xref: /freebsd/sys/powerpc/booke/pmap.c (revision daceb336172a6b0572de864b97e70b28451ca636)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - ptbl_bufs-1              : message buffer
63   *               ptbl_bufs - kernel_pdir-1            : kernel page tables
64   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
65   *             kernel_pp2d - .                        : kernel pointers to page directory
66   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
67   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
68   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
69   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
70   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
71   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
72   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
73   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
74   *                         - 0xffff_ffff_ffff_ffff    : device direct map
75   */
76 
77 #include <sys/cdefs.h>
78 __FBSDID("$FreeBSD$");
79 
80 #include "opt_ddb.h"
81 #include "opt_kstack_pages.h"
82 
83 #include <sys/param.h>
84 #include <sys/conf.h>
85 #include <sys/malloc.h>
86 #include <sys/ktr.h>
87 #include <sys/proc.h>
88 #include <sys/user.h>
89 #include <sys/queue.h>
90 #include <sys/systm.h>
91 #include <sys/kernel.h>
92 #include <sys/kerneldump.h>
93 #include <sys/linker.h>
94 #include <sys/msgbuf.h>
95 #include <sys/lock.h>
96 #include <sys/mutex.h>
97 #include <sys/rwlock.h>
98 #include <sys/sched.h>
99 #include <sys/smp.h>
100 #include <sys/vmmeter.h>
101 
102 #include <vm/vm.h>
103 #include <vm/vm_page.h>
104 #include <vm/vm_kern.h>
105 #include <vm/vm_pageout.h>
106 #include <vm/vm_extern.h>
107 #include <vm/vm_object.h>
108 #include <vm/vm_param.h>
109 #include <vm/vm_map.h>
110 #include <vm/vm_pager.h>
111 #include <vm/vm_phys.h>
112 #include <vm/vm_pagequeue.h>
113 #include <vm/uma.h>
114 
115 #include <machine/_inttypes.h>
116 #include <machine/cpu.h>
117 #include <machine/pcb.h>
118 #include <machine/platform.h>
119 
120 #include <machine/tlb.h>
121 #include <machine/spr.h>
122 #include <machine/md_var.h>
123 #include <machine/mmuvar.h>
124 #include <machine/pmap.h>
125 #include <machine/pte.h>
126 
127 #include <ddb/ddb.h>
128 
129 #include "mmu_if.h"
130 
131 #define	SPARSE_MAPDEV
132 #ifdef  DEBUG
133 #define debugf(fmt, args...) printf(fmt, ##args)
134 #else
135 #define debugf(fmt, args...)
136 #endif
137 
138 #ifdef __powerpc64__
139 #define	PRI0ptrX	"016lx"
140 #else
141 #define	PRI0ptrX	"08x"
142 #endif
143 
144 #define TODO			panic("%s: not implemented", __func__);
145 
146 extern unsigned char _etext[];
147 extern unsigned char _end[];
148 
149 extern uint32_t *bootinfo;
150 
151 vm_paddr_t kernload;
152 vm_offset_t kernstart;
153 vm_size_t kernsize;
154 
155 /* Message buffer and tables. */
156 static vm_offset_t data_start;
157 static vm_size_t data_end;
158 
159 /* Phys/avail memory regions. */
160 static struct mem_region *availmem_regions;
161 static int availmem_regions_sz;
162 static struct mem_region *physmem_regions;
163 static int physmem_regions_sz;
164 
165 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
166 static vm_offset_t zero_page_va;
167 static struct mtx zero_page_mutex;
168 
169 static struct mtx tlbivax_mutex;
170 
171 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
172 static vm_offset_t copy_page_src_va;
173 static vm_offset_t copy_page_dst_va;
174 static struct mtx copy_page_mutex;
175 
176 /**************************************************************************/
177 /* PMAP */
178 /**************************************************************************/
179 
180 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
181     vm_prot_t, u_int flags, int8_t psind);
182 
183 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
184 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
185 #ifdef __powerpc64__
186 unsigned int kernel_pdirs;
187 #endif
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 static void ptbl_init(void);
266 static struct ptbl_buf *ptbl_buf_alloc(void);
267 static void ptbl_buf_free(struct ptbl_buf *);
268 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
269 
270 #ifdef __powerpc64__
271 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
272 			 unsigned int, boolean_t);
273 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int);
274 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
275 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
276 #else
277 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
278 static void ptbl_free(mmu_t, pmap_t, unsigned int);
279 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
280 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
281 #endif
282 
283 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
284 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
285 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
286 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
287 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
288 
289 static pv_entry_t pv_alloc(void);
290 static void pv_free(pv_entry_t);
291 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
292 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
293 
294 static void booke_pmap_init_qpages(void);
295 
296 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
297 #ifdef __powerpc64__
298 #define PTBL_BUFS               (16UL * 16 * 16)
299 #else
300 #define PTBL_BUFS		(128 * 16)
301 #endif
302 
303 struct ptbl_buf {
304 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
305 	vm_offset_t kva;		/* va of mapping */
306 };
307 
308 /* ptbl free list and a lock used for access synchronization. */
309 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
310 static struct mtx ptbl_buf_freelist_lock;
311 
312 /* Base address of kva space allocated fot ptbl bufs. */
313 static vm_offset_t ptbl_buf_pool_vabase;
314 
315 /* Pointer to ptbl_buf structures. */
316 static struct ptbl_buf *ptbl_bufs;
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 	int		i;
578 
579 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
580 	TAILQ_INIT(&ptbl_buf_freelist);
581 
582 	for (i = 0; i < PTBL_BUFS; i++) {
583 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase +
584 		    i * MAX(PTBL_PAGES,PDIR_PAGES) * PAGE_SIZE;
585 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
586 	}
587 }
588 
589 /* Get an sf_buf from the freelist. */
590 static struct ptbl_buf *
591 ptbl_buf_alloc(void)
592 {
593 	struct ptbl_buf *buf;
594 
595 	mtx_lock(&ptbl_buf_freelist_lock);
596 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
597 	if (buf != NULL)
598 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
599 	mtx_unlock(&ptbl_buf_freelist_lock);
600 
601 	return (buf);
602 }
603 
604 /* Return ptbl buff to free pool. */
605 static void
606 ptbl_buf_free(struct ptbl_buf *buf)
607 {
608 	mtx_lock(&ptbl_buf_freelist_lock);
609 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
610 	mtx_unlock(&ptbl_buf_freelist_lock);
611 }
612 
613 /*
614  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
615  */
616 static void
617 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t * ptbl)
618 {
619 	struct ptbl_buf *pbuf;
620 
621 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) {
622 		if (pbuf->kva == (vm_offset_t) ptbl) {
623 			/* Remove from pmap ptbl buf list. */
624 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
625 
626 			/* Free corresponding ptbl buf. */
627 			ptbl_buf_free(pbuf);
628 
629 			break;
630 		}
631 	}
632 }
633 
634 /* Get a pointer to a PTE in a page table. */
635 static __inline pte_t *
636 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
637 {
638 	pte_t         **pdir;
639 	pte_t          *ptbl;
640 
641 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
642 
643 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
644 	if (!pdir)
645 		return NULL;
646 	ptbl = pdir[PDIR_IDX(va)];
647 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
648 }
649 
650 /*
651  * Search the list of allocated pdir bufs and find on list of allocated pdirs
652  */
653 static void
654 ptbl_free_pmap_pdir(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
655 {
656 	struct ptbl_buf *pbuf;
657 
658 	TAILQ_FOREACH(pbuf, &pmap->pm_pdir_list, link) {
659 		if (pbuf->kva == (vm_offset_t) pdir) {
660 			/* Remove from pmap ptbl buf list. */
661 			TAILQ_REMOVE(&pmap->pm_pdir_list, pbuf, link);
662 
663 			/* Free corresponding pdir buf. */
664 			ptbl_buf_free(pbuf);
665 
666 			break;
667 		}
668 	}
669 }
670 /* Free pdir pages and invalidate pdir entry. */
671 static void
672 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
673 {
674 	pte_t         **pdir;
675 	vm_paddr_t	pa;
676 	vm_offset_t	va;
677 	vm_page_t	m;
678 	int		i;
679 
680 	pdir = pmap->pm_pp2d[pp2d_idx];
681 
682 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
683 
684 	pmap->pm_pp2d[pp2d_idx] = NULL;
685 
686 	for (i = 0; i < PDIR_PAGES; i++) {
687 		va = ((vm_offset_t) pdir + (i * PAGE_SIZE));
688 		pa = pte_vatopa(mmu, kernel_pmap, va);
689 		m = PHYS_TO_VM_PAGE(pa);
690 		vm_page_free_zero(m);
691 		vm_wire_sub(1);
692 		pmap_kremove(va);
693 	}
694 
695 	ptbl_free_pmap_pdir(mmu, pmap, pdir);
696 }
697 
698 /*
699  * Decrement pdir pages hold count and attempt to free pdir pages. Called
700  * when removing directory entry from pdir.
701  *
702  * Return 1 if pdir pages were freed.
703  */
704 static int
705 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
706 {
707 	pte_t         **pdir;
708 	vm_paddr_t	pa;
709 	vm_page_t	m;
710 	int		i;
711 
712 	KASSERT((pmap != kernel_pmap),
713 		("pdir_unhold: unholding kernel pdir!"));
714 
715 	pdir = pmap->pm_pp2d[pp2d_idx];
716 
717 	KASSERT(((vm_offset_t) pdir >= VM_MIN_KERNEL_ADDRESS),
718 	    ("pdir_unhold: non kva pdir"));
719 
720 	/* decrement hold count */
721 	for (i = 0; i < PDIR_PAGES; i++) {
722 		pa = pte_vatopa(mmu, kernel_pmap,
723 		    (vm_offset_t) pdir + (i * PAGE_SIZE));
724 		m = PHYS_TO_VM_PAGE(pa);
725 		m->wire_count--;
726 	}
727 
728 	/*
729 	 * Free pdir pages if there are no dir entries in this pdir.
730 	 * wire_count has the same value for all ptbl pages, so check the
731 	 * last page.
732 	 */
733 	if (m->wire_count == 0) {
734 		pdir_free(mmu, pmap, pp2d_idx);
735 		return (1);
736 	}
737 	return (0);
738 }
739 
740 /*
741  * Increment hold count for pdir pages. This routine is used when new ptlb
742  * entry is being inserted into pdir.
743  */
744 static void
745 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
746 {
747 	vm_paddr_t	pa;
748 	vm_page_t	m;
749 	int		i;
750 
751 	KASSERT((pmap != kernel_pmap),
752 		("pdir_hold: holding kernel pdir!"));
753 
754 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
755 
756 	for (i = 0; i < PDIR_PAGES; i++) {
757 		pa = pte_vatopa(mmu, kernel_pmap,
758 				(vm_offset_t) pdir + (i * PAGE_SIZE));
759 		m = PHYS_TO_VM_PAGE(pa);
760 		m->wire_count++;
761 	}
762 }
763 
764 /* Allocate page table. */
765 static pte_t   *
766 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
767     boolean_t nosleep)
768 {
769 	vm_page_t	mtbl  [PTBL_PAGES];
770 	vm_page_t	m;
771 	struct ptbl_buf *pbuf;
772 	unsigned int	pidx;
773 	pte_t          *ptbl;
774 	int		i, j;
775 	int		req;
776 
777 	KASSERT((pdir[pdir_idx] == NULL),
778 		("%s: valid ptbl entry exists!", __func__));
779 
780 	pbuf = ptbl_buf_alloc();
781 	if (pbuf == NULL)
782 		panic("%s: couldn't alloc kernel virtual memory", __func__);
783 
784 	ptbl = (pte_t *) pbuf->kva;
785 
786 	for (i = 0; i < PTBL_PAGES; i++) {
787 		pidx = (PTBL_PAGES * pdir_idx) + i;
788 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
789 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
790 			PMAP_UNLOCK(pmap);
791 			rw_wunlock(&pvh_global_lock);
792 			if (nosleep) {
793 				ptbl_free_pmap_ptbl(pmap, ptbl);
794 				for (j = 0; j < i; j++)
795 					vm_page_free(mtbl[j]);
796 				vm_wire_sub(i);
797 				return (NULL);
798 			}
799 			vm_wait(NULL);
800 			rw_wlock(&pvh_global_lock);
801 			PMAP_LOCK(pmap);
802 		}
803 		mtbl[i] = m;
804 	}
805 
806 	/* Mapin allocated pages into kernel_pmap. */
807 	mmu_booke_qenter(mmu, (vm_offset_t) ptbl, mtbl, PTBL_PAGES);
808 	/* Zero whole ptbl. */
809 	bzero((caddr_t) ptbl, PTBL_PAGES * PAGE_SIZE);
810 
811 	/* Add pbuf to the pmap ptbl bufs list. */
812 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
813 
814 	return (ptbl);
815 }
816 
817 /* Free ptbl pages and invalidate pdir entry. */
818 static void
819 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
820 {
821 	pte_t          *ptbl;
822 	vm_paddr_t	pa;
823 	vm_offset_t	va;
824 	vm_page_t	m;
825 	int		i;
826 
827 	ptbl = pdir[pdir_idx];
828 
829 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
830 
831 	pdir[pdir_idx] = NULL;
832 
833 	for (i = 0; i < PTBL_PAGES; i++) {
834 		va = ((vm_offset_t) ptbl + (i * PAGE_SIZE));
835 		pa = pte_vatopa(mmu, kernel_pmap, va);
836 		m = PHYS_TO_VM_PAGE(pa);
837 		vm_page_free_zero(m);
838 		vm_wire_sub(1);
839 		pmap_kremove(va);
840 	}
841 
842 	ptbl_free_pmap_ptbl(pmap, ptbl);
843 }
844 
845 /*
846  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
847  * when removing pte entry from ptbl.
848  *
849  * Return 1 if ptbl pages were freed.
850  */
851 static int
852 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
853 {
854 	pte_t          *ptbl;
855 	vm_paddr_t	pa;
856 	vm_page_t	m;
857 	u_int		pp2d_idx;
858 	pte_t         **pdir;
859 	u_int		pdir_idx;
860 	int		i;
861 
862 	pp2d_idx = PP2D_IDX(va);
863 	pdir_idx = PDIR_IDX(va);
864 
865 	KASSERT((pmap != kernel_pmap),
866 		("ptbl_unhold: unholding kernel ptbl!"));
867 
868 	pdir = pmap->pm_pp2d[pp2d_idx];
869 	ptbl = pdir[pdir_idx];
870 
871 	KASSERT(((vm_offset_t) ptbl >= VM_MIN_KERNEL_ADDRESS),
872 	    ("ptbl_unhold: non kva ptbl"));
873 
874 	/* decrement hold count */
875 	for (i = 0; i < PTBL_PAGES; i++) {
876 		pa = pte_vatopa(mmu, kernel_pmap,
877 		    (vm_offset_t) ptbl + (i * PAGE_SIZE));
878 		m = PHYS_TO_VM_PAGE(pa);
879 		m->wire_count--;
880 	}
881 
882 	/*
883 	 * Free ptbl pages if there are no pte entries in this ptbl.
884 	 * wire_count has the same value for all ptbl pages, so check the
885 	 * last page.
