xref: /freebsd/sys/powerpc/booke/pmap.c (revision d8a0fe102c0cfdfcd5b818f850eff09d8536c9bc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - ptbl_bufs-1              : message buffer
63   *               ptbl_bufs - kernel_pdir-1            : kernel page tables
64   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
65   *             kernel_pp2d - .                        : kernel pointers to page directory
66   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
67   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
68   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
69   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
70   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
71   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
72   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
73   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
74   *                         - 0xffff_ffff_ffff_ffff    : device direct map
75   */
76 
77 #include <sys/cdefs.h>
78 __FBSDID("$FreeBSD$");
79 
80 #include "opt_kstack_pages.h"
81 
82 #include <sys/param.h>
83 #include <sys/conf.h>
84 #include <sys/malloc.h>
85 #include <sys/ktr.h>
86 #include <sys/proc.h>
87 #include <sys/user.h>
88 #include <sys/queue.h>
89 #include <sys/systm.h>
90 #include <sys/kernel.h>
91 #include <sys/kerneldump.h>
92 #include <sys/linker.h>
93 #include <sys/msgbuf.h>
94 #include <sys/lock.h>
95 #include <sys/mutex.h>
96 #include <sys/rwlock.h>
97 #include <sys/sched.h>
98 #include <sys/smp.h>
99 #include <sys/vmmeter.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_kern.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_param.h>
108 #include <vm/vm_map.h>
109 #include <vm/vm_pager.h>
110 #include <vm/uma.h>
111 
112 #include <machine/_inttypes.h>
113 #include <machine/cpu.h>
114 #include <machine/pcb.h>
115 #include <machine/platform.h>
116 
117 #include <machine/tlb.h>
118 #include <machine/spr.h>
119 #include <machine/md_var.h>
120 #include <machine/mmuvar.h>
121 #include <machine/pmap.h>
122 #include <machine/pte.h>
123 
124 #include "mmu_if.h"
125 
126 #define	SPARSE_MAPDEV
127 #ifdef  DEBUG
128 #define debugf(fmt, args...) printf(fmt, ##args)
129 #else
130 #define debugf(fmt, args...)
131 #endif
132 
133 #ifdef __powerpc64__
134 #define	PRI0ptrX	"016lx"
135 #else
136 #define	PRI0ptrX	"08x"
137 #endif
138 
139 #define TODO			panic("%s: not implemented", __func__);
140 
141 extern unsigned char _etext[];
142 extern unsigned char _end[];
143 
144 extern uint32_t *bootinfo;
145 
146 vm_paddr_t kernload;
147 vm_offset_t kernstart;
148 vm_size_t kernsize;
149 
150 /* Message buffer and tables. */
151 static vm_offset_t data_start;
152 static vm_size_t data_end;
153 
154 /* Phys/avail memory regions. */
155 static struct mem_region *availmem_regions;
156 static int availmem_regions_sz;
157 static struct mem_region *physmem_regions;
158 static int physmem_regions_sz;
159 
160 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
161 static vm_offset_t zero_page_va;
162 static struct mtx zero_page_mutex;
163 
164 static struct mtx tlbivax_mutex;
165 
166 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
167 static vm_offset_t copy_page_src_va;
168 static vm_offset_t copy_page_dst_va;
169 static struct mtx copy_page_mutex;
170 
171 /**************************************************************************/
172 /* PMAP */
173 /**************************************************************************/
174 
175 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
176     vm_prot_t, u_int flags, int8_t psind);
177 
178 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
179 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
180 #ifdef __powerpc64__
181 unsigned int kernel_pdirs;
182 #endif
183 
184 /*
185  * If user pmap is processed with mmu_booke_remove and the resident count
186  * drops to 0, there are no more pages to remove, so we need not continue.
187  */
188 #define PMAP_REMOVE_DONE(pmap) \
189 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
190 
191 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
192 extern int elf32_nxstack;
193 #endif
194 
195 /**************************************************************************/
196 /* TLB and TID handling */
197 /**************************************************************************/
198 
199 /* Translation ID busy table */
200 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
201 
202 /*
203  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
204  * core revisions and should be read from h/w registers during early config.
205  */
206 uint32_t tlb0_entries;
207 uint32_t tlb0_ways;
208 uint32_t tlb0_entries_per_way;
209 uint32_t tlb1_entries;
210 
211 #define TLB0_ENTRIES		(tlb0_entries)
212 #define TLB0_WAYS		(tlb0_ways)
213 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
214 
215 #define TLB1_ENTRIES (tlb1_entries)
216 
217 static vm_offset_t tlb1_map_base = VM_MAXUSER_ADDRESS + PAGE_SIZE;
218 
219 static tlbtid_t tid_alloc(struct pmap *);
220 static void tid_flush(tlbtid_t tid);
221 
222 #ifdef __powerpc64__
223 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
224 #else
225 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
226 #endif
227 
228 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
229 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
230 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
231 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
232 
233 static vm_size_t tsize2size(unsigned int);
234 static unsigned int size2tsize(vm_size_t);
235 static unsigned int ilog2(unsigned int);
236 
237 static void set_mas4_defaults(void);
238 
239 static inline void tlb0_flush_entry(vm_offset_t);
240 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
241 
242 /**************************************************************************/
243 /* Page table management */
244 /**************************************************************************/
245 
246 static struct rwlock_padalign pvh_global_lock;
247 
248 /* Data for the pv entry allocation mechanism */
249 static uma_zone_t pvzone;
250 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
251 
252 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
253 
254 #ifndef PMAP_SHPGPERPROC
255 #define PMAP_SHPGPERPROC	200
256 #endif
257 
258 static void ptbl_init(void);
259 static struct ptbl_buf *ptbl_buf_alloc(void);
260 static void ptbl_buf_free(struct ptbl_buf *);
261 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
262 
263 #ifdef __powerpc64__
264 static pte_t *ptbl_alloc(mmu_t, pmap_t, pte_t **,
265 			 unsigned int, boolean_t);
266 static void ptbl_free(mmu_t, pmap_t, pte_t **, unsigned int);
267 static void ptbl_hold(mmu_t, pmap_t, pte_t **, unsigned int);
268 static int ptbl_unhold(mmu_t, pmap_t, vm_offset_t);
269 #else
270 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
271 static void ptbl_free(mmu_t, pmap_t, unsigned int);
272 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
273 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
274 #endif
275 
276 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
277 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
278 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
279 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
280 static void kernel_pte_alloc(vm_offset_t, vm_offset_t, vm_offset_t);
281 
282 static pv_entry_t pv_alloc(void);
283 static void pv_free(pv_entry_t);
284 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
285 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
286 
287 static void booke_pmap_init_qpages(void);
288 
289 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
290 #ifdef __powerpc64__
291 #define PTBL_BUFS               (16UL * 16 * 16)
292 #else
293 #define PTBL_BUFS		(128 * 16)
294 #endif
295 
296 struct ptbl_buf {
297 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
298 	vm_offset_t kva;		/* va of mapping */
299 };
300 
301 /* ptbl free list and a lock used for access synchronization. */
302 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
303 static struct mtx ptbl_buf_freelist_lock;
304 
305 /* Base address of kva space allocated fot ptbl bufs. */
306 static vm_offset_t ptbl_buf_pool_vabase;
307 
308 /* Pointer to ptbl_buf structures. */
309 static struct ptbl_buf *ptbl_bufs;
310 
311 #ifdef SMP
312 extern tlb_entry_t __boot_tlb1[];
313 void pmap_bootstrap_ap(volatile uint32_t *);
314 #endif
315 
316 /*
317  * Kernel MMU interface
318  */
319 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
320 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
321     vm_size_t, vm_offset_t);
322 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
323 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
324     vm_offset_t, vm_page_t *, vm_offset_t, int);
325 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
326     vm_prot_t, u_int flags, int8_t psind);
327 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
328     vm_page_t, vm_prot_t);
329 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
330     vm_prot_t);
331 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
332 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
333     vm_prot_t);
334 static void		mmu_booke_init(mmu_t);
335 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
336 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
337 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
338 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
339 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
340     int);
341 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
342     vm_paddr_t *);
343 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
344     vm_object_t, vm_pindex_t, vm_size_t);
345 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
346 static void		mmu_booke_page_init(mmu_t, vm_page_t);
347 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
348 static void		mmu_booke_pinit(mmu_t, pmap_t);
349 static void		mmu_booke_pinit0(mmu_t, pmap_t);
350 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
351     vm_prot_t);
352 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
353 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
354 static void		mmu_booke_release(mmu_t, pmap_t);
355 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
356 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
357 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
358 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
359 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
360 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
361 static void		mmu_booke_activate(mmu_t, struct thread *);
362 static void		mmu_booke_deactivate(mmu_t, struct thread *);
363 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
364 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
365 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
366 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
367 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
368 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
369 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
370 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
371 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
372 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
373     vm_size_t);
374 static void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
375     void **);
376 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
377     void *);
378 static void		mmu_booke_scan_init(mmu_t);
379 static vm_offset_t	mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m);
380 static void		mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr);
381 static int		mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr,
382     vm_size_t sz, vm_memattr_t mode);
383 
384 static mmu_method_t mmu_booke_methods[] = {
385 	/* pmap dispatcher interface */
386 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
387 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
388 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
389 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
390 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
391 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
392 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
393 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
394 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
395 	MMUMETHOD(mmu_init,		mmu_booke_init),
396 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
397 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
398 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
399 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
400 	MMUMETHOD(mmu_map,		mmu_booke_map),
401 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
402 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
403 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
404 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
405 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
406 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
407 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
408 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
409 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
410 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
411 	MMUMETHOD(mmu_release,		mmu_booke_release),
412 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
413 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
414 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
415 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
416 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
417 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
418 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
419 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
420 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
421 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
422 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
423 
424 	/* Internal interfaces */
425 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
426 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
427 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
428 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
429 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
430 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
431 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
432 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
433 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
434 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
435 
436 	/* dumpsys() support */
437 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
438 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
439 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
440 
441 	{ 0, 0 }
442 };
443 
444 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
445 
446 static __inline uint32_t
447 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
448 {
449 	uint32_t attrib;
450 	int i;
451 
452 	if (ma != VM_MEMATTR_DEFAULT) {
453 		switch (ma) {
454 		case VM_MEMATTR_UNCACHEABLE:
455 			return (MAS2_I | MAS2_G);
456 		case VM_MEMATTR_WRITE_COMBINING:
457 		case VM_MEMATTR_WRITE_BACK:
458 		case VM_MEMATTR_PREFETCHABLE:
459 			return (MAS2_I);
460 		case VM_MEMATTR_WRITE_THROUGH:
461 			return (MAS2_W | MAS2_M);
462 		case VM_MEMATTR_CACHEABLE:
463 			return (MAS2_M);
464 		}
465 	}
466 
467 	/*
468 	 * Assume the page is cache inhibited and access is guarded unless
469 	 * it's in our available memory array.
470 	 */
471 	attrib = _TLB_ENTRY_IO;
472 	for (i = 0; i < physmem_regions_sz; i++) {
473 		if ((pa >= physmem_regions[i].mr_start) &&
474 		    (pa < (physmem_regions[i].mr_start +
475 		     physmem_regions[i].mr_size))) {
476 			attrib = _TLB_ENTRY_MEM;
477 			break;
478 		}
479 	}
480 
481 	return (attrib);
482 }
483 
484 static inline void
485 tlb_miss_lock(void)
486 {
487 #ifdef SMP
488 	struct pcpu *pc;
489 
490 	if (!smp_started)
491 		return;
492 
493 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
494 		if (pc != pcpup) {
495 
496 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
497 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
498 
499 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
500 			    ("tlb_miss_lock: tried to lock self"));
501 
502 			tlb_lock(pc->pc_booke_tlb_lock);
503 
504 			CTR1(KTR_PMAP, "%s: locked", __func__);
505 		}
506 	}
507 #endif
508 }
509 
510 static inline void
511 tlb_miss_unlock(void)
512 {
513 #ifdef SMP
514 	struct pcpu *pc;
515 
516 	if (!smp_started)
517 		return;
518 
519 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
520 		if (pc != pcpup) {
521 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
522 			    __func__, pc->pc_cpuid);
523 
524 			tlb_unlock(pc->pc_booke_tlb_lock);
525 
526 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
527 		}
528 	}
529 #endif
530 }
531 
532 /* Return number of entries in TLB0. */
533 static __inline void
534 tlb0_get_tlbconf(void)
535 {
536 	uint32_t tlb0_cfg;
537 
538 	tlb0_cfg = mfspr(SPR_TLB0CFG);
539 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
540 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
541 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
542 }
543 
544 /* Return number of entries in TLB1. */
545 static __inline void
546 tlb1_get_tlbconf(void)
547 {
548 	uint32_t tlb1_cfg;
549 
550 	tlb1_cfg = mfspr(SPR_TLB1CFG);
551 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
552 }
553 
554 /**************************************************************************/
555 /* Page table related */
556 /**************************************************************************/
557 
558 #ifdef __powerpc64__
559 /* Initialize pool of kva ptbl buffers. */
560 static void
561 ptbl_init(void)
562 {
563 	int		i;
564 
565 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
566 	TAILQ_INIT(&ptbl_buf_freelist);
567 
568 	for (i = 0; i < PTBL_BUFS; i++) {
569 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase +
570 		    i * MAX(PTBL_PAGES,PDIR_PAGES) * PAGE_SIZE;
571 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
572 	}
573 }
574 
575 /* Get an sf_buf from the freelist. */
576 static struct ptbl_buf *
577 ptbl_buf_alloc(void)
578 {
579 	struct ptbl_buf *buf;
580 
581 	mtx_lock(&ptbl_buf_freelist_lock);
582 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
583 	if (buf != NULL)
584 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
585 	mtx_unlock(&ptbl_buf_freelist_lock);
586 
587 	return (buf);
588 }
589 
590 /* Return ptbl buff to free pool. */
591 static void
592 ptbl_buf_free(struct ptbl_buf *buf)
593 {
594 	mtx_lock(&ptbl_buf_freelist_lock);
595 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
596 	mtx_unlock(&ptbl_buf_freelist_lock);
597 }
598 
599 /*
600  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
601  */
602 static void
603 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t * ptbl)
604 {
605 	struct ptbl_buf *pbuf;
606 
607 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) {
608 		if (pbuf->kva == (vm_offset_t) ptbl) {
609 			/* Remove from pmap ptbl buf list. */
610 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
611 
612 			/* Free corresponding ptbl buf. */
613 			ptbl_buf_free(pbuf);
614 
615 			break;
616 		}
617 	}
618 }
619 
620 /* Get a pointer to a PTE in a page table. */
621 static __inline pte_t *
622 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
623 {
624 	pte_t         **pdir;
625 	pte_t          *ptbl;
626 
627 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
628 
629 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
630 	if (!pdir)
631 		return NULL;
632 	ptbl = pdir[PDIR_IDX(va)];
633 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
634 }
635 
636 /*
637  * Search the list of allocated pdir bufs and find on list of allocated pdirs
638  */
639 static void
640 ptbl_free_pmap_pdir(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
641 {
642 	struct ptbl_buf *pbuf;
643 
644 	TAILQ_FOREACH(pbuf, &pmap->pm_pdir_list, link) {
645 		if (pbuf->kva == (vm_offset_t) pdir) {
646 			/* Remove from pmap ptbl buf list. */
647 			TAILQ_REMOVE(&pmap->pm_pdir_list, pbuf, link);
648 
649 			/* Free corresponding pdir buf. */
650 			ptbl_buf_free(pbuf);
651 
652 			break;
653 		}
654 	}
655 }
656 /* Free pdir pages and invalidate pdir entry. */
657 static void
658 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
659 {
660 	pte_t         **pdir;
661 	vm_paddr_t	pa;
662 	vm_offset_t	va;
663 	vm_page_t	m;
664 	int		i;
665 
666 	pdir = pmap->pm_pp2d[pp2d_idx];
667 
668 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
669 
670 	pmap->pm_pp2d[pp2d_idx] = NULL;
671 
672 	for (i = 0; i < PDIR_PAGES; i++) {
673 		va = ((vm_offset_t) pdir + (i * PAGE_SIZE));
674 		pa = pte_vatopa(mmu, kernel_pmap, va);
675 		m = PHYS_TO_VM_PAGE(pa);
676 		vm_page_free_zero(m);
677 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
678 		pmap_kremove(va);
679 	}
680 
681 	ptbl_free_pmap_pdir(mmu, pmap, pdir);
682 }
683 
684 /*
685  * Decrement pdir pages hold count and attempt to free pdir pages. Called
686  * when removing directory entry from pdir.
