions. */
159 static struct mem_region *availmem_regions;
160 static int availmem_regions_sz;
161 static struct mem_region *physmem_regions;
162 static int physmem_regions_sz;
163 
164 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
165 static vm_offset_t zero_page_va;
166 static struct mtx zero_page_mutex;
167 
168 static struct mtx tlbivax_mutex;
169 
170 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
171 static vm_offset_t copy_page_src_va;
172 static vm_offset_t copy_page_dst_va;
173 static struct mtx copy_page_mutex;
174 
175 /**************************************************************************/
176 /* PMAP */
177 /**************************************************************************/
178 
179 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
180     vm_prot_t, u_int flags, int8_t psind);
181 
182 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
183 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
184 #ifdef __powerpc64__
185 unsigned int kernel_pdirs;
186 #endif
187 static uma_zone_t ptbl_root_zone;
188 
189 /*
190  * If user pmap is processed with mmu_booke_remove and the resident count
191  * drops to 0, there are no more pages to remove, so we need not continue.
192  */
193 #define PMAP_REMOVE_DONE(pmap) \
194 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
195 
196 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
197 extern int elf32_nxstack;
198 #endif
199 
200 /**************************************************************************/
201 /* TLB and TID handling */
202 /**************************************************************************/
203 
204 /* Translation ID busy table */
205 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
206 
207 /*
208  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
209  * core revisions and should be read from h/w registers during early config.
210  */
211 uint32_t tlb0_entries;
212 uint32_t tlb0_ways;
213 uint32_t tlb0_entries_per_way;
214 uint32_t tlb1_entries;
215 
216 #define TLB0_ENTRIES		(tlb0_entries)
217 #define TLB0_WAYS		(tlb0_ways)
218 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
219 
220 #define TLB1_ENTRIES (tlb1_entries)
221 
222 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
223 
224 static tlbtid_t tid_alloc(struct pmap *);
225 static void tid_flush(tlbtid_t tid);
226 
227 #ifdef DDB
228 #ifdef __powerpc64__
229 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
230 #else
231 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
232 #endif
233 #endif
234 
235 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
236 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
237 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
238 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
239 
240 static vm_size_t tsize2size(unsigned int);
241 static unsigned int size2tsize(vm_size_t);
242 static unsigned int ilog2(unsigned long);
243 
244 static void set_mas4_defaults(void);
245 
246 static inline void tlb0_flush_entry(vm_offset_t);
247 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
248 
249 /**************************************************************************/
250 /* Page table management */
251 /**************************************************************************/
252 
253 static struct rwlock_padalign pvh_global_lock;
254 
255 /* Data for the pv entry allocation mechanism */
256 static uma_zone_t pvzone;
257 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
258 
259 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
260 
261 #ifndef PMAP_SHPGPERPROC
262 #define PMAP_SHPGPERPROC	200
263 #endif
264 
265 #ifdef __powerpc64__
266 #define PMAP_ROOT_SIZE	(sizeof(pte_t***) * PP2D_NENTRIES)
267 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
268 			 unsigned int, boolean_t);
269 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int, vm_page_t);
270 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
271 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
272 #else
273 #define PMAP_ROOT_SIZE	(sizeof(pte_t**) * PDIR_NENTRIES)
274 static void ptbl_init(void);
275 static struct ptbl_buf *ptbl_buf_alloc(void);
276 static void ptbl_buf_free(struct ptbl_buf *);
277 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
278 
279 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
280 static void ptbl_free(mmu_t, pmap_t, unsigned int);
281 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
282 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
283 #endif
284 
285 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
286 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
287 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
288 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
289 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
290 
291 static pv_entry_t pv_alloc(void);
292 static void pv_free(pv_entry_t);
293 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
294 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
295 
296 static void booke_pmap_init_qpages(void);
297 
298 struct ptbl_buf {
299 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
300 	vm_offset_t kva;		/* va of mapping */
301 };
302 
303 #ifndef __powerpc64__
304 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
305 #define PTBL_BUFS		(128 * 16)
306 
307 /* ptbl free list and a lock used for access synchronization. */
308 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
309 static struct mtx ptbl_buf_freelist_lock;
310 
311 /* Base address of kva space allocated fot ptbl bufs. */
312 static vm_offset_t ptbl_buf_pool_vabase;
313 
314 /* Pointer to ptbl_buf structures. */
315 static struct ptbl_buf *ptbl_bufs;
316 #endif
317 
318 #ifdef SMP
319 extern tlb_entry_t __boot_tlb1[];
320 void pmap_bootstrap_ap(volatile uint32_t *);
321 #endif
322 
323 /*
324  * Kernel MMU interface
325  */
326 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
327 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
328     vm_size_t, vm_offset_t);
329 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
330 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
331     vm_offset_t, vm_page_t *, vm_offset_t, int);
332 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
333     vm_prot_t, u_int flags, int8_t psind);
334 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
335     vm_page_t, vm_prot_t);
336 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
337     vm_prot_t);
338 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
339 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
340     vm_prot_t);
341 static void		mmu_booke_init(mmu_t);
342 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
343 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
344 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
345 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
346 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
347     int);
348 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
349     vm_paddr_t *);
350 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
351     vm_object_t, vm_pindex_t, vm_size_t);
352 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
353 static void		mmu_booke_page_init(mmu_t, vm_page_t);
354 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
355 static void		mmu_booke_pinit(mmu_t, pmap_t);
356 static void		mmu_booke_pinit0(mmu_t, pmap_t);
357 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
358     vm_prot_t);
359 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
360 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
361 static void		mmu_booke_release(mmu_t, pmap_t);
362 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
363 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
364 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
365 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
366 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
367 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
368 static void		mmu_booke_activate(mmu_t, struct thread *);
369 static void		mmu_booke_deactivate(mmu_t, struct thread *);
370 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
371 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
372 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
373 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
374 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
375 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
376 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
377 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
378 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
379 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
380     vm_size_t);
381 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
382     void **);
383 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
384     void *);
385 static void		mmu_booke_scan_init(mmu_t);
386 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
387 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
388 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
389     vm_size_t sz, vm_memattr_t mode);
390 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
391     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
392 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
393     int *is_user, vm_offset_t *decoded_addr);
394 
395 
396 static mmu_method_t mmu_booke_methods[] = {
397 	/* pmap dispatcher interface */
398 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
399 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
400 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
401 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
402 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
403 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
404 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
405 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
406 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
407 	MMUMETHOD(mmu_init,		mmu_booke_init),
408 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
409 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
410 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
411 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
412 	MMUMETHOD(mmu_map,		mmu_booke_map),
413 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
414 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
415 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
416 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
417 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
418 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
419 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
420 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
421 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
422 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
423 	MMUMETHOD(mmu_release,		mmu_booke_release),
424 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
425 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
426 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
427 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
428 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
429 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
430 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
431 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
432 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
433 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
434 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
435 
436 	/* Internal interfaces */
437 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
438 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
439 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
440 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
441 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
442 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
443 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
444 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
445 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
446 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
447 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
448 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
449 
450 	/* dumpsys() support */
451 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
452 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
453 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
454 
455 	{ 0, 0 }
456 };
457 
458 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
459 
460 static __inline uint32_t
461 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
462 {
463 	uint32_t attrib;
464 	int i;
465 
466 	if (ma != VM_MEMATTR_DEFAULT) {
467 		switch (ma) {
468 		case VM_MEMATTR_UNCACHEABLE:
469 			return (MAS2_I | MAS2_G);
470 		case VM_MEMATTR_WRITE_COMBINING:
471 		case VM_MEMATTR_WRITE_BACK:
472 		case VM_MEMATTR_PREFETCHABLE:
473 			return (MAS2_I);
474 		case VM_MEMATTR_WRITE_THROUGH:
475 			return (MAS2_W | MAS2_M);
476 		case VM_MEMATTR_CACHEABLE:
477 			return (MAS2_M);
478 		}
479 	}
480 
481 	/*
482 	 * Assume the page is cache inhibited and access is guarded unless
483 	 * it's in our available memory array.
484 	 */
485 	attrib = _TLB_ENTRY_IO;
486 	for (i = 0; i < physmem_regions_sz; i++) {
487 		if ((pa >= physmem_regions[i].mr_start) &&
488 		    (pa < (physmem_regions[i].mr_start +
489 		     physmem_regions[i].mr_size))) {
490 			attrib = _TLB_ENTRY_MEM;
491 			break;
492 		}
493 	}
494 
495 	return (attrib);
496 }
497 
498 static inline void
499 tlb_miss_lock(void)
500 {
501 #ifdef SMP
502 	struct pcpu *pc;
503 
504 	if (!smp_started)
505 		return;
506 
507 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
508 		if (pc != pcpup) {
509 
510 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
511 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
512 
513 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
514 			    ("tlb_miss_lock: tried to lock self"));
515 
516 			tlb_lock(pc->pc_booke.tlb_lock);
517 
518 			CTR1(KTR_PMAP, "%s: locked", __func__);
519 		}
520 	}
521 #endif
522 }
523 
524 static inline void
525 tlb_miss_unlock(void)
526 {
527 #ifdef SMP
528 	struct pcpu *pc;
529 
530 	if (!smp_started)
531 		return;
532 
533 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
534 		if (pc != pcpup) {
535 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
536 			    __func__, pc->pc_cpuid);
537 
538 			tlb_unlock(pc->pc_booke.tlb_lock);
539 
540 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
541 		}
542 	}
543 #endif
544 }
545 
546 /* Return number of entries in TLB0. */
547 static __inline void
548 tlb0_get_tlbconf(void)
549 {
550 	uint32_t tlb0_cfg;
551 
552 	tlb0_cfg = mfspr(SPR_TLB0CFG);
553 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
554 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
555 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
556 }
557 
558 /* Return number of entries in TLB1. */
559 static __inline void
560 tlb1_get_tlbconf(void)
561 {
562 	uint32_t tlb1_cfg;
563 
564 	tlb1_cfg = mfspr(SPR_TLB1CFG);
565 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
566 }
567 
568 /**************************************************************************/
569 /* Page table related */
570 /**************************************************************************/
571 
572 #ifdef __powerpc64__
573 /* Initialize pool of kva ptbl buffers. */
574 static void
575 ptbl_init(void)
576 {
577 }
578 
579 /* Get a pointer to a PTE in a page table. */
580 static __inline pte_t *
581 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
582 {
583 	pte_t         **pdir;
584 	pte_t          *ptbl;
585 
586 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
587 
588 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
589 	if (!pdir)
590 		return NULL;
591 	ptbl = pdir[PDIR_IDX(va)];
592 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
593 }
594 
595 /*
596  * allocate a page of pointers to page directories, do not preallocate the
597  * page tables
598  */
599 static pte_t  **
600 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
601 {
602 	vm_page_t	m;
603 	pte_t          **pdir;
604 	int		req;
605 
606 	req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
607 	while ((m = vm_page_alloc(NULL, pp2d_idx, req)) == NULL) {
608 		PMAP_UNLOCK(pmap);
609 		if (nosleep) {
610 			return (NULL);
611 		}
612 		vm_wait(NULL);
613 		PMAP_LOCK(pmap);
614 	}
615 
616 	/* Zero whole ptbl. */
617 	pdir = (pte_t **)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
618 	mmu_booke_zero_page(mmu, m);
619 
620 	return (pdir);
621 }
622 
623 /* Free pdir pages and invalidate pdir entry. */
624 static void
625 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, vm_page_t m)
626 {
627 	pte_t         **pdir;
628 
629 	pdir = pmap->pm_pp2d[pp2d_idx];
630 
631 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
632 
633 	pmap->pm_pp2d[pp2d_idx] = NULL;
634 
635 	vm_wire_sub(1);
636 	vm_page_free_zero(m);
637 }
638 
639 /*
640  * Decrement pdir pages hold count and attempt to free pdir pages. Called
641  * when removing directory entry from pdir.
642  *
643  * Return 1 if pdir pages were freed.
644  */
645 static int
646 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
647 {
648 	pte_t         **pdir;
649 	vm_paddr_t	pa;
650 	vm_page_t	m;
651 
652 	KASSERT((pmap != kernel_pmap),
653 		("pdir_unhold: unholding kernel pdir!"));
654 
655 	pdir = pmap->pm_pp2d[pp2d_idx];
656 
657 	/* decrement hold count */
658 	pa = DMAP_TO_PHYS((vm_offset_t) pdir);
659 	m = PHYS_TO_VM_PAGE(pa);
660 
661 	/*
662 	 * Free pdir page if there are no dir entries in this pdir.
663 	 */
664 	m->wire_count--;
665 	if (m->wire_count == 0) {
666 		pdir_free(mmu, pmap, pp2d_idx, m);
667 		return (1);
668 	}
669 	return (0);
670 }
671 
672 /*
673  * Increment hold count for pdir pages. This routine is used when new ptlb
674  * entry is being inserted into pdir.
675  */
676 static void
677 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
678 {
679 	vm_page_t	m;
680 
681 	KASSERT((pmap != kernel_pmap),
682 		("pdir_hold: holding kernel pdir!"));
683 
684 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
685 
686 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pdir));
687 	m->wire_count++;
688 }
689 
690 /* Allocate page table. */
691 static pte_t   *
692 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
693     boolean_t nosleep)
694 {
695 	vm_page_t	m;
696 	pte_t          *ptbl;
697 	int		req;
698 
699 	KASSERT((pdir[pdir_idx] == NULL),
700 		("%s: valid ptbl entry exists!", __func__));
701 
702 	req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
703 	while ((m = vm_page_alloc(NULL, pdir_idx, req)) == NULL) {
704 		PMAP_UNLOCK(pmap);
705 		rw_wunlock(&pvh_global_lock);
706 		if (nosleep) {
707 			return (NULL);
708 		}
709 		vm_wait(NULL);
710 		rw_wlock(&pvh_global_lock);
711 		PMAP_LOCK(pmap);
712 	}
713 
714 	/* Zero whole ptbl. */
715 	ptbl = (pte_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
716 	mmu_booke_zero_page(mmu, m);
717 
718 	return (ptbl);
719 }
720 
721 /* Free ptbl pages and invalidate pdir entry. */
722 static void
723 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx, vm_page_t m)
724 {
725 	pte_t          *ptbl;
726 
727 	ptbl = pdir[pdir_idx];
728 
729 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
730 
731 	pdir[pdir_idx] = NULL;
732 
733 	vm_wire_sub(1);
734 	vm_page_free_zero(m);
735 }
736 
737 /*
738  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
739  * when removing pte entry from ptbl.
740  *
741  * Return 1 if ptbl pages were freed.
742  */
743 static int
744 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
745 {
746 	pte_t          *ptbl;
747 	vm_page_t	m;
748 	u_int		pp2d_idx;
749 	pte_t         **pdir;
750 	u_int		pdir_idx;
751 
752 	pp2d_idx = PP2D_IDX(va);
753 	pdir_idx = PDIR_IDX(va);
754 
755 	KASSERT((pmap != kernel_pmap),
756 		("ptbl_unhold: unholding kernel ptbl!"));
757 
758 	pdir = pmap->pm_pp2d[pp2d_idx];
759 	ptbl = pdir[pdir_idx];
760 
761 	/* decrement hold count */
762 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
763 
764 	/*
765 	 * Free ptbl pages if there are no pte entries in this ptbl.