886 	 */
887 	if (m->wire_count == 0) {
888 		/* A pair of indirect entries might point to this ptbl page */
889 #if 0
890 		tlb_flush_entry(pmap, va & ~((2UL * PAGE_SIZE_1M) - 1),
891 				TLB_SIZE_1M, MAS6_SIND);
892 		tlb_flush_entry(pmap, (va & ~((2UL * PAGE_SIZE_1M) - 1)) | PAGE_SIZE_1M,
893 				TLB_SIZE_1M, MAS6_SIND);
894 #endif
895 		ptbl_free(mmu, pmap, pdir, pdir_idx);
896 		pdir_unhold(mmu, pmap, pp2d_idx);
897 		return (1);
898 	}
899 	return (0);
900 }
901 
902 /*
903  * Increment hold count for ptbl pages. This routine is used when new pte
904  * entry is being inserted into ptbl.
905  */
906 static void
907 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
908 {
909 	vm_paddr_t	pa;
910 	pte_t          *ptbl;
911 	vm_page_t	m;
912 	int		i;
913 
914 	KASSERT((pmap != kernel_pmap),
915 		("ptbl_hold: holding kernel ptbl!"));
916 
917 	ptbl = pdir[pdir_idx];
918 
919 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
920 
921 	for (i = 0; i < PTBL_PAGES; i++) {
922 		pa = pte_vatopa(mmu, kernel_pmap,
923 				(vm_offset_t) ptbl + (i * PAGE_SIZE));
924 		m = PHYS_TO_VM_PAGE(pa);
925 		m->wire_count++;
926 	}
927 }
928 #else
929 
930 /* Initialize pool of kva ptbl buffers. */
931 static void
932 ptbl_init(void)
933 {
934 	int i;
935 
936 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
937 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
938 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
939 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
940 
941 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
942 	TAILQ_INIT(&ptbl_buf_freelist);
943 
944 	for (i = 0; i < PTBL_BUFS; i++) {
945 		ptbl_bufs[i].kva =
946 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
947 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
948 	}
949 }
950 
951 /* Get a ptbl_buf from the freelist. */
952 static struct ptbl_buf *
953 ptbl_buf_alloc(void)
954 {
955 	struct ptbl_buf *buf;
956 
957 	mtx_lock(&ptbl_buf_freelist_lock);
958 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
959 	if (buf != NULL)
960 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
961 	mtx_unlock(&ptbl_buf_freelist_lock);
962 
963 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
964 
965 	return (buf);
966 }
967 
968 /* Return ptbl buff to free pool. */
969 static void
970 ptbl_buf_free(struct ptbl_buf *buf)
971 {
972 
973 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
974 
975 	mtx_lock(&ptbl_buf_freelist_lock);
976 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
977 	mtx_unlock(&ptbl_buf_freelist_lock);
978 }
979 
980 /*
981  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
982  */
983 static void
984 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
985 {
986 	struct ptbl_buf *pbuf;
987 
988 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
989 
990 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
991 
992 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
993 		if (pbuf->kva == (vm_offset_t)ptbl) {
994 			/* Remove from pmap ptbl buf list. */
995 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
996 
997 			/* Free corresponding ptbl buf. */
998 			ptbl_buf_free(pbuf);
999 			break;
1000 		}
1001 }
1002 
1003 /* Allocate page table. */
1004 static pte_t *
1005 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
1006 {
1007 	vm_page_t mtbl[PTBL_PAGES];
1008 	vm_page_t m;
1009 	struct ptbl_buf *pbuf;
1010 	unsigned int pidx;
1011 	pte_t *ptbl;
1012 	int i, j;
1013 
1014 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1015 	    (pmap == kernel_pmap), pdir_idx);
1016 
1017 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1018 	    ("ptbl_alloc: invalid pdir_idx"));
1019 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
1020 	    ("pte_alloc: valid ptbl entry exists!"));
1021 
1022 	pbuf = ptbl_buf_alloc();
1023 	if (pbuf == NULL)
1024 		panic("pte_alloc: couldn't alloc kernel virtual memory");
1025 
1026 	ptbl = (pte_t *)pbuf->kva;
1027 
1028 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
1029 
1030 	for (i = 0; i < PTBL_PAGES; i++) {
1031 		pidx = (PTBL_PAGES * pdir_idx) + i;
1032 		while ((m = vm_page_alloc(NULL, pidx,
1033 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
1034 			PMAP_UNLOCK(pmap);
1035 			rw_wunlock(&pvh_global_lock);
1036 			if (nosleep) {
1037 				ptbl_free_pmap_ptbl(pmap, ptbl);
1038 				for (j = 0; j < i; j++)
1039 					vm_page_free(mtbl[j]);
1040 				vm_wire_sub(i);
1041 				return (NULL);
1042 			}
1043 			vm_wait(NULL);
1044 			rw_wlock(&pvh_global_lock);
1045 			PMAP_LOCK(pmap);
1046 		}
1047 		mtbl[i] = m;
1048 	}
1049 
1050 	/* Map allocated pages into kernel_pmap. */
1051 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
1052 
1053 	/* Zero whole ptbl. */
1054 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
1055 
1056 	/* Add pbuf to the pmap ptbl bufs list. */
1057 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
1058 
1059 	return (ptbl);
1060 }
1061 
1062 /* Free ptbl pages and invalidate pdir entry. */
1063 static void
1064 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1065 {
1066 	pte_t *ptbl;
1067 	vm_paddr_t pa;
1068 	vm_offset_t va;
1069 	vm_page_t m;
1070 	int i;
1071 
1072 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1073 	    (pmap == kernel_pmap), pdir_idx);
1074 
1075 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1076 	    ("ptbl_free: invalid pdir_idx"));
1077 
1078 	ptbl = pmap->pm_pdir[pdir_idx];
1079 
1080 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
1081 
1082 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
1083 
1084 	/*
1085 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
1086 	 * don't attempt to look up the page tables we are releasing.
1087 	 */
1088 	mtx_lock_spin(&tlbivax_mutex);
1089 	tlb_miss_lock();
1090 
1091 	pmap->pm_pdir[pdir_idx] = NULL;
1092 
1093 	tlb_miss_unlock();
1094 	mtx_unlock_spin(&tlbivax_mutex);
1095 
1096 	for (i = 0; i < PTBL_PAGES; i++) {
1097 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
1098 		pa = pte_vatopa(mmu, kernel_pmap, va);
1099 		m = PHYS_TO_VM_PAGE(pa);
1100 		vm_page_free_zero(m);
1101 		vm_wire_sub(1);
1102 		mmu_booke_kremove(mmu, va);
1103 	}
1104 
1105 	ptbl_free_pmap_ptbl(pmap, ptbl);
1106 }
1107 
1108 /*
1109  * Decrement ptbl pages hold count and attempt to free ptbl pages.
1110  * Called when removing pte entry from ptbl.
1111  *
1112  * Return 1 if ptbl pages were freed.
1113  */
1114 static int
1115 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1116 {
1117 	pte_t *ptbl;
1118 	vm_paddr_t pa;
1119 	vm_page_t m;
1120 	int i;
1121 
1122 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1123 	    (pmap == kernel_pmap), pdir_idx);
1124 
1125 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1126 	    ("ptbl_unhold: invalid pdir_idx"));
1127 	KASSERT((pmap != kernel_pmap),
1128 	    ("ptbl_unhold: unholding kernel ptbl!"));
1129 
1130 	ptbl = pmap->pm_pdir[pdir_idx];
1131 
1132 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1133 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1134 	    ("ptbl_unhold: non kva ptbl"));
1135 
1136 	/* decrement hold count */
1137 	for (i = 0; i < PTBL_PAGES; i++) {
1138 		pa = pte_vatopa(mmu, kernel_pmap,
1139 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1140 		m = PHYS_TO_VM_PAGE(pa);
1141 		m->wire_count--;
1142 	}
1143 
1144 	/*
1145 	 * Free ptbl pages if there are no pte etries in this ptbl.
1146 	 * wire_count has the same value for all ptbl pages, so check the last
1147 	 * page.
1148 	 */
1149 	if (m->wire_count == 0) {
1150 		ptbl_free(mmu, pmap, pdir_idx);
1151 
1152 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1153 		return (1);
1154 	}
1155 
1156 	return (0);
1157 }
1158 
1159 /*
1160  * Increment hold count for ptbl pages. This routine is used when a new pte
1161  * entry is being inserted into the ptbl.
1162  */
1163 static void
1164 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1165 {
1166 	vm_paddr_t pa;
1167 	pte_t *ptbl;
1168 	vm_page_t m;
1169 	int i;
1170 
1171 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1172 	    pdir_idx);
1173 
1174 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1175 	    ("ptbl_hold: invalid pdir_idx"));
1176 	KASSERT((pmap != kernel_pmap),
1177 	    ("ptbl_hold: holding kernel ptbl!"));
1178 
1179 	ptbl = pmap->pm_pdir[pdir_idx];
1180 
1181 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1182 
1183 	for (i = 0; i < PTBL_PAGES; i++) {
1184 		pa = pte_vatopa(mmu, kernel_pmap,
1185 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1186 		m = PHYS_TO_VM_PAGE(pa);
1187 		m->wire_count++;
1188 	}
1189 }
1190 #endif
1191 
1192 /* Allocate pv_entry structure. */
1193 pv_entry_t
1194 pv_alloc(void)
1195 {
1196 	pv_entry_t pv;
1197 
1198 	pv_entry_count++;
1199 	if (pv_entry_count > pv_entry_high_water)
1200 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1201 	pv = uma_zalloc(pvzone, M_NOWAIT);
1202 
1203 	return (pv);
1204 }
1205 
1206 /* Free pv_entry structure. */
1207 static __inline void
1208 pv_free(pv_entry_t pve)
1209 {
1210 
1211 	pv_entry_count--;
1212 	uma_zfree(pvzone, pve);
1213 }
1214 
1215 
1216 /* Allocate and initialize pv_entry structure. */
1217 static void
1218 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1219 {
1220 	pv_entry_t pve;
1221 
1222 	//int su = (pmap == kernel_pmap);
1223 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1224 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1225 
1226 	pve = pv_alloc();
1227 	if (pve == NULL)
1228 		panic("pv_insert: no pv entries!");
1229 
1230 	pve->pv_pmap = pmap;
1231 	pve->pv_va = va;
1232 
1233 	/* add to pv_list */
1234 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1235 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1236 
1237 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1238 
1239 	//debugf("pv_insert: e\n");
1240 }
1241 
1242 /* Destroy pv entry. */
1243 static void
1244 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1245 {
1246 	pv_entry_t pve;
1247 
1248 	//int su = (pmap == kernel_pmap);
1249 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1250 
1251 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1252 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1253 
1254 	/* find pv entry */
1255 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1256 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1257 			/* remove from pv_list */
1258 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1259 			if (TAILQ_EMPTY(&m->md.pv_list))
1260 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1261 
1262 			/* free pv entry struct */
1263 			pv_free(pve);
1264 			break;
1265 		}
1266 	}
1267 
1268 	//debugf("pv_remove: e\n");
1269 }
1270 
1271 #ifdef __powerpc64__
1272 /*
1273  * Clean pte entry, try to free page table page if requested.
1274  *
1275  * Return 1 if ptbl pages were freed, otherwise return 0.
1276  */
1277 static int
1278 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1279 {
1280 	vm_page_t	m;
1281 	pte_t          *pte;
1282 
1283 	pte = pte_find(mmu, pmap, va);
1284 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1285 
1286 	if (!PTE_ISVALID(pte))
1287 		return (0);
1288 
1289 	/* Get vm_page_t for mapped pte. */
1290 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1291 
1292 	if (PTE_ISWIRED(pte))
1293 		pmap->pm_stats.wired_count--;
1294 
1295 	/* Handle managed entry. */
1296 	if (PTE_ISMANAGED(pte)) {
1297 
1298 		/* Handle modified pages. */
1299 		if (PTE_ISMODIFIED(pte))
1300 			vm_page_dirty(m);
1301 
1302 		/* Referenced pages. */
1303 		if (PTE_ISREFERENCED(pte))
1304 			vm_page_aflag_set(m, PGA_REFERENCED);
1305 
1306 		/* Remove pv_entry from pv_list. */
1307 		pv_remove(pmap, va, m);
1308 	} else if (m->md.pv_tracked) {
1309 		pv_remove(pmap, va, m);
1310 		if (TAILQ_EMPTY(&m->md.pv_list))
1311 			m->md.pv_tracked = false;
1312 	}
1313 	mtx_lock_spin(&tlbivax_mutex);
1314 	tlb_miss_lock();
1315 
1316 	tlb0_flush_entry(va);
1317 	*pte = 0;
1318 
1319 	tlb_miss_unlock();
1320 	mtx_unlock_spin(&tlbivax_mutex);
1321 
1322 	pmap->pm_stats.resident_count--;
1323 
1324 	if (flags & PTBL_UNHOLD) {
1325 		return (ptbl_unhold(mmu, pmap, va));
1326 	}
1327 	return (0);
1328 }
1329 
1330 /*
1331  * allocate a page of pointers to page directories, do not preallocate the
1332  * page tables
1333  */
1334 static pte_t  **
1335 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
1336 {
1337 	vm_page_t	mtbl  [PDIR_PAGES];
1338 	vm_page_t	m;
1339 	struct ptbl_buf *pbuf;
1340 	pte_t         **pdir;
1341 	unsigned int	pidx;
1342 	int		i;
1343 	int		req;
1344 
1345 	pbuf = ptbl_buf_alloc();
1346 
1347 	if (pbuf == NULL)
1348 		panic("%s: couldn't alloc kernel virtual memory", __func__);
1349 
1350 	/* Allocate pdir pages, this will sleep! */
1351 	for (i = 0; i < PDIR_PAGES; i++) {
1352 		pidx = (PDIR_PAGES * pp2d_idx) + i;
1353 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
1354 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
1355 			PMAP_UNLOCK(pmap);
1356 			vm_wait(NULL);
1357 			PMAP_LOCK(pmap);
1358 		}
1359 		mtbl[i] = m;
1360 	}
1361 
1362 	/* Mapin allocated pages into kernel_pmap. */
1363 	pdir = (pte_t **) pbuf->kva;
1364 	pmap_qenter((vm_offset_t) pdir, mtbl, PDIR_PAGES);
1365 
1366 	/* Zero whole pdir. */
1367 	bzero((caddr_t) pdir, PDIR_PAGES * PAGE_SIZE);
1368 
1369 	/* Add pdir to the pmap pdir bufs list. */
1370 	TAILQ_INSERT_TAIL(&pmap->pm_pdir_list, pbuf, link);
1371 
1372 	return pdir;
1373 }
1374 
1375 /*
1376  * Insert PTE for a given page and virtual address.
1377  */
1378 static int
1379 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1380     boolean_t nosleep)
1381 {
1382 	unsigned int	pp2d_idx = PP2D_IDX(va);
1383 	unsigned int	pdir_idx = PDIR_IDX(va);
1384 	unsigned int	ptbl_idx = PTBL_IDX(va);
1385 	pte_t          *ptbl, *pte;
1386 	pte_t         **pdir;
1387 
1388 	/* Get the page directory pointer. */
1389 	pdir = pmap->pm_pp2d[pp2d_idx];
1390 	if (pdir == NULL)
1391 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1392 
1393 	/* Get the page table pointer. */
1394 	ptbl = pdir[pdir_idx];
1395 
1396 	if (ptbl == NULL) {
1397 		/* Allocate page table pages. */
1398 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1399 		if (ptbl == NULL) {
1400 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1401 			return (ENOMEM);
1402 		}
1403 	} else {
1404 		/*
1405 		 * Check if there is valid mapping for requested va, if there
1406 		 * is, remove it.