687  *
688  * Return 1 if pdir pages were freed.
689  */
690 static int
691 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
692 {
693 	pte_t         **pdir;
694 	vm_paddr_t	pa;
695 	vm_page_t	m;
696 	int		i;
697 
698 	KASSERT((pmap != kernel_pmap),
699 		("pdir_unhold: unholding kernel pdir!"));
700 
701 	pdir = pmap->pm_pp2d[pp2d_idx];
702 
703 	KASSERT(((vm_offset_t) pdir >= VM_MIN_KERNEL_ADDRESS),
704 	    ("pdir_unhold: non kva pdir"));
705 
706 	/* decrement hold count */
707 	for (i = 0; i < PDIR_PAGES; i++) {
708 		pa = pte_vatopa(mmu, kernel_pmap,
709 		    (vm_offset_t) pdir + (i * PAGE_SIZE));
710 		m = PHYS_TO_VM_PAGE(pa);
711 		m->wire_count--;
712 	}
713 
714 	/*
715 	 * Free pdir pages if there are no dir entries in this pdir.
716 	 * wire_count has the same value for all ptbl pages, so check the
717 	 * last page.
718 	 */
719 	if (m->wire_count == 0) {
720 		pdir_free(mmu, pmap, pp2d_idx);
721 		return (1);
722 	}
723 	return (0);
724 }
725 
726 /*
727  * Increment hold count for pdir pages. This routine is used when new ptlb
728  * entry is being inserted into pdir.
729  */
730 static void
731 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
732 {
733 	vm_paddr_t	pa;
734 	vm_page_t	m;
735 	int		i;
736 
737 	KASSERT((pmap != kernel_pmap),
738 		("pdir_hold: holding kernel pdir!"));
739 
740 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
741 
742 	for (i = 0; i < PDIR_PAGES; i++) {
743 		pa = pte_vatopa(mmu, kernel_pmap,
744 				(vm_offset_t) pdir + (i * PAGE_SIZE));
745 		m = PHYS_TO_VM_PAGE(pa);
746 		m->wire_count++;
747 	}
748 }
749 
750 /* Allocate page table. */
751 static pte_t   *
752 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
753     boolean_t nosleep)
754 {
755 	vm_page_t	mtbl  [PTBL_PAGES];
756 	vm_page_t	m;
757 	struct ptbl_buf *pbuf;
758 	unsigned int	pidx;
759 	pte_t          *ptbl;
760 	int		i, j;
761 	int		req;
762 
763 	KASSERT((pdir[pdir_idx] == NULL),
764 		("%s: valid ptbl entry exists!", __func__));
765 
766 	pbuf = ptbl_buf_alloc();
767 	if (pbuf == NULL)
768 		panic("%s: couldn't alloc kernel virtual memory", __func__);
769 
770 	ptbl = (pte_t *) pbuf->kva;
771 
772 	for (i = 0; i < PTBL_PAGES; i++) {
773 		pidx = (PTBL_PAGES * pdir_idx) + i;
774 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
775 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
776 			PMAP_UNLOCK(pmap);
777 			rw_wunlock(&pvh_global_lock);
778 			if (nosleep) {
779 				ptbl_free_pmap_ptbl(pmap, ptbl);
780 				for (j = 0; j < i; j++)
781 					vm_page_free(mtbl[j]);
782 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
783 				return (NULL);
784 			}
785 			VM_WAIT;
786 			rw_wlock(&pvh_global_lock);
787 			PMAP_LOCK(pmap);
788 		}
789 		mtbl[i] = m;
790 	}
791 
792 	/* Mapin allocated pages into kernel_pmap. */
793 	mmu_booke_qenter(mmu, (vm_offset_t) ptbl, mtbl, PTBL_PAGES);
794 	/* Zero whole ptbl. */
795 	bzero((caddr_t) ptbl, PTBL_PAGES * PAGE_SIZE);
796 
797 	/* Add pbuf to the pmap ptbl bufs list. */
798 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
799 
800 	return (ptbl);
801 }
802 
803 /* Free ptbl pages and invalidate pdir entry. */
804 static void
805 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
806 {
807 	pte_t          *ptbl;
808 	vm_paddr_t	pa;
809 	vm_offset_t	va;
810 	vm_page_t	m;
811 	int		i;
812 
813 	ptbl = pdir[pdir_idx];
814 
815 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
816 
817 	pdir[pdir_idx] = NULL;
818 
819 	for (i = 0; i < PTBL_PAGES; i++) {
820 		va = ((vm_offset_t) ptbl + (i * PAGE_SIZE));
821 		pa = pte_vatopa(mmu, kernel_pmap, va);
822 		m = PHYS_TO_VM_PAGE(pa);
823 		vm_page_free_zero(m);
824 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
825 		pmap_kremove(va);
826 	}
827 
828 	ptbl_free_pmap_ptbl(pmap, ptbl);
829 }
830 
831 /*
832  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
833  * when removing pte entry from ptbl.
834  *
835  * Return 1 if ptbl pages were freed.
836  */
837 static int
838 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
839 {
840 	pte_t          *ptbl;
841 	vm_paddr_t	pa;
842 	vm_page_t	m;
843 	u_int		pp2d_idx;
844 	pte_t         **pdir;
845 	u_int		pdir_idx;
846 	int		i;
847 
848 	pp2d_idx = PP2D_IDX(va);
849 	pdir_idx = PDIR_IDX(va);
850 
851 	KASSERT((pmap != kernel_pmap),
852 		("ptbl_unhold: unholding kernel ptbl!"));
853 
854 	pdir = pmap->pm_pp2d[pp2d_idx];
855 	ptbl = pdir[pdir_idx];
856 
857 	KASSERT(((vm_offset_t) ptbl >= VM_MIN_KERNEL_ADDRESS),
858 	    ("ptbl_unhold: non kva ptbl"));
859 
860 	/* decrement hold count */
861 	for (i = 0; i < PTBL_PAGES; i++) {
862 		pa = pte_vatopa(mmu, kernel_pmap,
863 		    (vm_offset_t) ptbl + (i * PAGE_SIZE));
864 		m = PHYS_TO_VM_PAGE(pa);
865 		m->wire_count--;
866 	}
867 
868 	/*
869 	 * Free ptbl pages if there are no pte entries in this ptbl.
870 	 * wire_count has the same value for all ptbl pages, so check the
871 	 * last page.
872 	 */
873 	if (m->wire_count == 0) {
874 		/* A pair of indirect entries might point to this ptbl page */
875 #if 0
876 		tlb_flush_entry(pmap, va & ~((2UL * PAGE_SIZE_1M) - 1),
877 				TLB_SIZE_1M, MAS6_SIND);
878 		tlb_flush_entry(pmap, (va & ~((2UL * PAGE_SIZE_1M) - 1)) | PAGE_SIZE_1M,
879 				TLB_SIZE_1M, MAS6_SIND);
880 #endif
881 		ptbl_free(mmu, pmap, pdir, pdir_idx);
882 		pdir_unhold(mmu, pmap, pp2d_idx);
883 		return (1);
884 	}
885 	return (0);
886 }
887 
888 /*
889  * Increment hold count for ptbl pages. This routine is used when new pte
890  * entry is being inserted into ptbl.
891  */
892 static void
893 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
894 {
895 	vm_paddr_t	pa;
896 	pte_t          *ptbl;
897 	vm_page_t	m;
898 	int		i;
899 
900 	KASSERT((pmap != kernel_pmap),
901 		("ptbl_hold: holding kernel ptbl!"));
902 
903 	ptbl = pdir[pdir_idx];
904 
905 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
906 
907 	for (i = 0; i < PTBL_PAGES; i++) {
908 		pa = pte_vatopa(mmu, kernel_pmap,
909 				(vm_offset_t) ptbl + (i * PAGE_SIZE));
910 		m = PHYS_TO_VM_PAGE(pa);
911 		m->wire_count++;
912 	}
913 }
914 #else
915 
916 /* Initialize pool of kva ptbl buffers. */
917 static void
918 ptbl_init(void)
919 {
920 	int i;
921 
922 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
923 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
924 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
925 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
926 
927 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
928 	TAILQ_INIT(&ptbl_buf_freelist);
929 
930 	for (i = 0; i < PTBL_BUFS; i++) {
931 		ptbl_bufs[i].kva =
932 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
933 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
934 	}
935 }
936 
937 /* Get a ptbl_buf from the freelist. */
938 static struct ptbl_buf *
939 ptbl_buf_alloc(void)
940 {
941 	struct ptbl_buf *buf;
942 
943 	mtx_lock(&ptbl_buf_freelist_lock);
944 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
945 	if (buf != NULL)
946 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
947 	mtx_unlock(&ptbl_buf_freelist_lock);
948 
949 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
950 
951 	return (buf);
952 }
953 
954 /* Return ptbl buff to free pool. */
955 static void
956 ptbl_buf_free(struct ptbl_buf *buf)
957 {
958 
959 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
960 
961 	mtx_lock(&ptbl_buf_freelist_lock);
962 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
963 	mtx_unlock(&ptbl_buf_freelist_lock);
964 }
965 
966 /*
967  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
968  */
969 static void
970 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
971 {
972 	struct ptbl_buf *pbuf;
973 
974 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
975 
976 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
977 
978 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
979 		if (pbuf->kva == (vm_offset_t)ptbl) {
980 			/* Remove from pmap ptbl buf list. */
981 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
982 
983 			/* Free corresponding ptbl buf. */
984 			ptbl_buf_free(pbuf);
985 			break;
986 		}
987 }
988 
989 /* Allocate page table. */
990 static pte_t *
991 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
992 {
993 	vm_page_t mtbl[PTBL_PAGES];
994 	vm_page_t m;
995 	struct ptbl_buf *pbuf;
996 	unsigned int pidx;
997 	pte_t *ptbl;
998 	int i, j;
999 
1000 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1001 	    (pmap == kernel_pmap), pdir_idx);
1002 
1003 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1004 	    ("ptbl_alloc: invalid pdir_idx"));
1005 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
1006 	    ("pte_alloc: valid ptbl entry exists!"));
1007 
1008 	pbuf = ptbl_buf_alloc();
1009 	if (pbuf == NULL)
1010 		panic("pte_alloc: couldn't alloc kernel virtual memory");
1011 
1012 	ptbl = (pte_t *)pbuf->kva;
1013 
1014 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
1015 
1016 	for (i = 0; i < PTBL_PAGES; i++) {
1017 		pidx = (PTBL_PAGES * pdir_idx) + i;
1018 		while ((m = vm_page_alloc(NULL, pidx,
1019 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
1020 			PMAP_UNLOCK(pmap);
1021 			rw_wunlock(&pvh_global_lock);
1022 			if (nosleep) {
1023 				ptbl_free_pmap_ptbl(pmap, ptbl);
1024 				for (j = 0; j < i; j++)
1025 					vm_page_free(mtbl[j]);
1026 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
1027 				return (NULL);
1028 			}
1029 			VM_WAIT;
1030 			rw_wlock(&pvh_global_lock);
1031 			PMAP_LOCK(pmap);
1032 		}
1033 		mtbl[i] = m;
1034 	}
1035 
1036 	/* Map allocated pages into kernel_pmap. */
1037 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
1038 
1039 	/* Zero whole ptbl. */
1040 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
1041 
1042 	/* Add pbuf to the pmap ptbl bufs list. */
1043 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
1044 
1045 	return (ptbl);
1046 }
1047 
1048 /* Free ptbl pages and invalidate pdir entry. */
1049 static void
1050 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1051 {
1052 	pte_t *ptbl;
1053 	vm_paddr_t pa;
1054 	vm_offset_t va;
1055 	vm_page_t m;
1056 	int i;
1057 
1058 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1059 	    (pmap == kernel_pmap), pdir_idx);
1060 
1061 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1062 	    ("ptbl_free: invalid pdir_idx"));
1063 
1064 	ptbl = pmap->pm_pdir[pdir_idx];
1065 
1066 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
1067 
1068 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
1069 
1070 	/*
1071 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
1072 	 * don't attempt to look up the page tables we are releasing.
1073 	 */
1074 	mtx_lock_spin(&tlbivax_mutex);
1075 	tlb_miss_lock();
1076 
1077 	pmap->pm_pdir[pdir_idx] = NULL;
1078 
1079 	tlb_miss_unlock();
1080 	mtx_unlock_spin(&tlbivax_mutex);
1081 
1082 	for (i = 0; i < PTBL_PAGES; i++) {
1083 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
1084 		pa = pte_vatopa(mmu, kernel_pmap, va);
1085 		m = PHYS_TO_VM_PAGE(pa);
1086 		vm_page_free_zero(m);
1087 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
1088 		mmu_booke_kremove(mmu, va);
1089 	}
1090 
1091 	ptbl_free_pmap_ptbl(pmap, ptbl);
1092 }
1093 
1094 /*
1095  * Decrement ptbl pages hold count and attempt to free ptbl pages.
1096  * Called when removing pte entry from ptbl.
1097  *
1098  * Return 1 if ptbl pages were freed.
1099  */
1100 static int
1101 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1102 {
1103 	pte_t *ptbl;
1104 	vm_paddr_t pa;
1105 	vm_page_t m;
1106 	int i;
1107 
1108 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1109 	    (pmap == kernel_pmap), pdir_idx);
1110 
1111 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1112 	    ("ptbl_unhold: invalid pdir_idx"));
1113 	KASSERT((pmap != kernel_pmap),
1114 	    ("ptbl_unhold: unholding kernel ptbl!"));
1115 
1116 	ptbl = pmap->pm_pdir[pdir_idx];
1117 
1118 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1119 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1120 	    ("ptbl_unhold: non kva ptbl"));
1121 
1122 	/* decrement hold count */
1123 	for (i = 0; i < PTBL_PAGES; i++) {
1124 		pa = pte_vatopa(mmu, kernel_pmap,
1125 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1126 		m = PHYS_TO_VM_PAGE(pa);
1127 		m->wire_count--;
1128 	}
1129 
1130 	/*
1131 	 * Free ptbl pages if there are no pte etries in this ptbl.
1132 	 * wire_count has the same value for all ptbl pages, so check the last
1133 	 * page.
1134 	 */
1135 	if (m->wire_count == 0) {
1136 		ptbl_free(mmu, pmap, pdir_idx);
1137 
1138 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1139 		return (1);
1140 	}
1141 
1142 	return (0);
1143 }
1144 
1145 /*
1146  * Increment hold count for ptbl pages. This routine is used when a new pte
1147  * entry is being inserted into the ptbl.