766 	 * wire_count has the same value for all ptbl pages, so check the
767 	 * last page.
768 	 */
769 	m->wire_count--;
770 	if (m->wire_count == 0) {
771 		ptbl_free(mmu, pmap, pdir, pdir_idx, m);
772 		pdir_unhold(mmu, pmap, pp2d_idx);
773 		return (1);
774 	}
775 	return (0);
776 }
777 
778 /*
779  * Increment hold count for ptbl pages. This routine is used when new pte
780  * entry is being inserted into ptbl.
781  */
782 static void
783 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
784 {
785 	pte_t          *ptbl;
786 	vm_page_t	m;
787 
788 	KASSERT((pmap != kernel_pmap),
789 		("ptbl_hold: holding kernel ptbl!"));
790 
791 	ptbl = pdir[pdir_idx];
792 
793 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
794 
795 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
796 	m->wire_count++;
797 }
798 #else
799 
800 /* Initialize pool of kva ptbl buffers. */
801 static void
802 ptbl_init(void)
803 {
804 	int i;
805 
806 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
807 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
808 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
809 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
810 
811 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
812 	TAILQ_INIT(&ptbl_buf_freelist);
813 
814 	for (i = 0; i < PTBL_BUFS; i++) {
815 		ptbl_bufs[i].kva =
816 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
817 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
818 	}
819 }
820 
821 /* Get a ptbl_buf from the freelist. */
822 static struct ptbl_buf *
823 ptbl_buf_alloc(void)
824 {
825 	struct ptbl_buf *buf;
826 
827 	mtx_lock(&ptbl_buf_freelist_lock);
828 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
829 	if (buf != NULL)
830 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
831 	mtx_unlock(&ptbl_buf_freelist_lock);
832 
833 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
834 
835 	return (buf);
836 }
837 
838 /* Return ptbl buff to free pool. */
839 static void
840 ptbl_buf_free(struct ptbl_buf *buf)
841 {
842 
843 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
844 
845 	mtx_lock(&ptbl_buf_freelist_lock);
846 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
847 	mtx_unlock(&ptbl_buf_freelist_lock);
848 }
849 
850 /*
851  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
852  */
853 static void
854 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
855 {
856 	struct ptbl_buf *pbuf;
857 
858 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
859 
860 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
861 
862 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
863 		if (pbuf->kva == (vm_offset_t)ptbl) {
864 			/* Remove from pmap ptbl buf list. */
865 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
866 
867 			/* Free corresponding ptbl buf. */
868 			ptbl_buf_free(pbuf);
869 			break;
870 		}
871 }
872 
873 /* Allocate page table. */
874 static pte_t *
875 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
876 {
877 	vm_page_t mtbl[PTBL_PAGES];
878 	vm_page_t m;
879 	struct ptbl_buf *pbuf;
880 	unsigned int pidx;
881 	pte_t *ptbl;
882 	int i, j;
883 
884 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
885 	    (pmap == kernel_pmap), pdir_idx);
886 
887 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
888 	    ("ptbl_alloc: invalid pdir_idx"));
889 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
890 	    ("pte_alloc: valid ptbl entry exists!"));
891 
892 	pbuf = ptbl_buf_alloc();
893 	if (pbuf == NULL)
894 		panic("pte_alloc: couldn't alloc kernel virtual memory");
895 
896 	ptbl = (pte_t *)pbuf->kva;
897 
898 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
899 
900 	for (i = 0; i < PTBL_PAGES; i++) {
901 		pidx = (PTBL_PAGES * pdir_idx) + i;
902 		while ((m = vm_page_alloc(NULL, pidx,
903 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
904 			PMAP_UNLOCK(pmap);
905 			rw_wunlock(&pvh_global_lock);
906 			if (nosleep) {
907 				ptbl_free_pmap_ptbl(pmap, ptbl);
908 				for (j = 0; j < i; j++)
909 					vm_page_free(mtbl[j]);
910 				vm_wire_sub(i);
911 				return (NULL);
912 			}
913 			vm_wait(NULL);
914 			rw_wlock(&pvh_global_lock);
915 			PMAP_LOCK(pmap);
916 		}
917 		mtbl[i] = m;
918 	}
919 
920 	/* Map allocated pages into kernel_pmap. */
921 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
922 
923 	/* Zero whole ptbl. */
924 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
925 
926 	/* Add pbuf to the pmap ptbl bufs list. */
927 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
928 
929 	return (ptbl);
930 }
931 
932 /* Free ptbl pages and invalidate pdir entry. */
933 static void
934 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
935 {
936 	pte_t *ptbl;
937 	vm_paddr_t pa;
938 	vm_offset_t va;
939 	vm_page_t m;
940 	int i;
941 
942 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
943 	    (pmap == kernel_pmap), pdir_idx);
944 
945 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
946 	    ("ptbl_free: invalid pdir_idx"));
947 
948 	ptbl = pmap->pm_pdir[pdir_idx];
949 
950 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
951 
952 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
953 
954 	/*
955 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
956 	 * don't attempt to look up the page tables we are releasing.
957 	 */
958 	mtx_lock_spin(&tlbivax_mutex);
959 	tlb_miss_lock();
960 
961 	pmap->pm_pdir[pdir_idx] = NULL;
962 
963 	tlb_miss_unlock();
964 	mtx_unlock_spin(&tlbivax_mutex);
965 
966 	for (i = 0; i < PTBL_PAGES; i++) {
967 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
968 		pa = pte_vatopa(mmu, kernel_pmap, va);
969 		m = PHYS_TO_VM_PAGE(pa);
970 		vm_page_free_zero(m);
971 		vm_wire_sub(1);
972 		mmu_booke_kremove(mmu, va);
973 	}
974 
975 	ptbl_free_pmap_ptbl(pmap, ptbl);
976 }
977 
978 /*
979  * Decrement ptbl pages hold count and attempt to free ptbl pages.
980  * Called when removing pte entry from ptbl.
981  *
982  * Return 1 if ptbl pages were freed.
983  */
984 static int
985 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
986 {
987 	pte_t *ptbl;
988 	vm_paddr_t pa;
989 	vm_page_t m;
990 	int i;
991 
992 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
993 	    (pmap == kernel_pmap), pdir_idx);
994 
995 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
996 	    ("ptbl_unhold: invalid pdir_idx"));
997 	KASSERT((pmap != kernel_pmap),
998 	    ("ptbl_unhold: unholding kernel ptbl!"));
999 
1000 	ptbl = pmap->pm_pdir[pdir_idx];
1001 
1002 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1003 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1004 	    ("ptbl_unhold: non kva ptbl"));
1005 
1006 	/* decrement hold count */
1007 	for (i = 0; i < PTBL_PAGES; i++) {
1008 		pa = pte_vatopa(mmu, kernel_pmap,
1009 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1010 		m = PHYS_TO_VM_PAGE(pa);
1011 		m->wire_count--;
1012 	}
1013 
1014 	/*
1015 	 * Free ptbl pages if there are no pte etries in this ptbl.
1016 	 * wire_count has the same value for all ptbl pages, so check the last
1017 	 * page.
1018 	 */
1019 	if (m->wire_count == 0) {
1020 		ptbl_free(mmu, pmap, pdir_idx);
1021 
1022 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1023 		return (1);
1024 	}
1025 
1026 	return (0);
1027 }
1028 
1029 /*
1030  * Increment hold count for ptbl pages. This routine is used when a new pte
1031  * entry is being inserted into the ptbl.
1032  */
1033 static void
1034 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1035 {
1036 	vm_paddr_t pa;
1037 	pte_t *ptbl;
1038 	vm_page_t m;
1039 	int i;
1040 
1041 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1042 	    pdir_idx);
1043 
1044 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1045 	    ("ptbl_hold: invalid pdir_idx"));
1046 	KASSERT((pmap != kernel_pmap),
1047 	    ("ptbl_hold: holding kernel ptbl!"));
1048 
1049 	ptbl = pmap->pm_pdir[pdir_idx];
1050 
1051 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1052 
1053 	for (i = 0; i < PTBL_PAGES; i++) {
1054 		pa = pte_vatopa(mmu, kernel_pmap,
1055 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1056 		m = PHYS_TO_VM_PAGE(pa);
1057 		m->wire_count++;
1058 	}
1059 }
1060 #endif
1061 
1062 /* Allocate pv_entry structure. */
1063 pv_entry_t
1064 pv_alloc(void)
1065 {
1066 	pv_entry_t pv;
1067 
1068 	pv_entry_count++;
1069 	if (pv_entry_count > pv_entry_high_water)
1070 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1071 	pv = uma_zalloc(pvzone, M_NOWAIT);
1072 
1073 	return (pv);
1074 }
1075 
1076 /* Free pv_entry structure. */
1077 static __inline void
1078 pv_free(pv_entry_t pve)
1079 {
1080 
1081 	pv_entry_count--;
1082 	uma_zfree(pvzone, pve);
1083 }
1084 
1085 
1086 /* Allocate and initialize pv_entry structure. */
1087 static void
1088 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1089 {
1090 	pv_entry_t pve;
1091 
1092 	//int su = (pmap == kernel_pmap);
1093 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1094 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1095 
1096 	pve = pv_alloc();
1097 	if (pve == NULL)
1098 		panic("pv_insert: no pv entries!");
1099 
1100 	pve->pv_pmap = pmap;
1101 	pve->pv_va = va;
1102 
1103 	/* add to pv_list */
1104 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1105 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1106 
1107 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1108 
1109 	//debugf("pv_insert: e\n");
1110 }
1111 
1112 /* Destroy pv entry. */
1113 static void
1114 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1115 {
1116 	pv_entry_t pve;
1117 
1118 	//int su = (pmap == kernel_pmap);
1119 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1120 
1121 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1122 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1123 
1124 	/* find pv entry */
1125 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1126 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1127 			/* remove from pv_list */
1128 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1129 			if (TAILQ_EMPTY(&m->md.pv_list))
1130 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1131 
1132 			/* free pv entry struct */
1133 			pv_free(pve);
1134 			break;
1135 		}
1136 	}
1137 
1138 	//debugf("pv_remove: e\n");
1139 }
1140 
1141 #ifdef __powerpc64__
1142 /*
1143  * Clean pte entry, try to free page table page if requested.
1144  *
1145  * Return 1 if ptbl pages were freed, otherwise return 0.
1146  */
1147 static int
1148 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1149 {
1150 	vm_page_t	m;
1151 	pte_t          *pte;
1152 
1153 	pte = pte_find(mmu, pmap, va);
1154 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1155 
1156 	if (!PTE_ISVALID(pte))
1157 		return (0);
1158 
1159 	/* Get vm_page_t for mapped pte. */
1160 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1161 
1162 	if (PTE_ISWIRED(pte))
1163 		pmap->pm_stats.wired_count--;
1164 
1165 	/* Handle managed entry. */
1166 	if (PTE_ISMANAGED(pte)) {
1167 
1168 		/* Handle modified pages. */
1169 		if (PTE_ISMODIFIED(pte))
1170 			vm_page_dirty(m);
1171 
1172 		/* Referenced pages. */
1173 		if (PTE_ISREFERENCED(pte))
1174 			vm_page_aflag_set(m, PGA_REFERENCED);
1175 
1176 		/* Remove pv_entry from pv_list. */
1177 		pv_remove(pmap, va, m);
1178 	} else if (m->md.pv_tracked) {
1179 		pv_remove(pmap, va, m);
1180 		if (TAILQ_EMPTY(&m->md.pv_list))
1181 			m->md.pv_tracked = false;
1182 	}
1183 	mtx_lock_spin(&tlbivax_mutex);
1184 	tlb_miss_lock();
1185 
1186 	tlb0_flush_entry(va);
1187 	*pte = 0;
1188 
1189 	tlb_miss_unlock();
1190 	mtx_unlock_spin(&tlbivax_mutex);
1191 
1192 	pmap->pm_stats.resident_count--;
1193 
1194 	if (flags & PTBL_UNHOLD) {
1195 		return (ptbl_unhold(mmu, pmap, va));
1196 	}
1197 	return (0);
1198 }
1199 
1200 /*
1201  * Insert PTE for a given page and virtual address.
1202  */
1203 static int
1204 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1205     boolean_t nosleep)
1206 {
1207 	unsigned int	pp2d_idx = PP2D_IDX(va);
1208 	unsigned int	pdir_idx = PDIR_IDX(va);
1209 	unsigned int	ptbl_idx = PTBL_IDX(va);
1210 	pte_t          *ptbl, *pte, pte_tmp;
1211 	pte_t         **pdir;
1212 
1213 	/* Get the page directory pointer. */
1214 	pdir = pmap->pm_pp2d[pp2d_idx];
1215 	if (pdir == NULL)
1216 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1217 
1218 	/* Get the page table pointer. */
1219 	ptbl = pdir[pdir_idx];
1220 
1221 	if (ptbl == NULL) {
1222 		/* Allocate page table pages. */
1223 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1224 		if (ptbl == NULL) {
1225 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1226 			return (ENOMEM);
1227 		}
1228 		pte = &ptbl[ptbl_idx];
1229 	} else {
1230 		/*
1231 		 * Check if there is valid mapping for requested va, if there
1232 		 * is, remove it.
1233 		 */
1234 		pte = &ptbl[ptbl_idx];
1235 		if (PTE_ISVALID(pte)) {
1236 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1237 		} else {
1238 			/*
1239 			 * pte is not used, increment hold count for ptbl
1240 			 * pages.
1241 			 */
1242 			if (pmap != kernel_pmap)
1243 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1244 		}
1245 	}
1246 
1247 	if (pdir[pdir_idx] == NULL) {
1248 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1249 			pdir_hold(mmu, pmap, pdir);
1250 		pdir[pdir_idx] = ptbl;
1251 	}
1252 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1253 		pmap->pm_pp2d[pp2d_idx] = pdir;
1254 
1255 	/*
1256 	 * Insert pv_entry into pv_list for mapped page if part of managed
1257 	 * memory.
1258 	 */
1259 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1260 		flags |= PTE_MANAGED;
1261 
1262 		/* Create and insert pv entry. */
1263 		pv_insert(pmap, va, m);
1264 	}
1265 
1266 	pmap->pm_stats.resident_count++;
1267 
1268 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1269 	pte_tmp |= (PTE_VALID | flags);
1270 
1271 	mtx_lock_spin(&tlbivax_mutex);
1272 	tlb_miss_lock();
1273 
1274 	tlb0_flush_entry(va);
1275 	*pte = pte_tmp;
1276 
1277 	tlb_miss_unlock();
1278 	mtx_unlock_spin(&tlbivax_mutex);
1279 
1280 	return (0);
1281 }
1282 
1283 /* Return the pa for the given pmap/va. */
1284 static	vm_paddr_t
1285 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1286 {
1287 	vm_paddr_t	pa = 0;
1288 	pte_t          *pte;
1289 
1290 	pte = pte_find(mmu, pmap, va);
1291 	if ((pte != NULL) && PTE_ISVALID(pte))
1292 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1293 	return (pa);
1294 }
1295 
1296 
1297 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1298 static void
1299 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1300 {
1301 	int		i, j;
1302 	vm_offset_t	va;
1303 	pte_t		*pte;
1304 
1305 	va = addr;
1306 	/* Initialize kernel pdir */
1307 	for (i = 0; i < kernel_pdirs; i++) {
1308 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1309 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1310 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1311 		    j < PDIR_NENTRIES; j++) {
1312 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1313 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE) +
1314 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE));
1315 		}
1316 	}
1317 
1318 	/*
1319 	 * Fill in PTEs covering kernel code and data. They are not required
1320 	 * for address translation, as this area is covered by static TLB1
1321 	 * entries, but for pte_vatopa() to work correctly with kernel area
1322 	 * addresses.
