xref: /freebsd/sys/powerpc/booke/pmap.c (revision 3982006ed5587cfd83cc1955186e0aa4b92b3fcd)
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 static int		mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm,
384     volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen);
385 static int		mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr,
386     int *is_user, vm_offset_t *decoded_addr);
387 
388 
389 static mmu_method_t mmu_booke_methods[] = {
390 	/* pmap dispatcher interface */
391 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
392 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
393 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
394 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
395 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
396 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
397 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
398 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
399 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
400 	MMUMETHOD(mmu_init,		mmu_booke_init),
401 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
402 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
403 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
404 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
405 	MMUMETHOD(mmu_map,		mmu_booke_map),
406 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
407 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
408 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
409 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
410 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
411 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
412 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
413 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
414 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
415 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
416 	MMUMETHOD(mmu_release,		mmu_booke_release),
417 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
418 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
419 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
420 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
421 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
422 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
423 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
424 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
425 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
426 	MMUMETHOD(mmu_quick_enter_page, mmu_booke_quick_enter_page),
427 	MMUMETHOD(mmu_quick_remove_page, mmu_booke_quick_remove_page),
428 
429 	/* Internal interfaces */
430 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
431 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
432 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
433 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
434 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
435 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
436 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
437 	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),
438 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
439 	MMUMETHOD(mmu_change_attr,	mmu_booke_change_attr),
440 	MMUMETHOD(mmu_map_user_ptr,	mmu_booke_map_user_ptr),
441 	MMUMETHOD(mmu_decode_kernel_ptr, mmu_booke_decode_kernel_ptr),
442 
443 	/* dumpsys() support */
444 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
445 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
446 	MMUMETHOD(mmu_scan_init,	mmu_booke_scan_init),
447 
448 	{ 0, 0 }
449 };
450 
451 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
452 
453 static __inline uint32_t
454 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
455 {
456 	uint32_t attrib;
457 	int i;
458 
459 	if (ma != VM_MEMATTR_DEFAULT) {
460 		switch (ma) {
461 		case VM_MEMATTR_UNCACHEABLE:
462 			return (MAS2_I | MAS2_G);
463 		case VM_MEMATTR_WRITE_COMBINING:
464 		case VM_MEMATTR_WRITE_BACK:
465 		case VM_MEMATTR_PREFETCHABLE:
466 			return (MAS2_I);
467 		case VM_MEMATTR_WRITE_THROUGH:
468 			return (MAS2_W | MAS2_M);
469 		case VM_MEMATTR_CACHEABLE:
470 			return (MAS2_M);
471 		}
472 	}
473 
474 	/*
475 	 * Assume the page is cache inhibited and access is guarded unless
476 	 * it's in our available memory array.
477 	 */
478 	attrib = _TLB_ENTRY_IO;
479 	for (i = 0; i < physmem_regions_sz; i++) {
480 		if ((pa >= physmem_regions[i].mr_start) &&
481 		    (pa < (physmem_regions[i].mr_start +
482 		     physmem_regions[i].mr_size))) {
483 			attrib = _TLB_ENTRY_MEM;
484 			break;
485 		}
486 	}
487 
488 	return (attrib);
489 }
490 
491 static inline void
492 tlb_miss_lock(void)
493 {
494 #ifdef SMP
495 	struct pcpu *pc;
496 
497 	if (!smp_started)
498 		return;
499 
500 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
501 		if (pc != pcpup) {
502 
503 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
504 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
505 
506 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
507 			    ("tlb_miss_lock: tried to lock self"));
508 
509 			tlb_lock(pc->pc_booke.tlb_lock);
510 
511 			CTR1(KTR_PMAP, "%s: locked", __func__);
512 		}
513 	}
514 #endif
515 }
516 
517 static inline void
518 tlb_miss_unlock(void)
519 {
520 #ifdef SMP
521 	struct pcpu *pc;
522 
523 	if (!smp_started)
524 		return;
525 
526 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
527 		if (pc != pcpup) {
528 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