1148  */
1149 static void
1150 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1151 {
1152 	vm_paddr_t pa;
1153 	pte_t *ptbl;
1154 	vm_page_t m;
1155 	int i;
1156 
1157 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1158 	    pdir_idx);
1159 
1160 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1161 	    ("ptbl_hold: invalid pdir_idx"));
1162 	KASSERT((pmap != kernel_pmap),
1163 	    ("ptbl_hold: holding kernel ptbl!"));
1164 
1165 	ptbl = pmap->pm_pdir[pdir_idx];
1166 
1167 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1168 
1169 	for (i = 0; i < PTBL_PAGES; i++) {
1170 		pa = pte_vatopa(mmu, kernel_pmap,
1171 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1172 		m = PHYS_TO_VM_PAGE(pa);
1173 		m->wire_count++;
1174 	}
1175 }
1176 #endif
1177 
1178 /* Allocate pv_entry structure. */
1179 pv_entry_t
1180 pv_alloc(void)
1181 {
1182 	pv_entry_t pv;
1183 
1184 	pv_entry_count++;
1185 	if (pv_entry_count > pv_entry_high_water)
1186 		pagedaemon_wakeup();
1187 	pv = uma_zalloc(pvzone, M_NOWAIT);
1188 
1189 	return (pv);
1190 }
1191 
1192 /* Free pv_entry structure. */
1193 static __inline void
1194 pv_free(pv_entry_t pve)
1195 {
1196 
1197 	pv_entry_count--;
1198 	uma_zfree(pvzone, pve);
1199 }
1200 
1201 
1202 /* Allocate and initialize pv_entry structure. */
1203 static void
1204 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1205 {
1206 	pv_entry_t pve;
1207 
1208 	//int su = (pmap == kernel_pmap);
1209 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1210 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1211 
1212 	pve = pv_alloc();
1213 	if (pve == NULL)
1214 		panic("pv_insert: no pv entries!");
1215 
1216 	pve->pv_pmap = pmap;
1217 	pve->pv_va = va;
1218 
1219 	/* add to pv_list */
1220 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1221 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1222 
1223 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1224 
1225 	//debugf("pv_insert: e\n");
1226 }
1227 
1228 /* Destroy pv entry. */
1229 static void
1230 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1231 {
1232 	pv_entry_t pve;
1233 
1234 	//int su = (pmap == kernel_pmap);
1235 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1236 
1237 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1238 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1239 
1240 	/* find pv entry */
1241 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1242 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1243 			/* remove from pv_list */
1244 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1245 			if (TAILQ_EMPTY(&m->md.pv_list))
1246 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1247 
1248 			/* free pv entry struct */
1249 			pv_free(pve);
1250 			break;
1251 		}
1252 	}
1253 
1254 	//debugf("pv_remove: e\n");
1255 }
1256 
1257 #ifdef __powerpc64__
1258 /*
1259  * Clean pte entry, try to free page table page if requested.
1260  *
1261  * Return 1 if ptbl pages were freed, otherwise return 0.
1262  */
1263 static int
1264 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1265 {
1266 	vm_page_t	m;
1267 	pte_t          *pte;
1268 
1269 	pte = pte_find(mmu, pmap, va);
1270 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1271 
1272 	if (!PTE_ISVALID(pte))
1273 		return (0);
1274 
1275 	/* Get vm_page_t for mapped pte. */
1276 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1277 
1278 	if (PTE_ISWIRED(pte))
1279 		pmap->pm_stats.wired_count--;
1280 
1281 	/* Handle managed entry. */
1282 	if (PTE_ISMANAGED(pte)) {
1283 
1284 		/* Handle modified pages. */
1285 		if (PTE_ISMODIFIED(pte))
1286 			vm_page_dirty(m);
1287 
1288 		/* Referenced pages. */
1289 		if (PTE_ISREFERENCED(pte))
1290 			vm_page_aflag_set(m, PGA_REFERENCED);
1291 
1292 		/* Remove pv_entry from pv_list. */
1293 		pv_remove(pmap, va, m);
1294 	} else if (m->md.pv_tracked) {
1295 		pv_remove(pmap, va, m);
1296 		if (TAILQ_EMPTY(&m->md.pv_list))
1297 			m->md.pv_tracked = false;
1298 	}
1299 	mtx_lock_spin(&tlbivax_mutex);
1300 	tlb_miss_lock();
1301 
1302 	tlb0_flush_entry(va);
1303 	*pte = 0;
1304 
1305 	tlb_miss_unlock();
1306 	mtx_unlock_spin(&tlbivax_mutex);
1307 
1308 	pmap->pm_stats.resident_count--;
1309 
1310 	if (flags & PTBL_UNHOLD) {
1311 		return (ptbl_unhold(mmu, pmap, va));
1312 	}
1313 	return (0);
1314 }
1315 
1316 /*
1317  * allocate a page of pointers to page directories, do not preallocate the
1318  * page tables
1319  */
1320 static pte_t  **
1321 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
1322 {
1323 	vm_page_t	mtbl  [PDIR_PAGES];
1324 	vm_page_t	m;
1325 	struct ptbl_buf *pbuf;
1326 	pte_t         **pdir;
1327 	unsigned int	pidx;
1328 	int		i;
1329 	int		req;
1330 
1331 	pbuf = ptbl_buf_alloc();
1332 
1333 	if (pbuf == NULL)
1334 		panic("%s: couldn't alloc kernel virtual memory", __func__);
1335 
1336 	/* Allocate pdir pages, this will sleep! */
1337 	for (i = 0; i < PDIR_PAGES; i++) {
1338 		pidx = (PDIR_PAGES * pp2d_idx) + i;
1339 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
1340 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
1341 			PMAP_UNLOCK(pmap);
1342 			VM_WAIT;
1343 			PMAP_LOCK(pmap);
1344 		}
1345 		mtbl[i] = m;
1346 	}
1347 
1348 	/* Mapin allocated pages into kernel_pmap. */
1349 	pdir = (pte_t **) pbuf->kva;
1350 	pmap_qenter((vm_offset_t) pdir, mtbl, PDIR_PAGES);
1351 
1352 	/* Zero whole pdir. */
1353 	bzero((caddr_t) pdir, PDIR_PAGES * PAGE_SIZE);
1354 
1355 	/* Add pdir to the pmap pdir bufs list. */
1356 	TAILQ_INSERT_TAIL(&pmap->pm_pdir_list, pbuf, link);
1357 
1358 	return pdir;
1359 }
1360 
1361 /*
1362  * Insert PTE for a given page and virtual address.
1363  */
1364 static int
1365 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1366     boolean_t nosleep)
1367 {
1368 	unsigned int	pp2d_idx = PP2D_IDX(va);
1369 	unsigned int	pdir_idx = PDIR_IDX(va);
1370 	unsigned int	ptbl_idx = PTBL_IDX(va);
1371 	pte_t          *ptbl, *pte;
1372 	pte_t         **pdir;
1373 
1374 	/* Get the page directory pointer. */
1375 	pdir = pmap->pm_pp2d[pp2d_idx];
1376 	if (pdir == NULL)
1377 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1378 
1379 	/* Get the page table pointer. */
1380 	ptbl = pdir[pdir_idx];
1381 
1382 	if (ptbl == NULL) {
1383 		/* Allocate page table pages. */
1384 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1385 		if (ptbl == NULL) {
1386 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1387 			return (ENOMEM);
1388 		}
1389 	} else {
1390 		/*
1391 		 * Check if there is valid mapping for requested va, if there
1392 		 * is, remove it.
1393 		 */
1394 		pte = &pdir[pdir_idx][ptbl_idx];
1395 		if (PTE_ISVALID(pte)) {
1396 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1397 		} else {
1398 			/*
1399 			 * pte is not used, increment hold count for ptbl
1400 			 * pages.
1401 			 */
1402 			if (pmap != kernel_pmap)
1403 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1404 		}
1405 	}
1406 
1407 	if (pdir[pdir_idx] == NULL) {
1408 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1409 			pdir_hold(mmu, pmap, pdir);
1410 		pdir[pdir_idx] = ptbl;
1411 	}
1412 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1413 		pmap->pm_pp2d[pp2d_idx] = pdir;
1414 
1415 	/*
1416 	 * Insert pv_entry into pv_list for mapped page if part of managed
1417 	 * memory.
1418 	 */
1419 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1420 		flags |= PTE_MANAGED;
1421 
1422 		/* Create and insert pv entry. */
1423 		pv_insert(pmap, va, m);
1424 	}
1425 
1426 	mtx_lock_spin(&tlbivax_mutex);
1427 	tlb_miss_lock();
1428 
1429 	tlb0_flush_entry(va);
1430 	pmap->pm_stats.resident_count++;
1431 	pte = &pdir[pdir_idx][ptbl_idx];
1432 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1433 	*pte |= (PTE_VALID | flags);
1434 
1435 	tlb_miss_unlock();
1436 	mtx_unlock_spin(&tlbivax_mutex);
1437 
1438 	return (0);
1439 }
1440 
1441 /* Return the pa for the given pmap/va. */
1442 static	vm_paddr_t
1443 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1444 {
1445 	vm_paddr_t	pa = 0;
1446 	pte_t          *pte;
1447 
1448 	pte = pte_find(mmu, pmap, va);
1449 	if ((pte != NULL) && PTE_ISVALID(pte))
1450 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1451 	return (pa);
1452 }
1453 
1454 
1455 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1456 static void
1457 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1458 {
1459 	int		i, j;
1460 	vm_offset_t	va;
1461 	pte_t		*pte;
1462 
1463 	va = addr;
1464 	/* Initialize kernel pdir */
1465 	for (i = 0; i < kernel_pdirs; i++) {
1466 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1467 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1468 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1469 		    j < PDIR_NENTRIES; j++) {
1470 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1471 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE * PDIR_PAGES) +
1472 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE * PTBL_PAGES));
1473 		}
1474 	}
1475 
1476 	/*
1477 	 * Fill in PTEs covering kernel code and data. They are not required
1478 	 * for address translation, as this area is covered by static TLB1
1479 	 * entries, but for pte_vatopa() to work correctly with kernel area
1480 	 * addresses.
1481 	 */
1482 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1483 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1484 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1485 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1486 		    PTE_VALID | PTE_PS_4KB;
1487 	}
1488 }
1489 #else
1490 /*
1491  * Clean pte entry, try to free page table page if requested.
1492  *
1493  * Return 1 if ptbl pages were freed, otherwise return 0.
1494  */
1495 static int
1496 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1497 {
1498 	unsigned int pdir_idx = PDIR_IDX(va);
1499 	unsigned int ptbl_idx = PTBL_IDX(va);
1500 	vm_page_t m;
1501 	pte_t *ptbl;
1502 	pte_t *pte;
1503 
1504 	//int su = (pmap == kernel_pmap);
1505 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1506 	//		su, (u_int32_t)pmap, va, flags);
1507 
1508 	ptbl = pmap->pm_pdir[pdir_idx];
1509 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1510 
1511 	pte = &ptbl[ptbl_idx];
1512 
1513 	if (pte == NULL || !PTE_ISVALID(pte))
1514 		return (0);
1515 
1516 	if (PTE_ISWIRED(pte))
1517 		pmap->pm_stats.wired_count--;
1518 
1519 	/* Get vm_page_t for mapped pte. */
1520 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1521 
1522 	/* Handle managed entry. */
1523 	if (PTE_ISMANAGED(pte)) {
1524 
1525 		if (PTE_ISMODIFIED(pte))
1526 			vm_page_dirty(m);
1527 
1528 		if (PTE_ISREFERENCED(pte))
1529 			vm_page_aflag_set(m, PGA_REFERENCED);
1530 
1531 		pv_remove(pmap, va, m);
1532 	} else if (m->md.pv_tracked) {
1533 		/*
1534 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1535 		 * used.  This is needed by the NCSW support code for fast
1536 		 * VA<->PA translation.
1537 		 */
1538 		pv_remove(pmap, va, m);
1539 		if (TAILQ_EMPTY(&m->md.pv_list))
1540 			m->md.pv_tracked = false;
1541 	}
1542 
1543 	mtx_lock_spin(&tlbivax_mutex);
1544 	tlb_miss_lock();
1545 
1546 	tlb0_flush_entry(va);
1547 	*pte = 0;
1548 
1549 	tlb_miss_unlock();
1550 	mtx_unlock_spin(&tlbivax_mutex);
1551 
1552 	pmap->pm_stats.resident_count--;
1553 
1554 	if (flags & PTBL_UNHOLD) {
1555 		//debugf("pte_remove: e (unhold)\n");
1556 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1557 	}
1558 
1559 	//debugf("pte_remove: e\n");
1560 	return (0);
1561 }
1562 
1563 /*
1564  * Insert PTE for a given page and virtual address.
1565  */
1566 static int
1567 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1568     boolean_t nosleep)
1569 {
1570 	unsigned int pdir_idx = PDIR_IDX(va);
1571 	unsigned int ptbl_idx = PTBL_IDX(va);
1572 	pte_t *ptbl, *pte;
1573 
1574 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1575 	    pmap == kernel_pmap, pmap, va);
1576 
1577 	/* Get the page table pointer. */
1578 	ptbl = pmap->pm_pdir[pdir_idx];
1579 
1580 	if (ptbl == NULL) {
1581 		/* Allocate page table pages. */
1582 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1583 		if (ptbl == NULL) {
1584 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1585 			return (ENOMEM);
1586 		}
1587 	} else {
1588 		/*
1589 		 * Check if there is valid mapping for requested
1590 		 * va, if there is, remove it.
1591 		 */
1592 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1593 		if (PTE_ISVALID(pte)) {
1594 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1595 		} else {
1596 			/*
1597 			 * pte is not used, increment hold count
1598 			 * for ptbl pages.
1599 			 */
1600 			if (pmap != kernel_pmap)
1601 				ptbl_hold(mmu, pmap, pdir_idx);
1602 		}
1603 	}
1604 
1605 	/*
1606 	 * Insert pv_entry into pv_list for mapped page if part of managed
1607 	 * memory.
1608 	 */
1609 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1610 		flags |= PTE_MANAGED;
1611 
1612 		/* Create and insert pv entry. */
1613 		pv_insert(pmap, va, m);
1614 	}
1615 
1616 	pmap->pm_stats.resident_count++;
1617 
1618 	mtx_lock_spin(&tlbivax_mutex);
1619 	tlb_miss_lock();
1620 
1621 	tlb0_flush_entry(va);
1622 	if (pmap->pm_pdir[pdir_idx] == NULL) {
1623 		/*
1624 		 * If we just allocated a new page table, hook it in
1625 		 * the pdir.
1626 		 */
1627 		pmap->pm_pdir[pdir_idx] = ptbl;
1628 	}
1629 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
1630 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1631 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1632 
1633 	tlb_miss_unlock();
1634 	mtx_unlock_spin(&tlbivax_mutex);
1635 	return (0);
1636 }
1637 
1638 /* Return the pa for the given pmap/va. */
1639 static vm_paddr_t
1640 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1641 {
1642 	vm_paddr_t pa = 0;
1643 	pte_t *pte;
1644 
1645 	pte = pte_find(mmu, pmap, va);
1646 	if ((pte != NULL) && PTE_ISVALID(pte))
1647 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1648 	return (pa);
1649 }
1650 
1651 /* Get a pointer to a PTE in a page table. */
1652 static pte_t *
1653 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1654 {
1655 	unsigned int pdir_idx = PDIR_IDX(va);
1656 	unsigned int ptbl_idx = PTBL_IDX(va);
1657 
1658 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1659 
1660 	if (pmap->pm_pdir[pdir_idx])
1661 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1662 
1663 	return (NULL);
1664 }
1665 
1666 /* Set up kernel page tables. */
1667 static void
1668 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1669 {
1670 	int		i;
1671 	vm_offset_t	va;
1672 	pte_t		*pte;
1673 
1674 	/* Initialize kernel pdir */
1675 	for (i = 0; i < kernel_ptbls; i++)
1676 		kernel_pmap->pm_pdir[kptbl_min + i] =
1677 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1678 
1679 	/*
1680 	 * Fill in PTEs covering kernel code and data. They are not required
1681 	 * for address translation, as this area is covered by static TLB1
1682 	 * entries, but for pte_vatopa() to work correctly with kernel area
1683 	 * addresses.