1323 	 */
1324 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1325 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1326 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1327 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1328 		    PTE_VALID | PTE_PS_4KB;
1329 	}
1330 }
1331 #else
1332 /*
1333  * Clean pte entry, try to free page table page if requested.
1334  *
1335  * Return 1 if ptbl pages were freed, otherwise return 0.
1336  */
1337 static int
1338 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1339 {
1340 	unsigned int pdir_idx = PDIR_IDX(va);
1341 	unsigned int ptbl_idx = PTBL_IDX(va);
1342 	vm_page_t m;
1343 	pte_t *ptbl;
1344 	pte_t *pte;
1345 
1346 	//int su = (pmap == kernel_pmap);
1347 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1348 	//		su, (u_int32_t)pmap, va, flags);
1349 
1350 	ptbl = pmap->pm_pdir[pdir_idx];
1351 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1352 
1353 	pte = &ptbl[ptbl_idx];
1354 
1355 	if (pte == NULL || !PTE_ISVALID(pte))
1356 		return (0);
1357 
1358 	if (PTE_ISWIRED(pte))
1359 		pmap->pm_stats.wired_count--;
1360 
1361 	/* Get vm_page_t for mapped pte. */
1362 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1363 
1364 	/* Handle managed entry. */
1365 	if (PTE_ISMANAGED(pte)) {
1366 
1367 		if (PTE_ISMODIFIED(pte))
1368 			vm_page_dirty(m);
1369 
1370 		if (PTE_ISREFERENCED(pte))
1371 			vm_page_aflag_set(m, PGA_REFERENCED);
1372 
1373 		pv_remove(pmap, va, m);
1374 	} else if (m->md.pv_tracked) {
1375 		/*
1376 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1377 		 * used.  This is needed by the NCSW support code for fast
1378 		 * VA<->PA translation.
1379 		 */
1380 		pv_remove(pmap, va, m);
1381 		if (TAILQ_EMPTY(&m->md.pv_list))
1382 			m->md.pv_tracked = false;
1383 	}
1384 
1385 	mtx_lock_spin(&tlbivax_mutex);
1386 	tlb_miss_lock();
1387 
1388 	tlb0_flush_entry(va);
1389 	*pte = 0;
1390 
1391 	tlb_miss_unlock();
1392 	mtx_unlock_spin(&tlbivax_mutex);
1393 
1394 	pmap->pm_stats.resident_count--;
1395 
1396 	if (flags & PTBL_UNHOLD) {
1397 		//debugf("pte_remove: e (unhold)\n");
1398 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1399 	}
1400 
1401 	//debugf("pte_remove: e\n");
1402 	return (0);
1403 }
1404 
1405 /*
1406  * Insert PTE for a given page and virtual address.
1407  */
1408 static int
1409 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1410     boolean_t nosleep)
1411 {
1412 	unsigned int pdir_idx = PDIR_IDX(va);
1413 	unsigned int ptbl_idx = PTBL_IDX(va);
1414 	pte_t *ptbl, *pte, pte_tmp;
1415 
1416 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1417 	    pmap == kernel_pmap, pmap, va);
1418 
1419 	/* Get the page table pointer. */
1420 	ptbl = pmap->pm_pdir[pdir_idx];
1421 
1422 	if (ptbl == NULL) {
1423 		/* Allocate page table pages. */
1424 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1425 		if (ptbl == NULL) {
1426 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1427 			return (ENOMEM);
1428 		}
1429 		pmap->pm_pdir[pdir_idx] = ptbl;
1430 		pte = &ptbl[ptbl_idx];
1431 	} else {
1432 		/*
1433 		 * Check if there is valid mapping for requested
1434 		 * va, if there is, remove it.
1435 		 */
1436 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1437 		if (PTE_ISVALID(pte)) {
1438 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1439 		} else {
1440 			/*
1441 			 * pte is not used, increment hold count
1442 			 * for ptbl pages.
1443 			 */
1444 			if (pmap != kernel_pmap)
1445 				ptbl_hold(mmu, pmap, pdir_idx);
1446 		}
1447 	}
1448 
1449 	/*
1450 	 * Insert pv_entry into pv_list for mapped page if part of managed
1451 	 * memory.
1452 	 */
1453 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1454 		flags |= PTE_MANAGED;
1455 
1456 		/* Create and insert pv entry. */
1457 		pv_insert(pmap, va, m);
1458 	}
1459 
1460 	pmap->pm_stats.resident_count++;
1461 
1462 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1463 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1464 
1465 	mtx_lock_spin(&tlbivax_mutex);
1466 	tlb_miss_lock();
1467 
1468 	tlb0_flush_entry(va);
1469 	*pte = pte_tmp;
1470 
1471 	tlb_miss_unlock();
1472 	mtx_unlock_spin(&tlbivax_mutex);
1473 	return (0);
1474 }
1475 
1476 /* Return the pa for the given pmap/va. */
1477 static vm_paddr_t
1478 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1479 {
1480 	vm_paddr_t pa = 0;
1481 	pte_t *pte;
1482 
1483 	pte = pte_find(mmu, pmap, va);
1484 	if ((pte != NULL) && PTE_ISVALID(pte))
1485 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1486 	return (pa);
1487 }
1488 
1489 /* Get a pointer to a PTE in a page table. */
1490 static pte_t *
1491 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1492 {
1493 	unsigned int pdir_idx = PDIR_IDX(va);
1494 	unsigned int ptbl_idx = PTBL_IDX(va);
1495 
1496 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1497 
1498 	if (pmap->pm_pdir[pdir_idx])
1499 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1500 
1501 	return (NULL);
1502 }
1503 
1504 /* Set up kernel page tables. */
1505 static void
1506 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1507 {
1508 	int		i;
1509 	vm_offset_t	va;
1510 	pte_t		*pte;
1511 
1512 	/* Initialize kernel pdir */
1513 	for (i = 0; i < kernel_ptbls; i++)
1514 		kernel_pmap->pm_pdir[kptbl_min + i] =
1515 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1516 
1517 	/*
1518 	 * Fill in PTEs covering kernel code and data. They are not required
1519 	 * for address translation, as this area is covered by static TLB1
1520 	 * entries, but for pte_vatopa() to work correctly with kernel area
1521 	 * addresses.
1522 	 */
1523 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1524 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1525 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1526 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1527 		    PTE_VALID | PTE_PS_4KB;
1528 	}
1529 }
1530 #endif
1531 
1532 /**************************************************************************/
1533 /* PMAP related */
1534 /**************************************************************************/
1535 
1536 /*
1537  * This is called during booke_init, before the system is really initialized.
1538  */
1539 static void
1540 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1541 {
1542 	vm_paddr_t phys_kernelend;
1543 	struct mem_region *mp, *mp1;
1544 	int cnt, i, j;
1545 	vm_paddr_t s, e, sz;
1546 	vm_paddr_t physsz, hwphyssz;
1547 	u_int phys_avail_count;
1548 	vm_size_t kstack0_sz;
1549 	vm_offset_t kernel_pdir, kstack0;
1550 	vm_paddr_t kstack0_phys;
1551 	void *dpcpu;
1552 	vm_offset_t kernel_ptbl_root;
1553 
1554 	debugf("mmu_booke_bootstrap: entered\n");
1555 
1556 	/* Set interesting system properties */
1557 #ifdef __powerpc64__
1558 	hw_direct_map = 1;
1559 #else
1560 	hw_direct_map = 0;
1561 #endif
1562 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1563 	elf32_nxstack = 1;
1564 #endif
1565 
1566 	/* Initialize invalidation mutex */
1567 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1568 
1569 	/* Read TLB0 size and associativity. */
1570 	tlb0_get_tlbconf();
1571 
1572 	/*
1573 	 * Align kernel start and end address (kernel image).
1574 	 * Note that kernel end does not necessarily relate to kernsize.
1575 	 * kernsize is the size of the kernel that is actually mapped.
1576 	 */
1577 	kernstart = trunc_page(start);
1578 	data_start = round_page(kernelend);
1579 	data_end = data_start;
1580 
1581 	/* Allocate the dynamic per-cpu area. */
1582 	dpcpu = (void *)data_end;
1583 	data_end += DPCPU_SIZE;
1584 
1585 	/* Allocate space for the message buffer. */
1586 	msgbufp = (struct msgbuf *)data_end;
1587 	data_end += msgbufsize;
1588 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1589 	    (uintptr_t)msgbufp, data_end);
1590 
1591 	data_end = round_page(data_end);
1592 
1593 #ifdef __powerpc64__
1594 	kernel_ptbl_root = data_end;
1595 	data_end += PP2D_NENTRIES * sizeof(pte_t**);
1596 #else
1597 	/* Allocate space for ptbl_bufs. */
1598 	ptbl_bufs = (struct ptbl_buf *)data_end;
1599 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1600 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1601 	    (uintptr_t)ptbl_bufs, data_end);
1602 
1603 	data_end = round_page(data_end);
1604 	kernel_ptbl_root = data_end;
1605 	data_end += PDIR_NENTRIES * sizeof(pte_t*);
1606 #endif
1607 
1608 	/* Allocate PTE tables for kernel KVA. */
1609 	kernel_pdir = data_end;
1610 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1611 	    PDIR_SIZE);
1612 #ifdef __powerpc64__
1613 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1614 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1615 #endif
1616 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1617 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1618 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1619 	    kernel_pdir, data_end);
1620 
1621 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1622 	if (data_end - kernstart > kernsize) {
1623 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1624 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1625 	}
1626 	data_end = kernstart + kernsize;
1627 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1628 
1629 	/*
1630 	 * Clear the structures - note we can only do it safely after the
1631 	 * possible additional TLB1 translations are in place (above) so that
1632 	 * all range up to the currently calculated 'data_end' is covered.
1633 	 */
1634 	dpcpu_init(dpcpu, 0);
1635 #ifdef __powerpc64__
1636 	memset((void *)kernel_pdir, 0,
1637 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1638 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1639 #else
1640 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1641 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1642 #endif
1643 
1644 	/*******************************************************/
1645 	/* Set the start and end of kva. */
1646 	/*******************************************************/
1647 	virtual_avail = round_page(data_end);
1648 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1649 
1650 	/* Allocate KVA space for page zero/copy operations. */
1651 	zero_page_va = virtual_avail;
1652 	virtual_avail += PAGE_SIZE;
1653 	copy_page_src_va = virtual_avail;
1654 	virtual_avail += PAGE_SIZE;
1655 	copy_page_dst_va = virtual_avail;
1656 	virtual_avail += PAGE_SIZE;
1657 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
1658 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
1659 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
1660 
1661 	/* Initialize page zero/copy mutexes. */
1662 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1663 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1664 
1665 #ifndef __powerpc64__
1666 	/* Allocate KVA space for ptbl bufs. */
1667 	ptbl_buf_pool_vabase = virtual_avail;
1668 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1669 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1670 	    ptbl_buf_pool_vabase, virtual_avail);
1671 #endif
1672 
1673 	/* Calculate corresponding physical addresses for the kernel region. */
1674 	phys_kernelend = kernload + kernsize;
1675 	debugf("kernel image and allocated data:\n");
1676 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1677 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
1678 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
1679 
1680 	/*
1681 	 * Remove kernel physical address range from avail regions list. Page
1682 	 * align all regions.  Non-page aligned memory isn't very interesting
1683 	 * to us.  Also, sort the entries for ascending addresses.
1684 	 */
1685 
1686 	/* Retrieve phys/avail mem regions */
1687 	mem_regions(&physmem_regions, &physmem_regions_sz,
1688 	    &availmem_regions, &availmem_regions_sz);
1689 
1690 	if (nitems(phys_avail) < availmem_regions_sz)
1691 		panic("mmu_booke_bootstrap: phys_avail too small");
1692 
1693 	sz = 0;
1694 	cnt = availmem_regions_sz;
1695 	debugf("processing avail regions:\n");
1696 	for (mp = availmem_regions; mp->mr_size; mp++) {
1697 		s = mp->mr_start;
1698 		e = mp->mr_start + mp->mr_size;
1699 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1700 		/* Check whether this region holds all of the kernel. */
1701 		if (s < kernload && e > phys_kernelend) {
1702 			availmem_regions[cnt].mr_start = phys_kernelend;
1703 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1704 			e = kernload;
1705 		}
1706 		/* Look whether this regions starts within the kernel. */
1707 		if (s >= kernload && s < phys_kernelend) {
1708 			if (e <= phys_kernelend)
1709 				goto empty;
1710 			s = phys_kernelend;
1711 		}
1712 		/* Now look whether this region ends within the kernel. */
1713 		if (e > kernload && e <= phys_kernelend) {
1714 			if (s >= kernload)
1715 				goto empty;
1716 			e = kernload;
1717 		}
1718 		/* Now page align the start and size of the region. */
1719 		s = round_page(s);
1720 		e = trunc_page(e);
1721 		if (e < s)
1722 			e = s;
1723 		sz = e - s;
1724 		debugf("%09jx-%09jx = %jx\n",
1725 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1726 
1727 		/* Check whether some memory is left here. */
1728 		if (sz == 0) {
1729 		empty:
1730 			memmove(mp, mp + 1,
1731 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1732 			cnt--;
1733 			mp--;
1734 			continue;
1735 		}
1736 
1737 		/* Do an insertion sort. */
1738 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1739 			if (s < mp1->mr_start)
1740 				break;
1741 		if (mp1 < mp) {
1742 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1743 			mp1->mr_start = s;
1744 			mp1->mr_size = sz;
1745 		} else {
1746 			mp->mr_start = s;
1747 			mp->mr_size = sz;
1748 		}
1749 	}
1750 	availmem_regions_sz = cnt;
1751 
1752 	/*******************************************************/
1753 	/* Steal physical memory for kernel stack from the end */
1754 	/* of the first avail region                           */
1755 	/*******************************************************/
1756 	kstack0_sz = kstack_pages * PAGE_SIZE;
1757 	kstack0_phys = availmem_regions[0].mr_start +
1758 	    availmem_regions[0].mr_size;
1759 	kstack0_phys -= kstack0_sz;
1760 	availmem_regions[0].mr_size -= kstack0_sz;
1761 
1762 	/*******************************************************/
1763 	/* Fill in phys_avail table, based on availmem_regions */
1764 	/*******************************************************/
1765 	phys_avail_count = 0;
1766 	physsz = 0;
1767 	hwphyssz = 0;
1768 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1769 
1770 	debugf("fill in phys_avail:\n");
1771 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1772 
1773 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1774 		    (uintmax_t)availmem_regions[i].mr_start,
1775 		    (uintmax_t)availmem_regions[i].mr_start +
1776 		        availmem_regions[i].mr_size,
1777 		    (uintmax_t)availmem_regions[i].mr_size);
1778 
1779 		if (hwphyssz != 0 &&
1780 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1781 			debugf(" hw.physmem adjust\n");
1782 			if (physsz < hwphyssz) {
1783 				phys_avail[j] = availmem_regions[i].mr_start;
1784 				phys_avail[j + 1] =
1785 				    availmem_regions[i].mr_start +
1786 				    hwphyssz - physsz;
1787 				physsz = hwphyssz;
1788 				phys_avail_count++;
1789 			}
1790 			break;
1791 		}
1792 
1793 		phys_avail[j] = availmem_regions[i].mr_start;
1794 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1795 		    availmem_regions[i].mr_size;
1796 		phys_avail_count++;
1797 		physsz += availmem_regions[i].mr_size;
1798 	}
1799 	physmem = btoc(physsz);
1800 
1801 	/* Calculate the last available physical address. */
1802 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1803 		;
1804 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1805 
1806 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1807 	debugf("phys_avail_count = %d\n", phys_avail_count);
1808 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1809 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1810 
1811 #ifdef __powerpc64__
1812 	/*
1813 	 * Map the physical memory contiguously in TLB1.