529 			    __func__, pc->pc_cpuid);
530 
531 			tlb_unlock(pc->pc_booke.tlb_lock);
532 
533 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
534 		}
535 	}
536 #endif
537 }
538 
539 /* Return number of entries in TLB0. */
540 static __inline void
541 tlb0_get_tlbconf(void)
542 {
543 	uint32_t tlb0_cfg;
544 
545 	tlb0_cfg = mfspr(SPR_TLB0CFG);
546 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
547 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
548 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
549 }
550 
551 /* Return number of entries in TLB1. */
552 static __inline void
553 tlb1_get_tlbconf(void)
554 {
555 	uint32_t tlb1_cfg;
556 
557 	tlb1_cfg = mfspr(SPR_TLB1CFG);
558 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
559 }
560 
561 /**************************************************************************/
562 /* Page table related */
563 /**************************************************************************/
564 
565 #ifdef __powerpc64__
566 /* Initialize pool of kva ptbl buffers. */
567 static void
568 ptbl_init(void)
569 {
570 	int		i;
571 
572 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
573 	TAILQ_INIT(&ptbl_buf_freelist);
574 
575 	for (i = 0; i < PTBL_BUFS; i++) {
576 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase +
577 		    i * MAX(PTBL_PAGES,PDIR_PAGES) * PAGE_SIZE;
578 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
579 	}
580 }
581 
582 /* Get an sf_buf from the freelist. */
583 static struct ptbl_buf *
584 ptbl_buf_alloc(void)
585 {
586 	struct ptbl_buf *buf;
587 
588 	mtx_lock(&ptbl_buf_freelist_lock);
589 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
590 	if (buf != NULL)
591 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
592 	mtx_unlock(&ptbl_buf_freelist_lock);
593 
594 	return (buf);
595 }
596 
597 /* Return ptbl buff to free pool. */
598 static void
599 ptbl_buf_free(struct ptbl_buf *buf)
600 {
601 	mtx_lock(&ptbl_buf_freelist_lock);
602 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
603 	mtx_unlock(&ptbl_buf_freelist_lock);
604 }
605 
606 /*
607  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
608  */
609 static void
610 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t * ptbl)
611 {
612 	struct ptbl_buf *pbuf;
613 
614 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) {
615 		if (pbuf->kva == (vm_offset_t) ptbl) {
616 			/* Remove from pmap ptbl buf list. */
617 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
618 
619 			/* Free corresponding ptbl buf. */
620 			ptbl_buf_free(pbuf);
621 
622 			break;
623 		}
624 	}
625 }
626 
627 /* Get a pointer to a PTE in a page table. */
628 static __inline pte_t *
629 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
630 {
631 	pte_t         **pdir;
632 	pte_t          *ptbl;
633 
634 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
635 
636 	pdir = pmap->pm_pp2d[PP2D_IDX(va)];
637 	if (!pdir)
638 		return NULL;
639 	ptbl = pdir[PDIR_IDX(va)];
640 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
641 }
642 
643 /*
644  * Search the list of allocated pdir bufs and find on list of allocated pdirs
645  */
646 static void
647 ptbl_free_pmap_pdir(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
648 {
649 	struct ptbl_buf *pbuf;
650 
651 	TAILQ_FOREACH(pbuf, &pmap->pm_pdir_list, link) {
652 		if (pbuf->kva == (vm_offset_t) pdir) {
653 			/* Remove from pmap ptbl buf list. */
654 			TAILQ_REMOVE(&pmap->pm_pdir_list, pbuf, link);
655 
656 			/* Free corresponding pdir buf. */
657 			ptbl_buf_free(pbuf);
658 
659 			break;
660 		}
661 	}
662 }
663 /* Free pdir pages and invalidate pdir entry. */
664 static void
665 pdir_free(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx)
666 {
667 	pte_t         **pdir;
668 	vm_paddr_t	pa;
669 	vm_offset_t	va;
670 	vm_page_t	m;
671 	int		i;
672 
673 	pdir = pmap->pm_pp2d[pp2d_idx];
674 
675 	KASSERT((pdir != NULL), ("pdir_free: null pdir"));
676 
677 	pmap->pm_pp2d[pp2d_idx] = NULL;
678 
679 	for (i = 0; i < PDIR_PAGES; i++) {
680 		va = ((vm_offset_t) pdir + (i * PAGE_SIZE));
681 		pa = pte_vatopa(mmu, kernel_pmap, va);
682 		m = PHYS_TO_VM_PAGE(pa);
683 		vm_page_free_zero(m);
684 		vm_wire_sub(1);
685 		pmap_kremove(va);
686 	}
687 
688 	ptbl_free_pmap_pdir(mmu, pmap, pdir);
689 }
690 
691 /*
692  * Decrement pdir pages hold count and attempt to free pdir pages. Called
693  * when removing directory entry from pdir.
694  *
695  * Return 1 if pdir pages were freed.
696  */
697 static int
698 pdir_unhold(mmu_t mmu, pmap_t pmap, u_int pp2d_idx)
699 {
700 	pte_t         **pdir;
701 	vm_paddr_t	pa;
702 	vm_page_t	m;
703 	int		i;
704 
705 	KASSERT((pmap != kernel_pmap),
706 		("pdir_unhold: unholding kernel pdir!"));
707 
708 	pdir = pmap->pm_pp2d[pp2d_idx];
709 
710 	KASSERT(((vm_offset_t) pdir >= VM_MIN_KERNEL_ADDRESS),
711 	    ("pdir_unhold: non kva pdir"));
712 
713 	/* decrement hold count */
714 	for (i = 0; i < PDIR_PAGES; i++) {
715 		pa = pte_vatopa(mmu, kernel_pmap,
716 		    (vm_offset_t) pdir + (i * PAGE_SIZE));
717 		m = PHYS_TO_VM_PAGE(pa);
718 		m->wire_count--;
719 	}
720 
721 	/*
722 	 * Free pdir pages if there are no dir entries in this pdir.
723 	 * wire_count has the same value for all ptbl pages, so check the
724 	 * last page.
725 	 */
726 	if (m->wire_count == 0) {
727 		pdir_free(mmu, pmap, pp2d_idx);
728 		return (1);
729 	}
730 	return (0);
731 }
732 
733 /*
734  * Increment hold count for pdir pages. This routine is used when new ptlb
735  * entry is being inserted into pdir.