1684 	 */
1685 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1686 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1687 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1688 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1689 		    PTE_VALID | PTE_PS_4KB;
1690 	}
1691 }
1692 #endif
1693 
1694 /**************************************************************************/
1695 /* PMAP related */
1696 /**************************************************************************/
1697 
1698 /*
1699  * This is called during booke_init, before the system is really initialized.
1700  */
1701 static void
1702 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1703 {
1704 	vm_paddr_t phys_kernelend;
1705 	struct mem_region *mp, *mp1;
1706 	int cnt, i, j;
1707 	vm_paddr_t s, e, sz;
1708 	vm_paddr_t physsz, hwphyssz;
1709 	u_int phys_avail_count;
1710 	vm_size_t kstack0_sz;
1711 	vm_offset_t kernel_pdir, kstack0;
1712 	vm_paddr_t kstack0_phys;
1713 	void *dpcpu;
1714 
1715 	debugf("mmu_booke_bootstrap: entered\n");
1716 
1717 	/* Set interesting system properties */
1718 	hw_direct_map = 0;
1719 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1720 	elf32_nxstack = 1;
1721 #endif
1722 
1723 	/* Initialize invalidation mutex */
1724 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1725 
1726 	/* Read TLB0 size and associativity. */
1727 	tlb0_get_tlbconf();
1728 
1729 	/*
1730 	 * Align kernel start and end address (kernel image).
1731 	 * Note that kernel end does not necessarily relate to kernsize.
1732 	 * kernsize is the size of the kernel that is actually mapped.
1733 	 */
1734 	kernstart = trunc_page(start);
1735 	data_start = round_page(kernelend);
1736 	data_end = data_start;
1737 
1738 	/*
1739 	 * Addresses of preloaded modules (like file systems) use
1740 	 * physical addresses. Make sure we relocate those into
1741 	 * virtual addresses.
1742 	 */
1743 	preload_addr_relocate = kernstart - kernload;
1744 
1745 	/* Allocate the dynamic per-cpu area. */
1746 	dpcpu = (void *)data_end;
1747 	data_end += DPCPU_SIZE;
1748 
1749 	/* Allocate space for the message buffer. */
1750 	msgbufp = (struct msgbuf *)data_end;
1751 	data_end += msgbufsize;
1752 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1753 	    (uintptr_t)msgbufp, data_end);
1754 
1755 	data_end = round_page(data_end);
1756 
1757 	/* Allocate space for ptbl_bufs. */
1758 	ptbl_bufs = (struct ptbl_buf *)data_end;
1759 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1760 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1761 	    (uintptr_t)ptbl_bufs, data_end);
1762 
1763 	data_end = round_page(data_end);
1764 
1765 	/* Allocate PTE tables for kernel KVA. */
1766 	kernel_pdir = data_end;
1767 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1768 	    PDIR_SIZE);
1769 #ifdef __powerpc64__
1770 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1771 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1772 #endif
1773 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1774 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1775 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1776 	    kernel_pdir, data_end);
1777 
1778 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1779 	if (data_end - kernstart > kernsize) {
1780 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1781 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1782 	}
1783 	data_end = kernstart + kernsize;
1784 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1785 
1786 	/*
1787 	 * Clear the structures - note we can only do it safely after the
1788 	 * possible additional TLB1 translations are in place (above) so that
1789 	 * all range up to the currently calculated 'data_end' is covered.
1790 	 */
1791 	dpcpu_init(dpcpu, 0);
1792 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1793 #ifdef __powerpc64__
1794 	memset((void *)kernel_pdir, 0,
1795 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1796 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1797 #else
1798 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1799 #endif
1800 
1801 	/*******************************************************/
1802 	/* Set the start and end of kva. */
1803 	/*******************************************************/
1804 	virtual_avail = round_page(data_end);
1805 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1806 
1807 	/* Allocate KVA space for page zero/copy operations. */
1808 	zero_page_va = virtual_avail;
1809 	virtual_avail += PAGE_SIZE;
1810 	copy_page_src_va = virtual_avail;
1811 	virtual_avail += PAGE_SIZE;
1812 	copy_page_dst_va = virtual_avail;
1813 	virtual_avail += PAGE_SIZE;
1814 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1815 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1816 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1817 
1818 	/* Initialize page zero/copy mutexes. */
1819 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1820 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1821 
1822 	/* Allocate KVA space for ptbl bufs. */
1823 	ptbl_buf_pool_vabase = virtual_avail;
1824 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1825 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1826 	    ptbl_buf_pool_vabase, virtual_avail);
1827 
1828 	/* Calculate corresponding physical addresses for the kernel region. */
1829 	phys_kernelend = kernload + kernsize;
1830 	debugf("kernel image and allocated data:\n");
1831 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1832 	debugf(" kernstart   = 0x%08x\n", kernstart);
1833 	debugf(" kernsize    = 0x%08x\n", kernsize);
1834 
1835 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1836 		panic("mmu_booke_bootstrap: phys_avail too small");
1837 
1838 	/*
1839 	 * Remove kernel physical address range from avail regions list. Page
1840 	 * align all regions.  Non-page aligned memory isn't very interesting
1841 	 * to us.  Also, sort the entries for ascending addresses.
1842 	 */
1843 
1844 	/* Retrieve phys/avail mem regions */
1845 	mem_regions(&physmem_regions, &physmem_regions_sz,
1846 	    &availmem_regions, &availmem_regions_sz);
1847 	sz = 0;
1848 	cnt = availmem_regions_sz;
1849 	debugf("processing avail regions:\n");
1850 	for (mp = availmem_regions; mp->mr_size; mp++) {
1851 		s = mp->mr_start;
1852 		e = mp->mr_start + mp->mr_size;
1853 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1854 		/* Check whether this region holds all of the kernel. */
1855 		if (s < kernload && e > phys_kernelend) {
1856 			availmem_regions[cnt].mr_start = phys_kernelend;
1857 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1858 			e = kernload;
1859 		}
1860 		/* Look whether this regions starts within the kernel. */
1861 		if (s >= kernload && s < phys_kernelend) {
1862 			if (e <= phys_kernelend)
1863 				goto empty;
1864 			s = phys_kernelend;
1865 		}
1866 		/* Now look whether this region ends within the kernel. */
1867 		if (e > kernload && e <= phys_kernelend) {
1868 			if (s >= kernload)
1869 				goto empty;
1870 			e = kernload;
1871 		}
1872 		/* Now page align the start and size of the region. */
1873 		s = round_page(s);
1874 		e = trunc_page(e);
1875 		if (e < s)
1876 			e = s;
1877 		sz = e - s;
1878 		debugf("%09jx-%09jx = %jx\n",
1879 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1880 
1881 		/* Check whether some memory is left here. */
1882 		if (sz == 0) {
1883 		empty:
1884 			memmove(mp, mp + 1,
1885 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1886 			cnt--;
1887 			mp--;
1888 			continue;
1889 		}
1890 
1891 		/* Do an insertion sort. */
1892 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1893 			if (s < mp1->mr_start)
1894 				break;
1895 		if (mp1 < mp) {
1896 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1897 			mp1->mr_start = s;
1898 			mp1->mr_size = sz;
1899 		} else {
1900 			mp->mr_start = s;
1901 			mp->mr_size = sz;
1902 		}
1903 	}
1904 	availmem_regions_sz = cnt;
1905 
1906 	/*******************************************************/
1907 	/* Steal physical memory for kernel stack from the end */
1908 	/* of the first avail region                           */
1909 	/*******************************************************/
1910 	kstack0_sz = kstack_pages * PAGE_SIZE;
1911 	kstack0_phys = availmem_regions[0].mr_start +
1912 	    availmem_regions[0].mr_size;
1913 	kstack0_phys -= kstack0_sz;
1914 	availmem_regions[0].mr_size -= kstack0_sz;
1915 
1916 	/*******************************************************/
1917 	/* Fill in phys_avail table, based on availmem_regions */
1918 	/*******************************************************/
1919 	phys_avail_count = 0;
1920 	physsz = 0;
1921 	hwphyssz = 0;
1922 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1923 
1924 	debugf("fill in phys_avail:\n");
1925 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1926 
1927 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1928 		    (uintmax_t)availmem_regions[i].mr_start,
1929 		    (uintmax_t)availmem_regions[i].mr_start +
1930 		        availmem_regions[i].mr_size,
1931 		    (uintmax_t)availmem_regions[i].mr_size);
1932 
1933 		if (hwphyssz != 0 &&
1934 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1935 			debugf(" hw.physmem adjust\n");
1936 			if (physsz < hwphyssz) {
1937 				phys_avail[j] = availmem_regions[i].mr_start;
1938 				phys_avail[j + 1] =
1939 				    availmem_regions[i].mr_start +
1940 				    hwphyssz - physsz;
1941 				physsz = hwphyssz;
1942 				phys_avail_count++;
1943 			}
1944 			break;
1945 		}
1946 
1947 		phys_avail[j] = availmem_regions[i].mr_start;
1948 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1949 		    availmem_regions[i].mr_size;
1950 		phys_avail_count++;
1951 		physsz += availmem_regions[i].mr_size;
1952 	}
1953 	physmem = btoc(physsz);
1954 
1955 	/* Calculate the last available physical address. */
1956 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1957 		;
1958 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1959 
1960 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1961 	debugf("phys_avail_count = %d\n", phys_avail_count);
1962 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1963 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1964 
1965 	/*******************************************************/
1966 	/* Initialize (statically allocated) kernel pmap. */
1967 	/*******************************************************/
1968 	PMAP_LOCK_INIT(kernel_pmap);
1969 #ifndef __powerpc64__
1970 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1971 #endif
1972 
1973 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1974 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1975 	for (i = 0; i < MAXCPU; i++) {
1976 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1977 
1978 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1979 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1980 	}
1981 
1982 	/* Mark kernel_pmap active on all CPUs */
1983 	CPU_FILL(&kernel_pmap->pm_active);
1984 
1985  	/*
1986 	 * Initialize the global pv list lock.
1987 	 */
1988 	rw_init(&pvh_global_lock, "pmap pv global");
1989 
1990 	/*******************************************************/
1991 	/* Final setup */
1992 	/*******************************************************/
1993 
1994 	/* Enter kstack0 into kernel map, provide guard page */
1995 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1996 	thread0.td_kstack = kstack0;
1997 	thread0.td_kstack_pages = kstack_pages;
1998 
1999 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
2000 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
2001 	    kstack0_phys, kstack0_phys + kstack0_sz);
2002 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
2003 	    kstack0, kstack0 + kstack0_sz);
2004 
2005 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
2006 	for (i = 0; i < kstack_pages; i++) {
2007 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
2008 		kstack0 += PAGE_SIZE;
2009 		kstack0_phys += PAGE_SIZE;
2010 	}
2011 
2012 	pmap_bootstrapped = 1;
2013 
2014 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
2015 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
2016 
2017 	debugf("mmu_booke_bootstrap: exit\n");
2018 }
2019 
2020 #ifdef SMP
2021  void
2022 tlb1_ap_prep(void)
2023 {
2024 	tlb_entry_t *e, tmp;
2025 	unsigned int i;
2026 
2027 	/* Prepare TLB1 image for AP processors */
2028 	e = __boot_tlb1;
2029 	for (i = 0; i < TLB1_ENTRIES; i++) {
2030 		tlb1_read_entry(&tmp, i);
2031 
2032 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
2033 			memcpy(e++, &tmp, sizeof(tmp));
2034 	}
2035 }
2036 
2037 void
2038 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
2039 {
2040 	int i;
2041 
2042 	/*
2043 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
2044 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
2045 	 * created by tlb1_ap_prep(), so use these values directly to
2046 	 * (re)program AP's TLB1 hardware.
2047 	 *
2048 	 * Start at index 1 because index 0 has the kernel map.
2049 	 */
2050 	for (i = 1; i < TLB1_ENTRIES; i++) {
2051 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
2052 			tlb1_write_entry(&__boot_tlb1[i], i);
2053 	}
2054 
2055 	set_mas4_defaults();
2056 }
2057 #endif
2058 
2059 static void
2060 booke_pmap_init_qpages(void)
2061 {
2062 	struct pcpu *pc;
2063 	int i;
2064 
2065 	CPU_FOREACH(i) {
2066 		pc = pcpu_find(i);
2067 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
2068 		if (pc->pc_qmap_addr == 0)
2069 			panic("pmap_init_qpages: unable to allocate KVA");
2070 	}
2071 }
2072 
2073 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
2074 
2075 /*
2076  * Get the physical page address for the given pmap/virtual address.
2077  */
2078 static vm_paddr_t
2079 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
2080 {
2081 	vm_paddr_t pa;
2082 
2083 	PMAP_LOCK(pmap);
2084 	pa = pte_vatopa(mmu, pmap, va);
2085 	PMAP_UNLOCK(pmap);
2086 
2087 	return (pa);
2088 }
2089 
2090 /*
2091  * Extract the physical page address associated with the given
2092  * kernel virtual address.
2093  */
2094 static vm_paddr_t
2095 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
2096 {
2097 	tlb_entry_t e;
2098 	vm_paddr_t p = 0;
2099 	int i;
2100 
2101 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
2102 		p = pte_vatopa(mmu, kernel_pmap, va);
2103 
2104 	if (p == 0) {
2105 		/* Check TLB1 mappings */
2106 		for (i = 0; i < TLB1_ENTRIES; i++) {
2107 			tlb1_read_entry(&e, i);
2108 			if (!(e.mas1 & MAS1_VALID))
2109 				continue;
2110 			if (va >= e.virt && va < e.virt + e.size)
2111 				return (e.phys + (va - e.virt));
2112 		}
2113 	}
2114 
2115 	return (p);
2116 }
2117 
2118 /*
2119  * Initialize the pmap module.
2120  * Called by vm_init, to initialize any structures that the pmap
2121  * system needs to map virtual memory.
2122  */
2123 static void
2124 mmu_booke_init(mmu_t mmu)
2125 {
2126 	int shpgperproc = PMAP_SHPGPERPROC;
2127 
2128 	/*
2129 	 * Initialize the address space (zone) for the pv entries.  Set a
2130 	 * high water mark so that the system can recover from excessive
2131 	 * numbers of pv entries.
2132 	 */
2133 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2134 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2135 
2136 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2137 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2138 
2139 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2140 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2141 
2142 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2143 
2144 	/* Pre-fill pvzone with initial number of pv entries. */
2145 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2146 
2147 	/* Initialize ptbl allocation. */
2148 	ptbl_init();
2149 }
2150 
2151 /*
2152  * Map a list of wired pages into kernel virtual address space.  This is
2153  * intended for temporary mappings which do not need page modification or
2154  * references recorded.  Existing mappings in the region are overwritten.
2155  */
2156 static void
2157 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2158 {
2159 	vm_offset_t va;
2160 
2161 	va = sva;
2162 	while (count-- > 0) {
2163 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2164 		va += PAGE_SIZE;
2165 		m++;
2166 	}
2167 }
2168 
2169 /*
2170  * Remove page mappings from kernel virtual address space.  Intended for
2171  * temporary mappings entered by mmu_booke_qenter.
2172  */
2173 static void
2174 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2175 {
2176 	vm_offset_t va;
2177 
2178 	va = sva;
2179 	while (count-- > 0) {
2180 		mmu_booke_kremove(mmu, va);
2181 		va += PAGE_SIZE;
2182 	}
2183 }
2184 
2185 /*
2186  * Map a wired page into kernel virtual address space.