1814 	 * Round so it fits into a single mapping.
1815 	 */
1816 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
1817 	    phys_avail[i + 1]);
1818 #endif
1819 
1820 	/*******************************************************/
1821 	/* Initialize (statically allocated) kernel pmap. */
1822 	/*******************************************************/
1823 	PMAP_LOCK_INIT(kernel_pmap);
1824 #ifndef __powerpc64__
1825 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1826 #endif
1827 #ifdef __powerpc64__
1828 	kernel_pmap->pm_pp2d = (pte_t ***)kernel_ptbl_root;
1829 #else
1830 	kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
1831 #endif
1832 
1833 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1834 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1835 	for (i = 0; i < MAXCPU; i++) {
1836 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1837 
1838 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1839 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1840 	}
1841 
1842 	/* Mark kernel_pmap active on all CPUs */
1843 	CPU_FILL(&kernel_pmap->pm_active);
1844 
1845  	/*
1846 	 * Initialize the global pv list lock.
1847 	 */
1848 	rw_init(&pvh_global_lock, "pmap pv global");
1849 
1850 	/*******************************************************/
1851 	/* Final setup */
1852 	/*******************************************************/
1853 
1854 	/* Enter kstack0 into kernel map, provide guard page */
1855 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1856 	thread0.td_kstack = kstack0;
1857 	thread0.td_kstack_pages = kstack_pages;
1858 
1859 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1860 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
1861 	    kstack0_phys, kstack0_phys + kstack0_sz);
1862 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
1863 	    kstack0, kstack0 + kstack0_sz);
1864 
1865 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1866 	for (i = 0; i < kstack_pages; i++) {
1867 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1868 		kstack0 += PAGE_SIZE;
1869 		kstack0_phys += PAGE_SIZE;
1870 	}
1871 
1872 	pmap_bootstrapped = 1;
1873 
1874 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
1875 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
1876 
1877 	debugf("mmu_booke_bootstrap: exit\n");
1878 }
1879 
1880 #ifdef SMP
1881  void
1882 tlb1_ap_prep(void)
1883 {
1884 	tlb_entry_t *e, tmp;
1885 	unsigned int i;
1886 
1887 	/* Prepare TLB1 image for AP processors */
1888 	e = __boot_tlb1;
1889 	for (i = 0; i < TLB1_ENTRIES; i++) {
1890 		tlb1_read_entry(&tmp, i);
1891 
1892 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1893 			memcpy(e++, &tmp, sizeof(tmp));
1894 	}
1895 }
1896 
1897 void
1898 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1899 {
1900 	int i;
1901 
1902 	/*
1903 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1904 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1905 	 * created by tlb1_ap_prep(), so use these values directly to
1906 	 * (re)program AP's TLB1 hardware.
1907 	 *
1908 	 * Start at index 1 because index 0 has the kernel map.
1909 	 */
1910 	for (i = 1; i < TLB1_ENTRIES; i++) {
1911 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1912 			tlb1_write_entry(&__boot_tlb1[i], i);
1913 	}
1914 
1915 	set_mas4_defaults();
1916 }
1917 #endif
1918 
1919 static void
1920 booke_pmap_init_qpages(void)
1921 {
1922 	struct pcpu *pc;
1923 	int i;
1924 
1925 	CPU_FOREACH(i) {
1926 		pc = pcpu_find(i);
1927 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1928 		if (pc->pc_qmap_addr == 0)
1929 			panic("pmap_init_qpages: unable to allocate KVA");
1930 	}
1931 }
1932 
1933 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1934 
1935 /*
1936  * Get the physical page address for the given pmap/virtual address.
1937  */
1938 static vm_paddr_t
1939 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1940 {
1941 	vm_paddr_t pa;
1942 
1943 	PMAP_LOCK(pmap);
1944 	pa = pte_vatopa(mmu, pmap, va);
1945 	PMAP_UNLOCK(pmap);
1946 
1947 	return (pa);
1948 }
1949 
1950 /*
1951  * Extract the physical page address associated with the given
1952  * kernel virtual address.
1953  */
1954 static vm_paddr_t
1955 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1956 {
1957 	tlb_entry_t e;
1958 	vm_paddr_t p = 0;
1959 	int i;
1960 
1961 #ifdef __powerpc64__
1962 	if (va >= DMAP_BASE_ADDRESS && va <= DMAP_MAX_ADDRESS)
1963 		return (DMAP_TO_PHYS(va));
1964 #endif
1965 
1966 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
1967 		p = pte_vatopa(mmu, kernel_pmap, va);
1968 
1969 	if (p == 0) {
1970 		/* Check TLB1 mappings */
1971 		for (i = 0; i < TLB1_ENTRIES; i++) {
1972 			tlb1_read_entry(&e, i);
1973 			if (!(e.mas1 & MAS1_VALID))
1974 				continue;
1975 			if (va >= e.virt && va < e.virt + e.size)
1976 				return (e.phys + (va - e.virt));
1977 		}
1978 	}
1979 
1980 	return (p);
1981 }
1982 
1983 /*
1984  * Initialize the pmap module.
1985  * Called by vm_init, to initialize any structures that the pmap
1986  * system needs to map virtual memory.
1987  */
1988 static void
1989 mmu_booke_init(mmu_t mmu)
1990 {
1991 	int shpgperproc = PMAP_SHPGPERPROC;
1992 
1993 	/*
1994 	 * Initialize the address space (zone) for the pv entries.  Set a
1995 	 * high water mark so that the system can recover from excessive
1996 	 * numbers of pv entries.
1997 	 */
1998 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1999 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2000 
2001 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2002 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2003 
2004 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2005 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2006 
2007 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2008 
2009 	/* Pre-fill pvzone with initial number of pv entries. */
2010 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2011 
2012 	/* Create a UMA zone for page table roots. */
2013 	ptbl_root_zone = uma_zcreate("pmap root", PMAP_ROOT_SIZE,
2014 	    NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, UMA_ZONE_VM);
2015 
2016 	/* Initialize ptbl allocation. */
2017 	ptbl_init();
2018 }
2019 
2020 /*
2021  * Map a list of wired pages into kernel virtual address space.  This is
2022  * intended for temporary mappings which do not need page modification or
2023  * references recorded.  Existing mappings in the region are overwritten.
2024  */
2025 static void
2026 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2027 {
2028 	vm_offset_t va;
2029 
2030 	va = sva;
2031 	while (count-- > 0) {
2032 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2033 		va += PAGE_SIZE;
2034 		m++;
2035 	}
2036 }
2037 
2038 /*
2039  * Remove page mappings from kernel virtual address space.  Intended for
2040  * temporary mappings entered by mmu_booke_qenter.
2041  */
2042 static void
2043 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2044 {
2045 	vm_offset_t va;
2046 
2047 	va = sva;
2048 	while (count-- > 0) {
2049 		mmu_booke_kremove(mmu, va);
2050 		va += PAGE_SIZE;
2051 	}
2052 }
2053 
2054 /*
2055  * Map a wired page into kernel virtual address space.
2056  */
2057 static void
2058 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2059 {
2060 
2061 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2062 }
2063 
2064 static void
2065 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2066 {
2067 	uint32_t flags;
2068 	pte_t *pte;
2069 
2070 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2071 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2072 
2073 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2074 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2075 	flags |= PTE_PS_4KB;
2076 
2077 	pte = pte_find(mmu, kernel_pmap, va);
2078 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2079 
2080 	mtx_lock_spin(&tlbivax_mutex);
2081 	tlb_miss_lock();
2082 
2083 	if (PTE_ISVALID(pte)) {
2084 
2085 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2086 
2087 		/* Flush entry from TLB0 */
2088 		tlb0_flush_entry(va);
2089 	}
2090 
2091 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2092 
2093 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2094 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2095 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2096 
2097 	/* Flush the real memory from the instruction cache. */
2098 	if ((flags & (PTE_I | PTE_G)) == 0)
2099 		__syncicache((void *)va, PAGE_SIZE);
2100 
2101 	tlb_miss_unlock();
2102 	mtx_unlock_spin(&tlbivax_mutex);
2103 }
2104 
2105 /*
2106  * Remove a page from kernel page table.
2107  */
2108 static void
2109 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2110 {
2111 	pte_t *pte;
2112 
2113 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
2114 
2115 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2116 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2117 	    ("mmu_booke_kremove: invalid va"));
2118 
2119 	pte = pte_find(mmu, kernel_pmap, va);
2120 
2121 	if (!PTE_ISVALID(pte)) {
2122 
2123 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2124 
2125 		return;
2126 	}
2127 
2128 	mtx_lock_spin(&tlbivax_mutex);
2129 	tlb_miss_lock();
2130 
2131 	/* Invalidate entry in TLB0, update PTE. */
2132 	tlb0_flush_entry(va);
2133 	*pte = 0;
2134 
2135 	tlb_miss_unlock();
2136 	mtx_unlock_spin(&tlbivax_mutex);
2137 }
2138 
2139 /*
2140  * Provide a kernel pointer corresponding to a given userland pointer.
2141  * The returned pointer is valid until the next time this function is
2142  * called in this thread. This is used internally in copyin/copyout.
2143  */
2144 int
2145 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2146     void **kaddr, size_t ulen, size_t *klen)
2147 {
2148 
2149 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2150 		return (EFAULT);
2151 
2152 	*kaddr = (void *)(uintptr_t)uaddr;
2153 	if (klen)
2154 		*klen = ulen;
2155 
2156 	return (0);
2157 }
2158 
2159 /*
2160  * Figure out where a given kernel pointer (usually in a fault) points
2161  * to from the VM's perspective, potentially remapping into userland's
2162  * address space.
2163  */
2164 static int
2165 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2166     vm_offset_t *decoded_addr)
2167 {
2168 
2169 	if (addr < VM_MAXUSER_ADDRESS)
2170 		*is_user = 1;
2171 	else
2172 		*is_user = 0;
2173 
2174 	*decoded_addr = addr;
2175 	return (0);
2176 }
2177 
2178 /*
2179  * Initialize pmap associated with process 0.
2180  */
2181 static void
2182 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2183 {
2184 
2185 	PMAP_LOCK_INIT(pmap);
2186 	mmu_booke_pinit(mmu, pmap);
2187 	PCPU_SET(curpmap, pmap);
2188 }
2189 
2190 /*
2191  * Initialize a preallocated and zeroed pmap structure,
2192  * such as one in a vmspace structure.
2193  */
2194 static void
2195 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2196 {
2197 	int i;
2198 
2199 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2200 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2201 
2202 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2203 
2204 	for (i = 0; i < MAXCPU; i++)
2205 		pmap->pm_tid[i] = TID_NONE;
2206 	CPU_ZERO(&kernel_pmap->pm_active);
2207 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2208 #ifdef __powerpc64__
2209 	pmap->pm_pp2d = uma_zalloc(ptbl_root_zone, M_WAITOK);
2210 	bzero(pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2211 #else
2212 	pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
2213 	bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2214 	TAILQ_INIT(&pmap->pm_ptbl_list);
2215 #endif
2216 }
2217 
2218 /*
2219  * Release any resources held by the given physical map.
2220  * Called when a pmap initialized by mmu_booke_pinit is being released.
2221  * Should only be called if the map contains no valid mappings.
2222  */
2223 static void
2224 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2225 {
2226 
2227 	KASSERT(pmap->pm_stats.resident_count == 0,
2228 	    ("pmap_release: pmap resident count %ld != 0",
2229 	    pmap->pm_stats.resident_count));
2230 #ifdef __powerpc64__
2231 	uma_zfree(ptbl_root_zone, pmap->pm_pp2d);
2232 #else
2233 	uma_zfree(ptbl_root_zone, pmap->pm_pdir);
2234 #endif
2235 }
2236 
2237 /*
2238  * Insert the given physical page at the specified virtual address in the
2239  * target physical map with the protection requested. If specified the page
2240  * will be wired down.
2241  */
2242 static int
2243 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2244     vm_prot_t prot, u_int flags, int8_t psind)
2245 {
2246 	int error;
2247 
2248 	rw_wlock(&pvh_global_lock);
2249 	PMAP_LOCK(pmap);
2250 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2251 	PMAP_UNLOCK(pmap);
2252 	rw_wunlock(&pvh_global_lock);
2253 	return (error);
2254 }
2255 
2256 static int
2257 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2258     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2259 {
2260 	pte_t *pte;
2261 	vm_paddr_t pa;
2262 	uint32_t flags;
2263 	int error, su, sync;
2264 
2265 	pa = VM_PAGE_TO_PHYS(m);
2266 	su = (pmap == kernel_pmap);
2267 	sync = 0;
2268 
2269 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2270 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2271 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2272 	//		(u_int32_t)m, va, pa, prot, flags);
2273 
2274 	if (su) {
2275 		KASSERT(((va >= virtual_avail) &&
2276 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2277 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2278 	} else {
2279 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2280 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2281 	}
2282 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2283 		VM_OBJECT_ASSERT_LOCKED(m->object);
2284 
2285 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2286 
2287 	/*
2288 	 * If there is an existing mapping, and the physical address has not
2289 	 * changed, must be protection or wiring change.
2290 	 */
2291 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2292 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2293 
2294 		/*
2295 		 * Before actually updating pte->flags we calculate and
2296 		 * prepare its new value in a helper var.