736  */
737 static void
738 pdir_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir)
739 {
740 	vm_paddr_t	pa;
741 	vm_page_t	m;
742 	int		i;
743 
744 	KASSERT((pmap != kernel_pmap),
745 		("pdir_hold: holding kernel pdir!"));
746 
747 	KASSERT((pdir != NULL), ("pdir_hold: null pdir"));
748 
749 	for (i = 0; i < PDIR_PAGES; i++) {
750 		pa = pte_vatopa(mmu, kernel_pmap,
751 				(vm_offset_t) pdir + (i * PAGE_SIZE));
752 		m = PHYS_TO_VM_PAGE(pa);
753 		m->wire_count++;
754 	}
755 }
756 
757 /* Allocate page table. */
758 static pte_t   *
759 ptbl_alloc(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx,
760     boolean_t nosleep)
761 {
762 	vm_page_t	mtbl  [PTBL_PAGES];
763 	vm_page_t	m;
764 	struct ptbl_buf *pbuf;
765 	unsigned int	pidx;
766 	pte_t          *ptbl;
767 	int		i, j;
768 	int		req;
769 
770 	KASSERT((pdir[pdir_idx] == NULL),
771 		("%s: valid ptbl entry exists!", __func__));
772 
773 	pbuf = ptbl_buf_alloc();
774 	if (pbuf == NULL)
775 		panic("%s: couldn't alloc kernel virtual memory", __func__);
776 
777 	ptbl = (pte_t *) pbuf->kva;
778 
779 	for (i = 0; i < PTBL_PAGES; i++) {
780 		pidx = (PTBL_PAGES * pdir_idx) + i;
781 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
782 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
783 			PMAP_UNLOCK(pmap);
784 			rw_wunlock(&pvh_global_lock);
785 			if (nosleep) {
786 				ptbl_free_pmap_ptbl(pmap, ptbl);
787 				for (j = 0; j < i; j++)
788 					vm_page_free(mtbl[j]);
789 				vm_wire_sub(i);
790 				return (NULL);
791 			}
792 			vm_wait(NULL);
793 			rw_wlock(&pvh_global_lock);
794 			PMAP_LOCK(pmap);
795 		}
796 		mtbl[i] = m;
797 	}
798 
799 	/* Mapin allocated pages into kernel_pmap. */
800 	mmu_booke_qenter(mmu, (vm_offset_t) ptbl, mtbl, PTBL_PAGES);
801 	/* Zero whole ptbl. */
802 	bzero((caddr_t) ptbl, PTBL_PAGES * PAGE_SIZE);
803 
804 	/* Add pbuf to the pmap ptbl bufs list. */
805 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
806 
807 	return (ptbl);
808 }
809 
810 /* Free ptbl pages and invalidate pdir entry. */
811 static void
812 ptbl_free(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
813 {
814 	pte_t          *ptbl;
815 	vm_paddr_t	pa;
816 	vm_offset_t	va;
817 	vm_page_t	m;
818 	int		i;
819 
820 	ptbl = pdir[pdir_idx];
821 
822 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
823 
824 	pdir[pdir_idx] = NULL;
825 
826 	for (i = 0; i < PTBL_PAGES; i++) {
827 		va = ((vm_offset_t) ptbl + (i * PAGE_SIZE));
828 		pa = pte_vatopa(mmu, kernel_pmap, va);
829 		m = PHYS_TO_VM_PAGE(pa);
830 		vm_page_free_zero(m);
831 		vm_wire_sub(1);
832 		pmap_kremove(va);
833 	}
834 
835 	ptbl_free_pmap_ptbl(pmap, ptbl);
836 }
837 
838 /*
839  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
840  * when removing pte entry from ptbl.
841  *
842  * Return 1 if ptbl pages were freed.
843  */
844 static int
845 ptbl_unhold(mmu_t mmu, pmap_t pmap, vm_offset_t va)
846 {
847 	pte_t          *ptbl;
848 	vm_paddr_t	pa;
849 	vm_page_t	m;
850 	u_int		pp2d_idx;
851 	pte_t         **pdir;
852 	u_int		pdir_idx;
853 	int		i;
854 
855 	pp2d_idx = PP2D_IDX(va);
856 	pdir_idx = PDIR_IDX(va);
857 
858 	KASSERT((pmap != kernel_pmap),
859 		("ptbl_unhold: unholding kernel ptbl!"));
860 
861 	pdir = pmap->pm_pp2d[pp2d_idx];
862 	ptbl = pdir[pdir_idx];
863 
864 	KASSERT(((vm_offset_t) ptbl >= VM_MIN_KERNEL_ADDRESS),
865 	    ("ptbl_unhold: non kva ptbl"));
866 
867 	/* decrement hold count */
868 	for (i = 0; i < PTBL_PAGES; i++) {
869 		pa = pte_vatopa(mmu, kernel_pmap,
870 		    (vm_offset_t) ptbl + (i * PAGE_SIZE));
871 		m = PHYS_TO_VM_PAGE(pa);
872 		m->wire_count--;
873 	}
874 
875 	/*
876 	 * Free ptbl pages if there are no pte entries in this ptbl.
877 	 * wire_count has the same value for all ptbl pages, so check the
878 	 * last page.
879 	 */
880 	if (m->wire_count == 0) {
881 		/* A pair of indirect entries might point to this ptbl page */
882 #if 0
883 		tlb_flush_entry(pmap, va & ~((2UL * PAGE_SIZE_1M) - 1),
884 				TLB_SIZE_1M, MAS6_SIND);
885 		tlb_flush_entry(pmap, (va & ~((2UL * PAGE_SIZE_1M) - 1)) | PAGE_SIZE_1M,
886 				TLB_SIZE_1M, MAS6_SIND);
887 #endif
888 		ptbl_free(mmu, pmap, pdir, pdir_idx);
889 		pdir_unhold(mmu, pmap, pp2d_idx);
890 		return (1);
891 	}
892 	return (0);
893 }
894 
895 /*
896  * Increment hold count for ptbl pages. This routine is used when new pte
897  * entry is being inserted into ptbl.