1407 		 */
1408 		pte = &pdir[pdir_idx][ptbl_idx];
1409 		if (PTE_ISVALID(pte)) {
1410 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1411 		} else {
1412 			/*
1413 			 * pte is not used, increment hold count for ptbl
1414 			 * pages.
1415 			 */
1416 			if (pmap != kernel_pmap)
1417 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1418 		}
1419 	}
1420 
1421 	if (pdir[pdir_idx] == NULL) {
1422 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1423 			pdir_hold(mmu, pmap, pdir);
1424 		pdir[pdir_idx] = ptbl;
1425 	}
1426 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1427 		pmap->pm_pp2d[pp2d_idx] = pdir;
1428 
1429 	/*
1430 	 * Insert pv_entry into pv_list for mapped page if part of managed
1431 	 * memory.
1432 	 */
1433 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1434 		flags |= PTE_MANAGED;
1435 
1436 		/* Create and insert pv entry. */
1437 		pv_insert(pmap, va, m);
1438 	}
1439 
1440 	mtx_lock_spin(&tlbivax_mutex);
1441 	tlb_miss_lock();
1442 
1443 	tlb0_flush_entry(va);
1444 	pmap->pm_stats.resident_count++;
1445 	pte = &pdir[pdir_idx][ptbl_idx];
1446 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1447 	*pte |= (PTE_VALID | flags);
1448 
1449 	tlb_miss_unlock();
1450 	mtx_unlock_spin(&tlbivax_mutex);
1451 
1452 	return (0);
1453 }
1454 
1455 /* Return the pa for the given pmap/va. */
1456 static	vm_paddr_t
1457 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1458 {
1459 	vm_paddr_t	pa = 0;
1460 	pte_t          *pte;
1461 
1462 	pte = pte_find(mmu, pmap, va);
1463 	if ((pte != NULL) && PTE_ISVALID(pte))
1464 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1465 	return (pa);
1466 }
1467 
1468 
1469 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1470 static void
1471 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1472 {
1473 	int		i, j;
1474 	vm_offset_t	va;
1475 	pte_t		*pte;
1476 
1477 	va = addr;
1478 	/* Initialize kernel pdir */
1479 	for (i = 0; i < kernel_pdirs; i++) {
1480 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1481 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1482 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1483 		    j < PDIR_NENTRIES; j++) {
1484 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1485 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE * PDIR_PAGES) +
1486 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE * PTBL_PAGES));
1487 		}
1488 	}
1489 
1490 	/*
1491 	 * Fill in PTEs covering kernel code and data. They are not required
1492 	 * for address translation, as this area is covered by static TLB1
1493 	 * entries, but for pte_vatopa() to work correctly with kernel area
1494 	 * addresses.
1495 	 */
1496 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1497 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1498 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1499 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1500 		    PTE_VALID | PTE_PS_4KB;
1501 	}
1502 }
1503 #else
1504 /*
1505  * Clean pte entry, try to free page table page if requested.
1506  *
1507  * Return 1 if ptbl pages were freed, otherwise return 0.
1508  */
1509 static int
1510 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1511 {
1512 	unsigned int pdir_idx = PDIR_IDX(va);
1513 	unsigned int ptbl_idx = PTBL_IDX(va);
1514 	vm_page_t m;
1515 	pte_t *ptbl;
1516 	pte_t *pte;
1517 
1518 	//int su = (pmap == kernel_pmap);
1519 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1520 	//		su, (u_int32_t)pmap, va, flags);
1521 
1522 	ptbl = pmap->pm_pdir[pdir_idx];
1523 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1524 
1525 	pte = &ptbl[ptbl_idx];
1526 
1527 	if (pte == NULL || !PTE_ISVALID(pte))
1528 		return (0);
1529 
1530 	if (PTE_ISWIRED(pte))
1531 		pmap->pm_stats.wired_count--;
1532 
1533 	/* Get vm_page_t for mapped pte. */
1534 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1535 
1536 	/* Handle managed entry. */
1537 	if (PTE_ISMANAGED(pte)) {
1538 
1539 		if (PTE_ISMODIFIED(pte))
1540 			vm_page_dirty(m);
1541 
1542 		if (PTE_ISREFERENCED(pte))
1543 			vm_page_aflag_set(m, PGA_REFERENCED);
1544 
1545 		pv_remove(pmap, va, m);
1546 	} else if (m->md.pv_tracked) {
1547 		/*
1548 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1549 		 * used.  This is needed by the NCSW support code for fast
1550 		 * VA<->PA translation.
1551 		 */
1552 		pv_remove(pmap, va, m);
1553 		if (TAILQ_EMPTY(&m->md.pv_list))
1554 			m->md.pv_tracked = false;
1555 	}
1556 
1557 	mtx_lock_spin(&tlbivax_mutex);
1558 	tlb_miss_lock();
1559 
1560 	tlb0_flush_entry(va);
1561 	*pte = 0;
1562 
1563 	tlb_miss_unlock();
1564 	mtx_unlock_spin(&tlbivax_mutex);
1565 
1566 	pmap->pm_stats.resident_count--;
1567 
1568 	if (flags & PTBL_UNHOLD) {
1569 		//debugf("pte_remove: e (unhold)\n");
1570 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1571 	}
1572 
1573 	//debugf("pte_remove: e\n");
1574 	return (0);
1575 }
1576 
1577 /*
1578  * Insert PTE for a given page and virtual address.
1579  */
1580 static int
1581 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1582     boolean_t nosleep)
1583 {
1584 	unsigned int pdir_idx = PDIR_IDX(va);
1585 	unsigned int ptbl_idx = PTBL_IDX(va);
1586 	pte_t *ptbl, *pte;
1587 
1588 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1589 	    pmap == kernel_pmap, pmap, va);
1590 
1591 	/* Get the page table pointer. */
1592 	ptbl = pmap->pm_pdir[pdir_idx];
1593 
1594 	if (ptbl == NULL) {
1595 		/* Allocate page table pages. */
1596 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1597 		if (ptbl == NULL) {
1598 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1599 			return (ENOMEM);
1600 		}
1601 	} else {
1602 		/*
1603 		 * Check if there is valid mapping for requested
1604 		 * va, if there is, remove it.
1605 		 */
1606 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1607 		if (PTE_ISVALID(pte)) {
1608 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1609 		} else {
1610 			/*
1611 			 * pte is not used, increment hold count
1612 			 * for ptbl pages.
1613 			 */
1614 			if (pmap != kernel_pmap)
1615 				ptbl_hold(mmu, pmap, pdir_idx);
1616 		}
1617 	}
1618 
1619 	/*
1620 	 * Insert pv_entry into pv_list for mapped page if part of managed
1621 	 * memory.
1622 	 */
1623 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1624 		flags |= PTE_MANAGED;
1625 
1626 		/* Create and insert pv entry. */
1627 		pv_insert(pmap, va, m);
1628 	}
1629 
1630 	pmap->pm_stats.resident_count++;
1631 
1632 	mtx_lock_spin(&tlbivax_mutex);
1633 	tlb_miss_lock();
1634 
1635 	tlb0_flush_entry(va);
1636 	if (pmap->pm_pdir[pdir_idx] == NULL) {
1637 		/*
1638 		 * If we just allocated a new page table, hook it in
1639 		 * the pdir.
1640 		 */
1641 		pmap->pm_pdir[pdir_idx] = ptbl;
1642 	}
1643 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
1644 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1645 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1646 
1647 	tlb_miss_unlock();
1648 	mtx_unlock_spin(&tlbivax_mutex);
1649 	return (0);
1650 }
1651 
1652 /* Return the pa for the given pmap/va. */
1653 static vm_paddr_t
1654 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1655 {
1656 	vm_paddr_t pa = 0;
1657 	pte_t *pte;
1658 
1659 	pte = pte_find(mmu, pmap, va);
1660 	if ((pte != NULL) && PTE_ISVALID(pte))
1661 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1662 	return (pa);
1663 }
1664 
1665 /* Get a pointer to a PTE in a page table. */
1666 static pte_t *
1667 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1668 {
1669 	unsigned int pdir_idx = PDIR_IDX(va);
1670 	unsigned int ptbl_idx = PTBL_IDX(va);
1671 
1672 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1673 
1674 	if (pmap->pm_pdir[pdir_idx])
1675 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1676 
1677 	return (NULL);
1678 }
1679 
1680 /* Set up kernel page tables. */
1681 static void
1682 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1683 {
1684 	int		i;
1685 	vm_offset_t	va;
1686 	pte_t		*pte;
1687 
1688 	/* Initialize kernel pdir */
1689 	for (i = 0; i < kernel_ptbls; i++)
1690 		kernel_pmap->pm_pdir[kptbl_min + i] =
1691 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1692 
1693 	/*
1694 	 * Fill in PTEs covering kernel code and data. They are not required
1695 	 * for address translation, as this area is covered by static TLB1
1696 	 * entries, but for pte_vatopa() to work correctly with kernel area
1697 	 * addresses.
1698 	 */
1699 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1700 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1701 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1702 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1703 		    PTE_VALID | PTE_PS_4KB;
1704 	}
1705 }
1706 #endif
1707 
1708 /**************************************************************************/
1709 /* PMAP related */
1710 /**************************************************************************/
1711 
1712 /*
1713  * This is called during booke_init, before the system is really initialized.
1714  */
1715 static void
1716 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1717 {
1718 	vm_paddr_t phys_kernelend;
1719 	struct mem_region *mp, *mp1;
1720 	int cnt, i, j;
1721 	vm_paddr_t s, e, sz;
1722 	vm_paddr_t physsz, hwphyssz;
1723 	u_int phys_avail_count;
1724 	vm_size_t kstack0_sz;
1725 	vm_offset_t kernel_pdir, kstack0;
1726 	vm_paddr_t kstack0_phys;
1727 	void *dpcpu;
1728 
1729 	debugf("mmu_booke_bootstrap: entered\n");
1730 
1731 	/* Set interesting system properties */
1732 #ifdef __powerpc64__
1733 	hw_direct_map = 1;
1734 #else
1735 	hw_direct_map = 0;
1736 #endif
1737 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1738 	elf32_nxstack = 1;
1739 #endif
1740 
1741 	/* Initialize invalidation mutex */
1742 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1743 
1744 	/* Read TLB0 size and associativity. */
1745 	tlb0_get_tlbconf();
1746 
1747 	/*
1748 	 * Align kernel start and end address (kernel image).
1749 	 * Note that kernel end does not necessarily relate to kernsize.
1750 	 * kernsize is the size of the kernel that is actually mapped.
1751 	 */
1752 	kernstart = trunc_page(start);
1753 	data_start = round_page(kernelend);
1754 	data_end = data_start;
1755 
1756 	/* Allocate the dynamic per-cpu area. */
1757 	dpcpu = (void *)data_end;
1758 	data_end += DPCPU_SIZE;
1759 
1760 	/* Allocate space for the message buffer. */
1761 	msgbufp = (struct msgbuf *)data_end;
1762 	data_end += msgbufsize;
1763 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1764 	    (uintptr_t)msgbufp, data_end);
1765 
1766 	data_end = round_page(data_end);
1767 
1768 	/* Allocate space for ptbl_bufs. */
1769 	ptbl_bufs = (struct ptbl_buf *)data_end;
1770 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1771 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1772 	    (uintptr_t)ptbl_bufs, data_end);
1773 
1774 	data_end = round_page(data_end);
1775 
1776 	/* Allocate PTE tables for kernel KVA. */
1777 	kernel_pdir = data_end;
1778 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1779 	    PDIR_SIZE);
1780 #ifdef __powerpc64__
1781 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1782 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1783 #endif
1784 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1785 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1786 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1787 	    kernel_pdir, data_end);
1788 
1789 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1790 	if (data_end - kernstart > kernsize) {
1791 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1792 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1793 	}
1794 	data_end = kernstart + kernsize;
1795 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1796 
1797 	/*
1798 	 * Clear the structures - note we can only do it safely after the
1799 	 * possible additional TLB1 translations are in place (above) so that
1800 	 * all range up to the currently calculated 'data_end' is covered.
1801 	 */
1802 	dpcpu_init(dpcpu, 0);
1803 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1804 #ifdef __powerpc64__
1805 	memset((void *)kernel_pdir, 0,
1806 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1807 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1808 #else
1809 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1810 #endif
1811 
1812 	/*******************************************************/
1813 	/* Set the start and end of kva. */
1814 	/*******************************************************/
1815 	virtual_avail = round_page(data_end);
1816 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1817 
1818 	/* Allocate KVA space for page zero/copy operations. */
1819 	zero_page_va = virtual_avail;
1820 	virtual_avail += PAGE_SIZE;
1821 	copy_page_src_va = virtual_avail;
1822 	virtual_avail += PAGE_SIZE;
1823 	copy_page_dst_va = virtual_avail;
1824 	virtual_avail += PAGE_SIZE;
1825 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
1826 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
1827 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
1828 
1829 	/* Initialize page zero/copy mutexes. */
1830 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1831 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1832 
1833 	/* Allocate KVA space for ptbl bufs. */
1834 	ptbl_buf_pool_vabase = virtual_avail;
1835 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1836 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1837 	    ptbl_buf_pool_vabase, virtual_avail);
1838 
1839 	/* Calculate corresponding physical addresses for the kernel region. */
1840 	phys_kernelend = kernload + kernsize;
1841 	debugf("kernel image and allocated data:\n");
1842 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1843 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
1844 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
1845 
1846 	/*
1847 	 * Remove kernel physical address range from avail regions list. Page
1848 	 * align all regions.  Non-page aligned memory isn't very interesting
1849 	 * to us.  Also, sort the entries for ascending addresses.