2297 		 */
2298 		flags = *pte;
2299 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2300 
2301 		/* Wiring change, just update stats. */
2302 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2303 			if (!PTE_ISWIRED(pte)) {
2304 				flags |= PTE_WIRED;
2305 				pmap->pm_stats.wired_count++;
2306 			}
2307 		} else {
2308 			if (PTE_ISWIRED(pte)) {
2309 				flags &= ~PTE_WIRED;
2310 				pmap->pm_stats.wired_count--;
2311 			}
2312 		}
2313 
2314 		if (prot & VM_PROT_WRITE) {
2315 			/* Add write permissions. */
2316 			flags |= PTE_SW;
2317 			if (!su)
2318 				flags |= PTE_UW;
2319 
2320 			if ((flags & PTE_MANAGED) != 0)
2321 				vm_page_aflag_set(m, PGA_WRITEABLE);
2322 		} else {
2323 			/* Handle modified pages, sense modify status. */
2324 
2325 			/*
2326 			 * The PTE_MODIFIED flag could be set by underlying
2327 			 * TLB misses since we last read it (above), possibly
2328 			 * other CPUs could update it so we check in the PTE
2329 			 * directly rather than rely on that saved local flags
2330 			 * copy.
2331 			 */
2332 			if (PTE_ISMODIFIED(pte))
2333 				vm_page_dirty(m);
2334 		}
2335 
2336 		if (prot & VM_PROT_EXECUTE) {
2337 			flags |= PTE_SX;
2338 			if (!su)
2339 				flags |= PTE_UX;
2340 
2341 			/*
2342 			 * Check existing flags for execute permissions: if we
2343 			 * are turning execute permissions on, icache should
2344 			 * be flushed.
2345 			 */
2346 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2347 				sync++;
2348 		}
2349 
2350 		flags &= ~PTE_REFERENCED;
2351 
2352 		/*
2353 		 * The new flags value is all calculated -- only now actually
2354 		 * update the PTE.
2355 		 */
2356 		mtx_lock_spin(&tlbivax_mutex);
2357 		tlb_miss_lock();
2358 
2359 		tlb0_flush_entry(va);
2360 		*pte &= ~PTE_FLAGS_MASK;
2361 		*pte |= flags;
2362 
2363 		tlb_miss_unlock();
2364 		mtx_unlock_spin(&tlbivax_mutex);
2365 
2366 	} else {
2367 		/*
2368 		 * If there is an existing mapping, but it's for a different
2369 		 * physical address, pte_enter() will delete the old mapping.
2370 		 */
2371 		//if ((pte != NULL) && PTE_ISVALID(pte))
2372 		//	debugf("mmu_booke_enter_locked: replace\n");
2373 		//else
2374 		//	debugf("mmu_booke_enter_locked: new\n");
2375 
2376 		/* Now set up the flags and install the new mapping. */
2377 		flags = (PTE_SR | PTE_VALID);
2378 		flags |= PTE_M;
2379 
2380 		if (!su)
2381 			flags |= PTE_UR;
2382 
2383 		if (prot & VM_PROT_WRITE) {
2384 			flags |= PTE_SW;
2385 			if (!su)
2386 				flags |= PTE_UW;
2387 
2388 			if ((m->oflags & VPO_UNMANAGED) == 0)
2389 				vm_page_aflag_set(m, PGA_WRITEABLE);
2390 		}
2391 
2392 		if (prot & VM_PROT_EXECUTE) {
2393 			flags |= PTE_SX;
2394 			if (!su)
2395 				flags |= PTE_UX;
2396 		}
2397 
2398 		/* If its wired update stats. */
2399 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2400 			flags |= PTE_WIRED;
2401 
2402 		error = pte_enter(mmu, pmap, m, va, flags,
2403 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2404 		if (error != 0)
2405 			return (KERN_RESOURCE_SHORTAGE);
2406 
2407 		if ((flags & PMAP_ENTER_WIRED) != 0)
2408 			pmap->pm_stats.wired_count++;
2409 
2410 		/* Flush the real memory from the instruction cache. */
2411 		if (prot & VM_PROT_EXECUTE)
2412 			sync++;
2413 	}
2414 
2415 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2416 		__syncicache((void *)va, PAGE_SIZE);
2417 		sync = 0;
2418 	}
2419 
2420 	return (KERN_SUCCESS);
2421 }
2422 
2423 /*
2424  * Maps a sequence of resident pages belonging to the same object.
2425  * The sequence begins with the given page m_start.  This page is
2426  * mapped at the given virtual address start.  Each subsequent page is
2427  * mapped at a virtual address that is offset from start by the same
2428  * amount as the page is offset from m_start within the object.  The
2429  * last page in the sequence is the page with the largest offset from
2430  * m_start that can be mapped at a virtual address less than the given
2431  * virtual address end.  Not every virtual page between start and end
2432  * is mapped; only those for which a resident page exists with the
2433  * corresponding offset from m_start are mapped.
2434  */
2435 static void
2436 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2437     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2438 {
2439 	vm_page_t m;
2440 	vm_pindex_t diff, psize;
2441 
2442 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2443 
2444 	psize = atop(end - start);
2445 	m = m_start;
2446 	rw_wlock(&pvh_global_lock);
2447 	PMAP_LOCK(pmap);
2448 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2449 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2450 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2451 		    PMAP_ENTER_NOSLEEP, 0);
2452 		m = TAILQ_NEXT(m, listq);
2453 	}
2454 	rw_wunlock(&pvh_global_lock);
2455 	PMAP_UNLOCK(pmap);
2456 }
2457 
2458 static void
2459 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2460     vm_prot_t prot)
2461 {
2462 
2463 	rw_wlock(&pvh_global_lock);
2464 	PMAP_LOCK(pmap);
2465 	mmu_booke_enter_locked(mmu, pmap, va, m,
2466 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2467 	    0);
2468 	rw_wunlock(&pvh_global_lock);
2469 	PMAP_UNLOCK(pmap);
2470 }
2471 
2472 /*
2473  * Remove the given range of addresses from the specified map.
2474  *
2475  * It is assumed that the start and end are properly rounded to the page size.
2476  */
2477 static void
2478 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2479 {
2480 	pte_t *pte;
2481 	uint8_t hold_flag;
2482 
2483 	int su = (pmap == kernel_pmap);
2484 
2485 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2486 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2487 
2488 	if (su) {
2489 		KASSERT(((va >= virtual_avail) &&
2490 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2491 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2492 	} else {
2493 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2494 		    ("mmu_booke_remove: user pmap, non user va"));
2495 	}
2496 
2497 	if (PMAP_REMOVE_DONE(pmap)) {
2498 		//debugf("mmu_booke_remove: e (empty)\n");
2499 		return;
2500 	}
2501 
2502 	hold_flag = PTBL_HOLD_FLAG(pmap);
2503 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2504 
2505 	rw_wlock(&pvh_global_lock);
2506 	PMAP_LOCK(pmap);
2507 	for (; va < endva; va += PAGE_SIZE) {
2508 		pte = pte_find(mmu, pmap, va);
2509 		if ((pte != NULL) && PTE_ISVALID(pte))
2510 			pte_remove(mmu, pmap, va, hold_flag);
2511 	}
2512 	PMAP_UNLOCK(pmap);
2513 	rw_wunlock(&pvh_global_lock);
2514 
2515 	//debugf("mmu_booke_remove: e\n");
2516 }
2517 
2518 /*
2519  * Remove physical page from all pmaps in which it resides.
2520  */
2521 static void
2522 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2523 {
2524 	pv_entry_t pv, pvn;
2525 	uint8_t hold_flag;
2526 
2527 	rw_wlock(&pvh_global_lock);
2528 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2529 		pvn = TAILQ_NEXT(pv, pv_link);
2530 
2531 		PMAP_LOCK(pv->pv_pmap);
2532 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2533 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2534 		PMAP_UNLOCK(pv->pv_pmap);
2535 	}
2536 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2537 	rw_wunlock(&pvh_global_lock);
2538 }
2539 
2540 /*
2541  * Map a range of physical addresses into kernel virtual address space.
2542  */
2543 static vm_offset_t
2544 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2545     vm_paddr_t pa_end, int prot)
2546 {
2547 	vm_offset_t sva = *virt;
2548 	vm_offset_t va = sva;
2549 
2550 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2551 	//		sva, pa_start, pa_end);
2552 
2553 	while (pa_start < pa_end) {
2554 		mmu_booke_kenter(mmu, va, pa_start);
2555 		va += PAGE_SIZE;
2556 		pa_start += PAGE_SIZE;
2557 	}
2558 	*virt = va;
2559 
2560 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2561 	return (sva);
2562 }
2563 
2564 /*
2565  * The pmap must be activated before it's address space can be accessed in any
2566  * way.
2567  */
2568 static void
2569 mmu_booke_activate(mmu_t mmu, struct thread *td)
2570 {
2571 	pmap_t pmap;
2572 	u_int cpuid;
2573 
2574 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2575 
2576 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
2577 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2578 
2579 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2580 
2581 	sched_pin();
2582 
2583 	cpuid = PCPU_GET(cpuid);
2584 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2585 	PCPU_SET(curpmap, pmap);
2586 
2587 	if (pmap->pm_tid[cpuid] == TID_NONE)
2588 		tid_alloc(pmap);
2589 
2590 	/* Load PID0 register with pmap tid value. */
2591 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2592 	__asm __volatile("isync");
2593 
2594 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2595 
2596 	sched_unpin();
2597 
2598 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2599 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2600 }
2601 
2602 /*
2603  * Deactivate the specified process's address space.
2604  */
2605 static void
2606 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2607 {
2608 	pmap_t pmap;
2609 
2610 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2611 
2612 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
2613 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2614 
2615 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2616 
2617 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2618 	PCPU_SET(curpmap, NULL);
2619 }
2620 
2621 /*
2622  * Copy the range specified by src_addr/len
2623  * from the source map to the range dst_addr/len
2624  * in the destination map.
2625  *
2626  * This routine is only advisory and need not do anything.
2627  */
2628 static void
2629 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2630     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2631 {
2632 
2633 }
2634 
2635 /*
2636  * Set the physical protection on the specified range of this map as requested.
2637  */
2638 static void
2639 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2640     vm_prot_t prot)
2641 {
2642 	vm_offset_t va;
2643 	vm_page_t m;
2644 	pte_t *pte;
2645 
2646 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2647 		mmu_booke_remove(mmu, pmap, sva, eva);
2648 		return;
2649 	}
2650 
2651 	if (prot & VM_PROT_WRITE)
2652 		return;
2653 
2654 	PMAP_LOCK(pmap);
2655 	for (va = sva; va < eva; va += PAGE_SIZE) {
2656 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2657 			if (PTE_ISVALID(pte)) {
2658 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2659 
2660 				mtx_lock_spin(&tlbivax_mutex);
2661 				tlb_miss_lock();
2662 
2663 				/* Handle modified pages. */
2664 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2665 					vm_page_dirty(m);
2666 
2667 				tlb0_flush_entry(va);
2668 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2669 
2670 				tlb_miss_unlock();
2671 				mtx_unlock_spin(&tlbivax_mutex);
2672 			}
2673 		}
2674 	}
2675 	PMAP_UNLOCK(pmap);
2676 }
2677 
2678 /*
2679  * Clear the write and modified bits in each of the given page's mappings.
2680  */
2681 static void
2682 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2683 {
2684 	pv_entry_t pv;
2685 	pte_t *pte;
2686 
2687 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2688 	    ("mmu_booke_remove_write: page %p is not managed", m));
2689 
2690 	/*
2691 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2692 	 * set by another thread while the object is locked.  Thus,
2693 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2694 	 */
2695 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2696 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2697 		return;
2698 	rw_wlock(&pvh_global_lock);
2699 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2700 		PMAP_LOCK(pv->pv_pmap);
2701 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2702 			if (PTE_ISVALID(pte)) {
2703 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2704 
2705 				mtx_lock_spin(&tlbivax_mutex);
2706 				tlb_miss_lock();
2707 
2708 				/* Handle modified pages. */
2709 				if (PTE_ISMODIFIED(pte))
2710 					vm_page_dirty(m);
2711 
2712 				/* Flush mapping from TLB0. */
2713 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2714 
2715 				tlb_miss_unlock();
2716 				mtx_unlock_spin(&tlbivax_mutex);
2717 			}
2718 		}
2719 		PMAP_UNLOCK(pv->pv_pmap);
2720 	}
2721 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2722 	rw_wunlock(&pvh_global_lock);
2723 }
2724 
2725 static void
2726 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2727 {
2728 	pte_t *pte;
2729 	vm_paddr_t pa = 0;
2730 	int sync_sz, valid;
2731 #ifndef __powerpc64__
2732 	pmap_t pmap;
2733 	vm_page_t m;
2734 	vm_offset_t addr;
2735 	int active;
2736 #endif
2737 
2738 #ifndef __powerpc64__
2739 	rw_wlock(&pvh_global_lock);
2740 	pmap = PCPU_GET(curpmap);
2741 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2742 #endif
2743 	while (sz > 0) {
2744 		PMAP_LOCK(pm);
2745 		pte = pte_find(mmu, pm, va);
2746 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2747 		if (valid)
2748 			pa = PTE_PA(pte);
2749 		PMAP_UNLOCK(pm);
2750 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
2751 		sync_sz = min(sync_sz, sz);
2752 		if (valid) {
2753 #ifdef __powerpc64__
2754 			pa += (va & PAGE_MASK);
2755 			__syncicache((void *)PHYS_TO_DMAP(pa), sync_sz);
2756 #else
2757 			if (!active) {
2758 				/* Create a mapping in the active pmap. */
2759 				addr = 0;
2760 				m = PHYS_TO_VM_PAGE(pa);
2761 				PMAP_LOCK(pmap);
2762 				pte_enter(mmu, pmap, m, addr,
2763 				    PTE_SR | PTE_VALID, FALSE);
2764 				addr += (va & PAGE_MASK);
2765 				__syncicache((void *)addr, sync_sz);
2766 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2767 				PMAP_UNLOCK(pmap);
2768 			} else
2769 				__syncicache((void *)va, sync_sz);
2770 #endif
2771 		}
2772 		va += sync_sz;
2773 		sz -= sync_sz;
2774 	}
2775 #ifndef __powerpc64__
2776 	rw_wunlock(&pvh_global_lock);
2777 #endif
2778 }
2779 
2780 /*
2781  * Atomically extract and hold the physical page with the given
2782  * pmap and virtual address pair if that mapping permits the given
2783  * protection.
2784  */
2785 static vm_page_t
2786 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2787     vm_prot_t prot)
2788 {
2789 	pte_t *pte;
2790 	vm_page_t m;
2791 	uint32_t pte_wbit;
2792 	vm_paddr_t pa;
2793 
2794 	m = NULL;
2795 	pa = 0;
2796 	PMAP_LOCK(pmap);
2797 retry:
2798 	pte = pte_find(mmu, pmap, va);
2799 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2800 		if (pmap == kernel_pmap)
2801 			pte_wbit = PTE_SW;
2802 		else
2803 			pte_wbit = PTE_UW;
2804 
2805 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2806 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2807 				goto retry;
2808 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2809 			m->wire_count++;
2810 		}
2811 	}
2812 
2813 	PA_UNLOCK_COND(pa);
2814 	PMAP_UNLOCK(pmap);
2815 	return (m);
2816 }
2817 
2818 /*
2819  * Initialize a vm_page's machine-dependent fields.
2820  */
2821 static void
2822 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2823 {
2824 
2825 	m->md.pv_tracked = 0;
2826 	TAILQ_INIT(&m->md.pv_list);
2827 }
2828 
2829 /*
2830  * mmu_booke_zero_page_area zeros the specified hardware page by
2831  * mapping it into virtual memory and using bzero to clear
2832  * its contents.