898  */
899 static void
900 ptbl_hold(mmu_t mmu, pmap_t pmap, pte_t ** pdir, unsigned int pdir_idx)
901 {
902 	vm_paddr_t	pa;
903 	pte_t          *ptbl;
904 	vm_page_t	m;
905 	int		i;
906 
907 	KASSERT((pmap != kernel_pmap),
908 		("ptbl_hold: holding kernel ptbl!"));
909 
910 	ptbl = pdir[pdir_idx];
911 
912 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
913 
914 	for (i = 0; i < PTBL_PAGES; i++) {
915 		pa = pte_vatopa(mmu, kernel_pmap,
916 				(vm_offset_t) ptbl + (i * PAGE_SIZE));
917 		m = PHYS_TO_VM_PAGE(pa);
918 		m->wire_count++;
919 	}
920 }
921 #else
922 
923 /* Initialize pool of kva ptbl buffers. */
924 static void
925 ptbl_init(void)
926 {
927 	int i;
928 
929 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
930 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
931 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
932 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
933 
934 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
935 	TAILQ_INIT(&ptbl_buf_freelist);
936 
937 	for (i = 0; i < PTBL_BUFS; i++) {
938 		ptbl_bufs[i].kva =
939 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
940 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
941 	}
942 }
943 
944 /* Get a ptbl_buf from the freelist. */
945 static struct ptbl_buf *
946 ptbl_buf_alloc(void)
947 {
948 	struct ptbl_buf *buf;
949 
950 	mtx_lock(&ptbl_buf_freelist_lock);
951 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
952 	if (buf != NULL)
953 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
954 	mtx_unlock(&ptbl_buf_freelist_lock);
955 
956 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
957 
958 	return (buf);
959 }
960 
961 /* Return ptbl buff to free pool. */
962 static void
963 ptbl_buf_free(struct ptbl_buf *buf)
964 {
965 
966 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
967 
968 	mtx_lock(&ptbl_buf_freelist_lock);
969 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
970 	mtx_unlock(&ptbl_buf_freelist_lock);
971 }
972 
973 /*
974  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
975  */
976 static void
977 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
978 {
979 	struct ptbl_buf *pbuf;
980 
981 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
982 
983 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
984 
985 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
986 		if (pbuf->kva == (vm_offset_t)ptbl) {
987 			/* Remove from pmap ptbl buf list. */
988 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
989 
990 			/* Free corresponding ptbl buf. */
991 			ptbl_buf_free(pbuf);
992 			break;
993 		}
994 }
995 
996 /* Allocate page table. */
997 static pte_t *
998 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
999 {
1000 	vm_page_t mtbl[PTBL_PAGES];
1001 	vm_page_t m;
1002 	struct ptbl_buf *pbuf;
1003 	unsigned int pidx;
1004 	pte_t *ptbl;
1005 	int i, j;
1006 
1007 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1008 	    (pmap == kernel_pmap), pdir_idx);
1009 
1010 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1011 	    ("ptbl_alloc: invalid pdir_idx"));
1012 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
1013 	    ("pte_alloc: valid ptbl entry exists!"));
1014 
1015 	pbuf = ptbl_buf_alloc();
1016 	if (pbuf == NULL)
1017 		panic("pte_alloc: couldn't alloc kernel virtual memory");
1018 
1019 	ptbl = (pte_t *)pbuf->kva;
1020 
1021 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
1022 
1023 	for (i = 0; i < PTBL_PAGES; i++) {
1024 		pidx = (PTBL_PAGES * pdir_idx) + i;
1025 		while ((m = vm_page_alloc(NULL, pidx,
1026 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
1027 			PMAP_UNLOCK(pmap);
1028 			rw_wunlock(&pvh_global_lock);
1029 			if (nosleep) {
1030 				ptbl_free_pmap_ptbl(pmap, ptbl);
1031 				for (j = 0; j < i; j++)
1032 					vm_page_free(mtbl[j]);
1033 				vm_wire_sub(i);
1034 				return (NULL);
1035 			}
1036 			vm_wait(NULL);
1037 			rw_wlock(&pvh_global_lock);
1038 			PMAP_LOCK(pmap);
1039 		}
1040 		mtbl[i] = m;
1041 	}
1042 
1043 	/* Map allocated pages into kernel_pmap. */
1044 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
1045 
1046 	/* Zero whole ptbl. */
1047 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
1048 
1049 	/* Add pbuf to the pmap ptbl bufs list. */
1050 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
1051 
1052 	return (ptbl);
1053 }
1054 
1055 /* Free ptbl pages and invalidate pdir entry. */
1056 static void
1057 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1058 {
1059 	pte_t *ptbl;
1060 	vm_paddr_t pa;
1061 	vm_offset_t va;
1062 	vm_page_t m;
1063 	int i;
1064 
1065 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1066 	    (pmap == kernel_pmap), pdir_idx);
1067 
1068 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1069 	    ("ptbl_free: invalid pdir_idx"));
1070 
1071 	ptbl = pmap->pm_pdir[pdir_idx];
1072 
1073 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
1074 
1075 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
1076 
1077 	/*
1078 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
1079 	 * don't attempt to look up the page tables we are releasing.
1080 	 */
1081 	mtx_lock_spin(&tlbivax_mutex);
1082 	tlb_miss_lock();
1083 
1084 	pmap->pm_pdir[pdir_idx] = NULL;
1085 
1086 	tlb_miss_unlock();
1087 	mtx_unlock_spin(&tlbivax_mutex);
1088 
1089 	for (i = 0; i < PTBL_PAGES; i++) {
1090 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
1091 		pa = pte_vatopa(mmu, kernel_pmap, va);
1092 		m = PHYS_TO_VM_PAGE(pa);
1093 		vm_page_free_zero(m);
1094 		vm_wire_sub(1);
1095 		mmu_booke_kremove(mmu, va);
1096 	}
1097 
1098 	ptbl_free_pmap_ptbl(pmap, ptbl);
1099 }
1100 
1101 /*
1102  * Decrement ptbl pages hold count and attempt to free ptbl pages.