1850 	 */
1851 
1852 	/* Retrieve phys/avail mem regions */
1853 	mem_regions(&physmem_regions, &physmem_regions_sz,
1854 	    &availmem_regions, &availmem_regions_sz);
1855 
1856 	if (nitems(phys_avail) < availmem_regions_sz)
1857 		panic("mmu_booke_bootstrap: phys_avail too small");
1858 
1859 	sz = 0;
1860 	cnt = availmem_regions_sz;
1861 	debugf("processing avail regions:\n");
1862 	for (mp = availmem_regions; mp->mr_size; mp++) {
1863 		s = mp->mr_start;
1864 		e = mp->mr_start + mp->mr_size;
1865 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1866 		/* Check whether this region holds all of the kernel. */
1867 		if (s < kernload && e > phys_kernelend) {
1868 			availmem_regions[cnt].mr_start = phys_kernelend;
1869 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1870 			e = kernload;
1871 		}
1872 		/* Look whether this regions starts within the kernel. */
1873 		if (s >= kernload && s < phys_kernelend) {
1874 			if (e <= phys_kernelend)
1875 				goto empty;
1876 			s = phys_kernelend;
1877 		}
1878 		/* Now look whether this region ends within the kernel. */
1879 		if (e > kernload && e <= phys_kernelend) {
1880 			if (s >= kernload)
1881 				goto empty;
1882 			e = kernload;
1883 		}
1884 		/* Now page align the start and size of the region. */
1885 		s = round_page(s);
1886 		e = trunc_page(e);
1887 		if (e < s)
1888 			e = s;
1889 		sz = e - s;
1890 		debugf("%09jx-%09jx = %jx\n",
1891 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1892 
1893 		/* Check whether some memory is left here. */
1894 		if (sz == 0) {
1895 		empty:
1896 			memmove(mp, mp + 1,
1897 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1898 			cnt--;
1899 			mp--;
1900 			continue;
1901 		}
1902 
1903 		/* Do an insertion sort. */
1904 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1905 			if (s < mp1->mr_start)
1906 				break;
1907 		if (mp1 < mp) {
1908 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1909 			mp1->mr_start = s;
1910 			mp1->mr_size = sz;
1911 		} else {
1912 			mp->mr_start = s;
1913 			mp->mr_size = sz;
1914 		}
1915 	}
1916 	availmem_regions_sz = cnt;
1917 
1918 	/*******************************************************/
1919 	/* Steal physical memory for kernel stack from the end */
1920 	/* of the first avail region                           */
1921 	/*******************************************************/
1922 	kstack0_sz = kstack_pages * PAGE_SIZE;
1923 	kstack0_phys = availmem_regions[0].mr_start +
1924 	    availmem_regions[0].mr_size;
1925 	kstack0_phys -= kstack0_sz;
1926 	availmem_regions[0].mr_size -= kstack0_sz;
1927 
1928 	/*******************************************************/
1929 	/* Fill in phys_avail table, based on availmem_regions */
1930 	/*******************************************************/
1931 	phys_avail_count = 0;
1932 	physsz = 0;
1933 	hwphyssz = 0;
1934 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1935 
1936 	debugf("fill in phys_avail:\n");
1937 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1938 
1939 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1940 		    (uintmax_t)availmem_regions[i].mr_start,
1941 		    (uintmax_t)availmem_regions[i].mr_start +
1942 		        availmem_regions[i].mr_size,
1943 		    (uintmax_t)availmem_regions[i].mr_size);
1944 
1945 		if (hwphyssz != 0 &&
1946 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1947 			debugf(" hw.physmem adjust\n");
1948 			if (physsz < hwphyssz) {
1949 				phys_avail[j] = availmem_regions[i].mr_start;
1950 				phys_avail[j + 1] =
1951 				    availmem_regions[i].mr_start +
1952 				    hwphyssz - physsz;
1953 				physsz = hwphyssz;
1954 				phys_avail_count++;
1955 			}
1956 			break;
1957 		}
1958 
1959 		phys_avail[j] = availmem_regions[i].mr_start;
1960 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1961 		    availmem_regions[i].mr_size;
1962 		phys_avail_count++;
1963 		physsz += availmem_regions[i].mr_size;
1964 	}
1965 	physmem = btoc(physsz);
1966 
1967 	/* Calculate the last available physical address. */
1968 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1969 		;
1970 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1971 
1972 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1973 	debugf("phys_avail_count = %d\n", phys_avail_count);
1974 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1975 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1976 
1977 #ifdef __powerpc64__
1978 	/*
1979 	 * Map the physical memory contiguously in TLB1.
1980 	 * Round so it fits into a single mapping.
1981 	 */
1982 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
1983 	    phys_avail[i + 1]);
1984 #endif
1985 
1986 	/*******************************************************/
1987 	/* Initialize (statically allocated) kernel pmap. */
1988 	/*******************************************************/
1989 	PMAP_LOCK_INIT(kernel_pmap);
1990 #ifndef __powerpc64__
1991 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1992 #endif
1993 
1994 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1995 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1996 	for (i = 0; i < MAXCPU; i++) {
1997 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1998 
1999 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
2000 		tidbusy[i][TID_KERNEL] = kernel_pmap;
2001 	}
2002 
2003 	/* Mark kernel_pmap active on all CPUs */
2004 	CPU_FILL(&kernel_pmap->pm_active);
2005 
2006  	/*
2007 	 * Initialize the global pv list lock.
2008 	 */
2009 	rw_init(&pvh_global_lock, "pmap pv global");
2010 
2011 	/*******************************************************/
2012 	/* Final setup */
2013 	/*******************************************************/
2014 
2015 	/* Enter kstack0 into kernel map, provide guard page */
2016 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
2017 	thread0.td_kstack = kstack0;
2018 	thread0.td_kstack_pages = kstack_pages;
2019 
2020 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
2021 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
2022 	    kstack0_phys, kstack0_phys + kstack0_sz);
2023 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
2024 	    kstack0, kstack0 + kstack0_sz);
2025 
2026 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
2027 	for (i = 0; i < kstack_pages; i++) {
2028 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
2029 		kstack0 += PAGE_SIZE;
2030 		kstack0_phys += PAGE_SIZE;
2031 	}
2032 
2033 	pmap_bootstrapped = 1;
2034 
2035 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
2036 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
2037 
2038 	debugf("mmu_booke_bootstrap: exit\n");
2039 }
2040 
2041 #ifdef SMP
2042  void
2043 tlb1_ap_prep(void)
2044 {
2045 	tlb_entry_t *e, tmp;
2046 	unsigned int i;
2047 
2048 	/* Prepare TLB1 image for AP processors */
2049 	e = __boot_tlb1;
2050 	for (i = 0; i < TLB1_ENTRIES; i++) {
2051 		tlb1_read_entry(&tmp, i);
2052 
2053 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
2054 			memcpy(e++, &tmp, sizeof(tmp));
2055 	}
2056 }
2057 
2058 void
2059 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
2060 {
2061 	int i;
2062 
2063 	/*
2064 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
2065 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
2066 	 * created by tlb1_ap_prep(), so use these values directly to
2067 	 * (re)program AP's TLB1 hardware.
2068 	 *
2069 	 * Start at index 1 because index 0 has the kernel map.
2070 	 */
2071 	for (i = 1; i < TLB1_ENTRIES; i++) {
2072 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
2073 			tlb1_write_entry(&__boot_tlb1[i], i);
2074 	}
2075 
2076 	set_mas4_defaults();
2077 }
2078 #endif
2079 
2080 static void
2081 booke_pmap_init_qpages(void)
2082 {
2083 	struct pcpu *pc;
2084 	int i;
2085 
2086 	CPU_FOREACH(i) {
2087 		pc = pcpu_find(i);
2088 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
2089 		if (pc->pc_qmap_addr == 0)
2090 			panic("pmap_init_qpages: unable to allocate KVA");
2091 	}
2092 }
2093 
2094 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
2095 
2096 /*
2097  * Get the physical page address for the given pmap/virtual address.
2098  */
2099 static vm_paddr_t
2100 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
2101 {
2102 	vm_paddr_t pa;
2103 
2104 	PMAP_LOCK(pmap);
2105 	pa = pte_vatopa(mmu, pmap, va);
2106 	PMAP_UNLOCK(pmap);
2107 
2108 	return (pa);
2109 }
2110 
2111 /*
2112  * Extract the physical page address associated with the given
2113  * kernel virtual address.
2114  */
2115 static vm_paddr_t
2116 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
2117 {
2118 	tlb_entry_t e;
2119 	vm_paddr_t p = 0;
2120 	int i;
2121 
2122 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
2123 		p = pte_vatopa(mmu, kernel_pmap, va);
2124 
2125 	if (p == 0) {
2126 		/* Check TLB1 mappings */
2127 		for (i = 0; i < TLB1_ENTRIES; i++) {
2128 			tlb1_read_entry(&e, i);
2129 			if (!(e.mas1 & MAS1_VALID))
2130 				continue;
2131 			if (va >= e.virt && va < e.virt + e.size)
2132 				return (e.phys + (va - e.virt));
2133 		}
2134 	}
2135 
2136 	return (p);
2137 }
2138 
2139 /*
2140  * Initialize the pmap module.
2141  * Called by vm_init, to initialize any structures that the pmap
2142  * system needs to map virtual memory.
2143  */
2144 static void
2145 mmu_booke_init(mmu_t mmu)
2146 {
2147 	int shpgperproc = PMAP_SHPGPERPROC;
2148 
2149 	/*
2150 	 * Initialize the address space (zone) for the pv entries.  Set a
2151 	 * high water mark so that the system can recover from excessive
2152 	 * numbers of pv entries.
2153 	 */
2154 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2155 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2156 
2157 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2158 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2159 
2160 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2161 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2162 
2163 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2164 
2165 	/* Pre-fill pvzone with initial number of pv entries. */
2166 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2167 
2168 	/* Initialize ptbl allocation. */
2169 	ptbl_init();
2170 }
2171 
2172 /*
2173  * Map a list of wired pages into kernel virtual address space.  This is
2174  * intended for temporary mappings which do not need page modification or
2175  * references recorded.  Existing mappings in the region are overwritten.
2176  */
2177 static void
2178 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2179 {
2180 	vm_offset_t va;
2181 
2182 	va = sva;
2183 	while (count-- > 0) {
2184 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2185 		va += PAGE_SIZE;
2186 		m++;
2187 	}
2188 }
2189 
2190 /*
2191  * Remove page mappings from kernel virtual address space.  Intended for
2192  * temporary mappings entered by mmu_booke_qenter.
2193  */
2194 static void
2195 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2196 {
2197 	vm_offset_t va;
2198 
2199 	va = sva;
2200 	while (count-- > 0) {
2201 		mmu_booke_kremove(mmu, va);
2202 		va += PAGE_SIZE;
2203 	}
2204 }
2205 
2206 /*
2207  * Map a wired page into kernel virtual address space.
2208  */
2209 static void
2210 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2211 {
2212 
2213 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2214 }
2215 
2216 static void
2217 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2218 {
2219 	uint32_t flags;
2220 	pte_t *pte;
2221 
2222 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2223 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2224 
2225 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2226 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2227 	flags |= PTE_PS_4KB;
2228 
2229 	pte = pte_find(mmu, kernel_pmap, va);
2230 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2231 
2232 	mtx_lock_spin(&tlbivax_mutex);
2233 	tlb_miss_lock();
2234 
2235 	if (PTE_ISVALID(pte)) {
2236 
2237 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2238 
2239 		/* Flush entry from TLB0 */
2240 		tlb0_flush_entry(va);
2241 	}
2242 
2243 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2244 
2245 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2246 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2247 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2248 
2249 	/* Flush the real memory from the instruction cache. */
2250 	if ((flags & (PTE_I | PTE_G)) == 0)
2251 		__syncicache((void *)va, PAGE_SIZE);
2252 
2253 	tlb_miss_unlock();
2254 	mtx_unlock_spin(&tlbivax_mutex);
2255 }
2256 
2257 /*
2258  * Remove a page from kernel page table.
2259  */
2260 static void
2261 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2262 {
2263 	pte_t *pte;
2264 
2265 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
2266 
2267 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2268 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2269 	    ("mmu_booke_kremove: invalid va"));
2270 
2271 	pte = pte_find(mmu, kernel_pmap, va);
2272 
2273 	if (!PTE_ISVALID(pte)) {
2274 
2275 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2276 
2277 		return;
2278 	}
2279 
2280 	mtx_lock_spin(&tlbivax_mutex);
2281 	tlb_miss_lock();
2282 
2283 	/* Invalidate entry in TLB0, update PTE. */
2284 	tlb0_flush_entry(va);
2285 	*pte = 0;
2286 
2287 	tlb_miss_unlock();
2288 	mtx_unlock_spin(&tlbivax_mutex);
2289 }
2290 
2291 /*
2292  * Provide a kernel pointer corresponding to a given userland pointer.
2293  * The returned pointer is valid until the next time this function is
2294  * called in this thread. This is used internally in copyin/copyout.
2295  */
2296 int
2297 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2298     void **kaddr, size_t ulen, size_t *klen)
2299 {
2300 
2301 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2302 		return (EFAULT);
2303 
2304 	*kaddr = (void *)(uintptr_t)uaddr;
2305 	if (klen)
2306 		*klen = ulen;
2307 
2308 	return (0);
2309 }
2310 
2311 /*
2312  * Figure out where a given kernel pointer (usually in a fault) points
2313  * to from the VM's perspective, potentially remapping into userland's
2314  * address space.
2315  */
2316 static int
2317 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2318     vm_offset_t *decoded_addr)
2319 {
2320 
2321 	if (addr < VM_MAXUSER_ADDRESS)
2322 		*is_user = 1;
2323 	else
2324 		*is_user = 0;
2325 
2326 	*decoded_addr = addr;
2327 	return (0);
2328 }
2329 
2330 /*
2331  * Initialize pmap associated with process 0.
2332  */
2333 static void
2334 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2335 {
2336 
2337 	PMAP_LOCK_INIT(pmap);
2338 	mmu_booke_pinit(mmu, pmap);
2339 	PCPU_SET(curpmap, pmap);
2340 }
2341 
2342 /*
2343  * Initialize a preallocated and zeroed pmap structure,
2344  * such as one in a vmspace structure.
2345  */
2346 static void
2347 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2348 {
2349 	int i;
2350 
2351 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2352 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2353 
2354 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2355 
2356 	for (i = 0; i < MAXCPU; i++)
2357 		pmap->pm_tid[i] = TID_NONE;
2358 	CPU_ZERO(&kernel_pmap->pm_active);
2359 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2360 #ifdef __powerpc64__
2361 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2362 	TAILQ_INIT(&pmap->pm_pdir_list);
2363 #else
2364 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2365 #endif
2366 	TAILQ_INIT(&pmap->pm_ptbl_list);
2367 }
2368 
2369 /*
2370  * Release any resources held by the given physical map.
2371  * Called when a pmap initialized by mmu_booke_pinit is being released.
2372  * Should only be called if the map contains no valid mappings.
2373  */
2374 static void
2375 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2376 {
2377 
2378 	KASSERT(pmap->pm_stats.resident_count == 0,
2379 	    ("pmap_release: pmap resident count %ld != 0",
2380 	    pmap->pm_stats.resident_count));
2381 }
2382 
2383 /*
2384  * Insert the given physical page at the specified virtual address in the
2385  * target physical map with the protection requested. If specified the page
2386  * will be wired down.
2387  */
2388 static int
2389 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2390     vm_prot_t prot, u_int flags, int8_t psind)
2391 {
2392 	int error;
2393 
2394 	rw_wlock(&pvh_global_lock);
2395 	PMAP_LOCK(pmap);
2396 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2397 	PMAP_UNLOCK(pmap);
2398 	rw_wunlock(&pvh_global_lock);
2399 	return (error);
2400 }
2401 
2402 static int
2403 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2404     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2405 {
2406 	pte_t *pte;
2407 	vm_paddr_t pa;
2408 	uint32_t flags;
2409 	int error, su, sync;
2410 
2411 	pa = VM_PAGE_TO_PHYS(m);
2412 	su = (pmap == kernel_pmap);
2413 	sync = 0;
2414 
2415 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2416 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2417 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2418 	//		(u_int32_t)m, va, pa, prot, flags);
2419 
2420 	if (su) {
2421 		KASSERT(((va >= virtual_avail) &&
2422 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2423 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2424 	} else {
2425 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2426 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2427 	}
2428 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2429 		VM_OBJECT_ASSERT_LOCKED(m->object);
2430 
2431 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2432 
2433 	/*
2434 	 * If there is an existing mapping, and the physical address has not
2435 	 * changed, must be protection or wiring change.