2187  */
2188 static void
2189 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2190 {
2191 
2192 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2193 }
2194 
2195 static void
2196 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2197 {
2198 	uint32_t flags;
2199 	pte_t *pte;
2200 
2201 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2202 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2203 
2204 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2205 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2206 	flags |= PTE_PS_4KB;
2207 
2208 	pte = pte_find(mmu, kernel_pmap, va);
2209 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2210 
2211 	mtx_lock_spin(&tlbivax_mutex);
2212 	tlb_miss_lock();
2213 
2214 	if (PTE_ISVALID(pte)) {
2215 
2216 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2217 
2218 		/* Flush entry from TLB0 */
2219 		tlb0_flush_entry(va);
2220 	}
2221 
2222 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2223 
2224 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2225 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2226 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2227 
2228 	/* Flush the real memory from the instruction cache. */
2229 	if ((flags & (PTE_I | PTE_G)) == 0)
2230 		__syncicache((void *)va, PAGE_SIZE);
2231 
2232 	tlb_miss_unlock();
2233 	mtx_unlock_spin(&tlbivax_mutex);
2234 }
2235 
2236 /*
2237  * Remove a page from kernel page table.
2238  */
2239 static void
2240 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2241 {
2242 	pte_t *pte;
2243 
2244 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
2245 
2246 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2247 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2248 	    ("mmu_booke_kremove: invalid va"));
2249 
2250 	pte = pte_find(mmu, kernel_pmap, va);
2251 
2252 	if (!PTE_ISVALID(pte)) {
2253 
2254 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2255 
2256 		return;
2257 	}
2258 
2259 	mtx_lock_spin(&tlbivax_mutex);
2260 	tlb_miss_lock();
2261 
2262 	/* Invalidate entry in TLB0, update PTE. */
2263 	tlb0_flush_entry(va);
2264 	*pte = 0;
2265 
2266 	tlb_miss_unlock();
2267 	mtx_unlock_spin(&tlbivax_mutex);
2268 }
2269 
2270 /*
2271  * Initialize pmap associated with process 0.
2272  */
2273 static void
2274 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2275 {
2276 
2277 	PMAP_LOCK_INIT(pmap);
2278 	mmu_booke_pinit(mmu, pmap);
2279 	PCPU_SET(curpmap, pmap);
2280 }
2281 
2282 /*
2283  * Initialize a preallocated and zeroed pmap structure,
2284  * such as one in a vmspace structure.
2285  */
2286 static void
2287 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2288 {
2289 	int i;
2290 
2291 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2292 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2293 
2294 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2295 
2296 	for (i = 0; i < MAXCPU; i++)
2297 		pmap->pm_tid[i] = TID_NONE;
2298 	CPU_ZERO(&kernel_pmap->pm_active);
2299 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2300 #ifdef __powerpc64__
2301 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2302 	TAILQ_INIT(&pmap->pm_pdir_list);
2303 #else
2304 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2305 #endif
2306 	TAILQ_INIT(&pmap->pm_ptbl_list);
2307 }
2308 
2309 /*
2310  * Release any resources held by the given physical map.
2311  * Called when a pmap initialized by mmu_booke_pinit is being released.
2312  * Should only be called if the map contains no valid mappings.
2313  */
2314 static void
2315 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2316 {
2317 
2318 	KASSERT(pmap->pm_stats.resident_count == 0,
2319 	    ("pmap_release: pmap resident count %ld != 0",
2320 	    pmap->pm_stats.resident_count));
2321 }
2322 
2323 /*
2324  * Insert the given physical page at the specified virtual address in the
2325  * target physical map with the protection requested. If specified the page
2326  * will be wired down.
2327  */
2328 static int
2329 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2330     vm_prot_t prot, u_int flags, int8_t psind)
2331 {
2332 	int error;
2333 
2334 	rw_wlock(&pvh_global_lock);
2335 	PMAP_LOCK(pmap);
2336 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2337 	PMAP_UNLOCK(pmap);
2338 	rw_wunlock(&pvh_global_lock);
2339 	return (error);
2340 }
2341 
2342 static int
2343 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2344     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2345 {
2346 	pte_t *pte;
2347 	vm_paddr_t pa;
2348 	uint32_t flags;
2349 	int error, su, sync;
2350 
2351 	pa = VM_PAGE_TO_PHYS(m);
2352 	su = (pmap == kernel_pmap);
2353 	sync = 0;
2354 
2355 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2356 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2357 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2358 	//		(u_int32_t)m, va, pa, prot, flags);
2359 
2360 	if (su) {
2361 		KASSERT(((va >= virtual_avail) &&
2362 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2363 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2364 	} else {
2365 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2366 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2367 	}
2368 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2369 		VM_OBJECT_ASSERT_LOCKED(m->object);
2370 
2371 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2372 
2373 	/*
2374 	 * If there is an existing mapping, and the physical address has not
2375 	 * changed, must be protection or wiring change.
2376 	 */
2377 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2378 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2379 
2380 		/*
2381 		 * Before actually updating pte->flags we calculate and
2382 		 * prepare its new value in a helper var.
2383 		 */
2384 		flags = *pte;
2385 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2386 
2387 		/* Wiring change, just update stats. */
2388 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2389 			if (!PTE_ISWIRED(pte)) {
2390 				flags |= PTE_WIRED;
2391 				pmap->pm_stats.wired_count++;
2392 			}
2393 		} else {
2394 			if (PTE_ISWIRED(pte)) {
2395 				flags &= ~PTE_WIRED;
2396 				pmap->pm_stats.wired_count--;
2397 			}
2398 		}
2399 
2400 		if (prot & VM_PROT_WRITE) {
2401 			/* Add write permissions. */
2402 			flags |= PTE_SW;
2403 			if (!su)
2404 				flags |= PTE_UW;
2405 
2406 			if ((flags & PTE_MANAGED) != 0)
2407 				vm_page_aflag_set(m, PGA_WRITEABLE);
2408 		} else {
2409 			/* Handle modified pages, sense modify status. */
2410 
2411 			/*
2412 			 * The PTE_MODIFIED flag could be set by underlying
2413 			 * TLB misses since we last read it (above), possibly
2414 			 * other CPUs could update it so we check in the PTE
2415 			 * directly rather than rely on that saved local flags
2416 			 * copy.
2417 			 */
2418 			if (PTE_ISMODIFIED(pte))
2419 				vm_page_dirty(m);
2420 		}
2421 
2422 		if (prot & VM_PROT_EXECUTE) {
2423 			flags |= PTE_SX;
2424 			if (!su)
2425 				flags |= PTE_UX;
2426 
2427 			/*
2428 			 * Check existing flags for execute permissions: if we
2429 			 * are turning execute permissions on, icache should
2430 			 * be flushed.
2431 			 */
2432 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2433 				sync++;
2434 		}
2435 
2436 		flags &= ~PTE_REFERENCED;
2437 
2438 		/*
2439 		 * The new flags value is all calculated -- only now actually
2440 		 * update the PTE.
2441 		 */
2442 		mtx_lock_spin(&tlbivax_mutex);
2443 		tlb_miss_lock();
2444 
2445 		tlb0_flush_entry(va);
2446 		*pte &= ~PTE_FLAGS_MASK;
2447 		*pte |= flags;
2448 
2449 		tlb_miss_unlock();
2450 		mtx_unlock_spin(&tlbivax_mutex);
2451 
2452 	} else {
2453 		/*
2454 		 * If there is an existing mapping, but it's for a different
2455 		 * physical address, pte_enter() will delete the old mapping.
2456 		 */
2457 		//if ((pte != NULL) && PTE_ISVALID(pte))
2458 		//	debugf("mmu_booke_enter_locked: replace\n");
2459 		//else
2460 		//	debugf("mmu_booke_enter_locked: new\n");
2461 
2462 		/* Now set up the flags and install the new mapping. */
2463 		flags = (PTE_SR | PTE_VALID);
2464 		flags |= PTE_M;
2465 
2466 		if (!su)
2467 			flags |= PTE_UR;
2468 
2469 		if (prot & VM_PROT_WRITE) {
2470 			flags |= PTE_SW;
2471 			if (!su)
2472 				flags |= PTE_UW;
2473 
2474 			if ((m->oflags & VPO_UNMANAGED) == 0)
2475 				vm_page_aflag_set(m, PGA_WRITEABLE);
2476 		}
2477 
2478 		if (prot & VM_PROT_EXECUTE) {
2479 			flags |= PTE_SX;
2480 			if (!su)
2481 				flags |= PTE_UX;
2482 		}
2483 
2484 		/* If its wired update stats. */
2485 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2486 			flags |= PTE_WIRED;
2487 
2488 		error = pte_enter(mmu, pmap, m, va, flags,
2489 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2490 		if (error != 0)
2491 			return (KERN_RESOURCE_SHORTAGE);
2492 
2493 		if ((flags & PMAP_ENTER_WIRED) != 0)
2494 			pmap->pm_stats.wired_count++;
2495 
2496 		/* Flush the real memory from the instruction cache. */
2497 		if (prot & VM_PROT_EXECUTE)
2498 			sync++;
2499 	}
2500 
2501 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2502 		__syncicache((void *)va, PAGE_SIZE);
2503 		sync = 0;
2504 	}
2505 
2506 	return (KERN_SUCCESS);
2507 }
2508 
2509 /*
2510  * Maps a sequence of resident pages belonging to the same object.
2511  * The sequence begins with the given page m_start.  This page is
2512  * mapped at the given virtual address start.  Each subsequent page is
2513  * mapped at a virtual address that is offset from start by the same
2514  * amount as the page is offset from m_start within the object.  The
2515  * last page in the sequence is the page with the largest offset from
2516  * m_start that can be mapped at a virtual address less than the given
2517  * virtual address end.  Not every virtual page between start and end
2518  * is mapped; only those for which a resident page exists with the
2519  * corresponding offset from m_start are mapped.
2520  */
2521 static void
2522 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2523     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2524 {
2525 	vm_page_t m;
2526 	vm_pindex_t diff, psize;
2527 
2528 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2529 
2530 	psize = atop(end - start);
2531 	m = m_start;
2532 	rw_wlock(&pvh_global_lock);
2533 	PMAP_LOCK(pmap);
2534 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2535 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2536 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2537 		    PMAP_ENTER_NOSLEEP, 0);
2538 		m = TAILQ_NEXT(m, listq);
2539 	}
2540 	rw_wunlock(&pvh_global_lock);
2541 	PMAP_UNLOCK(pmap);
2542 }
2543 
2544 static void
2545 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2546     vm_prot_t prot)
2547 {
2548 
2549 	rw_wlock(&pvh_global_lock);
2550 	PMAP_LOCK(pmap);
2551 	mmu_booke_enter_locked(mmu, pmap, va, m,
2552 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2553 	    0);
2554 	rw_wunlock(&pvh_global_lock);
2555 	PMAP_UNLOCK(pmap);
2556 }
2557 
2558 /*
2559  * Remove the given range of addresses from the specified map.
2560  *
2561  * It is assumed that the start and end are properly rounded to the page size.
2562  */
2563 static void
2564 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2565 {
2566 	pte_t *pte;
2567 	uint8_t hold_flag;
2568 
2569 	int su = (pmap == kernel_pmap);
2570 
2571 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2572 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2573 
2574 	if (su) {
2575 		KASSERT(((va >= virtual_avail) &&
2576 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2577 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2578 	} else {
2579 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2580 		    ("mmu_booke_remove: user pmap, non user va"));
2581 	}
2582 
2583 	if (PMAP_REMOVE_DONE(pmap)) {
2584 		//debugf("mmu_booke_remove: e (empty)\n");
2585 		return;
2586 	}
2587 
2588 	hold_flag = PTBL_HOLD_FLAG(pmap);
2589 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2590 
2591 	rw_wlock(&pvh_global_lock);
2592 	PMAP_LOCK(pmap);
2593 	for (; va < endva; va += PAGE_SIZE) {
2594 		pte = pte_find(mmu, pmap, va);
2595 		if ((pte != NULL) && PTE_ISVALID(pte))
2596 			pte_remove(mmu, pmap, va, hold_flag);
2597 	}
2598 	PMAP_UNLOCK(pmap);
2599 	rw_wunlock(&pvh_global_lock);
2600 
2601 	//debugf("mmu_booke_remove: e\n");
2602 }
2603 
2604 /*
2605  * Remove physical page from all pmaps in which it resides.
2606  */
2607 static void
2608 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2609 {
2610 	pv_entry_t pv, pvn;
2611 	uint8_t hold_flag;
2612 
2613 	rw_wlock(&pvh_global_lock);
2614 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2615 		pvn = TAILQ_NEXT(pv, pv_link);
2616 
2617 		PMAP_LOCK(pv->pv_pmap);
2618 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2619 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2620 		PMAP_UNLOCK(pv->pv_pmap);
2621 	}
2622 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2623 	rw_wunlock(&pvh_global_lock);
2624 }
2625 
2626 /*
2627  * Map a range of physical addresses into kernel virtual address space.
2628  */
2629 static vm_offset_t
2630 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2631     vm_paddr_t pa_end, int prot)
2632 {
2633 	vm_offset_t sva = *virt;
2634 	vm_offset_t va = sva;
2635 
2636 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2637 	//		sva, pa_start, pa_end);
2638 
2639 	while (pa_start < pa_end) {
2640 		mmu_booke_kenter(mmu, va, pa_start);
2641 		va += PAGE_SIZE;
2642 		pa_start += PAGE_SIZE;
2643 	}
2644 	*virt = va;
2645 
2646 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2647 	return (sva);
2648 }
2649 
2650 /*
2651  * The pmap must be activated before it's address space can be accessed in any
2652  * way.
2653  */
2654 static void
2655 mmu_booke_activate(mmu_t mmu, struct thread *td)
2656 {
2657 	pmap_t pmap;
2658 	u_int cpuid;
2659 
2660 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2661 
2662 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
2663 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2664 
2665 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2666 
2667 	sched_pin();
2668 
2669 	cpuid = PCPU_GET(cpuid);
2670 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2671 	PCPU_SET(curpmap, pmap);
2672 
2673 	if (pmap->pm_tid[cpuid] == TID_NONE)
2674 		tid_alloc(pmap);
2675 
2676 	/* Load PID0 register with pmap tid value. */
2677 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2678 	__asm __volatile("isync");
2679 
2680 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2681 
2682 	sched_unpin();
2683 
2684 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2685 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2686 }
2687 
2688 /*
2689  * Deactivate the specified process's address space.
2690  */
2691 static void
2692 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2693 {
2694 	pmap_t pmap;
2695 
2696 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2697 
2698 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2699 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2700 
2701 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2702 
2703 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2704 	PCPU_SET(curpmap, NULL);
2705 }
2706 
2707 /*
2708  * Copy the range specified by src_addr/len
2709  * from the source map to the range dst_addr/len
2710  * in the destination map.
2711  *
2712  * This routine is only advisory and need not do anything.
2713  */
2714 static void
2715 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2716     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2717 {
2718 
2719 }
2720 
2721 /*
2722  * Set the physical protection on the specified range of this map as requested.
2723  */
2724 static void
2725 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2726     vm_prot_t prot)
2727 {
2728 	vm_offset_t va;
2729 	vm_page_t m;
2730 	pte_t *pte;
2731 
2732 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2733 		mmu_booke_remove(mmu, pmap, sva, eva);
2734 		return;
2735 	}
2736 
2737 	if (prot & VM_PROT_WRITE)
2738 		return;
2739 
2740 	PMAP_LOCK(pmap);
2741 	for (va = sva; va < eva; va += PAGE_SIZE) {
2742 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2743 			if (PTE_ISVALID(pte)) {
2744 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2745 
2746 				mtx_lock_spin(&tlbivax_mutex);
2747 				tlb_miss_lock();
2748 
2749 				/* Handle modified pages. */
2750 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2751 					vm_page_dirty(m);
2752 
2753 				tlb0_flush_entry(va);
2754 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2755 
2756 				tlb_miss_unlock();
2757 				mtx_unlock_spin(&tlbivax_mutex);
2758 			}
2759 		}
2760 	}
2761 	PMAP_UNLOCK(pmap);
2762 }
2763 
2764 /*
2765  * Clear the write and modified bits in each of the given page's mappings.