1103  * Called when removing pte entry from ptbl.
1104  *
1105  * Return 1 if ptbl pages were freed.
1106  */
1107 static int
1108 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1109 {
1110 	pte_t *ptbl;
1111 	vm_paddr_t pa;
1112 	vm_page_t m;
1113 	int i;
1114 
1115 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
1116 	    (pmap == kernel_pmap), pdir_idx);
1117 
1118 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1119 	    ("ptbl_unhold: invalid pdir_idx"));
1120 	KASSERT((pmap != kernel_pmap),
1121 	    ("ptbl_unhold: unholding kernel ptbl!"));
1122 
1123 	ptbl = pmap->pm_pdir[pdir_idx];
1124 
1125 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
1126 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
1127 	    ("ptbl_unhold: non kva ptbl"));
1128 
1129 	/* decrement hold count */
1130 	for (i = 0; i < PTBL_PAGES; i++) {
1131 		pa = pte_vatopa(mmu, kernel_pmap,
1132 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1133 		m = PHYS_TO_VM_PAGE(pa);
1134 		m->wire_count--;
1135 	}
1136 
1137 	/*
1138 	 * Free ptbl pages if there are no pte etries in this ptbl.
1139 	 * wire_count has the same value for all ptbl pages, so check the last
1140 	 * page.
1141 	 */
1142 	if (m->wire_count == 0) {
1143 		ptbl_free(mmu, pmap, pdir_idx);
1144 
1145 		//debugf("ptbl_unhold: e (freed ptbl)\n");
1146 		return (1);
1147 	}
1148 
1149 	return (0);
1150 }
1151 
1152 /*
1153  * Increment hold count for ptbl pages. This routine is used when a new pte
1154  * entry is being inserted into the ptbl.
1155  */
1156 static void
1157 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
1158 {
1159 	vm_paddr_t pa;
1160 	pte_t *ptbl;
1161 	vm_page_t m;
1162 	int i;
1163 
1164 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
1165 	    pdir_idx);
1166 
1167 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
1168 	    ("ptbl_hold: invalid pdir_idx"));
1169 	KASSERT((pmap != kernel_pmap),
1170 	    ("ptbl_hold: holding kernel ptbl!"));
1171 
1172 	ptbl = pmap->pm_pdir[pdir_idx];
1173 
1174 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
1175 
1176 	for (i = 0; i < PTBL_PAGES; i++) {
1177 		pa = pte_vatopa(mmu, kernel_pmap,
1178 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
1179 		m = PHYS_TO_VM_PAGE(pa);
1180 		m->wire_count++;
1181 	}
1182 }
1183 #endif
1184 
1185 /* Allocate pv_entry structure. */
1186 pv_entry_t
1187 pv_alloc(void)
1188 {
1189 	pv_entry_t pv;
1190 
1191 	pv_entry_count++;
1192 	if (pv_entry_count > pv_entry_high_water)
1193 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
1194 	pv = uma_zalloc(pvzone, M_NOWAIT);
1195 
1196 	return (pv);
1197 }
1198 
1199 /* Free pv_entry structure. */
1200 static __inline void
1201 pv_free(pv_entry_t pve)
1202 {
1203 
1204 	pv_entry_count--;
1205 	uma_zfree(pvzone, pve);
1206 }
1207 
1208 
1209 /* Allocate and initialize pv_entry structure. */
1210 static void
1211 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
1212 {
1213 	pv_entry_t pve;
1214 
1215 	//int su = (pmap == kernel_pmap);
1216 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
1217 	//	(u_int32_t)pmap, va, (u_int32_t)m);
1218 
1219 	pve = pv_alloc();
1220 	if (pve == NULL)
1221 		panic("pv_insert: no pv entries!");
1222 
1223 	pve->pv_pmap = pmap;
1224 	pve->pv_va = va;
1225 
1226 	/* add to pv_list */
1227 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1228 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1229 
1230 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
1231 
1232 	//debugf("pv_insert: e\n");
1233 }
1234 
1235 /* Destroy pv entry. */
1236 static void
1237 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
1238 {
1239 	pv_entry_t pve;
1240 
1241 	//int su = (pmap == kernel_pmap);
1242 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
1243 
1244 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1245 	rw_assert(&pvh_global_lock, RA_WLOCKED);
1246 
1247 	/* find pv entry */
1248 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
1249 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
1250 			/* remove from pv_list */
1251 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
1252 			if (TAILQ_EMPTY(&m->md.pv_list))
1253 				vm_page_aflag_clear(m, PGA_WRITEABLE);
1254 
1255 			/* free pv entry struct */
1256 			pv_free(pve);
1257 			break;
1258 		}
1259 	}
1260 
1261 	//debugf("pv_remove: e\n");
1262 }
1263 
1264 #ifdef __powerpc64__
1265 /*
1266  * Clean pte entry, try to free page table page if requested.
1267  *
1268  * Return 1 if ptbl pages were freed, otherwise return 0.