2436 	 */
2437 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2438 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2439 
2440 		/*
2441 		 * Before actually updating pte->flags we calculate and
2442 		 * prepare its new value in a helper var.
2443 		 */
2444 		flags = *pte;
2445 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2446 
2447 		/* Wiring change, just update stats. */
2448 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2449 			if (!PTE_ISWIRED(pte)) {
2450 				flags |= PTE_WIRED;
2451 				pmap->pm_stats.wired_count++;
2452 			}
2453 		} else {
2454 			if (PTE_ISWIRED(pte)) {
2455 				flags &= ~PTE_WIRED;
2456 				pmap->pm_stats.wired_count--;
2457 			}
2458 		}
2459 
2460 		if (prot & VM_PROT_WRITE) {
2461 			/* Add write permissions. */
2462 			flags |= PTE_SW;
2463 			if (!su)
2464 				flags |= PTE_UW;
2465 
2466 			if ((flags & PTE_MANAGED) != 0)
2467 				vm_page_aflag_set(m, PGA_WRITEABLE);
2468 		} else {
2469 			/* Handle modified pages, sense modify status. */
2470 
2471 			/*
2472 			 * The PTE_MODIFIED flag could be set by underlying
2473 			 * TLB misses since we last read it (above), possibly
2474 			 * other CPUs could update it so we check in the PTE
2475 			 * directly rather than rely on that saved local flags
2476 			 * copy.
2477 			 */
2478 			if (PTE_ISMODIFIED(pte))
2479 				vm_page_dirty(m);
2480 		}
2481 
2482 		if (prot & VM_PROT_EXECUTE) {
2483 			flags |= PTE_SX;
2484 			if (!su)
2485 				flags |= PTE_UX;
2486 
2487 			/*
2488 			 * Check existing flags for execute permissions: if we
2489 			 * are turning execute permissions on, icache should
2490 			 * be flushed.
2491 			 */
2492 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2493 				sync++;
2494 		}
2495 
2496 		flags &= ~PTE_REFERENCED;
2497 
2498 		/*
2499 		 * The new flags value is all calculated -- only now actually
2500 		 * update the PTE.
2501 		 */
2502 		mtx_lock_spin(&tlbivax_mutex);
2503 		tlb_miss_lock();
2504 
2505 		tlb0_flush_entry(va);
2506 		*pte &= ~PTE_FLAGS_MASK;
2507 		*pte |= flags;
2508 
2509 		tlb_miss_unlock();
2510 		mtx_unlock_spin(&tlbivax_mutex);
2511 
2512 	} else {
2513 		/*
2514 		 * If there is an existing mapping, but it's for a different
2515 		 * physical address, pte_enter() will delete the old mapping.
2516 		 */
2517 		//if ((pte != NULL) && PTE_ISVALID(pte))
2518 		//	debugf("mmu_booke_enter_locked: replace\n");
2519 		//else
2520 		//	debugf("mmu_booke_enter_locked: new\n");
2521 
2522 		/* Now set up the flags and install the new mapping. */
2523 		flags = (PTE_SR | PTE_VALID);
2524 		flags |= PTE_M;
2525 
2526 		if (!su)
2527 			flags |= PTE_UR;
2528 
2529 		if (prot & VM_PROT_WRITE) {
2530 			flags |= PTE_SW;
2531 			if (!su)
2532 				flags |= PTE_UW;
2533 
2534 			if ((m->oflags & VPO_UNMANAGED) == 0)
2535 				vm_page_aflag_set(m, PGA_WRITEABLE);
2536 		}
2537 
2538 		if (prot & VM_PROT_EXECUTE) {
2539 			flags |= PTE_SX;
2540 			if (!su)
2541 				flags |= PTE_UX;
2542 		}
2543 
2544 		/* If its wired update stats. */
2545 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2546 			flags |= PTE_WIRED;
2547 
2548 		error = pte_enter(mmu, pmap, m, va, flags,
2549 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2550 		if (error != 0)
2551 			return (KERN_RESOURCE_SHORTAGE);
2552 
2553 		if ((flags & PMAP_ENTER_WIRED) != 0)
2554 			pmap->pm_stats.wired_count++;
2555 
2556 		/* Flush the real memory from the instruction cache. */
2557 		if (prot & VM_PROT_EXECUTE)
2558 			sync++;
2559 	}
2560 
2561 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2562 		__syncicache((void *)va, PAGE_SIZE);
2563 		sync = 0;
2564 	}
2565 
2566 	return (KERN_SUCCESS);
2567 }
2568 
2569 /*
2570  * Maps a sequence of resident pages belonging to the same object.
2571  * The sequence begins with the given page m_start.  This page is
2572  * mapped at the given virtual address start.  Each subsequent page is
2573  * mapped at a virtual address that is offset from start by the same
2574  * amount as the page is offset from m_start within the object.  The
2575  * last page in the sequence is the page with the largest offset from
2576  * m_start that can be mapped at a virtual address less than the given
2577  * virtual address end.  Not every virtual page between start and end
2578  * is mapped; only those for which a resident page exists with the
2579  * corresponding offset from m_start are mapped.
2580  */
2581 static void
2582 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2583     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2584 {
2585 	vm_page_t m;
2586 	vm_pindex_t diff, psize;
2587 
2588 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2589 
2590 	psize = atop(end - start);
2591 	m = m_start;
2592 	rw_wlock(&pvh_global_lock);
2593 	PMAP_LOCK(pmap);
2594 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2595 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2596 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2597 		    PMAP_ENTER_NOSLEEP, 0);
2598 		m = TAILQ_NEXT(m, listq);
2599 	}
2600 	rw_wunlock(&pvh_global_lock);
2601 	PMAP_UNLOCK(pmap);
2602 }
2603 
2604 static void
2605 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2606     vm_prot_t prot)
2607 {
2608 
2609 	rw_wlock(&pvh_global_lock);
2610 	PMAP_LOCK(pmap);
2611 	mmu_booke_enter_locked(mmu, pmap, va, m,
2612 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2613 	    0);
2614 	rw_wunlock(&pvh_global_lock);
2615 	PMAP_UNLOCK(pmap);
2616 }
2617 
2618 /*
2619  * Remove the given range of addresses from the specified map.
2620  *
2621  * It is assumed that the start and end are properly rounded to the page size.
2622  */
2623 static void
2624 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2625 {
2626 	pte_t *pte;
2627 	uint8_t hold_flag;
2628 
2629 	int su = (pmap == kernel_pmap);
2630 
2631 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2632 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2633 
2634 	if (su) {
2635 		KASSERT(((va >= virtual_avail) &&
2636 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2637 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2638 	} else {
2639 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2640 		    ("mmu_booke_remove: user pmap, non user va"));
2641 	}
2642 
2643 	if (PMAP_REMOVE_DONE(pmap)) {
2644 		//debugf("mmu_booke_remove: e (empty)\n");
2645 		return;
2646 	}
2647 
2648 	hold_flag = PTBL_HOLD_FLAG(pmap);
2649 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2650 
2651 	rw_wlock(&pvh_global_lock);
2652 	PMAP_LOCK(pmap);
2653 	for (; va < endva; va += PAGE_SIZE) {
2654 		pte = pte_find(mmu, pmap, va);
2655 		if ((pte != NULL) && PTE_ISVALID(pte))
2656 			pte_remove(mmu, pmap, va, hold_flag);
2657 	}
2658 	PMAP_UNLOCK(pmap);
2659 	rw_wunlock(&pvh_global_lock);
2660 
2661 	//debugf("mmu_booke_remove: e\n");
2662 }
2663 
2664 /*
2665  * Remove physical page from all pmaps in which it resides.
2666  */
2667 static void
2668 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2669 {
2670 	pv_entry_t pv, pvn;
2671 	uint8_t hold_flag;
2672 
2673 	rw_wlock(&pvh_global_lock);
2674 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2675 		pvn = TAILQ_NEXT(pv, pv_link);
2676 
2677 		PMAP_LOCK(pv->pv_pmap);
2678 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2679 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2680 		PMAP_UNLOCK(pv->pv_pmap);
2681 	}
2682 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2683 	rw_wunlock(&pvh_global_lock);
2684 }
2685 
2686 /*
2687  * Map a range of physical addresses into kernel virtual address space.
2688  */
2689 static vm_offset_t
2690 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2691     vm_paddr_t pa_end, int prot)
2692 {
2693 	vm_offset_t sva = *virt;
2694 	vm_offset_t va = sva;
2695 
2696 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2697 	//		sva, pa_start, pa_end);
2698 
2699 	while (pa_start < pa_end) {
2700 		mmu_booke_kenter(mmu, va, pa_start);
2701 		va += PAGE_SIZE;
2702 		pa_start += PAGE_SIZE;
2703 	}
2704 	*virt = va;
2705 
2706 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2707 	return (sva);
2708 }
2709 
2710 /*
2711  * The pmap must be activated before it's address space can be accessed in any
2712  * way.
2713  */
2714 static void
2715 mmu_booke_activate(mmu_t mmu, struct thread *td)
2716 {
2717 	pmap_t pmap;
2718 	u_int cpuid;
2719 
2720 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2721 
2722 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
2723 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2724 
2725 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2726 
2727 	sched_pin();
2728 
2729 	cpuid = PCPU_GET(cpuid);
2730 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2731 	PCPU_SET(curpmap, pmap);
2732 
2733 	if (pmap->pm_tid[cpuid] == TID_NONE)
2734 		tid_alloc(pmap);
2735 
2736 	/* Load PID0 register with pmap tid value. */
2737 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2738 	__asm __volatile("isync");
2739 
2740 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2741 
2742 	sched_unpin();
2743 
2744 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2745 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2746 }
2747 
2748 /*
2749  * Deactivate the specified process's address space.
2750  */
2751 static void
2752 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2753 {
2754 	pmap_t pmap;
2755 
2756 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2757 
2758 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
2759 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2760 
2761 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2762 
2763 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2764 	PCPU_SET(curpmap, NULL);
2765 }
2766 
2767 /*
2768  * Copy the range specified by src_addr/len
2769  * from the source map to the range dst_addr/len
2770  * in the destination map.
2771  *
2772  * This routine is only advisory and need not do anything.
2773  */
2774 static void
2775 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2776     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2777 {
2778 
2779 }
2780 
2781 /*
2782  * Set the physical protection on the specified range of this map as requested.
2783  */
2784 static void
2785 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2786     vm_prot_t prot)
2787 {
2788 	vm_offset_t va;
2789 	vm_page_t m;
2790 	pte_t *pte;
2791 
2792 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2793 		mmu_booke_remove(mmu, pmap, sva, eva);
2794 		return;
2795 	}
2796 
2797 	if (prot & VM_PROT_WRITE)
2798 		return;
2799 
2800 	PMAP_LOCK(pmap);
2801 	for (va = sva; va < eva; va += PAGE_SIZE) {
2802 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2803 			if (PTE_ISVALID(pte)) {
2804 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2805 
2806 				mtx_lock_spin(&tlbivax_mutex);
2807 				tlb_miss_lock();
2808 
2809 				/* Handle modified pages. */
2810 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2811 					vm_page_dirty(m);
2812 
2813 				tlb0_flush_entry(va);
2814 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2815 
2816 				tlb_miss_unlock();
2817 				mtx_unlock_spin(&tlbivax_mutex);
2818 			}
2819 		}
2820 	}
2821 	PMAP_UNLOCK(pmap);
2822 }
2823 
2824 /*
2825  * Clear the write and modified bits in each of the given page's mappings.
2826  */
2827 static void
2828 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2829 {
2830 	pv_entry_t pv;
2831 	pte_t *pte;
2832 
2833 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2834 	    ("mmu_booke_remove_write: page %p is not managed", m));
2835 
2836 	/*
2837 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2838 	 * set by another thread while the object is locked.  Thus,
2839 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2840 	 */
2841 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2842 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2843 		return;
2844 	rw_wlock(&pvh_global_lock);
2845 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2846 		PMAP_LOCK(pv->pv_pmap);
2847 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2848 			if (PTE_ISVALID(pte)) {
2849 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2850 
2851 				mtx_lock_spin(&tlbivax_mutex);
2852 				tlb_miss_lock();
2853 
2854 				/* Handle modified pages. */
2855 				if (PTE_ISMODIFIED(pte))
2856 					vm_page_dirty(m);
2857 
2858 				/* Flush mapping from TLB0. */
2859 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2860 
2861 				tlb_miss_unlock();
2862 				mtx_unlock_spin(&tlbivax_mutex);
2863 			}
2864 		}
2865 		PMAP_UNLOCK(pv->pv_pmap);
2866 	}
2867 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2868 	rw_wunlock(&pvh_global_lock);
2869 }
2870 
2871 static void
2872 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2873 {
2874 	pte_t *pte;
2875 	pmap_t pmap;
2876 	vm_page_t m;
2877 	vm_offset_t addr;
2878 	vm_paddr_t pa = 0;
2879 	int active, valid;
2880 
2881 	va = trunc_page(va);
2882 	sz = round_page(sz);
2883 
2884 	rw_wlock(&pvh_global_lock);
2885 	pmap = PCPU_GET(curpmap);
2886 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2887 	while (sz > 0) {
2888 		PMAP_LOCK(pm);
2889 		pte = pte_find(mmu, pm, va);
2890 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2891 		if (valid)
2892 			pa = PTE_PA(pte);
2893 		PMAP_UNLOCK(pm);
2894 		if (valid) {
2895 			if (!active) {
2896 				/* Create a mapping in the active pmap. */
2897 				addr = 0;
2898 				m = PHYS_TO_VM_PAGE(pa);
2899 				PMAP_LOCK(pmap);
2900 				pte_enter(mmu, pmap, m, addr,
2901 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2902 				__syncicache((void *)addr, PAGE_SIZE);
2903 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2904 				PMAP_UNLOCK(pmap);
2905 			} else
2906 				__syncicache((void *)va, PAGE_SIZE);
2907 		}
2908 		va += PAGE_SIZE;
2909 		sz -= PAGE_SIZE;
2910 	}
2911 	rw_wunlock(&pvh_global_lock);
2912 }
2913 
2914 /*
2915  * Atomically extract and hold the physical page with the given
2916  * pmap and virtual address pair if that mapping permits the given
2917  * protection.
2918  */
2919 static vm_page_t
2920 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2921     vm_prot_t prot)
2922 {
2923 	pte_t *pte;
2924 	vm_page_t m;
2925 	uint32_t pte_wbit;
2926 	vm_paddr_t pa;
2927 
2928 	m = NULL;
2929 	pa = 0;
2930 	PMAP_LOCK(pmap);
2931 retry:
2932 	pte = pte_find(mmu, pmap, va);
2933 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2934 		if (pmap == kernel_pmap)
2935 			pte_wbit = PTE_SW;
2936 		else
2937 			pte_wbit = PTE_UW;
2938 
2939 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2940 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2941 				goto retry;
2942 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2943 			vm_page_hold(m);
2944 		}
2945 	}
2946 
2947 	PA_UNLOCK_COND(pa);
2948 	PMAP_UNLOCK(pmap);
2949 	return (m);
2950 }
2951 
2952 /*
2953  * Initialize a vm_page's machine-dependent fields.