2766  */
2767 static void
2768 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2769 {
2770 	pv_entry_t pv;
2771 	pte_t *pte;
2772 
2773 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2774 	    ("mmu_booke_remove_write: page %p is not managed", m));
2775 
2776 	/*
2777 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2778 	 * set by another thread while the object is locked.  Thus,
2779 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2780 	 */
2781 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2782 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2783 		return;
2784 	rw_wlock(&pvh_global_lock);
2785 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2786 		PMAP_LOCK(pv->pv_pmap);
2787 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2788 			if (PTE_ISVALID(pte)) {
2789 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2790 
2791 				mtx_lock_spin(&tlbivax_mutex);
2792 				tlb_miss_lock();
2793 
2794 				/* Handle modified pages. */
2795 				if (PTE_ISMODIFIED(pte))
2796 					vm_page_dirty(m);
2797 
2798 				/* Flush mapping from TLB0. */
2799 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2800 
2801 				tlb_miss_unlock();
2802 				mtx_unlock_spin(&tlbivax_mutex);
2803 			}
2804 		}
2805 		PMAP_UNLOCK(pv->pv_pmap);
2806 	}
2807 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2808 	rw_wunlock(&pvh_global_lock);
2809 }
2810 
2811 static void
2812 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2813 {
2814 	pte_t *pte;
2815 	pmap_t pmap;
2816 	vm_page_t m;
2817 	vm_offset_t addr;
2818 	vm_paddr_t pa = 0;
2819 	int active, valid;
2820 
2821 	va = trunc_page(va);
2822 	sz = round_page(sz);
2823 
2824 	rw_wlock(&pvh_global_lock);
2825 	pmap = PCPU_GET(curpmap);
2826 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2827 	while (sz > 0) {
2828 		PMAP_LOCK(pm);
2829 		pte = pte_find(mmu, pm, va);
2830 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2831 		if (valid)
2832 			pa = PTE_PA(pte);
2833 		PMAP_UNLOCK(pm);
2834 		if (valid) {
2835 			if (!active) {
2836 				/* Create a mapping in the active pmap. */
2837 				addr = 0;
2838 				m = PHYS_TO_VM_PAGE(pa);
2839 				PMAP_LOCK(pmap);
2840 				pte_enter(mmu, pmap, m, addr,
2841 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2842 				__syncicache((void *)addr, PAGE_SIZE);
2843 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2844 				PMAP_UNLOCK(pmap);
2845 			} else
2846 				__syncicache((void *)va, PAGE_SIZE);
2847 		}
2848 		va += PAGE_SIZE;
2849 		sz -= PAGE_SIZE;
2850 	}
2851 	rw_wunlock(&pvh_global_lock);
2852 }
2853 
2854 /*
2855  * Atomically extract and hold the physical page with the given
2856  * pmap and virtual address pair if that mapping permits the given
2857  * protection.
2858  */
2859 static vm_page_t
2860 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2861     vm_prot_t prot)
2862 {
2863 	pte_t *pte;
2864 	vm_page_t m;
2865 	uint32_t pte_wbit;
2866 	vm_paddr_t pa;
2867 
2868 	m = NULL;
2869 	pa = 0;
2870 	PMAP_LOCK(pmap);
2871 retry:
2872 	pte = pte_find(mmu, pmap, va);
2873 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2874 		if (pmap == kernel_pmap)
2875 			pte_wbit = PTE_SW;
2876 		else
2877 			pte_wbit = PTE_UW;
2878 
2879 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2880 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2881 				goto retry;
2882 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2883 			vm_page_hold(m);
2884 		}
2885 	}
2886 
2887 	PA_UNLOCK_COND(pa);
2888 	PMAP_UNLOCK(pmap);
2889 	return (m);
2890 }
2891 
2892 /*
2893  * Initialize a vm_page's machine-dependent fields.
2894  */
2895 static void
2896 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2897 {
2898 
2899 	m->md.pv_tracked = 0;
2900 	TAILQ_INIT(&m->md.pv_list);
2901 }
2902 
2903 /*
2904  * mmu_booke_zero_page_area zeros the specified hardware page by
2905  * mapping it into virtual memory and using bzero to clear
2906  * its contents.
2907  *
2908  * off and size must reside within a single page.
2909  */
2910 static void
2911 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2912 {
2913 	vm_offset_t va;
2914 
2915 	/* XXX KASSERT off and size are within a single page? */
2916 
2917 	mtx_lock(&zero_page_mutex);
2918 	va = zero_page_va;
2919 
2920 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2921 	bzero((caddr_t)va + off, size);
2922 	mmu_booke_kremove(mmu, va);
2923 
2924 	mtx_unlock(&zero_page_mutex);
2925 }
2926 
2927 /*
2928  * mmu_booke_zero_page zeros the specified hardware page.
2929  */
2930 static void
2931 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2932 {
2933 	vm_offset_t off, va;
2934 
2935 	mtx_lock(&zero_page_mutex);
2936 	va = zero_page_va;
2937 
2938 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2939 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2940 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2941 	mmu_booke_kremove(mmu, va);
2942 
2943 	mtx_unlock(&zero_page_mutex);
2944 }
2945 
2946 /*
2947  * mmu_booke_copy_page copies the specified (machine independent) page by
2948  * mapping the page into virtual memory and using memcopy to copy the page,
2949  * one machine dependent page at a time.
2950  */
2951 static void
2952 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2953 {
2954 	vm_offset_t sva, dva;
2955 
2956 	sva = copy_page_src_va;
2957 	dva = copy_page_dst_va;
2958 
2959 	mtx_lock(&copy_page_mutex);
2960 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2961 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2962 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2963 	mmu_booke_kremove(mmu, dva);
2964 	mmu_booke_kremove(mmu, sva);
2965 	mtx_unlock(&copy_page_mutex);
2966 }
2967 
2968 static inline void
2969 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2970     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2971 {
2972 	void *a_cp, *b_cp;
2973 	vm_offset_t a_pg_offset, b_pg_offset;
2974 	int cnt;
2975 
2976 	mtx_lock(&copy_page_mutex);
2977 	while (xfersize > 0) {
2978 		a_pg_offset = a_offset & PAGE_MASK;
2979 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2980 		mmu_booke_kenter(mmu, copy_page_src_va,
2981 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2982 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2983 		b_pg_offset = b_offset & PAGE_MASK;
2984 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2985 		mmu_booke_kenter(mmu, copy_page_dst_va,
2986 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2987 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2988 		bcopy(a_cp, b_cp, cnt);
2989 		mmu_booke_kremove(mmu, copy_page_dst_va);
2990 		mmu_booke_kremove(mmu, copy_page_src_va);
2991 		a_offset += cnt;
2992 		b_offset += cnt;
2993 		xfersize -= cnt;
2994 	}
2995 	mtx_unlock(&copy_page_mutex);
2996 }
2997 
2998 static vm_offset_t
2999 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
3000 {
3001 	vm_paddr_t paddr;
3002 	vm_offset_t qaddr;
3003 	uint32_t flags;
3004 	pte_t *pte;
3005 
3006 	paddr = VM_PAGE_TO_PHYS(m);
3007 
3008 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3009 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3010 	flags |= PTE_PS_4KB;
3011 
3012 	critical_enter();
3013 	qaddr = PCPU_GET(qmap_addr);
3014 
3015 	pte = pte_find(mmu, kernel_pmap, qaddr);
3016 
3017 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3018 
3019 	/*
3020 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3021  	 * not be present in other TLBs.  Is there a better instruction
3022 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3023 	 */
3024 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3025 	__asm __volatile("isync; msync");
3026 
3027 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3028 
3029 	/* Flush the real memory from the instruction cache. */
3030 	if ((flags & (PTE_I | PTE_G)) == 0)
3031 		__syncicache((void *)qaddr, PAGE_SIZE);
3032 
3033 	return (qaddr);
3034 }
3035 
3036 static void
3037 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3038 {
3039 	pte_t *pte;
3040 
3041 	pte = pte_find(mmu, kernel_pmap, addr);
3042 
3043 	KASSERT(PCPU_GET(qmap_addr) == addr,
3044 	    ("mmu_booke_quick_remove_page: invalid address"));
3045 	KASSERT(*pte != 0,
3046 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3047 
3048 	*pte = 0;
3049 	critical_exit();
3050 }
3051 
3052 /*
3053  * Return whether or not the specified physical page was modified
3054  * in any of physical maps.
3055  */
3056 static boolean_t
3057 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3058 {
3059 	pte_t *pte;
3060 	pv_entry_t pv;
3061 	boolean_t rv;
3062 
3063 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3064 	    ("mmu_booke_is_modified: page %p is not managed", m));
3065 	rv = FALSE;
3066 
3067 	/*
3068 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3069 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3070 	 * is clear, no PTEs can be modified.
3071 	 */
3072 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3073 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3074 		return (rv);
3075 	rw_wlock(&pvh_global_lock);
3076 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3077 		PMAP_LOCK(pv->pv_pmap);
3078 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3079 		    PTE_ISVALID(pte)) {
3080 			if (PTE_ISMODIFIED(pte))
3081 				rv = TRUE;
3082 		}
3083 		PMAP_UNLOCK(pv->pv_pmap);
3084 		if (rv)
3085 			break;
3086 	}
3087 	rw_wunlock(&pvh_global_lock);
3088 	return (rv);
3089 }
3090 
3091 /*
3092  * Return whether or not the specified virtual address is eligible
3093  * for prefault.
3094  */
3095 static boolean_t
3096 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3097 {
3098 
3099 	return (FALSE);
3100 }
3101 
3102 /*
3103  * Return whether or not the specified physical page was referenced
3104  * in any physical maps.
3105  */
3106 static boolean_t
3107 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3108 {
3109 	pte_t *pte;
3110 	pv_entry_t pv;
3111 	boolean_t rv;
3112 
3113 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3114 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3115 	rv = FALSE;
3116 	rw_wlock(&pvh_global_lock);
3117 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3118 		PMAP_LOCK(pv->pv_pmap);
3119 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3120 		    PTE_ISVALID(pte)) {
3121 			if (PTE_ISREFERENCED(pte))
3122 				rv = TRUE;
3123 		}
3124 		PMAP_UNLOCK(pv->pv_pmap);
3125 		if (rv)
3126 			break;
3127 	}
3128 	rw_wunlock(&pvh_global_lock);
3129 	return (rv);
3130 }
3131 
3132 /*
3133  * Clear the modify bits on the specified physical page.
3134  */
3135 static void
3136 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3137 {
3138 	pte_t *pte;
3139 	pv_entry_t pv;
3140 
3141 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3142 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3143 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3144 	KASSERT(!vm_page_xbusied(m),
3145 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3146 
3147 	/*
3148 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3149 	 * If the object containing the page is locked and the page is not
3150 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3151 	 */
3152 	if ((m->aflags & PGA_WRITEABLE) == 0)
3153 		return;
3154 	rw_wlock(&pvh_global_lock);
3155 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3156 		PMAP_LOCK(pv->pv_pmap);
3157 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3158 		    PTE_ISVALID(pte)) {
3159 			mtx_lock_spin(&tlbivax_mutex);
3160 			tlb_miss_lock();
3161 
3162 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3163 				tlb0_flush_entry(pv->pv_va);
3164 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3165 				    PTE_REFERENCED);
3166 			}
3167 
3168 			tlb_miss_unlock();
3169 			mtx_unlock_spin(&tlbivax_mutex);
3170 		}
3171 		PMAP_UNLOCK(pv->pv_pmap);
3172 	}
3173 	rw_wunlock(&pvh_global_lock);
3174 }
3175 
3176 /*
3177  * Return a count of reference bits for a page, clearing those bits.
3178  * It is not necessary for every reference bit to be cleared, but it
3179  * is necessary that 0 only be returned when there are truly no
3180  * reference bits set.
3181  *
3182  * As an optimization, update the page's dirty field if a modified bit is
3183  * found while counting reference bits.  This opportunistic update can be
3184  * performed at low cost and can eliminate the need for some future calls
3185  * to pmap_is_modified().  However, since this function stops after
3186  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3187  * dirty pages.  Those dirty pages will only be detected by a future call
3188  * to pmap_is_modified().
3189  */
3190 static int
3191 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3192 {
3193 	pte_t *pte;
3194 	pv_entry_t pv;
3195 	int count;
3196 
3197 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3198 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3199 	count = 0;
3200 	rw_wlock(&pvh_global_lock);
3201 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3202 		PMAP_LOCK(pv->pv_pmap);
3203 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3204 		    PTE_ISVALID(pte)) {
3205 			if (PTE_ISMODIFIED(pte))
3206 				vm_page_dirty(m);
3207 			if (PTE_ISREFERENCED(pte)) {
3208 				mtx_lock_spin(&tlbivax_mutex);
3209 				tlb_miss_lock();
3210 
3211 				tlb0_flush_entry(pv->pv_va);
3212 				*pte &= ~PTE_REFERENCED;
3213 
3214 				tlb_miss_unlock();
3215 				mtx_unlock_spin(&tlbivax_mutex);
3216 
3217 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3218 					PMAP_UNLOCK(pv->pv_pmap);
3219 					break;
3220 				}
3221 			}
3222 		}
3223 		PMAP_UNLOCK(pv->pv_pmap);
3224 	}
3225 	rw_wunlock(&pvh_global_lock);
3226 	return (count);
3227 }
3228 
3229 /*
3230  * Clear the wired attribute from the mappings for the specified range of
3231  * addresses in the given pmap.  Every valid mapping within that range must
3232  * have the wired attribute set.  In contrast, invalid mappings cannot have
3233  * the wired attribute set, so they are ignored.
3234  *
3235  * The wired attribute of the page table entry is not a hardware feature, so
3236  * there is no need to invalidate any TLB entries.
3237  */
3238 static void
3239 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3240 {
3241 	vm_offset_t va;
3242 	pte_t *pte;
3243 
3244 	PMAP_LOCK(pmap);
3245 	for (va = sva; va < eva; va += PAGE_SIZE) {
3246 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3247 		    PTE_ISVALID(pte)) {
3248 			if (!PTE_ISWIRED(pte))
3249 				panic("mmu_booke_unwire: pte %p isn't wired",
3250 				    pte);
3251 			*pte &= ~PTE_WIRED;
3252 			pmap->pm_stats.wired_count--;
3253 		}
3254 	}
3255 	PMAP_UNLOCK(pmap);
3256 
3257 }
3258 
3259 /*
3260  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3261  * page.  This count may be changed upwards or downwards in the future; it is
3262  * only necessary that true be returned for a small subset of pmaps for proper
3263  * page aging.
3264  */
3265 static boolean_t
3266 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3267 {
3268 	pv_entry_t pv;
3269 	int loops;
3270 	boolean_t rv;
3271 
3272 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3273 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3274 	loops = 0;
3275 	rv = FALSE;
3276 	rw_wlock(&pvh_global_lock);
3277 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3278 		if (pv->pv_pmap == pmap) {
3279 			rv = TRUE;
3280 			break;
3281 		}
3282 		if (++loops >= 16)
3283 			break;
3284 	}
3285 	rw_wunlock(&pvh_global_lock);
3286 	return (rv);
3287 }
3288 
3289 /*
3290  * Return the number of managed mappings to the given physical page that are
3291  * wired.
3292  */
3293 static int
3294 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3295 {
3296 	pv_entry_t pv;
3297 	pte_t *pte;
3298 	int count = 0;
3299 
3300 	if ((m->oflags & VPO_UNMANAGED) != 0)
3301 		return (count);
3302 	rw_wlock(&pvh_global_lock);
3303 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3304 		PMAP_LOCK(pv->pv_pmap);
3305 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3306 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3307 				count++;
3308 		PMAP_UNLOCK(pv->pv_pmap);
3309 	}
3310 	rw_wunlock(&pvh_global_lock);
3311 	return (count);
3312 }
3313 
3314 static int
3315 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3316 {
3317 	int i;
3318 	vm_offset_t va;
3319 
3320 	/*
3321 	 * This currently does not work for entries that
3322 	 * overlap TLB1 entries.