1269  */
1270 static int
1271 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, u_int8_t flags)
1272 {
1273 	vm_page_t	m;
1274 	pte_t          *pte;
1275 
1276 	pte = pte_find(mmu, pmap, va);
1277 	KASSERT(pte != NULL, ("%s: NULL pte", __func__));
1278 
1279 	if (!PTE_ISVALID(pte))
1280 		return (0);
1281 
1282 	/* Get vm_page_t for mapped pte. */
1283 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1284 
1285 	if (PTE_ISWIRED(pte))
1286 		pmap->pm_stats.wired_count--;
1287 
1288 	/* Handle managed entry. */
1289 	if (PTE_ISMANAGED(pte)) {
1290 
1291 		/* Handle modified pages. */
1292 		if (PTE_ISMODIFIED(pte))
1293 			vm_page_dirty(m);
1294 
1295 		/* Referenced pages. */
1296 		if (PTE_ISREFERENCED(pte))
1297 			vm_page_aflag_set(m, PGA_REFERENCED);
1298 
1299 		/* Remove pv_entry from pv_list. */
1300 		pv_remove(pmap, va, m);
1301 	} else if (m->md.pv_tracked) {
1302 		pv_remove(pmap, va, m);
1303 		if (TAILQ_EMPTY(&m->md.pv_list))
1304 			m->md.pv_tracked = false;
1305 	}
1306 	mtx_lock_spin(&tlbivax_mutex);
1307 	tlb_miss_lock();
1308 
1309 	tlb0_flush_entry(va);
1310 	*pte = 0;
1311 
1312 	tlb_miss_unlock();
1313 	mtx_unlock_spin(&tlbivax_mutex);
1314 
1315 	pmap->pm_stats.resident_count--;
1316 
1317 	if (flags & PTBL_UNHOLD) {
1318 		return (ptbl_unhold(mmu, pmap, va));
1319 	}
1320 	return (0);
1321 }
1322 
1323 /*
1324  * allocate a page of pointers to page directories, do not preallocate the
1325  * page tables
1326  */
1327 static pte_t  **
1328 pdir_alloc(mmu_t mmu, pmap_t pmap, unsigned int pp2d_idx, bool nosleep)
1329 {
1330 	vm_page_t	mtbl  [PDIR_PAGES];
1331 	vm_page_t	m;
1332 	struct ptbl_buf *pbuf;
1333 	pte_t         **pdir;
1334 	unsigned int	pidx;
1335 	int		i;
1336 	int		req;
1337 
1338 	pbuf = ptbl_buf_alloc();
1339 
1340 	if (pbuf == NULL)
1341 		panic("%s: couldn't alloc kernel virtual memory", __func__);
1342 
1343 	/* Allocate pdir pages, this will sleep! */
1344 	for (i = 0; i < PDIR_PAGES; i++) {
1345 		pidx = (PDIR_PAGES * pp2d_idx) + i;
1346 		req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED;
1347 		while ((m = vm_page_alloc(NULL, pidx, req)) == NULL) {
1348 			PMAP_UNLOCK(pmap);
1349 			vm_wait(NULL);
1350 			PMAP_LOCK(pmap);
1351 		}
1352 		mtbl[i] = m;
1353 	}
1354 
1355 	/* Mapin allocated pages into kernel_pmap. */
1356 	pdir = (pte_t **) pbuf->kva;
1357 	pmap_qenter((vm_offset_t) pdir, mtbl, PDIR_PAGES);
1358 
1359 	/* Zero whole pdir. */
1360 	bzero((caddr_t) pdir, PDIR_PAGES * PAGE_SIZE);
1361 
1362 	/* Add pdir to the pmap pdir bufs list. */
1363 	TAILQ_INSERT_TAIL(&pmap->pm_pdir_list, pbuf, link);
1364 
1365 	return pdir;
1366 }
1367 
1368 /*
1369  * Insert PTE for a given page and virtual address.
1370  */
1371 static int
1372 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1373     boolean_t nosleep)
1374 {
1375 	unsigned int	pp2d_idx = PP2D_IDX(va);
1376 	unsigned int	pdir_idx = PDIR_IDX(va);
1377 	unsigned int	ptbl_idx = PTBL_IDX(va);
1378 	pte_t          *ptbl, *pte;
1379 	pte_t         **pdir;
1380 
1381 	/* Get the page directory pointer. */
1382 	pdir = pmap->pm_pp2d[pp2d_idx];
1383 	if (pdir == NULL)
1384 		pdir = pdir_alloc(mmu, pmap, pp2d_idx, nosleep);
1385 
1386 	/* Get the page table pointer. */
1387 	ptbl = pdir[pdir_idx];
1388 
1389 	if (ptbl == NULL) {
1390 		/* Allocate page table pages. */
1391 		ptbl = ptbl_alloc(mmu, pmap, pdir, pdir_idx, nosleep);
1392 		if (ptbl == NULL) {
1393 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1394 			return (ENOMEM);
1395 		}
1396 	} else {
1397 		/*
1398 		 * Check if there is valid mapping for requested va, if there
1399 		 * is, remove it.
1400 		 */
1401 		pte = &pdir[pdir_idx][ptbl_idx];
1402 		if (PTE_ISVALID(pte)) {
1403 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1404 		} else {
1405 			/*
1406 			 * pte is not used, increment hold count for ptbl
1407 			 * pages.