2954  */
2955 static void
2956 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2957 {
2958 
2959 	m->md.pv_tracked = 0;
2960 	TAILQ_INIT(&m->md.pv_list);
2961 }
2962 
2963 /*
2964  * mmu_booke_zero_page_area zeros the specified hardware page by
2965  * mapping it into virtual memory and using bzero to clear
2966  * its contents.
2967  *
2968  * off and size must reside within a single page.
2969  */
2970 static void
2971 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2972 {
2973 	vm_offset_t va;
2974 
2975 	/* XXX KASSERT off and size are within a single page? */
2976 
2977 #ifdef __powerpc64__
2978 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
2979 	bzero((caddr_t)va + off, size);
2980 #else
2981 	mtx_lock(&zero_page_mutex);
2982 	va = zero_page_va;
2983 
2984 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2985 	bzero((caddr_t)va + off, size);
2986 	mmu_booke_kremove(mmu, va);
2987 
2988 	mtx_unlock(&zero_page_mutex);
2989 #endif
2990 }
2991 
2992 /*
2993  * mmu_booke_zero_page zeros the specified hardware page.
2994  */
2995 static void
2996 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2997 {
2998 	vm_offset_t off, va;
2999 
3000 #ifdef __powerpc64__
3001 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
3002 
3003 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
3004 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
3005 #else
3006 	va = zero_page_va;
3007 	mtx_lock(&zero_page_mutex);
3008 
3009 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
3010 
3011 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
3012 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
3013 
3014 	mmu_booke_kremove(mmu, va);
3015 
3016 	mtx_unlock(&zero_page_mutex);
3017 #endif
3018 }
3019 
3020 /*
3021  * mmu_booke_copy_page copies the specified (machine independent) page by
3022  * mapping the page into virtual memory and using memcopy to copy the page,
3023  * one machine dependent page at a time.
3024  */
3025 static void
3026 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
3027 {
3028 	vm_offset_t sva, dva;
3029 
3030 #ifdef __powerpc64__
3031 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
3032 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
3033 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3034 #else
3035 	mtx_lock(&copy_page_mutex);
3036 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
3037 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
3038 	sva = copy_page_src_va;
3039 	dva = copy_page_dst_va;
3040 
3041 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3042 
3043 	mmu_booke_kremove(mmu, dva);
3044 	mmu_booke_kremove(mmu, sva);
3045 	mtx_unlock(&copy_page_mutex);
3046 #endif
3047 }
3048 
3049 static inline void
3050 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
3051     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
3052 {
3053 	void *a_cp, *b_cp;
3054 	vm_offset_t a_pg_offset, b_pg_offset;
3055 	int cnt;
3056 
3057 #ifdef __powerpc64__
3058 	vm_page_t pa, pb;
3059 
3060 	while (xfersize > 0) {
3061 		a_pg_offset = a_offset & PAGE_MASK;
3062 		pa = ma[a_offset >> PAGE_SHIFT];
3063 		b_pg_offset = b_offset & PAGE_MASK;
3064 		pb = mb[b_offset >> PAGE_SHIFT];
3065 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3066 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3067 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
3068 		    a_pg_offset);
3069 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
3070 		    b_pg_offset);
3071 		bcopy(a_cp, b_cp, cnt);
3072 		a_offset += cnt;
3073 		b_offset += cnt;
3074 		xfersize -= cnt;
3075 	}
3076 #else
3077 	mtx_lock(&copy_page_mutex);
3078 	while (xfersize > 0) {
3079 		a_pg_offset = a_offset & PAGE_MASK;
3080 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3081 		mmu_booke_kenter(mmu, copy_page_src_va,
3082 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
3083 		a_cp = (char *)copy_page_src_va + a_pg_offset;
3084 		b_pg_offset = b_offset & PAGE_MASK;
3085 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3086 		mmu_booke_kenter(mmu, copy_page_dst_va,
3087 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
3088 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
3089 		bcopy(a_cp, b_cp, cnt);
3090 		mmu_booke_kremove(mmu, copy_page_dst_va);
3091 		mmu_booke_kremove(mmu, copy_page_src_va);
3092 		a_offset += cnt;
3093 		b_offset += cnt;
3094 		xfersize -= cnt;
3095 	}
3096 	mtx_unlock(&copy_page_mutex);
3097 #endif
3098 }
3099 
3100 static vm_offset_t
3101 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
3102 {
3103 #ifdef __powerpc64__
3104 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
3105 #else
3106 	vm_paddr_t paddr;
3107 	vm_offset_t qaddr;
3108 	uint32_t flags;
3109 	pte_t *pte;
3110 
3111 	paddr = VM_PAGE_TO_PHYS(m);
3112 
3113 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3114 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3115 	flags |= PTE_PS_4KB;
3116 
3117 	critical_enter();
3118 	qaddr = PCPU_GET(qmap_addr);
3119 
3120 	pte = pte_find(mmu, kernel_pmap, qaddr);
3121 
3122 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3123 
3124 	/*
3125 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3126  	 * not be present in other TLBs.  Is there a better instruction
3127 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3128 	 */
3129 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3130 	__asm __volatile("isync; msync");
3131 
3132 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3133 
3134 	/* Flush the real memory from the instruction cache. */
3135 	if ((flags & (PTE_I | PTE_G)) == 0)
3136 		__syncicache((void *)qaddr, PAGE_SIZE);
3137 
3138 	return (qaddr);
3139 #endif
3140 }
3141 
3142 static void
3143 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3144 {
3145 #ifndef __powerpc64__
3146 	pte_t *pte;
3147 
3148 	pte = pte_find(mmu, kernel_pmap, addr);
3149 
3150 	KASSERT(PCPU_GET(qmap_addr) == addr,
3151 	    ("mmu_booke_quick_remove_page: invalid address"));
3152 	KASSERT(*pte != 0,
3153 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3154 
3155 	*pte = 0;
3156 	critical_exit();
3157 #endif
3158 }
3159 
3160 /*
3161  * Return whether or not the specified physical page was modified
3162  * in any of physical maps.
3163  */
3164 static boolean_t
3165 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3166 {
3167 	pte_t *pte;
3168 	pv_entry_t pv;
3169 	boolean_t rv;
3170 
3171 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3172 	    ("mmu_booke_is_modified: page %p is not managed", m));
3173 	rv = FALSE;
3174 
3175 	/*
3176 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3177 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3178 	 * is clear, no PTEs can be modified.
3179 	 */
3180 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3181 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3182 		return (rv);
3183 	rw_wlock(&pvh_global_lock);
3184 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3185 		PMAP_LOCK(pv->pv_pmap);
3186 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3187 		    PTE_ISVALID(pte)) {
3188 			if (PTE_ISMODIFIED(pte))
3189 				rv = TRUE;
3190 		}
3191 		PMAP_UNLOCK(pv->pv_pmap);
3192 		if (rv)
3193 			break;
3194 	}
3195 	rw_wunlock(&pvh_global_lock);
3196 	return (rv);
3197 }
3198 
3199 /*
3200  * Return whether or not the specified virtual address is eligible
3201  * for prefault.
3202  */
3203 static boolean_t
3204 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3205 {
3206 
3207 	return (FALSE);
3208 }
3209 
3210 /*
3211  * Return whether or not the specified physical page was referenced
3212  * in any physical maps.
3213  */
3214 static boolean_t
3215 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3216 {
3217 	pte_t *pte;
3218 	pv_entry_t pv;
3219 	boolean_t rv;
3220 
3221 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3222 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3223 	rv = FALSE;
3224 	rw_wlock(&pvh_global_lock);
3225 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3226 		PMAP_LOCK(pv->pv_pmap);
3227 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3228 		    PTE_ISVALID(pte)) {
3229 			if (PTE_ISREFERENCED(pte))
3230 				rv = TRUE;
3231 		}
3232 		PMAP_UNLOCK(pv->pv_pmap);
3233 		if (rv)
3234 			break;
3235 	}
3236 	rw_wunlock(&pvh_global_lock);
3237 	return (rv);
3238 }
3239 
3240 /*
3241  * Clear the modify bits on the specified physical page.
3242  */
3243 static void
3244 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3245 {
3246 	pte_t *pte;
3247 	pv_entry_t pv;
3248 
3249 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3250 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3251 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3252 	KASSERT(!vm_page_xbusied(m),
3253 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3254 
3255 	/*
3256 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3257 	 * If the object containing the page is locked and the page is not
3258 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3259 	 */
3260 	if ((m->aflags & PGA_WRITEABLE) == 0)
3261 		return;
3262 	rw_wlock(&pvh_global_lock);
3263 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3264 		PMAP_LOCK(pv->pv_pmap);
3265 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3266 		    PTE_ISVALID(pte)) {
3267 			mtx_lock_spin(&tlbivax_mutex);
3268 			tlb_miss_lock();
3269 
3270 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3271 				tlb0_flush_entry(pv->pv_va);
3272 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3273 				    PTE_REFERENCED);
3274 			}
3275 
3276 			tlb_miss_unlock();
3277 			mtx_unlock_spin(&tlbivax_mutex);
3278 		}
3279 		PMAP_UNLOCK(pv->pv_pmap);
3280 	}
3281 	rw_wunlock(&pvh_global_lock);
3282 }
3283 
3284 /*
3285  * Return a count of reference bits for a page, clearing those bits.
3286  * It is not necessary for every reference bit to be cleared, but it
3287  * is necessary that 0 only be returned when there are truly no
3288  * reference bits set.
3289  *
3290  * As an optimization, update the page's dirty field if a modified bit is
3291  * found while counting reference bits.  This opportunistic update can be
3292  * performed at low cost and can eliminate the need for some future calls
3293  * to pmap_is_modified().  However, since this function stops after
3294  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3295  * dirty pages.  Those dirty pages will only be detected by a future call
3296  * to pmap_is_modified().
3297  */
3298 static int
3299 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3300 {
3301 	pte_t *pte;
3302 	pv_entry_t pv;
3303 	int count;
3304 
3305 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3306 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3307 	count = 0;
3308 	rw_wlock(&pvh_global_lock);
3309 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3310 		PMAP_LOCK(pv->pv_pmap);
3311 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3312 		    PTE_ISVALID(pte)) {
3313 			if (PTE_ISMODIFIED(pte))
3314 				vm_page_dirty(m);
3315 			if (PTE_ISREFERENCED(pte)) {
3316 				mtx_lock_spin(&tlbivax_mutex);
3317 				tlb_miss_lock();
3318 
3319 				tlb0_flush_entry(pv->pv_va);
3320 				*pte &= ~PTE_REFERENCED;
3321 
3322 				tlb_miss_unlock();
3323 				mtx_unlock_spin(&tlbivax_mutex);
3324 
3325 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3326 					PMAP_UNLOCK(pv->pv_pmap);
3327 					break;
3328 				}
3329 			}
3330 		}
3331 		PMAP_UNLOCK(pv->pv_pmap);
3332 	}
3333 	rw_wunlock(&pvh_global_lock);
3334 	return (count);
3335 }
3336 
3337 /*
3338  * Clear the wired attribute from the mappings for the specified range of
3339  * addresses in the given pmap.  Every valid mapping within that range must
3340  * have the wired attribute set.  In contrast, invalid mappings cannot have
3341  * the wired attribute set, so they are ignored.
3342  *
3343  * The wired attribute of the page table entry is not a hardware feature, so
3344  * there is no need to invalidate any TLB entries.
3345  */
3346 static void
3347 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3348 {
3349 	vm_offset_t va;
3350 	pte_t *pte;
3351 
3352 	PMAP_LOCK(pmap);
3353 	for (va = sva; va < eva; va += PAGE_SIZE) {
3354 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3355 		    PTE_ISVALID(pte)) {
3356 			if (!PTE_ISWIRED(pte))
3357 				panic("mmu_booke_unwire: pte %p isn't wired",
3358 				    pte);
3359 			*pte &= ~PTE_WIRED;
3360 			pmap->pm_stats.wired_count--;
3361 		}
3362 	}
3363 	PMAP_UNLOCK(pmap);
3364 
3365 }
3366 
3367 /*
3368  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3369  * page.  This count may be changed upwards or downwards in the future; it is
3370  * only necessary that true be returned for a small subset of pmaps for proper
3371  * page aging.
3372  */
3373 static boolean_t
3374 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3375 {
3376 	pv_entry_t pv;
3377 	int loops;
3378 	boolean_t rv;
3379 
3380 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3381 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3382 	loops = 0;
3383 	rv = FALSE;
3384 	rw_wlock(&pvh_global_lock);
3385 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3386 		if (pv->pv_pmap == pmap) {
3387 			rv = TRUE;
3388 			break;
3389 		}
3390 		if (++loops >= 16)
3391 			break;
3392 	}
3393 	rw_wunlock(&pvh_global_lock);
3394 	return (rv);
3395 }
3396 
3397 /*
3398  * Return the number of managed mappings to the given physical page that are
3399  * wired.
3400  */
3401 static int
3402 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3403 {
3404 	pv_entry_t pv;
3405 	pte_t *pte;
3406 	int count = 0;
3407 
3408 	if ((m->oflags & VPO_UNMANAGED) != 0)
3409 		return (count);
3410 	rw_wlock(&pvh_global_lock);
3411 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3412 		PMAP_LOCK(pv->pv_pmap);
3413 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3414 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3415 				count++;
3416 		PMAP_UNLOCK(pv->pv_pmap);
3417 	}
3418 	rw_wunlock(&pvh_global_lock);
3419 	return (count);
3420 }
3421 
3422 static int
3423 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3424 {
3425 	int i;
3426 	vm_offset_t va;
3427 
3428 	/*
3429 	 * This currently does not work for entries that
3430 	 * overlap TLB1 entries.