2833  *
2834  * off and size must reside within a single page.
2835  */
2836 static void
2837 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2838 {
2839 	vm_offset_t va;
2840 
2841 	/* XXX KASSERT off and size are within a single page? */
2842 
2843 #ifdef __powerpc64__
2844 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2845 	bzero((caddr_t)va + off, size);
2846 #else
2847 	mtx_lock(&zero_page_mutex);
2848 	va = zero_page_va;
2849 
2850 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2851 	bzero((caddr_t)va + off, size);
2852 	mmu_booke_kremove(mmu, va);
2853 
2854 	mtx_unlock(&zero_page_mutex);
2855 #endif
2856 }
2857 
2858 /*
2859  * mmu_booke_zero_page zeros the specified hardware page.
2860  */
2861 static void
2862 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2863 {
2864 	vm_offset_t off, va;
2865 
2866 #ifdef __powerpc64__
2867 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2868 
2869 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2870 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2871 #else
2872 	va = zero_page_va;
2873 	mtx_lock(&zero_page_mutex);
2874 
2875 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2876 
2877 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2878 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2879 
2880 	mmu_booke_kremove(mmu, va);
2881 
2882 	mtx_unlock(&zero_page_mutex);
2883 #endif
2884 }
2885 
2886 /*
2887  * mmu_booke_copy_page copies the specified (machine independent) page by
2888  * mapping the page into virtual memory and using memcopy to copy the page,
2889  * one machine dependent page at a time.
2890  */
2891 static void
2892 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2893 {
2894 	vm_offset_t sva, dva;
2895 
2896 #ifdef __powerpc64__
2897 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
2898 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
2899 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2900 #else
2901 	sva = copy_page_src_va;
2902 	dva = copy_page_dst_va;
2903 
2904 	mtx_lock(&copy_page_mutex);
2905 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2906 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2907 
2908 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2909 
2910 	mmu_booke_kremove(mmu, dva);
2911 	mmu_booke_kremove(mmu, sva);
2912 	mtx_unlock(&copy_page_mutex);
2913 #endif
2914 }
2915 
2916 static inline void
2917 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2918     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2919 {
2920 	void *a_cp, *b_cp;
2921 	vm_offset_t a_pg_offset, b_pg_offset;
2922 	int cnt;
2923 
2924 #ifdef __powerpc64__
2925 	vm_page_t pa, pb;
2926 
2927 	while (xfersize > 0) {
2928 		a_pg_offset = a_offset & PAGE_MASK;
2929 		pa = ma[a_offset >> PAGE_SHIFT];
2930 		b_pg_offset = b_offset & PAGE_MASK;
2931 		pb = mb[b_offset >> PAGE_SHIFT];
2932 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2933 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2934 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
2935 		    a_pg_offset);
2936 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
2937 		    b_pg_offset);
2938 		bcopy(a_cp, b_cp, cnt);
2939 		a_offset += cnt;
2940 		b_offset += cnt;
2941 		xfersize -= cnt;
2942 	}
2943 #else
2944 	mtx_lock(&copy_page_mutex);
2945 	while (xfersize > 0) {
2946 		a_pg_offset = a_offset & PAGE_MASK;
2947 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2948 		mmu_booke_kenter(mmu, copy_page_src_va,
2949 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2950 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2951 		b_pg_offset = b_offset & PAGE_MASK;
2952 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2953 		mmu_booke_kenter(mmu, copy_page_dst_va,
2954 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2955 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2956 		bcopy(a_cp, b_cp, cnt);
2957 		mmu_booke_kremove(mmu, copy_page_dst_va);
2958 		mmu_booke_kremove(mmu, copy_page_src_va);
2959 		a_offset += cnt;
2960 		b_offset += cnt;
2961 		xfersize -= cnt;
2962 	}
2963 	mtx_unlock(&copy_page_mutex);
2964 #endif
2965 }
2966 
2967 static vm_offset_t
2968 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
2969 {
2970 #ifdef __powerpc64__
2971 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
2972 #else
2973 	vm_paddr_t paddr;
2974 	vm_offset_t qaddr;
2975 	uint32_t flags;
2976 	pte_t *pte;
2977 
2978 	paddr = VM_PAGE_TO_PHYS(m);
2979 
2980 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2981 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
2982 	flags |= PTE_PS_4KB;
2983 
2984 	critical_enter();
2985 	qaddr = PCPU_GET(qmap_addr);
2986 
2987 	pte = pte_find(mmu, kernel_pmap, qaddr);
2988 
2989 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
2990 
2991 	/*
2992 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
2993  	 * not be present in other TLBs.  Is there a better instruction
2994 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
2995 	 */
2996 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
2997 	__asm __volatile("isync; msync");
2998 
2999 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3000 
3001 	/* Flush the real memory from the instruction cache. */
3002 	if ((flags & (PTE_I | PTE_G)) == 0)
3003 		__syncicache((void *)qaddr, PAGE_SIZE);
3004 
3005 	return (qaddr);
3006 #endif
3007 }
3008 
3009 static void
3010 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3011 {
3012 #ifndef __powerpc64__
3013 	pte_t *pte;
3014 
3015 	pte = pte_find(mmu, kernel_pmap, addr);
3016 
3017 	KASSERT(PCPU_GET(qmap_addr) == addr,
3018 	    ("mmu_booke_quick_remove_page: invalid address"));
3019 	KASSERT(*pte != 0,
3020 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3021 
3022 	*pte = 0;
3023 	critical_exit();
3024 #endif
3025 }
3026 
3027 /*
3028  * Return whether or not the specified physical page was modified
3029  * in any of physical maps.
3030  */
3031 static boolean_t
3032 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3033 {
3034 	pte_t *pte;
3035 	pv_entry_t pv;
3036 	boolean_t rv;
3037 
3038 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3039 	    ("mmu_booke_is_modified: page %p is not managed", m));
3040 	rv = FALSE;
3041 
3042 	/*
3043 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3044 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3045 	 * is clear, no PTEs can be modified.
3046 	 */
3047 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3048 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3049 		return (rv);
3050 	rw_wlock(&pvh_global_lock);
3051 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3052 		PMAP_LOCK(pv->pv_pmap);
3053 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3054 		    PTE_ISVALID(pte)) {
3055 			if (PTE_ISMODIFIED(pte))
3056 				rv = TRUE;
3057 		}
3058 		PMAP_UNLOCK(pv->pv_pmap);
3059 		if (rv)
3060 			break;
3061 	}
3062 	rw_wunlock(&pvh_global_lock);
3063 	return (rv);
3064 }
3065 
3066 /*
3067  * Return whether or not the specified virtual address is eligible
3068  * for prefault.
3069  */
3070 static boolean_t
3071 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3072 {
3073 
3074 	return (FALSE);
3075 }
3076 
3077 /*
3078  * Return whether or not the specified physical page was referenced
3079  * in any physical maps.
3080  */
3081 static boolean_t
3082 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3083 {
3084 	pte_t *pte;
3085 	pv_entry_t pv;
3086 	boolean_t rv;
3087 
3088 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3089 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3090 	rv = FALSE;
3091 	rw_wlock(&pvh_global_lock);
3092 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3093 		PMAP_LOCK(pv->pv_pmap);
3094 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3095 		    PTE_ISVALID(pte)) {
3096 			if (PTE_ISREFERENCED(pte))
3097 				rv = TRUE;
3098 		}
3099 		PMAP_UNLOCK(pv->pv_pmap);
3100 		if (rv)
3101 			break;
3102 	}
3103 	rw_wunlock(&pvh_global_lock);
3104 	return (rv);
3105 }
3106 
3107 /*
3108  * Clear the modify bits on the specified physical page.
3109  */
3110 static void
3111 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3112 {
3113 	pte_t *pte;
3114 	pv_entry_t pv;
3115 
3116 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3117 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3118 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3119 	KASSERT(!vm_page_xbusied(m),
3120 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3121 
3122 	/*
3123 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3124 	 * If the object containing the page is locked and the page is not
3125 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3126 	 */
3127 	if ((m->aflags & PGA_WRITEABLE) == 0)
3128 		return;
3129 	rw_wlock(&pvh_global_lock);
3130 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3131 		PMAP_LOCK(pv->pv_pmap);
3132 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3133 		    PTE_ISVALID(pte)) {
3134 			mtx_lock_spin(&tlbivax_mutex);
3135 			tlb_miss_lock();
3136 
3137 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3138 				tlb0_flush_entry(pv->pv_va);
3139 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3140 				    PTE_REFERENCED);
3141 			}
3142 
3143 			tlb_miss_unlock();
3144 			mtx_unlock_spin(&tlbivax_mutex);
3145 		}
3146 		PMAP_UNLOCK(pv->pv_pmap);
3147 	}
3148 	rw_wunlock(&pvh_global_lock);
3149 }
3150 
3151 /*
3152  * Return a count of reference bits for a page, clearing those bits.
3153  * It is not necessary for every reference bit to be cleared, but it
3154  * is necessary that 0 only be returned when there are truly no
3155  * reference bits set.
3156  *
3157  * As an optimization, update the page's dirty field if a modified bit is
3158  * found while counting reference bits.  This opportunistic update can be
3159  * performed at low cost and can eliminate the need for some future calls
3160  * to pmap_is_modified().  However, since this function stops after
3161  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3162  * dirty pages.  Those dirty pages will only be detected by a future call
3163  * to pmap_is_modified().
3164  */
3165 static int
3166 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3167 {
3168 	pte_t *pte;
3169 	pv_entry_t pv;
3170 	int count;
3171 
3172 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3173 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3174 	count = 0;
3175 	rw_wlock(&pvh_global_lock);
3176 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3177 		PMAP_LOCK(pv->pv_pmap);
3178 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3179 		    PTE_ISVALID(pte)) {
3180 			if (PTE_ISMODIFIED(pte))
3181 				vm_page_dirty(m);
3182 			if (PTE_ISREFERENCED(pte)) {
3183 				mtx_lock_spin(&tlbivax_mutex);
3184 				tlb_miss_lock();
3185 
3186 				tlb0_flush_entry(pv->pv_va);
3187 				*pte &= ~PTE_REFERENCED;
3188 
3189 				tlb_miss_unlock();
3190 				mtx_unlock_spin(&tlbivax_mutex);
3191 
3192 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3193 					PMAP_UNLOCK(pv->pv_pmap);
3194 					break;
3195 				}
3196 			}
3197 		}
3198 		PMAP_UNLOCK(pv->pv_pmap);
3199 	}
3200 	rw_wunlock(&pvh_global_lock);
3201 	return (count);
3202 }
3203 
3204 /*
3205  * Clear the wired attribute from the mappings for the specified range of
3206  * addresses in the given pmap.  Every valid mapping within that range must
3207  * have the wired attribute set.  In contrast, invalid mappings cannot have
3208  * the wired attribute set, so they are ignored.
3209  *
3210  * The wired attribute of the page table entry is not a hardware feature, so
3211  * there is no need to invalidate any TLB entries.
3212  */
3213 static void
3214 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3215 {
3216 	vm_offset_t va;
3217 	pte_t *pte;
3218 
3219 	PMAP_LOCK(pmap);
3220 	for (va = sva; va < eva; va += PAGE_SIZE) {
3221 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3222 		    PTE_ISVALID(pte)) {
3223 			if (!PTE_ISWIRED(pte))
3224 				panic("mmu_booke_unwire: pte %p isn't wired",
3225 				    pte);
3226 			*pte &= ~PTE_WIRED;
3227 			pmap->pm_stats.wired_count--;
3228 		}
3229 	}
3230 	PMAP_UNLOCK(pmap);
3231 
3232 }
3233 
3234 /*
3235  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3236  * page.  This count may be changed upwards or downwards in the future; it is
3237  * only necessary that true be returned for a small subset of pmaps for proper
3238  * page aging.
3239  */
3240 static boolean_t
3241 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3242 {
3243 	pv_entry_t pv;
3244 	int loops;
3245 	boolean_t rv;
3246 
3247 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3248 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3249 	loops = 0;
3250 	rv = FALSE;
3251 	rw_wlock(&pvh_global_lock);
3252 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3253 		if (pv->pv_pmap == pmap) {
3254 			rv = TRUE;
3255 			break;
3256 		}
3257 		if (++loops >= 16)
3258 			break;
3259 	}
3260 	rw_wunlock(&pvh_global_lock);
3261 	return (rv);
3262 }
3263 
3264 /*
3265  * Return the number of managed mappings to the given physical page that are
3266  * wired.
3267  */
3268 static int
3269 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3270 {
3271 	pv_entry_t pv;
3272 	pte_t *pte;
3273 	int count = 0;
3274 
3275 	if ((m->oflags & VPO_UNMANAGED) != 0)
3276 		return (count);
3277 	rw_wlock(&pvh_global_lock);
3278 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3279 		PMAP_LOCK(pv->pv_pmap);
3280 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3281 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3282 				count++;
3283 		PMAP_UNLOCK(pv->pv_pmap);
3284 	}
3285 	rw_wunlock(&pvh_global_lock);
3286 	return (count);
3287 }
3288 
3289 static int
3290 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3291 {
3292 	int i;
3293 	vm_offset_t va;
3294 
3295 	/*
3296 	 * This currently does not work for entries that
3297 	 * overlap TLB1 entries.