1408 			 */
1409 			if (pmap != kernel_pmap)
1410 				ptbl_hold(mmu, pmap, pdir, pdir_idx);
1411 		}
1412 	}
1413 
1414 	if (pdir[pdir_idx] == NULL) {
1415 		if (pmap != kernel_pmap && pmap->pm_pp2d[pp2d_idx] != NULL)
1416 			pdir_hold(mmu, pmap, pdir);
1417 		pdir[pdir_idx] = ptbl;
1418 	}
1419 	if (pmap->pm_pp2d[pp2d_idx] == NULL)
1420 		pmap->pm_pp2d[pp2d_idx] = pdir;
1421 
1422 	/*
1423 	 * Insert pv_entry into pv_list for mapped page if part of managed
1424 	 * memory.
1425 	 */
1426 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1427 		flags |= PTE_MANAGED;
1428 
1429 		/* Create and insert pv entry. */
1430 		pv_insert(pmap, va, m);
1431 	}
1432 
1433 	mtx_lock_spin(&tlbivax_mutex);
1434 	tlb_miss_lock();
1435 
1436 	tlb0_flush_entry(va);
1437 	pmap->pm_stats.resident_count++;
1438 	pte = &pdir[pdir_idx][ptbl_idx];
1439 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1440 	*pte |= (PTE_VALID | flags);
1441 
1442 	tlb_miss_unlock();
1443 	mtx_unlock_spin(&tlbivax_mutex);
1444 
1445 	return (0);
1446 }
1447 
1448 /* Return the pa for the given pmap/va. */
1449 static	vm_paddr_t
1450 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1451 {
1452 	vm_paddr_t	pa = 0;
1453 	pte_t          *pte;
1454 
1455 	pte = pte_find(mmu, pmap, va);
1456 	if ((pte != NULL) && PTE_ISVALID(pte))
1457 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1458 	return (pa);
1459 }
1460 
1461 
1462 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
1463 static void
1464 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1465 {
1466 	int		i, j;
1467 	vm_offset_t	va;
1468 	pte_t		*pte;
1469 
1470 	va = addr;
1471 	/* Initialize kernel pdir */
1472 	for (i = 0; i < kernel_pdirs; i++) {
1473 		kernel_pmap->pm_pp2d[i + PP2D_IDX(va)] =
1474 		    (pte_t **)(pdir + (i * PAGE_SIZE * PDIR_PAGES));
1475 		for (j = PDIR_IDX(va + (i * PAGE_SIZE * PDIR_NENTRIES * PTBL_NENTRIES));
1476 		    j < PDIR_NENTRIES; j++) {
1477 			kernel_pmap->pm_pp2d[i + PP2D_IDX(va)][j] =
1478 			    (pte_t *)(pdir + (kernel_pdirs * PAGE_SIZE * PDIR_PAGES) +
1479 			     (((i * PDIR_NENTRIES) + j) * PAGE_SIZE * PTBL_PAGES));
1480 		}
1481 	}
1482 
1483 	/*
1484 	 * Fill in PTEs covering kernel code and data. They are not required
1485 	 * for address translation, as this area is covered by static TLB1
1486 	 * entries, but for pte_vatopa() to work correctly with kernel area
1487 	 * addresses.
1488 	 */
1489 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1490 		pte = &(kernel_pmap->pm_pp2d[PP2D_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
1491 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1492 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1493 		    PTE_VALID | PTE_PS_4KB;
1494 	}
1495 }
1496 #else
1497 /*
1498  * Clean pte entry, try to free page table page if requested.
1499  *
1500  * Return 1 if ptbl pages were freed, otherwise return 0.
1501  */
1502 static int
1503 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
1504 {
1505 	unsigned int pdir_idx = PDIR_IDX(va);
1506 	unsigned int ptbl_idx = PTBL_IDX(va);
1507 	vm_page_t m;
1508 	pte_t *ptbl;
1509 	pte_t *pte;
1510 
1511 	//int su = (pmap == kernel_pmap);
1512 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
1513 	//		su, (u_int32_t)pmap, va, flags);
1514 
1515 	ptbl = pmap->pm_pdir[pdir_idx];
1516 	KASSERT(ptbl, ("pte_remove: null ptbl"));
1517 
1518 	pte = &ptbl[ptbl_idx];
1519 
1520 	if (pte == NULL || !PTE_ISVALID(pte))
1521 		return (0);
1522 
1523 	if (PTE_ISWIRED(pte))
1524 		pmap->pm_stats.wired_count--;
1525 
1526 	/* Get vm_page_t for mapped pte. */
1527 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1528 
1529 	/* Handle managed entry. */
1530 	if (PTE_ISMANAGED(pte)) {
1531 
1532 		if (PTE_ISMODIFIED(pte))
1533 			vm_page_dirty(m);
1534 
1535 		if (PTE_ISREFERENCED(pte))
1536 			vm_page_aflag_set(m, PGA_REFERENCED);
1537 
1538 		pv_remove(pmap, va, m);
1539 	} else if (m->md.pv_tracked) {
1540 		/*
1541 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
1542 		 * used.  This is needed by the NCSW support code for fast
1543 		 * VA<->PA translation.
1544 		 */
1545 		pv_remove(pmap, va, m);
1546 		if (TAILQ_EMPTY(&m->md.pv_list))
1547 			m->md.pv_tracked = false;
1548 	}
1549 
1550 	mtx_lock_spin(&tlbivax_mutex);
1551 	tlb_miss_lock();
1552 
1553 	tlb0_flush_entry(va);
1554 	*pte = 0;
1555 
1556 	tlb_miss_unlock();
1557 	mtx_unlock_spin(&tlbivax_mutex);
1558 
1559 	pmap->pm_stats.resident_count--;
1560 
1561 	if (flags & PTBL_UNHOLD) {
1562 		//debugf("pte_remove: e (unhold)\n");
1563 		return (ptbl_unhold(mmu, pmap, pdir_idx));
1564 	}
1565 
1566 	//debugf("pte_remove: e\n");
1567 	return (0);
1568 }
1569 
1570 /*
1571  * Insert PTE for a given page and virtual address.