3431 	 */
3432 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3433 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3434 			return (0);
3435 	}
3436 
3437 	return (EFAULT);
3438 }
3439 
3440 void
3441 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3442 {
3443 	vm_paddr_t ppa;
3444 	vm_offset_t ofs;
3445 	vm_size_t gran;
3446 
3447 	/* Minidumps are based on virtual memory addresses. */
3448 	if (do_minidump) {
3449 		*va = (void *)(vm_offset_t)pa;
3450 		return;
3451 	}
3452 
3453 	/* Raw physical memory dumps don't have a virtual address. */
3454 	/* We always map a 256MB page at 256M. */
3455 	gran = 256 * 1024 * 1024;
3456 	ppa = rounddown2(pa, gran);
3457 	ofs = pa - ppa;
3458 	*va = (void *)gran;
3459 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3460 
3461 	if (sz > (gran - ofs))
3462 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3463 		    _TLB_ENTRY_IO);
3464 }
3465 
3466 void
3467 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3468 {
3469 	vm_paddr_t ppa;
3470 	vm_offset_t ofs;
3471 	vm_size_t gran;
3472 	tlb_entry_t e;
3473 	int i;
3474 
3475 	/* Minidumps are based on virtual memory addresses. */
3476 	/* Nothing to do... */
3477 	if (do_minidump)
3478 		return;
3479 
3480 	for (i = 0; i < TLB1_ENTRIES; i++) {
3481 		tlb1_read_entry(&e, i);
3482 		if (!(e.mas1 & MAS1_VALID))
3483 			break;
3484 	}
3485 
3486 	/* Raw physical memory dumps don't have a virtual address. */
3487 	i--;
3488 	e.mas1 = 0;
3489 	e.mas2 = 0;
3490 	e.mas3 = 0;
3491 	tlb1_write_entry(&e, i);
3492 
3493 	gran = 256 * 1024 * 1024;
3494 	ppa = rounddown2(pa, gran);
3495 	ofs = pa - ppa;
3496 	if (sz > (gran - ofs)) {
3497 		i--;
3498 		e.mas1 = 0;
3499 		e.mas2 = 0;
3500 		e.mas3 = 0;
3501 		tlb1_write_entry(&e, i);
3502 	}
3503 }
3504 
3505 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3506 
3507 void
3508 mmu_booke_scan_init(mmu_t mmu)
3509 {
3510 	vm_offset_t va;
3511 	pte_t *pte;
3512 	int i;
3513 
3514 	if (!do_minidump) {
3515 		/* Initialize phys. segments for dumpsys(). */
3516 		memset(&dump_map, 0, sizeof(dump_map));
3517 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3518 		    &availmem_regions_sz);
3519 		for (i = 0; i < physmem_regions_sz; i++) {
3520 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3521 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3522 		}
3523 		return;
3524 	}
3525 
3526 	/* Virtual segments for minidumps: */
3527 	memset(&dump_map, 0, sizeof(dump_map));
3528 
3529 	/* 1st: kernel .data and .bss. */
3530 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3531 	dump_map[0].pa_size =
3532 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3533 
3534 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3535 	dump_map[1].pa_start = data_start;
3536 	dump_map[1].pa_size = data_end - data_start;
3537 
3538 	/* 3rd: kernel VM. */
3539 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3540 	/* Find start of next chunk (from va). */
3541 	while (va < virtual_end) {
3542 		/* Don't dump the buffer cache. */
3543 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3544 			va = kmi.buffer_eva;
3545 			continue;
3546 		}
3547 		pte = pte_find(mmu, kernel_pmap, va);
3548 		if (pte != NULL && PTE_ISVALID(pte))
3549 			break;
3550 		va += PAGE_SIZE;
3551 	}
3552 	if (va < virtual_end) {
3553 		dump_map[2].pa_start = va;
3554 		va += PAGE_SIZE;
3555 		/* Find last page in chunk. */
3556 		while (va < virtual_end) {
3557 			/* Don't run into the buffer cache. */
3558 			if (va == kmi.buffer_sva)
3559 				break;
3560 			pte = pte_find(mmu, kernel_pmap, va);
3561 			if (pte == NULL || !PTE_ISVALID(pte))
3562 				break;
3563 			va += PAGE_SIZE;
3564 		}
3565 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3566 	}
3567 }
3568 
3569 /*
3570  * Map a set of physical memory pages into the kernel virtual address space.
3571  * Return a pointer to where it is mapped. This routine is intended to be used
3572  * for mapping device memory, NOT real memory.
3573  */
3574 static void *
3575 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3576 {
3577 
3578 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3579 }
3580 
3581 static void *
3582 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3583 {
3584 	tlb_entry_t e;
3585 	void *res;
3586 	uintptr_t va, tmpva;
3587 	vm_size_t sz;
3588 	int i;
3589 
3590 	/*
3591 	 * Check if this is premapped in TLB1. Note: this should probably also
3592 	 * check whether a sequence of TLB1 entries exist that match the
3593 	 * requirement, but now only checks the easy case.
3594 	 */
3595 	for (i = 0; i < TLB1_ENTRIES; i++) {
3596 		tlb1_read_entry(&e, i);
3597 		if (!(e.mas1 & MAS1_VALID))
3598 			continue;
3599 		if (pa >= e.phys &&
3600 		    (pa + size) <= (e.phys + e.size) &&
3601 		    (ma == VM_MEMATTR_DEFAULT ||
3602 		     tlb_calc_wimg(pa, ma) ==
3603 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3604 			return (void *)(e.virt +
3605 			    (vm_offset_t)(pa - e.phys));
3606 	}
3607 
3608 	size = roundup(size, PAGE_SIZE);
3609 
3610 	/*
3611 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3612 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3613 	 */
3614 #ifdef SPARSE_MAPDEV
3615 	/*
3616 	 * With a sparse mapdev, align to the largest starting region.  This
3617 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3618 	 * calculation could be very costly.
3619 	 * Align to the smaller of:
3620 	 * - first set bit in overlap of (pa & size mask)
3621 	 * - largest size envelope
3622 	 *
3623 	 * It's possible the device mapping may start at a PA that's not larger
3624 	 * than the size mask, so we need to offset in to maximize the TLB entry
3625 	 * range and minimize the number of used TLB entries.
3626 	 */
3627 	do {
3628 	    tmpva = tlb1_map_base;
3629 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3630 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3631 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3632 #ifdef __powerpc64__
3633 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3634 #else
3635 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3636 #endif
3637 #else
3638 #ifdef __powerpc64__
3639 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3640 #else
3641 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3642 #endif
3643 #endif
3644 	res = (void *)va;
3645 
3646 	do {
3647 		sz = 1 << (ilog2(size) & ~1);
3648 		/* Align size to PA */
3649 		if (pa % sz != 0) {
3650 			do {
3651 				sz >>= 2;
3652 			} while (pa % sz != 0);
3653 		}
3654 		/* Now align from there to VA */
3655 		if (va % sz != 0) {
3656 			do {
3657 				sz >>= 2;
3658 			} while (va % sz != 0);
3659 		}
3660 		if (bootverbose)
3661 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3662 			    va, (uintmax_t)pa, sz);
3663 		if (tlb1_set_entry(va, pa, sz,
3664 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3665 			return (NULL);
3666 		size -= sz;
3667 		pa += sz;
3668 		va += sz;
3669 	} while (size > 0);
3670 
3671 	return (res);
3672 }
3673 
3674 /*
3675  * 'Unmap' a range mapped by mmu_booke_mapdev().
3676  */
3677 static void
3678 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3679 {
3680 #ifdef SUPPORTS_SHRINKING_TLB1
3681 	vm_offset_t base, offset;
3682 
3683 	/*
3684 	 * Unmap only if this is inside kernel virtual space.
3685 	 */
3686 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3687 		base = trunc_page(va);
3688 		offset = va & PAGE_MASK;
3689 		size = roundup(offset + size, PAGE_SIZE);
3690 		kva_free(base, size);
3691 	}
3692 #endif
3693 }
3694 
3695 /*
3696  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3697  * specified pmap. This eliminates the blast of soft faults on process startup
3698  * and immediately after an mmap.
3699  */
3700 static void
3701 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3702     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3703 {
3704 
3705 	VM_OBJECT_ASSERT_WLOCKED(object);
3706 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3707 	    ("mmu_booke_object_init_pt: non-device object"));
3708 }
3709 
3710 /*
3711  * Perform the pmap work for mincore.
3712  */
3713 static int
3714 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3715     vm_paddr_t *locked_pa)
3716 {
3717 
3718 	/* XXX: this should be implemented at some point */
3719 	return (0);
3720 }
3721 
3722 static int
3723 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3724     vm_memattr_t mode)
3725 {
3726 	vm_offset_t va;
3727 	pte_t *pte;
3728 	int i, j;
3729 	tlb_entry_t e;
3730 
3731 	/* Check TLB1 mappings */
3732 	for (i = 0; i < TLB1_ENTRIES; i++) {
3733 		tlb1_read_entry(&e, i);
3734 		if (!(e.mas1 & MAS1_VALID))
3735 			continue;
3736 		if (addr >= e.virt && addr < e.virt + e.size)
3737 			break;
3738 	}
3739 	if (i < TLB1_ENTRIES) {
3740 		/* Only allow full mappings to be modified for now. */
3741 		/* Validate the range. */
3742 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3743 			tlb1_read_entry(&e, j);
3744 			if (va != e.virt || (sz - (va - addr) < e.size))
3745 				return (EINVAL);
3746 		}
3747 		for (va = addr; va < addr + sz; va += e.size, i++) {
3748 			tlb1_read_entry(&e, i);
3749 			e.mas2 &= ~MAS2_WIMGE_MASK;
3750 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3751 
3752 			/*
3753 			 * Write it out to the TLB.  Should really re-sync with other
3754 			 * cores.
3755 			 */
3756 			tlb1_write_entry(&e, i);
3757 		}
3758 		return (0);
3759 	}
3760 
3761 	/* Not in TLB1, try through pmap */
3762 	/* First validate the range. */
3763 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3764 		pte = pte_find(mmu, kernel_pmap, va);
3765 		if (pte == NULL || !PTE_ISVALID(pte))
3766 			return (EINVAL);
3767 	}
3768 
3769 	mtx_lock_spin(&tlbivax_mutex);
3770 	tlb_miss_lock();
3771 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3772 		pte = pte_find(mmu, kernel_pmap, va);
3773 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3774 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3775 		tlb0_flush_entry(va);
3776 	}
3777 	tlb_miss_unlock();
3778 	mtx_unlock_spin(&tlbivax_mutex);
3779 
3780 	return (0);
3781 }
3782 
3783 /**************************************************************************/
3784 /* TID handling */
3785 /**************************************************************************/
3786 
3787 /*
3788  * Allocate a TID. If necessary, steal one from someone else.
3789  * The new TID is flushed from the TLB before returning.
3790  */
3791 static tlbtid_t
3792 tid_alloc(pmap_t pmap)
3793 {
3794 	tlbtid_t tid;
3795 	int thiscpu;
3796 
3797 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3798 
3799 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3800 
3801 	thiscpu = PCPU_GET(cpuid);
3802 
3803 	tid = PCPU_GET(booke.tid_next);
3804 	if (tid > TID_MAX)
3805 		tid = TID_MIN;
3806 	PCPU_SET(booke.tid_next, tid + 1);
3807 
3808 	/* If we are stealing TID then clear the relevant pmap's field */
3809 	if (tidbusy[thiscpu][tid] != NULL) {
3810 
3811 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3812 
3813 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3814 
3815 		/* Flush all entries from TLB0 matching this TID. */
3816 		tid_flush(tid);
3817 	}
3818 
3819 	tidbusy[thiscpu][tid] = pmap;
3820 	pmap->pm_tid[thiscpu] = tid;
3821 	__asm __volatile("msync; isync");
3822 
3823 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3824 	    PCPU_GET(booke.tid_next));
3825 
3826 	return (tid);
3827 }
3828 
3829 /**************************************************************************/
3830 /* TLB0 handling */
3831 /**************************************************************************/
3832 
3833 /* Convert TLB0 va and way number to tlb0[] table index. */
3834 static inline unsigned int
3835 tlb0_tableidx(vm_offset_t va, unsigned int way)
3836 {
3837 	unsigned int idx;
3838 
3839 	idx = (way * TLB0_ENTRIES_PER_WAY);
3840 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3841 	return (idx);
3842 }
3843 
3844 /*
3845  * Invalidate TLB0 entry.
3846  */
3847 static inline void
3848 tlb0_flush_entry(vm_offset_t va)
3849 {
3850 
3851 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3852 
3853 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3854 
3855 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3856 	__asm __volatile("isync; msync");
3857 	__asm __volatile("tlbsync; msync");
3858 
3859 	CTR1(KTR_PMAP, "%s: e", __func__);
3860 }
3861 
3862 
3863 /**************************************************************************/
3864 /* TLB1 handling */
3865 /**************************************************************************/
3866 
3867 /*
3868  * TLB1 mapping notes:
3869  *
3870  * TLB1[0]	Kernel text and data.
3871  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3872  *		windows, other devices mappings.
3873  */
3874 
3875  /*
3876  * Read an entry from given TLB1 slot.
3877  */
3878 void
3879 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3880 {
3881 	register_t msr;
3882 	uint32_t mas0;
3883 
3884 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3885 
3886 	msr = mfmsr();
3887 	__asm __volatile("wrteei 0");
3888 
3889 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3890 	mtspr(SPR_MAS0, mas0);
3891 	__asm __volatile("isync; tlbre");
3892 
3893 	entry->mas1 = mfspr(SPR_MAS1);
3894 	entry->mas2 = mfspr(SPR_MAS2);
3895 	entry->mas3 = mfspr(SPR_MAS3);
3896 
3897 	switch ((mfpvr() >> 16) & 0xFFFF) {
3898 	case FSL_E500v2:
3899 	case FSL_E500mc:
3900 	case FSL_E5500:
3901 	case FSL_E6500:
3902 		entry->mas7 = mfspr(SPR_MAS7);
3903 		break;
3904 	default:
3905 		entry->mas7 = 0;
3906 		break;
3907 	}
3908 	mtmsr(msr);
3909 
3910 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3911 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3912 	    (entry->mas3 & MAS3_RPN);
3913 	entry->size =
3914 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3915 }
3916 
3917 struct tlbwrite_args {
3918 	tlb_entry_t *e;
3919 	unsigned int idx;
3920 };
3921 
3922 static void
3923 tlb1_write_entry_int(void *arg)
3924 {
3925 	struct tlbwrite_args *args = arg;
3926 	uint32_t mas0;
3927 
3928 	/* Select entry */
3929 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3930 
3931 	mtspr(SPR_MAS0, mas0);
3932 	mtspr(SPR_MAS1, args->e->mas1);
3933 	mtspr(SPR_MAS2, args->e->mas2);
3934 	mtspr(SPR_MAS3, args->e->mas3);
3935 	switch ((mfpvr() >> 16) & 0xFFFF) {
3936 	case FSL_E500mc:
3937 	case FSL_E5500:
3938 	case FSL_E6500:
3939 		mtspr(SPR_MAS8, 0);
3940 		/* FALLTHROUGH */
3941 	case FSL_E500v2:
3942 		mtspr(SPR_MAS7, args->e->mas7);
3943 		break;
3944 	default:
3945 		break;
3946 	}
3947 
3948 	__asm __volatile("isync; tlbwe; isync; msync");
3949 
3950 }
3951 
3952 static void
3953 tlb1_write_entry_sync(void *arg)
3954 {
3955 	/* Empty synchronization point for smp_rendezvous(). */
3956 }
3957 
3958 /*
3959  * Write given entry to TLB1 hardware.
3960  */
3961 static void
3962 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3963 {
3964 	struct tlbwrite_args args;
3965 
3966 	args.e = e;
3967 	args.idx = idx;
3968 
3969 #ifdef SMP
3970 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3971 		mb();
3972 		smp_rendezvous(tlb1_write_entry_sync,
3973 		    tlb1_write_entry_int,
3974 		    tlb1_write_entry_sync, &args);
3975 	} else
3976 #endif
3977 	{
3978 		register_t msr;
3979 
3980 		msr = mfmsr();
3981 		__asm __volatile("wrteei 0");
3982 		tlb1_write_entry_int(&args);
3983 		mtmsr(msr);
3984 	}
3985 }
3986 
3987 /*
3988  * Return the largest uint value log such that 2^log <= num.
3989  */
3990 static unsigned int
3991 ilog2(unsigned long num)
3992 {
3993 	long lz;
3994 
3995 #ifdef __powerpc64__
3996 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
3997 	return (63 - lz);
3998 #else
3999 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
4000 	return (31 - lz);
4001 #endif
4002 }
4003 
4004 /*
4005  * Convert TLB TSIZE value to mapped region size.