3298 	 */
3299 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3300 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3301 			return (0);
3302 	}
3303 
3304 	return (EFAULT);
3305 }
3306 
3307 void
3308 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3309 {
3310 	vm_paddr_t ppa;
3311 	vm_offset_t ofs;
3312 	vm_size_t gran;
3313 
3314 	/* Minidumps are based on virtual memory addresses. */
3315 	if (do_minidump) {
3316 		*va = (void *)(vm_offset_t)pa;
3317 		return;
3318 	}
3319 
3320 	/* Raw physical memory dumps don't have a virtual address. */
3321 	/* We always map a 256MB page at 256M. */
3322 	gran = 256 * 1024 * 1024;
3323 	ppa = rounddown2(pa, gran);
3324 	ofs = pa - ppa;
3325 	*va = (void *)gran;
3326 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3327 
3328 	if (sz > (gran - ofs))
3329 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3330 		    _TLB_ENTRY_IO);
3331 }
3332 
3333 void
3334 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3335 {
3336 	vm_paddr_t ppa;
3337 	vm_offset_t ofs;
3338 	vm_size_t gran;
3339 	tlb_entry_t e;
3340 	int i;
3341 
3342 	/* Minidumps are based on virtual memory addresses. */
3343 	/* Nothing to do... */
3344 	if (do_minidump)
3345 		return;
3346 
3347 	for (i = 0; i < TLB1_ENTRIES; i++) {
3348 		tlb1_read_entry(&e, i);
3349 		if (!(e.mas1 & MAS1_VALID))
3350 			break;
3351 	}
3352 
3353 	/* Raw physical memory dumps don't have a virtual address. */
3354 	i--;
3355 	e.mas1 = 0;
3356 	e.mas2 = 0;
3357 	e.mas3 = 0;
3358 	tlb1_write_entry(&e, i);
3359 
3360 	gran = 256 * 1024 * 1024;
3361 	ppa = rounddown2(pa, gran);
3362 	ofs = pa - ppa;
3363 	if (sz > (gran - ofs)) {
3364 		i--;
3365 		e.mas1 = 0;
3366 		e.mas2 = 0;
3367 		e.mas3 = 0;
3368 		tlb1_write_entry(&e, i);
3369 	}
3370 }
3371 
3372 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3373 
3374 void
3375 mmu_booke_scan_init(mmu_t mmu)
3376 {
3377 	vm_offset_t va;
3378 	pte_t *pte;
3379 	int i;
3380 
3381 	if (!do_minidump) {
3382 		/* Initialize phys. segments for dumpsys(). */
3383 		memset(&dump_map, 0, sizeof(dump_map));
3384 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3385 		    &availmem_regions_sz);
3386 		for (i = 0; i < physmem_regions_sz; i++) {
3387 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3388 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3389 		}
3390 		return;
3391 	}
3392 
3393 	/* Virtual segments for minidumps: */
3394 	memset(&dump_map, 0, sizeof(dump_map));
3395 
3396 	/* 1st: kernel .data and .bss. */
3397 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3398 	dump_map[0].pa_size =
3399 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3400 
3401 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3402 	dump_map[1].pa_start = data_start;
3403 	dump_map[1].pa_size = data_end - data_start;
3404 
3405 	/* 3rd: kernel VM. */
3406 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3407 	/* Find start of next chunk (from va). */
3408 	while (va < virtual_end) {
3409 		/* Don't dump the buffer cache. */
3410 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3411 			va = kmi.buffer_eva;
3412 			continue;
3413 		}
3414 		pte = pte_find(mmu, kernel_pmap, va);
3415 		if (pte != NULL && PTE_ISVALID(pte))
3416 			break;
3417 		va += PAGE_SIZE;
3418 	}
3419 	if (va < virtual_end) {
3420 		dump_map[2].pa_start = va;
3421 		va += PAGE_SIZE;
3422 		/* Find last page in chunk. */
3423 		while (va < virtual_end) {
3424 			/* Don't run into the buffer cache. */
3425 			if (va == kmi.buffer_sva)
3426 				break;
3427 			pte = pte_find(mmu, kernel_pmap, va);
3428 			if (pte == NULL || !PTE_ISVALID(pte))
3429 				break;
3430 			va += PAGE_SIZE;
3431 		}
3432 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3433 	}
3434 }
3435 
3436 /*
3437  * Map a set of physical memory pages into the kernel virtual address space.
3438  * Return a pointer to where it is mapped. This routine is intended to be used
3439  * for mapping device memory, NOT real memory.
3440  */
3441 static void *
3442 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3443 {
3444 
3445 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3446 }
3447 
3448 static void *
3449 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3450 {
3451 	tlb_entry_t e;
3452 	void *res;
3453 	uintptr_t va, tmpva;
3454 	vm_size_t sz;
3455 	int i;
3456 
3457 	/*
3458 	 * Check if this is premapped in TLB1. Note: this should probably also
3459 	 * check whether a sequence of TLB1 entries exist that match the
3460 	 * requirement, but now only checks the easy case.
3461 	 */
3462 	for (i = 0; i < TLB1_ENTRIES; i++) {
3463 		tlb1_read_entry(&e, i);
3464 		if (!(e.mas1 & MAS1_VALID))
3465 			continue;
3466 		if (pa >= e.phys &&
3467 		    (pa + size) <= (e.phys + e.size) &&
3468 		    (ma == VM_MEMATTR_DEFAULT ||
3469 		     tlb_calc_wimg(pa, ma) ==
3470 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3471 			return (void *)(e.virt +
3472 			    (vm_offset_t)(pa - e.phys));
3473 	}
3474 
3475 	size = roundup(size, PAGE_SIZE);
3476 
3477 	/*
3478 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3479 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3480 	 */
3481 #ifdef SPARSE_MAPDEV
3482 	/*
3483 	 * With a sparse mapdev, align to the largest starting region.  This
3484 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3485 	 * calculation could be very costly.
3486 	 * Align to the smaller of:
3487 	 * - first set bit in overlap of (pa & size mask)
3488 	 * - largest size envelope
3489 	 *
3490 	 * It's possible the device mapping may start at a PA that's not larger
3491 	 * than the size mask, so we need to offset in to maximize the TLB entry
3492 	 * range and minimize the number of used TLB entries.
3493 	 */
3494 	do {
3495 	    tmpva = tlb1_map_base;
3496 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3497 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3498 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3499 #ifdef __powerpc64__
3500 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3501 #else
3502 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3503 #endif
3504 #else
3505 #ifdef __powerpc64__
3506 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3507 #else
3508 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3509 #endif
3510 #endif
3511 	res = (void *)va;
3512 
3513 	do {
3514 		sz = 1 << (ilog2(size) & ~1);
3515 		/* Align size to PA */
3516 		if (pa % sz != 0) {
3517 			do {
3518 				sz >>= 2;
3519 			} while (pa % sz != 0);
3520 		}
3521 		/* Now align from there to VA */
3522 		if (va % sz != 0) {
3523 			do {
3524 				sz >>= 2;
3525 			} while (va % sz != 0);
3526 		}
3527 		if (bootverbose)
3528 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3529 			    va, (uintmax_t)pa, sz);
3530 		if (tlb1_set_entry(va, pa, sz,
3531 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3532 			return (NULL);
3533 		size -= sz;
3534 		pa += sz;
3535 		va += sz;
3536 	} while (size > 0);
3537 
3538 	return (res);
3539 }
3540 
3541 /*
3542  * 'Unmap' a range mapped by mmu_booke_mapdev().
3543  */
3544 static void
3545 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3546 {
3547 #ifdef SUPPORTS_SHRINKING_TLB1
3548 	vm_offset_t base, offset;
3549 
3550 	/*
3551 	 * Unmap only if this is inside kernel virtual space.
3552 	 */
3553 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3554 		base = trunc_page(va);
3555 		offset = va & PAGE_MASK;
3556 		size = roundup(offset + size, PAGE_SIZE);
3557 		kva_free(base, size);
3558 	}
3559 #endif
3560 }
3561 
3562 /*
3563  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3564  * specified pmap. This eliminates the blast of soft faults on process startup
3565  * and immediately after an mmap.
3566  */
3567 static void
3568 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3569     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3570 {
3571 
3572 	VM_OBJECT_ASSERT_WLOCKED(object);
3573 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3574 	    ("mmu_booke_object_init_pt: non-device object"));
3575 }
3576 
3577 /*
3578  * Perform the pmap work for mincore.
3579  */
3580 static int
3581 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3582     vm_paddr_t *locked_pa)
3583 {
3584 
3585 	/* XXX: this should be implemented at some point */
3586 	return (0);
3587 }
3588 
3589 static int
3590 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3591     vm_memattr_t mode)
3592 {
3593 	vm_offset_t va;
3594 	pte_t *pte;
3595 	int i, j;
3596 	tlb_entry_t e;
3597 
3598 	/* Check TLB1 mappings */
3599 	for (i = 0; i < TLB1_ENTRIES; i++) {
3600 		tlb1_read_entry(&e, i);
3601 		if (!(e.mas1 & MAS1_VALID))
3602 			continue;
3603 		if (addr >= e.virt && addr < e.virt + e.size)
3604 			break;
3605 	}
3606 	if (i < TLB1_ENTRIES) {
3607 		/* Only allow full mappings to be modified for now. */
3608 		/* Validate the range. */
3609 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3610 			tlb1_read_entry(&e, j);
3611 			if (va != e.virt || (sz - (va - addr) < e.size))
3612 				return (EINVAL);
3613 		}
3614 		for (va = addr; va < addr + sz; va += e.size, i++) {
3615 			tlb1_read_entry(&e, i);
3616 			e.mas2 &= ~MAS2_WIMGE_MASK;
3617 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3618 
3619 			/*
3620 			 * Write it out to the TLB.  Should really re-sync with other
3621 			 * cores.
3622 			 */
3623 			tlb1_write_entry(&e, i);
3624 		}
3625 		return (0);
3626 	}
3627 
3628 	/* Not in TLB1, try through pmap */
3629 	/* First validate the range. */
3630 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3631 		pte = pte_find(mmu, kernel_pmap, va);
3632 		if (pte == NULL || !PTE_ISVALID(pte))
3633 			return (EINVAL);
3634 	}
3635 
3636 	mtx_lock_spin(&tlbivax_mutex);
3637 	tlb_miss_lock();
3638 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3639 		pte = pte_find(mmu, kernel_pmap, va);
3640 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3641 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3642 		tlb0_flush_entry(va);
3643 	}
3644 	tlb_miss_unlock();
3645 	mtx_unlock_spin(&tlbivax_mutex);
3646 
3647 	return (0);
3648 }
3649 
3650 /**************************************************************************/
3651 /* TID handling */
3652 /**************************************************************************/
3653 
3654 /*
3655  * Allocate a TID. If necessary, steal one from someone else.
3656  * The new TID is flushed from the TLB before returning.
3657  */
3658 static tlbtid_t
3659 tid_alloc(pmap_t pmap)
3660 {
3661 	tlbtid_t tid;
3662 	int thiscpu;
3663 
3664 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3665 
3666 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3667 
3668 	thiscpu = PCPU_GET(cpuid);
3669 
3670 	tid = PCPU_GET(booke.tid_next);
3671 	if (tid > TID_MAX)
3672 		tid = TID_MIN;
3673 	PCPU_SET(booke.tid_next, tid + 1);
3674 
3675 	/* If we are stealing TID then clear the relevant pmap's field */
3676 	if (tidbusy[thiscpu][tid] != NULL) {
3677 
3678 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3679 
3680 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3681 
3682 		/* Flush all entries from TLB0 matching this TID. */
3683 		tid_flush(tid);
3684 	}
3685 
3686 	tidbusy[thiscpu][tid] = pmap;
3687 	pmap->pm_tid[thiscpu] = tid;
3688 	__asm __volatile("msync; isync");
3689 
3690 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3691 	    PCPU_GET(booke.tid_next));
3692 
3693 	return (tid);
3694 }
3695 
3696 /**************************************************************************/
3697 /* TLB0 handling */
3698 /**************************************************************************/
3699 
3700 /* Convert TLB0 va and way number to tlb0[] table index. */
3701 static inline unsigned int
3702 tlb0_tableidx(vm_offset_t va, unsigned int way)
3703 {
3704 	unsigned int idx;
3705 
3706 	idx = (way * TLB0_ENTRIES_PER_WAY);
3707 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3708 	return (idx);
3709 }
3710 
3711 /*
3712  * Invalidate TLB0 entry.
3713  */
3714 static inline void
3715 tlb0_flush_entry(vm_offset_t va)
3716 {
3717 
3718 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3719 
3720 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3721 
3722 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3723 	__asm __volatile("isync; msync");
3724 	__asm __volatile("tlbsync; msync");
3725 
3726 	CTR1(KTR_PMAP, "%s: e", __func__);
3727 }
3728 
3729 
3730 /**************************************************************************/
3731 /* TLB1 handling */
3732 /**************************************************************************/
3733 
3734 /*
3735  * TLB1 mapping notes:
3736  *
3737  * TLB1[0]	Kernel text and data.
3738  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3739  *		windows, other devices mappings.
3740  */
3741 
3742  /*
3743  * Read an entry from given TLB1 slot.
3744  */
3745 void
3746 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3747 {
3748 	register_t msr;
3749 	uint32_t mas0;
3750 
3751 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3752 
3753 	msr = mfmsr();
3754 	__asm __volatile("wrteei 0");
3755 
3756 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3757 	mtspr(SPR_MAS0, mas0);
3758 	__asm __volatile("isync; tlbre");
3759 
3760 	entry->mas1 = mfspr(SPR_MAS1);
3761 	entry->mas2 = mfspr(SPR_MAS2);
3762 	entry->mas3 = mfspr(SPR_MAS3);
3763 
3764 	switch ((mfpvr() >> 16) & 0xFFFF) {
3765 	case FSL_E500v2:
3766 	case FSL_E500mc:
3767 	case FSL_E5500:
3768 	case FSL_E6500:
3769 		entry->mas7 = mfspr(SPR_MAS7);
3770 		break;
3771 	default:
3772 		entry->mas7 = 0;
3773 		break;
3774 	}
3775 	__asm __volatile("wrtee %0" :: "r"(msr));
3776 
3777 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3778 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3779 	    (entry->mas3 & MAS3_RPN);
3780 	entry->size =
3781 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3782 }
3783 
3784 struct tlbwrite_args {
3785 	tlb_entry_t *e;
3786 	unsigned int idx;
3787 };
3788 
3789 static void
3790 tlb1_write_entry_int(void *arg)
3791 {
3792 	struct tlbwrite_args *args = arg;
3793 	uint32_t mas0;
3794 
3795 	/* Select entry */
3796 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3797 
3798 	mtspr(SPR_MAS0, mas0);
3799 	mtspr(SPR_MAS1, args->e->mas1);
3800 	mtspr(SPR_MAS2, args->e->mas2);
3801 	mtspr(SPR_MAS3, args->e->mas3);
3802 	switch ((mfpvr() >> 16) & 0xFFFF) {
3803 	case FSL_E500mc:
3804 	case FSL_E5500:
3805 	case FSL_E6500:
3806 		mtspr(SPR_MAS8, 0);
3807 		/* FALLTHROUGH */
3808 	case FSL_E500v2:
3809 		mtspr(SPR_MAS7, args->e->mas7);
3810 		break;
3811 	default:
3812 		break;
3813 	}
3814 
3815 	__asm __volatile("isync; tlbwe; isync; msync");
3816 
3817 }
3818 
3819 static void
3820 tlb1_write_entry_sync(void *arg)
3821 {
3822 	/* Empty synchronization point for smp_rendezvous(). */
3823 }
3824 
3825 /*
3826  * Write given entry to TLB1 hardware.
3827  */
3828 static void
3829 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3830 {
3831 	struct tlbwrite_args args;
3832 
3833 	args.e = e;
3834 	args.idx = idx;
3835 
3836 #ifdef SMP
3837 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3838 		mb();
3839 		smp_rendezvous(tlb1_write_entry_sync,
3840 		    tlb1_write_entry_int,
3841 		    tlb1_write_entry_sync, &args);
3842 	} else
3843 #endif
3844 	{
3845 		register_t msr;
3846 
3847 		msr = mfmsr();
3848 		__asm __volatile("wrteei 0");
3849 		tlb1_write_entry_int(&args);
3850 		__asm __volatile("wrtee %0" :: "r"(msr));
3851 	}
3852 }
3853 
3854 /*
3855  * Return the largest uint value log such that 2^log <= num.
3856  */
3857 static unsigned int
3858 ilog2(unsigned long num)
3859 {
3860 	long lz;
3861 
3862 #ifdef __powerpc64__
3863 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
3864 	return (63 - lz);
3865 #else
3866 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3867 	return (31 - lz);
3868 #endif
3869 }
3870 
3871 /*
3872  * Convert TLB TSIZE value to mapped region size.