3323 	 */
3324 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3325 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3326 			return (0);
3327 	}
3328 
3329 	return (EFAULT);
3330 }
3331 
3332 void
3333 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3334 {
3335 	vm_paddr_t ppa;
3336 	vm_offset_t ofs;
3337 	vm_size_t gran;
3338 
3339 	/* Minidumps are based on virtual memory addresses. */
3340 	if (do_minidump) {
3341 		*va = (void *)(vm_offset_t)pa;
3342 		return;
3343 	}
3344 
3345 	/* Raw physical memory dumps don't have a virtual address. */
3346 	/* We always map a 256MB page at 256M. */
3347 	gran = 256 * 1024 * 1024;
3348 	ppa = rounddown2(pa, gran);
3349 	ofs = pa - ppa;
3350 	*va = (void *)gran;
3351 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3352 
3353 	if (sz > (gran - ofs))
3354 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3355 		    _TLB_ENTRY_IO);
3356 }
3357 
3358 void
3359 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3360 {
3361 	vm_paddr_t ppa;
3362 	vm_offset_t ofs;
3363 	vm_size_t gran;
3364 	tlb_entry_t e;
3365 	int i;
3366 
3367 	/* Minidumps are based on virtual memory addresses. */
3368 	/* Nothing to do... */
3369 	if (do_minidump)
3370 		return;
3371 
3372 	for (i = 0; i < TLB1_ENTRIES; i++) {
3373 		tlb1_read_entry(&e, i);
3374 		if (!(e.mas1 & MAS1_VALID))
3375 			break;
3376 	}
3377 
3378 	/* Raw physical memory dumps don't have a virtual address. */
3379 	i--;
3380 	e.mas1 = 0;
3381 	e.mas2 = 0;
3382 	e.mas3 = 0;
3383 	tlb1_write_entry(&e, i);
3384 
3385 	gran = 256 * 1024 * 1024;
3386 	ppa = rounddown2(pa, gran);
3387 	ofs = pa - ppa;
3388 	if (sz > (gran - ofs)) {
3389 		i--;
3390 		e.mas1 = 0;
3391 		e.mas2 = 0;
3392 		e.mas3 = 0;
3393 		tlb1_write_entry(&e, i);
3394 	}
3395 }
3396 
3397 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3398 
3399 void
3400 mmu_booke_scan_init(mmu_t mmu)
3401 {
3402 	vm_offset_t va;
3403 	pte_t *pte;
3404 	int i;
3405 
3406 	if (!do_minidump) {
3407 		/* Initialize phys. segments for dumpsys(). */
3408 		memset(&dump_map, 0, sizeof(dump_map));
3409 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3410 		    &availmem_regions_sz);
3411 		for (i = 0; i < physmem_regions_sz; i++) {
3412 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3413 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3414 		}
3415 		return;
3416 	}
3417 
3418 	/* Virtual segments for minidumps: */
3419 	memset(&dump_map, 0, sizeof(dump_map));
3420 
3421 	/* 1st: kernel .data and .bss. */
3422 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3423 	dump_map[0].pa_size =
3424 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3425 
3426 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3427 	dump_map[1].pa_start = data_start;
3428 	dump_map[1].pa_size = data_end - data_start;
3429 
3430 	/* 3rd: kernel VM. */
3431 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3432 	/* Find start of next chunk (from va). */
3433 	while (va < virtual_end) {
3434 		/* Don't dump the buffer cache. */
3435 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3436 			va = kmi.buffer_eva;
3437 			continue;
3438 		}
3439 		pte = pte_find(mmu, kernel_pmap, va);
3440 		if (pte != NULL && PTE_ISVALID(pte))
3441 			break;
3442 		va += PAGE_SIZE;
3443 	}
3444 	if (va < virtual_end) {
3445 		dump_map[2].pa_start = va;
3446 		va += PAGE_SIZE;
3447 		/* Find last page in chunk. */
3448 		while (va < virtual_end) {
3449 			/* Don't run into the buffer cache. */
3450 			if (va == kmi.buffer_sva)
3451 				break;
3452 			pte = pte_find(mmu, kernel_pmap, va);
3453 			if (pte == NULL || !PTE_ISVALID(pte))
3454 				break;
3455 			va += PAGE_SIZE;
3456 		}
3457 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3458 	}
3459 }
3460 
3461 /*
3462  * Map a set of physical memory pages into the kernel virtual address space.
3463  * Return a pointer to where it is mapped. This routine is intended to be used
3464  * for mapping device memory, NOT real memory.
3465  */
3466 static void *
3467 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3468 {
3469 
3470 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3471 }
3472 
3473 static void *
3474 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3475 {
3476 	tlb_entry_t e;
3477 	void *res;
3478 	uintptr_t va, tmpva;
3479 	vm_size_t sz;
3480 	int i;
3481 
3482 	/*
3483 	 * Check if this is premapped in TLB1. Note: this should probably also
3484 	 * check whether a sequence of TLB1 entries exist that match the
3485 	 * requirement, but now only checks the easy case.
3486 	 */
3487 	for (i = 0; i < TLB1_ENTRIES; i++) {
3488 		tlb1_read_entry(&e, i);
3489 		if (!(e.mas1 & MAS1_VALID))
3490 			continue;
3491 		if (pa >= e.phys &&
3492 		    (pa + size) <= (e.phys + e.size) &&
3493 		    (ma == VM_MEMATTR_DEFAULT ||
3494 		     tlb_calc_wimg(pa, ma) ==
3495 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3496 			return (void *)(e.virt +
3497 			    (vm_offset_t)(pa - e.phys));
3498 	}
3499 
3500 	size = roundup(size, PAGE_SIZE);
3501 
3502 	/*
3503 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3504 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3505 	 */
3506 #ifdef SPARSE_MAPDEV
3507 	/*
3508 	 * With a sparse mapdev, align to the largest starting region.  This
3509 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3510 	 * calculation could be very costly.
3511 	 * Align to the smaller of:
3512 	 * - first set bit in overlap of (pa & size mask)
3513 	 * - largest size envelope
3514 	 *
3515 	 * It's possible the device mapping may start at a PA that's not larger
3516 	 * than the size mask, so we need to offset in to maximize the TLB entry
3517 	 * range and minimize the number of used TLB entries.
3518 	 */
3519 	do {
3520 	    tmpva = tlb1_map_base;
3521 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3522 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3523 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3524 #ifdef __powerpc64__
3525 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3526 #else
3527 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3528 #endif
3529 #else
3530 #ifdef __powerpc64__
3531 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3532 #else
3533 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3534 #endif
3535 #endif
3536 	res = (void *)va;
3537 
3538 	do {
3539 		sz = 1 << (ilog2(size) & ~1);
3540 		/* Align size to PA */
3541 		if (pa % sz != 0) {
3542 			do {
3543 				sz >>= 2;
3544 			} while (pa % sz != 0);
3545 		}
3546 		/* Now align from there to VA */
3547 		if (va % sz != 0) {
3548 			do {
3549 				sz >>= 2;
3550 			} while (va % sz != 0);
3551 		}
3552 		if (bootverbose)
3553 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3554 			    va, (uintmax_t)pa, sz);
3555 		if (tlb1_set_entry(va, pa, sz,
3556 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3557 			return (NULL);
3558 		size -= sz;
3559 		pa += sz;
3560 		va += sz;
3561 	} while (size > 0);
3562 
3563 	return (res);
3564 }
3565 
3566 /*
3567  * 'Unmap' a range mapped by mmu_booke_mapdev().
3568  */
3569 static void
3570 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3571 {
3572 #ifdef SUPPORTS_SHRINKING_TLB1
3573 	vm_offset_t base, offset;
3574 
3575 	/*
3576 	 * Unmap only if this is inside kernel virtual space.
3577 	 */
3578 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3579 		base = trunc_page(va);
3580 		offset = va & PAGE_MASK;
3581 		size = roundup(offset + size, PAGE_SIZE);
3582 		kva_free(base, size);
3583 	}
3584 #endif
3585 }
3586 
3587 /*
3588  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3589  * specified pmap. This eliminates the blast of soft faults on process startup
3590  * and immediately after an mmap.
3591  */
3592 static void
3593 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3594     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3595 {
3596 
3597 	VM_OBJECT_ASSERT_WLOCKED(object);
3598 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3599 	    ("mmu_booke_object_init_pt: non-device object"));
3600 }
3601 
3602 /*
3603  * Perform the pmap work for mincore.
3604  */
3605 static int
3606 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3607     vm_paddr_t *locked_pa)
3608 {
3609 
3610 	/* XXX: this should be implemented at some point */
3611 	return (0);
3612 }
3613 
3614 static int
3615 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3616     vm_memattr_t mode)
3617 {
3618 	vm_offset_t va;
3619 	pte_t *pte;
3620 	int i, j;
3621 	tlb_entry_t e;
3622 
3623 	/* Check TLB1 mappings */
3624 	for (i = 0; i < TLB1_ENTRIES; i++) {
3625 		tlb1_read_entry(&e, i);
3626 		if (!(e.mas1 & MAS1_VALID))
3627 			continue;
3628 		if (addr >= e.virt && addr < e.virt + e.size)
3629 			break;
3630 	}
3631 	if (i < TLB1_ENTRIES) {
3632 		/* Only allow full mappings to be modified for now. */
3633 		/* Validate the range. */
3634 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3635 			tlb1_read_entry(&e, j);
3636 			if (va != e.virt || (sz - (va - addr) < e.size))
3637 				return (EINVAL);
3638 		}
3639 		for (va = addr; va < addr + sz; va += e.size, i++) {
3640 			tlb1_read_entry(&e, i);
3641 			e.mas2 &= ~MAS2_WIMGE_MASK;
3642 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3643 
3644 			/*
3645 			 * Write it out to the TLB.  Should really re-sync with other
3646 			 * cores.
3647 			 */
3648 			tlb1_write_entry(&e, i);
3649 		}
3650 		return (0);
3651 	}
3652 
3653 	/* Not in TLB1, try through pmap */
3654 	/* First validate the range. */
3655 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3656 		pte = pte_find(mmu, kernel_pmap, va);
3657 		if (pte == NULL || !PTE_ISVALID(pte))
3658 			return (EINVAL);
3659 	}
3660 
3661 	mtx_lock_spin(&tlbivax_mutex);
3662 	tlb_miss_lock();
3663 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3664 		pte = pte_find(mmu, kernel_pmap, va);
3665 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3666 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3667 		tlb0_flush_entry(va);
3668 	}
3669 	tlb_miss_unlock();
3670 	mtx_unlock_spin(&tlbivax_mutex);
3671 
3672 	return (0);
3673 }
3674 
3675 /**************************************************************************/
3676 /* TID handling */
3677 /**************************************************************************/
3678 
3679 /*
3680  * Allocate a TID. If necessary, steal one from someone else.
3681  * The new TID is flushed from the TLB before returning.
3682  */
3683 static tlbtid_t
3684 tid_alloc(pmap_t pmap)
3685 {
3686 	tlbtid_t tid;
3687 	int thiscpu;
3688 
3689 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3690 
3691 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3692 
3693 	thiscpu = PCPU_GET(cpuid);
3694 
3695 	tid = PCPU_GET(tid_next);
3696 	if (tid > TID_MAX)
3697 		tid = TID_MIN;
3698 	PCPU_SET(tid_next, tid + 1);
3699 
3700 	/* If we are stealing TID then clear the relevant pmap's field */
3701 	if (tidbusy[thiscpu][tid] != NULL) {
3702 
3703 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3704 
3705 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3706 
3707 		/* Flush all entries from TLB0 matching this TID. */
3708 		tid_flush(tid);
3709 	}
3710 
3711 	tidbusy[thiscpu][tid] = pmap;
3712 	pmap->pm_tid[thiscpu] = tid;
3713 	__asm __volatile("msync; isync");
3714 
3715 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3716 	    PCPU_GET(tid_next));
3717 
3718 	return (tid);
3719 }
3720 
3721 /**************************************************************************/
3722 /* TLB0 handling */
3723 /**************************************************************************/
3724 
3725 static void
3726 #ifdef __powerpc64__
3727 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
3728 #else
3729 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3730 #endif
3731     uint32_t mas7)
3732 {
3733 	int as;
3734 	char desc[3];
3735 	tlbtid_t tid;
3736 	vm_size_t size;
3737 	unsigned int tsize;
3738 
3739 	desc[2] = '\0';
3740 	if (mas1 & MAS1_VALID)
3741 		desc[0] = 'V';
3742 	else
3743 		desc[0] = ' ';
3744 
3745 	if (mas1 & MAS1_IPROT)
3746 		desc[1] = 'P';
3747 	else
3748 		desc[1] = ' ';
3749 
3750 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3751 	tid = MAS1_GETTID(mas1);
3752 
3753 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3754 	size = 0;
3755 	if (tsize)
3756 		size = tsize2size(tsize);
3757 
3758 	debugf("%3d: (%s) [AS=%d] "
3759 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3760 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3761 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3762 }
3763 
3764 /* Convert TLB0 va and way number to tlb0[] table index. */
3765 static inline unsigned int
3766 tlb0_tableidx(vm_offset_t va, unsigned int way)
3767 {
3768 	unsigned int idx;
3769 
3770 	idx = (way * TLB0_ENTRIES_PER_WAY);
3771 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3772 	return (idx);
3773 }
3774 
3775 /*
3776  * Invalidate TLB0 entry.
3777  */
3778 static inline void
3779 tlb0_flush_entry(vm_offset_t va)
3780 {
3781 
3782 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3783 
3784 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3785 
3786 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3787 	__asm __volatile("isync; msync");
3788 	__asm __volatile("tlbsync; msync");
3789 
3790 	CTR1(KTR_PMAP, "%s: e", __func__);
3791 }
3792 
3793 /* Print out contents of the MAS registers for each TLB0 entry */
3794 void
3795 tlb0_print_tlbentries(void)
3796 {
3797 	uint32_t mas0, mas1, mas3, mas7;
3798 #ifdef __powerpc64__
3799 	uint64_t mas2;
3800 #else
3801 	uint32_t mas2;
3802 #endif
3803 	int entryidx, way, idx;
3804 
3805 	debugf("TLB0 entries:\n");
3806 	for (way = 0; way < TLB0_WAYS; way ++)
3807 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3808 
3809 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3810 			mtspr(SPR_MAS0, mas0);
3811 			__asm __volatile("isync");
3812 
3813 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3814 			mtspr(SPR_MAS2, mas2);
3815 
3816 			__asm __volatile("isync; tlbre");
3817 
3818 			mas1 = mfspr(SPR_MAS1);
3819 			mas2 = mfspr(SPR_MAS2);
3820 			mas3 = mfspr(SPR_MAS3);
3821 			mas7 = mfspr(SPR_MAS7);
3822 
3823 			idx = tlb0_tableidx(mas2, way);
3824 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3825 		}
3826 }
3827 
3828 /**************************************************************************/
3829 /* TLB1 handling */
3830 /**************************************************************************/
3831 
3832 /*
3833  * TLB1 mapping notes:
3834  *
3835  * TLB1[0]	Kernel text and data.
3836  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3837  *		windows, other devices mappings.
3838  */
3839 
3840  /*
3841  * Read an entry from given TLB1 slot.