4006  */
4007 static vm_size_t
4008 tsize2size(unsigned int tsize)
4009 {
4010 
4011 	/*
4012 	 * size = 4^tsize KB
4013 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
4014 	 */
4015 
4016 	return ((1 << (2 * tsize)) * 1024);
4017 }
4018 
4019 /*
4020  * Convert region size (must be power of 4) to TLB TSIZE value.
4021  */
4022 static unsigned int
4023 size2tsize(vm_size_t size)
4024 {
4025 
4026 	return (ilog2(size) / 2 - 5);
4027 }
4028 
4029 /*
4030  * Register permanent kernel mapping in TLB1.
4031  *
4032  * Entries are created starting from index 0 (current free entry is
4033  * kept in tlb1_idx) and are not supposed to be invalidated.
4034  */
4035 int
4036 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
4037     uint32_t flags)
4038 {
4039 	tlb_entry_t e;
4040 	uint32_t ts, tid;
4041 	int tsize, index;
4042 
4043 	for (index = 0; index < TLB1_ENTRIES; index++) {
4044 		tlb1_read_entry(&e, index);
4045 		if ((e.mas1 & MAS1_VALID) == 0)
4046 			break;
4047 		/* Check if we're just updating the flags, and update them. */
4048 		if (e.phys == pa && e.virt == va && e.size == size) {
4049 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
4050 			tlb1_write_entry(&e, index);
4051 			return (0);
4052 		}
4053 	}
4054 	if (index >= TLB1_ENTRIES) {
4055 		printf("tlb1_set_entry: TLB1 full!\n");
4056 		return (-1);
4057 	}
4058 
4059 	/* Convert size to TSIZE */
4060 	tsize = size2tsize(size);
4061 
4062 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
4063 	/* XXX TS is hard coded to 0 for now as we only use single address space */
4064 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
4065 
4066 	e.phys = pa;
4067 	e.virt = va;
4068 	e.size = size;
4069 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
4070 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
4071 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
4072 
4073 	/* Set supervisor RWX permission bits */
4074 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
4075 	e.mas7 = (pa >> 32) & MAS7_RPN;
4076 
4077 	tlb1_write_entry(&e, index);
4078 
4079 	/*
4080 	 * XXX in general TLB1 updates should be propagated between CPUs,
4081 	 * since current design assumes to have the same TLB1 set-up on all
4082 	 * cores.
4083 	 */
4084 	return (0);
4085 }
4086 
4087 /*
4088  * Map in contiguous RAM region into the TLB1 using maximum of
4089  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
4090  *
4091  * If necessary round up last entry size and return total size
4092  * used by all allocated entries.
4093  */
4094 vm_size_t
4095 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
4096 {
4097 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
4098 	vm_size_t mapped, pgsz, base, mask;
4099 	int idx, nents;
4100 
4101 	/* Round up to the next 1M */
4102 	size = roundup2(size, 1 << 20);
4103 
4104 	mapped = 0;
4105 	idx = 0;
4106 	base = va;
4107 	pgsz = 64*1024*1024;
4108 	while (mapped < size) {
4109 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
4110 			while (pgsz > (size - mapped))
4111 				pgsz >>= 2;
4112 			pgs[idx++] = pgsz;
4113 			mapped += pgsz;
4114 		}
4115 
4116 		/* We under-map. Correct for this. */
4117 		if (mapped < size) {
4118 			while (pgs[idx - 1] == pgsz) {
4119 				idx--;
4120 				mapped -= pgsz;
4121 			}
4122 			/* XXX We may increase beyond out starting point. */
4123 			pgsz <<= 2;
4124 			pgs[idx++] = pgsz;
4125 			mapped += pgsz;
4126 		}
4127 	}
4128 
4129 	nents = idx;
4130 	mask = pgs[0] - 1;
4131 	/* Align address to the boundary */
4132 	if (va & mask) {
4133 		va = (va + mask) & ~mask;
4134 		pa = (pa + mask) & ~mask;
4135 	}
4136 
4137 	for (idx = 0; idx < nents; idx++) {
4138 		pgsz = pgs[idx];
4139 		debugf("%u: %llx -> %jx, size=%jx\n", idx, pa,
4140 		    (uintmax_t)va, (uintmax_t)pgsz);
4141 		tlb1_set_entry(va, pa, pgsz,
4142 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4143 		pa += pgsz;
4144 		va += pgsz;
4145 	}
4146 
4147 	mapped = (va - base);
4148 	if (bootverbose)
4149 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
4150 		    mapped, mapped - size);
4151 	return (mapped);
4152 }
4153 
4154 /*
4155  * TLB1 initialization routine, to be called after the very first
4156  * assembler level setup done in locore.S.
4157  */
4158 void
4159 tlb1_init()
4160 {
4161 	uint32_t mas0, mas1, mas2, mas3, mas7;
4162 	uint32_t tsz;
4163 
4164 	tlb1_get_tlbconf();
4165 
4166 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4167 	mtspr(SPR_MAS0, mas0);
4168 	__asm __volatile("isync; tlbre");
4169 
4170 	mas1 = mfspr(SPR_MAS1);
4171 	mas2 = mfspr(SPR_MAS2);
4172 	mas3 = mfspr(SPR_MAS3);
4173 	mas7 = mfspr(SPR_MAS7);
4174 
4175 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4176 	    (mas3 & MAS3_RPN);
4177 
4178 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4179 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4180 
4181 	/* Setup TLB miss defaults */
4182 	set_mas4_defaults();
4183 }
4184 
4185 /*
4186  * pmap_early_io_unmap() should be used in short conjunction with
4187  * pmap_early_io_map(), as in the following snippet:
4188  *
4189  * x = pmap_early_io_map(...);
4190  * <do something with x>
4191  * pmap_early_io_unmap(x, size);
4192  *
4193  * And avoiding more allocations between.
4194  */
4195 void
4196 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4197 {
4198 	int i;
4199 	tlb_entry_t e;
4200 	vm_size_t isize;
4201 
4202 	size = roundup(size, PAGE_SIZE);
4203 	isize = size;
4204 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4205 		tlb1_read_entry(&e, i);
4206 		if (!(e.mas1 & MAS1_VALID))
4207 			continue;
4208 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4209 			size -= e.size;
4210 			e.mas1 &= ~MAS1_VALID;
4211 			tlb1_write_entry(&e, i);
4212 		}
4213 	}
4214 	if (tlb1_map_base == va + isize)
4215 		tlb1_map_base -= isize;
4216 }
4217 
4218 vm_offset_t
4219 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4220 {
4221 	vm_paddr_t pa_base;
4222 	vm_offset_t va, sz;
4223 	int i;
4224 	tlb_entry_t e;
4225 
4226 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4227 
4228 	for (i = 0; i < TLB1_ENTRIES; i++) {
4229 		tlb1_read_entry(&e, i);
4230 		if (!(e.mas1 & MAS1_VALID))
4231 			continue;
4232 		if (pa >= e.phys && (pa + size) <=
4233 		    (e.phys + e.size))
4234 			return (e.virt + (pa - e.phys));
4235 	}
4236 
4237 	pa_base = rounddown(pa, PAGE_SIZE);
4238 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4239 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4240 	va = tlb1_map_base + (pa - pa_base);
4241 
4242 	do {
4243 		sz = 1 << (ilog2(size) & ~1);
4244 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4245 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4246 		size -= sz;
4247 		pa_base += sz;
4248 		tlb1_map_base += sz;
4249 	} while (size > 0);
4250 
4251 	return (va);
4252 }
4253 
4254 void
4255 pmap_track_page(pmap_t pmap, vm_offset_t va)
4256 {
4257 	vm_paddr_t pa;
4258 	vm_page_t page;
4259 	struct pv_entry *pve;
4260 
4261 	va = trunc_page(va);
4262 	pa = pmap_kextract(va);
4263 	page = PHYS_TO_VM_PAGE(pa);
4264 
4265 	rw_wlock(&pvh_global_lock);
4266 	PMAP_LOCK(pmap);
4267 
4268 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4269 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4270 			goto out;
4271 		}
4272 	}
4273 	page->md.pv_tracked = true;
4274 	pv_insert(pmap, va, page);
4275 out:
4276 	PMAP_UNLOCK(pmap);
4277 	rw_wunlock(&pvh_global_lock);
4278 }
4279 
4280 
4281 /*
4282  * Setup MAS4 defaults.
4283  * These values are loaded to MAS0-2 on a TLB miss.
4284  */
4285 static void
4286 set_mas4_defaults(void)
4287 {
4288 	uint32_t mas4;
4289 
4290 	/* Defaults: TLB0, PID0, TSIZED=4K */
4291 	mas4 = MAS4_TLBSELD0;
4292 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4293 #ifdef SMP
4294 	mas4 |= MAS4_MD;
4295 #endif
4296 	mtspr(SPR_MAS4, mas4);
4297 	__asm __volatile("isync");
4298 }
4299 
4300 
4301 /*
4302  * Return 0 if the physical IO range is encompassed by one of the
4303  * the TLB1 entries, otherwise return related error code.
4304  */
4305 static int
4306 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4307 {
4308 	uint32_t prot;
4309 	vm_paddr_t pa_start;
4310 	vm_paddr_t pa_end;
4311 	unsigned int entry_tsize;
4312 	vm_size_t entry_size;
4313 	tlb_entry_t e;
4314 
4315 	*va = (vm_offset_t)NULL;
4316 
4317 	tlb1_read_entry(&e, i);
4318 	/* Skip invalid entries */
4319 	if (!(e.mas1 & MAS1_VALID))
4320 		return (EINVAL);
4321 
4322 	/*
4323 	 * The entry must be cache-inhibited, guarded, and r/w
4324 	 * so it can function as an i/o page
4325 	 */
4326 	prot = e.mas2 & (MAS2_I | MAS2_G);
4327 	if (prot != (MAS2_I | MAS2_G))
4328 		return (EPERM);
4329 
4330 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4331 	if (prot != (MAS3_SR | MAS3_SW))
4332 		return (EPERM);
4333 
4334 	/* The address should be within the entry range. */
4335 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4336 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4337 
4338 	entry_size = tsize2size(entry_tsize);
4339 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4340 	    (e.mas3 & MAS3_RPN);
4341 	pa_end = pa_start + entry_size;
4342 
4343 	if ((pa < pa_start) || ((pa + size) > pa_end))
4344 		return (ERANGE);
4345 
4346 	/* Return virtual address of this mapping. */
4347 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4348 	return (0);
4349 }
4350 
4351 /*
4352  * Invalidate all TLB0 entries which match the given TID. Note this is
4353  * dedicated for cases when invalidations should NOT be propagated to other
4354  * CPUs.
4355  */
4356 static void
4357 tid_flush(tlbtid_t tid)
4358 {
4359 	register_t msr;
4360 	uint32_t mas0, mas1, mas2;
4361 	int entry, way;
4362 
4363 
4364 	/* Don't evict kernel translations */
4365 	if (tid == TID_KERNEL)
4366 		return;
4367 
4368 	msr = mfmsr();
4369 	__asm __volatile("wrteei 0");
4370 
4371 	/*
4372 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
4373 	 * it for PID invalidation.
4374 	 */
4375 	switch ((mfpvr() >> 16) & 0xffff) {
4376 	case FSL_E500mc:
4377 	case FSL_E5500:
4378 	case FSL_E6500:
4379 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
4380 		/* tlbilxpid */
4381 		__asm __volatile("isync; .long 0x7c000024; isync; msync");
4382 		mtmsr(msr);
4383 		return;
4384 	}
4385 
4386 	for (way = 0; way < TLB0_WAYS; way++)
4387 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4388 
4389 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4390 			mtspr(SPR_MAS0, mas0);
4391 
4392 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4393 			mtspr(SPR_MAS2, mas2);
4394 
4395 			__asm __volatile("isync; tlbre");
4396 
4397 			mas1 = mfspr(SPR_MAS1);
4398 
4399 			if (!(mas1 & MAS1_VALID))
4400 				continue;
4401 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4402 				continue;
4403 			mas1 &= ~MAS1_VALID;
4404 			mtspr(SPR_MAS1, mas1);
4405 			__asm __volatile("isync; tlbwe; isync; msync");
4406 		}
4407 	mtmsr(msr);
4408 }
4409 
4410 #ifdef DDB
4411 /* Print out contents of the MAS registers for each TLB0 entry */
4412 static void
4413 #ifdef __powerpc64__
4414 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
4415 #else
4416 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
4417 #endif
4418     uint32_t mas7)
4419 {
4420 	int as;
4421 	char desc[3];
4422 	tlbtid_t tid;
4423 	vm_size_t size;
4424 	unsigned int tsize;
4425 
4426 	desc[2] = '\0';
4427 	if (mas1 & MAS1_VALID)
4428 		desc[0] = 'V';
4429 	else
4430 		desc[0] = ' ';
4431 
4432 	if (mas1 & MAS1_IPROT)
4433 		desc[1] = 'P';
4434 	else
4435 		desc[1] = ' ';
4436 
4437 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
4438 	tid = MAS1_GETTID(mas1);
4439 
4440 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4441 	size = 0;
4442 	if (tsize)
4443 		size = tsize2size(tsize);
4444 
4445 	printf("%3d: (%s) [AS=%d] "
4446 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
4447 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
4448 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
4449 }
4450 
4451 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
4452 {
4453 	uint32_t mas0, mas1, mas3, mas7;
4454 #ifdef __powerpc64__
4455 	uint64_t mas2;
4456 #else
4457 	uint32_t mas2;
4458 #endif
4459 	int entryidx, way, idx;
4460 
4461 	printf("TLB0 entries:\n");
4462 	for (way = 0; way < TLB0_WAYS; way ++)
4463 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
4464 
4465 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4466 			mtspr(SPR_MAS0, mas0);
4467 
4468 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
4469 			mtspr(SPR_MAS2, mas2);
4470 
4471 			__asm __volatile("isync; tlbre");
4472 
4473 			mas1 = mfspr(SPR_MAS1);
4474 			mas2 = mfspr(SPR_MAS2);
4475 			mas3 = mfspr(SPR_MAS3);
4476 			mas7 = mfspr(SPR_MAS7);
4477 
4478 			idx = tlb0_tableidx(mas2, way);
4479 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
4480 		}
4481 }
4482 
4483 /*
4484  * Print out contents of the MAS registers for each TLB1 entry
4485  */
4486 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
4487 {
4488 	uint32_t mas0, mas1, mas3, mas7;
4489 #ifdef __powerpc64__
4490 	uint64_t mas2;
4491 #else
4492 	uint32_t mas2;
4493 #endif
4494 	int i;
4495 
4496 	printf("TLB1 entries:\n");
4497 	for (i = 0; i < TLB1_ENTRIES; i++) {
4498 
4499 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4500 		mtspr(SPR_MAS0, mas0);
4501 
4502 		__asm __volatile("isync; tlbre");
4503 
4504 		mas1 = mfspr(SPR_MAS1);
4505 		mas2 = mfspr(SPR_MAS2);
4506 		mas3 = mfspr(SPR_MAS3);
4507 		mas7 = mfspr(SPR_MAS7);
4508 
4509 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4510 	}
4511 }
4512 #endif
4513