3873  */
3874 static vm_size_t
3875 tsize2size(unsigned int tsize)
3876 {
3877 
3878 	/*
3879 	 * size = 4^tsize KB
3880 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3881 	 */
3882 
3883 	return ((1 << (2 * tsize)) * 1024);
3884 }
3885 
3886 /*
3887  * Convert region size (must be power of 4) to TLB TSIZE value.
3888  */
3889 static unsigned int
3890 size2tsize(vm_size_t size)
3891 {
3892 
3893 	return (ilog2(size) / 2 - 5);
3894 }
3895 
3896 /*
3897  * Register permanent kernel mapping in TLB1.
3898  *
3899  * Entries are created starting from index 0 (current free entry is
3900  * kept in tlb1_idx) and are not supposed to be invalidated.
3901  */
3902 int
3903 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
3904     uint32_t flags)
3905 {
3906 	tlb_entry_t e;
3907 	uint32_t ts, tid;
3908 	int tsize, index;
3909 
3910 	for (index = 0; index < TLB1_ENTRIES; index++) {
3911 		tlb1_read_entry(&e, index);
3912 		if ((e.mas1 & MAS1_VALID) == 0)
3913 			break;
3914 		/* Check if we're just updating the flags, and update them. */
3915 		if (e.phys == pa && e.virt == va && e.size == size) {
3916 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
3917 			tlb1_write_entry(&e, index);
3918 			return (0);
3919 		}
3920 	}
3921 	if (index >= TLB1_ENTRIES) {
3922 		printf("tlb1_set_entry: TLB1 full!\n");
3923 		return (-1);
3924 	}
3925 
3926 	/* Convert size to TSIZE */
3927 	tsize = size2tsize(size);
3928 
3929 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3930 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3931 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3932 
3933 	e.phys = pa;
3934 	e.virt = va;
3935 	e.size = size;
3936 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3937 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3938 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
3939 
3940 	/* Set supervisor RWX permission bits */
3941 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3942 	e.mas7 = (pa >> 32) & MAS7_RPN;
3943 
3944 	tlb1_write_entry(&e, index);
3945 
3946 	/*
3947 	 * XXX in general TLB1 updates should be propagated between CPUs,
3948 	 * since current design assumes to have the same TLB1 set-up on all
3949 	 * cores.
3950 	 */
3951 	return (0);
3952 }
3953 
3954 /*
3955  * Map in contiguous RAM region into the TLB1 using maximum of
3956  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3957  *
3958  * If necessary round up last entry size and return total size
3959  * used by all allocated entries.
3960  */
3961 vm_size_t
3962 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3963 {
3964 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3965 	vm_size_t mapped, pgsz, base, mask;
3966 	int idx, nents;
3967 
3968 	/* Round up to the next 1M */
3969 	size = roundup2(size, 1 << 20);
3970 
3971 	mapped = 0;
3972 	idx = 0;
3973 	base = va;
3974 	pgsz = 64*1024*1024;
3975 	while (mapped < size) {
3976 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3977 			while (pgsz > (size - mapped))
3978 				pgsz >>= 2;
3979 			pgs[idx++] = pgsz;
3980 			mapped += pgsz;
3981 		}
3982 
3983 		/* We under-map. Correct for this. */
3984 		if (mapped < size) {
3985 			while (pgs[idx - 1] == pgsz) {
3986 				idx--;
3987 				mapped -= pgsz;
3988 			}
3989 			/* XXX We may increase beyond out starting point. */
3990 			pgsz <<= 2;
3991 			pgs[idx++] = pgsz;
3992 			mapped += pgsz;
3993 		}
3994 	}
3995 
3996 	nents = idx;
3997 	mask = pgs[0] - 1;
3998 	/* Align address to the boundary */
3999 	if (va & mask) {
4000 		va = (va + mask) & ~mask;
4001 		pa = (pa + mask) & ~mask;
4002 	}
4003 
4004 	for (idx = 0; idx < nents; idx++) {
4005 		pgsz = pgs[idx];
4006 		debugf("%u: %llx -> %jx, size=%jx\n", idx, pa,
4007 		    (uintmax_t)va, (uintmax_t)pgsz);
4008 		tlb1_set_entry(va, pa, pgsz,
4009 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4010 		pa += pgsz;
4011 		va += pgsz;
4012 	}
4013 
4014 	mapped = (va - base);
4015 	if (bootverbose)
4016 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
4017 		    mapped, mapped - size);
4018 	return (mapped);
4019 }
4020 
4021 /*
4022  * TLB1 initialization routine, to be called after the very first
4023  * assembler level setup done in locore.S.
4024  */
4025 void
4026 tlb1_init()
4027 {
4028 	uint32_t mas0, mas1, mas2, mas3, mas7;
4029 	uint32_t tsz;
4030 
4031 	tlb1_get_tlbconf();
4032 
4033 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4034 	mtspr(SPR_MAS0, mas0);
4035 	__asm __volatile("isync; tlbre");
4036 
4037 	mas1 = mfspr(SPR_MAS1);
4038 	mas2 = mfspr(SPR_MAS2);
4039 	mas3 = mfspr(SPR_MAS3);
4040 	mas7 = mfspr(SPR_MAS7);
4041 
4042 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4043 	    (mas3 & MAS3_RPN);
4044 
4045 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4046 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4047 
4048 	/* Setup TLB miss defaults */
4049 	set_mas4_defaults();
4050 }
4051 
4052 /*
4053  * pmap_early_io_unmap() should be used in short conjunction with
4054  * pmap_early_io_map(), as in the following snippet:
4055  *
4056  * x = pmap_early_io_map(...);
4057  * <do something with x>
4058  * pmap_early_io_unmap(x, size);
4059  *
4060  * And avoiding more allocations between.
4061  */
4062 void
4063 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4064 {
4065 	int i;
4066 	tlb_entry_t e;
4067 	vm_size_t isize;
4068 
4069 	size = roundup(size, PAGE_SIZE);
4070 	isize = size;
4071 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4072 		tlb1_read_entry(&e, i);
4073 		if (!(e.mas1 & MAS1_VALID))
4074 			continue;
4075 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4076 			size -= e.size;
4077 			e.mas1 &= ~MAS1_VALID;
4078 			tlb1_write_entry(&e, i);
4079 		}
4080 	}
4081 	if (tlb1_map_base == va + isize)
4082 		tlb1_map_base -= isize;
4083 }
4084 
4085 vm_offset_t
4086 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4087 {
4088 	vm_paddr_t pa_base;
4089 	vm_offset_t va, sz;
4090 	int i;
4091 	tlb_entry_t e;
4092 
4093 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4094 
4095 	for (i = 0; i < TLB1_ENTRIES; i++) {
4096 		tlb1_read_entry(&e, i);
4097 		if (!(e.mas1 & MAS1_VALID))
4098 			continue;
4099 		if (pa >= e.phys && (pa + size) <=
4100 		    (e.phys + e.size))
4101 			return (e.virt + (pa - e.phys));
4102 	}
4103 
4104 	pa_base = rounddown(pa, PAGE_SIZE);
4105 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4106 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4107 	va = tlb1_map_base + (pa - pa_base);
4108 
4109 	do {
4110 		sz = 1 << (ilog2(size) & ~1);
4111 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4112 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4113 		size -= sz;
4114 		pa_base += sz;
4115 		tlb1_map_base += sz;
4116 	} while (size > 0);
4117 
4118 	return (va);
4119 }
4120 
4121 void
4122 pmap_track_page(pmap_t pmap, vm_offset_t va)
4123 {
4124 	vm_paddr_t pa;
4125 	vm_page_t page;
4126 	struct pv_entry *pve;
4127 
4128 	va = trunc_page(va);
4129 	pa = pmap_kextract(va);
4130 	page = PHYS_TO_VM_PAGE(pa);
4131 
4132 	rw_wlock(&pvh_global_lock);
4133 	PMAP_LOCK(pmap);
4134 
4135 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4136 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4137 			goto out;
4138 		}
4139 	}
4140 	page->md.pv_tracked = true;
4141 	pv_insert(pmap, va, page);
4142 out:
4143 	PMAP_UNLOCK(pmap);
4144 	rw_wunlock(&pvh_global_lock);
4145 }
4146 
4147 
4148 /*
4149  * Setup MAS4 defaults.
4150  * These values are loaded to MAS0-2 on a TLB miss.
4151  */
4152 static void
4153 set_mas4_defaults(void)
4154 {
4155 	uint32_t mas4;
4156 
4157 	/* Defaults: TLB0, PID0, TSIZED=4K */
4158 	mas4 = MAS4_TLBSELD0;
4159 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4160 #ifdef SMP
4161 	mas4 |= MAS4_MD;
4162 #endif
4163 	mtspr(SPR_MAS4, mas4);
4164 	__asm __volatile("isync");
4165 }
4166 
4167 
4168 /*
4169  * Return 0 if the physical IO range is encompassed by one of the
4170  * the TLB1 entries, otherwise return related error code.
4171  */
4172 static int
4173 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4174 {
4175 	uint32_t prot;
4176 	vm_paddr_t pa_start;
4177 	vm_paddr_t pa_end;
4178 	unsigned int entry_tsize;
4179 	vm_size_t entry_size;
4180 	tlb_entry_t e;
4181 
4182 	*va = (vm_offset_t)NULL;
4183 
4184 	tlb1_read_entry(&e, i);
4185 	/* Skip invalid entries */
4186 	if (!(e.mas1 & MAS1_VALID))
4187 		return (EINVAL);
4188 
4189 	/*
4190 	 * The entry must be cache-inhibited, guarded, and r/w
4191 	 * so it can function as an i/o page
4192 	 */
4193 	prot = e.mas2 & (MAS2_I | MAS2_G);
4194 	if (prot != (MAS2_I | MAS2_G))
4195 		return (EPERM);
4196 
4197 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4198 	if (prot != (MAS3_SR | MAS3_SW))
4199 		return (EPERM);
4200 
4201 	/* The address should be within the entry range. */
4202 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4203 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4204 
4205 	entry_size = tsize2size(entry_tsize);
4206 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4207 	    (e.mas3 & MAS3_RPN);
4208 	pa_end = pa_start + entry_size;
4209 
4210 	if ((pa < pa_start) || ((pa + size) > pa_end))
4211 		return (ERANGE);
4212 
4213 	/* Return virtual address of this mapping. */
4214 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4215 	return (0);
4216 }
4217 
4218 /*
4219  * Invalidate all TLB0 entries which match the given TID. Note this is
4220  * dedicated for cases when invalidations should NOT be propagated to other
4221  * CPUs.
4222  */
4223 static void
4224 tid_flush(tlbtid_t tid)
4225 {
4226 	register_t msr;
4227 	uint32_t mas0, mas1, mas2;
4228 	int entry, way;
4229 
4230 
4231 	/* Don't evict kernel translations */
4232 	if (tid == TID_KERNEL)
4233 		return;
4234 
4235 	msr = mfmsr();
4236 	__asm __volatile("wrteei 0");
4237 
4238 	/*
4239 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
4240 	 * it for PID invalidation.
4241 	 */
4242 	switch ((mfpvr() >> 16) & 0xffff) {
4243 	case FSL_E500mc:
4244 	case FSL_E5500:
4245 	case FSL_E6500:
4246 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
4247 		/* tlbilxpid */
4248 		__asm __volatile("isync; .long 0x7c000024; isync; msync");
4249 		__asm __volatile("wrtee %0" :: "r"(msr));
4250 		return;
4251 	}
4252 
4253 	for (way = 0; way < TLB0_WAYS; way++)
4254 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4255 
4256 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4257 			mtspr(SPR_MAS0, mas0);
4258 
4259 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4260 			mtspr(SPR_MAS2, mas2);
4261 
4262 			__asm __volatile("isync; tlbre");
4263 
4264 			mas1 = mfspr(SPR_MAS1);
4265 
4266 			if (!(mas1 & MAS1_VALID))
4267 				continue;
4268 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4269 				continue;
4270 			mas1 &= ~MAS1_VALID;
4271 			mtspr(SPR_MAS1, mas1);
4272 			__asm __volatile("isync; tlbwe; isync; msync");
4273 		}
4274 	__asm __volatile("wrtee %0" :: "r"(msr));
4275 }
4276 
4277 #ifdef DDB
4278 /* Print out contents of the MAS registers for each TLB0 entry */
4279 static void
4280 #ifdef __powerpc64__
4281 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
4282 #else
4283 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
4284 #endif
4285     uint32_t mas7)
4286 {
4287 	int as;
4288 	char desc[3];
4289 	tlbtid_t tid;
4290 	vm_size_t size;
4291 	unsigned int tsize;
4292 
4293 	desc[2] = '\0';
4294 	if (mas1 & MAS1_VALID)
4295 		desc[0] = 'V';
4296 	else
4297 		desc[0] = ' ';
4298 
4299 	if (mas1 & MAS1_IPROT)
4300 		desc[1] = 'P';
4301 	else
4302 		desc[1] = ' ';
4303 
4304 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
4305 	tid = MAS1_GETTID(mas1);
4306 
4307 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4308 	size = 0;
4309 	if (tsize)
4310 		size = tsize2size(tsize);
4311 
4312 	printf("%3d: (%s) [AS=%d] "
4313 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
4314 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
4315 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
4316 }
4317 
4318 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
4319 {
4320 	uint32_t mas0, mas1, mas3, mas7;
4321 #ifdef __powerpc64__
4322 	uint64_t mas2;
4323 #else
4324 	uint32_t mas2;
4325 #endif
4326 	int entryidx, way, idx;
4327 
4328 	printf("TLB0 entries:\n");
4329 	for (way = 0; way < TLB0_WAYS; way ++)
4330 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
4331 
4332 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4333 			mtspr(SPR_MAS0, mas0);
4334 
4335 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
4336 			mtspr(SPR_MAS2, mas2);
4337 
4338 			__asm __volatile("isync; tlbre");
4339 
4340 			mas1 = mfspr(SPR_MAS1);
4341 			mas2 = mfspr(SPR_MAS2);
4342 			mas3 = mfspr(SPR_MAS3);
4343 			mas7 = mfspr(SPR_MAS7);
4344 
4345 			idx = tlb0_tableidx(mas2, way);
4346 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
4347 		}
4348 }
4349 
4350 /*
4351  * Print out contents of the MAS registers for each TLB1 entry
4352  */
4353 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
4354 {
4355 	uint32_t mas0, mas1, mas3, mas7;
4356 #ifdef __powerpc64__
4357 	uint64_t mas2;
4358 #else
4359 	uint32_t mas2;
4360 #endif
4361 	int i;
4362 
4363 	printf("TLB1 entries:\n");
4364 	for (i = 0; i < TLB1_ENTRIES; i++) {
4365 
4366 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4367 		mtspr(SPR_MAS0, mas0);
4368 
4369 		__asm __volatile("isync; tlbre");
4370 
4371 		mas1 = mfspr(SPR_MAS1);
4372 		mas2 = mfspr(SPR_MAS2);
4373 		mas3 = mfspr(SPR_MAS3);
4374 		mas7 = mfspr(SPR_MAS7);
4375 
4376 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4377 	}
4378 }
4379 #endif
4380