3842  */
3843 void
3844 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3845 {
3846 	register_t msr;
3847 	uint32_t mas0;
3848 
3849 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3850 
3851 	msr = mfmsr();
3852 	__asm __volatile("wrteei 0");
3853 
3854 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3855 	mtspr(SPR_MAS0, mas0);
3856 	__asm __volatile("isync; tlbre");
3857 
3858 	entry->mas1 = mfspr(SPR_MAS1);
3859 	entry->mas2 = mfspr(SPR_MAS2);
3860 	entry->mas3 = mfspr(SPR_MAS3);
3861 
3862 	switch ((mfpvr() >> 16) & 0xFFFF) {
3863 	case FSL_E500v2:
3864 	case FSL_E500mc:
3865 	case FSL_E5500:
3866 	case FSL_E6500:
3867 		entry->mas7 = mfspr(SPR_MAS7);
3868 		break;
3869 	default:
3870 		entry->mas7 = 0;
3871 		break;
3872 	}
3873 	mtmsr(msr);
3874 
3875 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3876 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3877 	    (entry->mas3 & MAS3_RPN);
3878 	entry->size =
3879 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3880 }
3881 
3882 struct tlbwrite_args {
3883 	tlb_entry_t *e;
3884 	unsigned int idx;
3885 };
3886 
3887 static void
3888 tlb1_write_entry_int(void *arg)
3889 {
3890 	struct tlbwrite_args *args = arg;
3891 	uint32_t mas0;
3892 
3893 	/* Select entry */
3894 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3895 
3896 	mtspr(SPR_MAS0, mas0);
3897 	__asm __volatile("isync");
3898 	mtspr(SPR_MAS1, args->e->mas1);
3899 	__asm __volatile("isync");
3900 	mtspr(SPR_MAS2, args->e->mas2);
3901 	__asm __volatile("isync");
3902 	mtspr(SPR_MAS3, args->e->mas3);
3903 	__asm __volatile("isync");
3904 	switch ((mfpvr() >> 16) & 0xFFFF) {
3905 	case FSL_E500mc:
3906 	case FSL_E5500:
3907 	case FSL_E6500:
3908 		mtspr(SPR_MAS8, 0);
3909 		__asm __volatile("isync");
3910 		/* FALLTHROUGH */
3911 	case FSL_E500v2:
3912 		mtspr(SPR_MAS7, args->e->mas7);
3913 		__asm __volatile("isync");
3914 		break;
3915 	default:
3916 		break;
3917 	}
3918 
3919 	__asm __volatile("tlbwe; isync; msync");
3920 
3921 }
3922 
3923 static void
3924 tlb1_write_entry_sync(void *arg)
3925 {
3926 	/* Empty synchronization point for smp_rendezvous(). */
3927 }
3928 
3929 /*
3930  * Write given entry to TLB1 hardware.
3931  */
3932 static void
3933 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3934 {
3935 	struct tlbwrite_args args;
3936 
3937 	args.e = e;
3938 	args.idx = idx;
3939 
3940 #ifdef SMP
3941 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3942 		mb();
3943 		smp_rendezvous(tlb1_write_entry_sync,
3944 		    tlb1_write_entry_int,
3945 		    tlb1_write_entry_sync, &args);
3946 	} else
3947 #endif
3948 	{
3949 		register_t msr;
3950 
3951 		msr = mfmsr();
3952 		__asm __volatile("wrteei 0");
3953 		tlb1_write_entry_int(&args);
3954 		mtmsr(msr);
3955 	}
3956 }
3957 
3958 /*
3959  * Return the largest uint value log such that 2^log <= num.
3960  */
3961 static unsigned int
3962 ilog2(unsigned int num)
3963 {
3964 	int lz;
3965 
3966 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
3967 	return (31 - lz);
3968 }
3969 
3970 /*
3971  * Convert TLB TSIZE value to mapped region size.
3972  */
3973 static vm_size_t
3974 tsize2size(unsigned int tsize)
3975 {
3976 
3977 	/*
3978 	 * size = 4^tsize KB
3979 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3980 	 */
3981 
3982 	return ((1 << (2 * tsize)) * 1024);
3983 }
3984 
3985 /*
3986  * Convert region size (must be power of 4) to TLB TSIZE value.
3987  */
3988 static unsigned int
3989 size2tsize(vm_size_t size)
3990 {
3991 
3992 	return (ilog2(size) / 2 - 5);
3993 }
3994 
3995 /*
3996  * Register permanent kernel mapping in TLB1.
3997  *
3998  * Entries are created starting from index 0 (current free entry is
3999  * kept in tlb1_idx) and are not supposed to be invalidated.
4000  */
4001 int
4002 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
4003     uint32_t flags)
4004 {
4005 	tlb_entry_t e;
4006 	uint32_t ts, tid;
4007 	int tsize, index;
4008 
4009 	for (index = 0; index < TLB1_ENTRIES; index++) {
4010 		tlb1_read_entry(&e, index);
4011 		if ((e.mas1 & MAS1_VALID) == 0)
4012 			break;
4013 		/* Check if we're just updating the flags, and update them. */
4014 		if (e.phys == pa && e.virt == va && e.size == size) {
4015 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
4016 			tlb1_write_entry(&e, index);
4017 			return (0);
4018 		}
4019 	}
4020 	if (index >= TLB1_ENTRIES) {
4021 		printf("tlb1_set_entry: TLB1 full!\n");
4022 		return (-1);
4023 	}
4024 
4025 	/* Convert size to TSIZE */
4026 	tsize = size2tsize(size);
4027 
4028 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
4029 	/* XXX TS is hard coded to 0 for now as we only use single address space */
4030 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
4031 
4032 	e.phys = pa;
4033 	e.virt = va;
4034 	e.size = size;
4035 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
4036 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
4037 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
4038 
4039 	/* Set supervisor RWX permission bits */
4040 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
4041 	e.mas7 = (pa >> 32) & MAS7_RPN;
4042 
4043 	tlb1_write_entry(&e, index);
4044 
4045 	/*
4046 	 * XXX in general TLB1 updates should be propagated between CPUs,
4047 	 * since current design assumes to have the same TLB1 set-up on all
4048 	 * cores.
4049 	 */
4050 	return (0);
4051 }
4052 
4053 /*
4054  * Map in contiguous RAM region into the TLB1 using maximum of
4055  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
4056  *
4057  * If necessary round up last entry size and return total size
4058  * used by all allocated entries.
4059  */
4060 vm_size_t
4061 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
4062 {
4063 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
4064 	vm_size_t mapped, pgsz, base, mask;
4065 	int idx, nents;
4066 
4067 	/* Round up to the next 1M */
4068 	size = roundup2(size, 1 << 20);
4069 
4070 	mapped = 0;
4071 	idx = 0;
4072 	base = va;
4073 	pgsz = 64*1024*1024;
4074 	while (mapped < size) {
4075 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
4076 			while (pgsz > (size - mapped))
4077 				pgsz >>= 2;
4078 			pgs[idx++] = pgsz;
4079 			mapped += pgsz;
4080 		}
4081 
4082 		/* We under-map. Correct for this. */
4083 		if (mapped < size) {
4084 			while (pgs[idx - 1] == pgsz) {
4085 				idx--;
4086 				mapped -= pgsz;
4087 			}
4088 			/* XXX We may increase beyond out starting point. */
4089 			pgsz <<= 2;
4090 			pgs[idx++] = pgsz;
4091 			mapped += pgsz;
4092 		}
4093 	}
4094 
4095 	nents = idx;
4096 	mask = pgs[0] - 1;
4097 	/* Align address to the boundary */
4098 	if (va & mask) {
4099 		va = (va + mask) & ~mask;
4100 		pa = (pa + mask) & ~mask;
4101 	}
4102 
4103 	for (idx = 0; idx < nents; idx++) {
4104 		pgsz = pgs[idx];
4105 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
4106 		tlb1_set_entry(va, pa, pgsz,
4107 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4108 		pa += pgsz;
4109 		va += pgsz;
4110 	}
4111 
4112 	mapped = (va - base);
4113 	printf("mapped size 0x%"PRI0ptrX" (wasted space 0x%"PRIxPTR")\n",
4114 	    mapped, mapped - size);
4115 	return (mapped);
4116 }
4117 
4118 /*
4119  * TLB1 initialization routine, to be called after the very first
4120  * assembler level setup done in locore.S.
4121  */
4122 void
4123 tlb1_init()
4124 {
4125 	uint32_t mas0, mas1, mas2, mas3, mas7;
4126 	uint32_t tsz;
4127 
4128 	tlb1_get_tlbconf();
4129 
4130 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4131 	mtspr(SPR_MAS0, mas0);
4132 	__asm __volatile("isync; tlbre");
4133 
4134 	mas1 = mfspr(SPR_MAS1);
4135 	mas2 = mfspr(SPR_MAS2);
4136 	mas3 = mfspr(SPR_MAS3);
4137 	mas7 = mfspr(SPR_MAS7);
4138 
4139 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4140 	    (mas3 & MAS3_RPN);
4141 
4142 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4143 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4144 
4145 	/* Setup TLB miss defaults */
4146 	set_mas4_defaults();
4147 }
4148 
4149 /*
4150  * pmap_early_io_unmap() should be used in short conjunction with
4151  * pmap_early_io_map(), as in the following snippet:
4152  *
4153  * x = pmap_early_io_map(...);
4154  * <do something with x>
4155  * pmap_early_io_unmap(x, size);
4156  *
4157  * And avoiding more allocations between.
4158  */
4159 void
4160 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4161 {
4162 	int i;
4163 	tlb_entry_t e;
4164 	vm_size_t isize;
4165 
4166 	size = roundup(size, PAGE_SIZE);
4167 	isize = size;
4168 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4169 		tlb1_read_entry(&e, i);
4170 		if (!(e.mas1 & MAS1_VALID))
4171 			continue;
4172 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4173 			size -= e.size;
4174 			e.mas1 &= ~MAS1_VALID;
4175 			tlb1_write_entry(&e, i);
4176 		}
4177 	}
4178 	if (tlb1_map_base == va + isize)
4179 		tlb1_map_base -= isize;
4180 }
4181 
4182 vm_offset_t
4183 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4184 {
4185 	vm_paddr_t pa_base;
4186 	vm_offset_t va, sz;
4187 	int i;
4188 	tlb_entry_t e;
4189 
4190 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4191 
4192 	for (i = 0; i < TLB1_ENTRIES; i++) {
4193 		tlb1_read_entry(&e, i);
4194 		if (!(e.mas1 & MAS1_VALID))
4195 			continue;
4196 		if (pa >= e.phys && (pa + size) <=
4197 		    (e.phys + e.size))
4198 			return (e.virt + (pa - e.phys));
4199 	}
4200 
4201 	pa_base = rounddown(pa, PAGE_SIZE);
4202 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4203 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4204 	va = tlb1_map_base + (pa - pa_base);
4205 
4206 	do {
4207 		sz = 1 << (ilog2(size) & ~1);
4208 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4209 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4210 		size -= sz;
4211 		pa_base += sz;
4212 		tlb1_map_base += sz;
4213 	} while (size > 0);
4214 
4215 	return (va);
4216 }
4217 
4218 void
4219 pmap_track_page(pmap_t pmap, vm_offset_t va)
4220 {
4221 	vm_paddr_t pa;
4222 	vm_page_t page;
4223 	struct pv_entry *pve;
4224 
4225 	va = trunc_page(va);
4226 	pa = pmap_kextract(va);
4227 	page = PHYS_TO_VM_PAGE(pa);
4228 
4229 	rw_wlock(&pvh_global_lock);
4230 	PMAP_LOCK(pmap);
4231 
4232 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4233 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4234 			goto out;
4235 		}
4236 	}
4237 	page->md.pv_tracked = true;
4238 	pv_insert(pmap, va, page);
4239 out:
4240 	PMAP_UNLOCK(pmap);
4241 	rw_wunlock(&pvh_global_lock);
4242 }
4243 
4244 
4245 /*
4246  * Setup MAS4 defaults.
4247  * These values are loaded to MAS0-2 on a TLB miss.
4248  */
4249 static void
4250 set_mas4_defaults(void)
4251 {
4252 	uint32_t mas4;
4253 
4254 	/* Defaults: TLB0, PID0, TSIZED=4K */
4255 	mas4 = MAS4_TLBSELD0;
4256 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4257 #ifdef SMP
4258 	mas4 |= MAS4_MD;
4259 #endif
4260 	mtspr(SPR_MAS4, mas4);
4261 	__asm __volatile("isync");
4262 }
4263 
4264 /*
4265  * Print out contents of the MAS registers for each TLB1 entry
4266  */
4267 void
4268 tlb1_print_tlbentries(void)
4269 {
4270 	uint32_t mas0, mas1, mas3, mas7;
4271 #ifdef __powerpc64__
4272 	uint64_t mas2;
4273 #else
4274 	uint32_t mas2;
4275 #endif
4276 	int i;
4277 
4278 	debugf("TLB1 entries:\n");
4279 	for (i = 0; i < TLB1_ENTRIES; i++) {
4280 
4281 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4282 		mtspr(SPR_MAS0, mas0);
4283 
4284 		__asm __volatile("isync; tlbre");
4285 
4286 		mas1 = mfspr(SPR_MAS1);
4287 		mas2 = mfspr(SPR_MAS2);
4288 		mas3 = mfspr(SPR_MAS3);
4289 		mas7 = mfspr(SPR_MAS7);
4290 
4291 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4292 	}
4293 }
4294 
4295 /*
4296  * Return 0 if the physical IO range is encompassed by one of the
4297  * the TLB1 entries, otherwise return related error code.
4298  */
4299 static int
4300 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4301 {
4302 	uint32_t prot;
4303 	vm_paddr_t pa_start;
4304 	vm_paddr_t pa_end;
4305 	unsigned int entry_tsize;
4306 	vm_size_t entry_size;
4307 	tlb_entry_t e;
4308 
4309 	*va = (vm_offset_t)NULL;
4310 
4311 	tlb1_read_entry(&e, i);
4312 	/* Skip invalid entries */
4313 	if (!(e.mas1 & MAS1_VALID))
4314 		return (EINVAL);
4315 
4316 	/*
4317 	 * The entry must be cache-inhibited, guarded, and r/w
4318 	 * so it can function as an i/o page
4319 	 */
4320 	prot = e.mas2 & (MAS2_I | MAS2_G);
4321 	if (prot != (MAS2_I | MAS2_G))
4322 		return (EPERM);
4323 
4324 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4325 	if (prot != (MAS3_SR | MAS3_SW))
4326 		return (EPERM);
4327 
4328 	/* The address should be within the entry range. */
4329 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4330 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4331 
4332 	entry_size = tsize2size(entry_tsize);
4333 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4334 	    (e.mas3 & MAS3_RPN);
4335 	pa_end = pa_start + entry_size;
4336 
4337 	if ((pa < pa_start) || ((pa + size) > pa_end))
4338 		return (ERANGE);
4339 
4340 	/* Return virtual address of this mapping. */
4341 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4342 	return (0);
4343 }
4344 
4345 /*
4346  * Invalidate all TLB0 entries which match the given TID. Note this is
4347  * dedicated for cases when invalidations should NOT be propagated to other
4348  * CPUs.
4349  */
4350 static void
4351 tid_flush(tlbtid_t tid)
4352 {
4353 	register_t msr;
4354 	uint32_t mas0, mas1, mas2;
4355 	int entry, way;
4356 
4357 
4358 	/* Don't evict kernel translations */
4359 	if (tid == TID_KERNEL)
4360 		return;
4361 
4362 	msr = mfmsr();
4363 	__asm __volatile("wrteei 0");
4364 
4365 	for (way = 0; way < TLB0_WAYS; way++)
4366 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4367 
4368 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4369 			mtspr(SPR_MAS0, mas0);
4370 			__asm __volatile("isync");
4371 
4372 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4373 			mtspr(SPR_MAS2, mas2);
4374 
4375 			__asm __volatile("isync; tlbre");
4376 
4377 			mas1 = mfspr(SPR_MAS1);
4378 
4379 			if (!(mas1 & MAS1_VALID))
4380 				continue;
4381 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4382 				continue;
4383 			mas1 &= ~MAS1_VALID;
4384 			mtspr(SPR_MAS1, mas1);
4385 			__asm __volatile("isync; tlbwe; isync; msync");
4386 		}
4387 	mtmsr(msr);
4388 }
4389