1572  */
1573 static int
1574 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
1575     boolean_t nosleep)
1576 {
1577 	unsigned int pdir_idx = PDIR_IDX(va);
1578 	unsigned int ptbl_idx = PTBL_IDX(va);
1579 	pte_t *ptbl, *pte;
1580 
1581 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
1582 	    pmap == kernel_pmap, pmap, va);
1583 
1584 	/* Get the page table pointer. */
1585 	ptbl = pmap->pm_pdir[pdir_idx];
1586 
1587 	if (ptbl == NULL) {
1588 		/* Allocate page table pages. */
1589 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
1590 		if (ptbl == NULL) {
1591 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
1592 			return (ENOMEM);
1593 		}
1594 	} else {
1595 		/*
1596 		 * Check if there is valid mapping for requested
1597 		 * va, if there is, remove it.
1598 		 */
1599 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
1600 		if (PTE_ISVALID(pte)) {
1601 			pte_remove(mmu, pmap, va, PTBL_HOLD);
1602 		} else {
1603 			/*
1604 			 * pte is not used, increment hold count
1605 			 * for ptbl pages.
1606 			 */
1607 			if (pmap != kernel_pmap)
1608 				ptbl_hold(mmu, pmap, pdir_idx);
1609 		}
1610 	}
1611 
1612 	/*
1613 	 * Insert pv_entry into pv_list for mapped page if part of managed
1614 	 * memory.
1615 	 */
1616 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1617 		flags |= PTE_MANAGED;
1618 
1619 		/* Create and insert pv entry. */
1620 		pv_insert(pmap, va, m);
1621 	}
1622 
1623 	pmap->pm_stats.resident_count++;
1624 
1625 	mtx_lock_spin(&tlbivax_mutex);
1626 	tlb_miss_lock();
1627 
1628 	tlb0_flush_entry(va);
1629 	if (pmap->pm_pdir[pdir_idx] == NULL) {
1630 		/*
1631 		 * If we just allocated a new page table, hook it in
1632 		 * the pdir.
1633 		 */
1634 		pmap->pm_pdir[pdir_idx] = ptbl;
1635 	}
1636 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
1637 	*pte = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
1638 	*pte |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
1639 
1640 	tlb_miss_unlock();
1641 	mtx_unlock_spin(&tlbivax_mutex);
1642 	return (0);
1643 }
1644 
1645 /* Return the pa for the given pmap/va. */
1646 static vm_paddr_t
1647 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1648 {
1649 	vm_paddr_t pa = 0;
1650 	pte_t *pte;
1651 
1652 	pte = pte_find(mmu, pmap, va);
1653 	if ((pte != NULL) && PTE_ISVALID(pte))
1654 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
1655 	return (pa);
1656 }
1657 
1658 /* Get a pointer to a PTE in a page table. */
1659 static pte_t *
1660 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1661 {
1662 	unsigned int pdir_idx = PDIR_IDX(va);
1663 	unsigned int ptbl_idx = PTBL_IDX(va);
1664 
1665 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
1666 
1667 	if (pmap->pm_pdir[pdir_idx])
1668 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
1669 
1670 	return (NULL);
1671 }
1672 
1673 /* Set up kernel page tables. */
1674 static void
1675 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr, vm_offset_t pdir)
1676 {
1677 	int		i;
1678 	vm_offset_t	va;
1679 	pte_t		*pte;
1680 
1681 	/* Initialize kernel pdir */
1682 	for (i = 0; i < kernel_ptbls; i++)
1683 		kernel_pmap->pm_pdir[kptbl_min + i] =
1684 		    (pte_t *)(pdir + (i * PAGE_SIZE * PTBL_PAGES));
1685 
1686 	/*
1687 	 * Fill in PTEs covering kernel code and data. They are not required
1688 	 * for address translation, as this area is covered by static TLB1
1689 	 * entries, but for pte_vatopa() to work correctly with kernel area
1690 	 * addresses.
1691 	 */
1692 	for (va = addr; va < data_end; va += PAGE_SIZE) {
1693 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1694 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
1695 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1696 		    PTE_VALID | PTE_PS_4KB;
1697 	}
1698 }
1699 #endif
1700 
1701 /**************************************************************************/
1702 /* PMAP related */
1703 /**************************************************************************/
1704 
1705 /*
1706  * This is called during booke_init, before the system is really initialized.
1707  */
1708 static void
1709 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1710 {
1711 	vm_paddr_t phys_kernelend;
1712 	struct mem_region *mp, *mp1;
1713 	int cnt, i, j;
1714 	vm_paddr_t s, e, sz;
1715 	vm_paddr_t physsz, hwphyssz;
1716 	u_int phys_avail_count;
1717 	vm_size_t kstack0_sz;
1718 	vm_offset_t kernel_pdir, kstack0;
1719 	vm_paddr_t kstack0_phys;
1720 	void *dpcpu;
1721 
1722 	debugf("mmu_booke_bootstrap: entered\n");
1723 
1724 	/* Set interesting system properties */
1725 	hw_direct_map = 0;
1726 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
1727 	elf32_nxstack = 1;
1728 #endif
1729 
1730 	/* Initialize invalidation mutex */
1731 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1732 
1733 	/* Read TLB0 size and associativity. */
1734 	tlb0_get_tlbconf();
1735 
1736 	/*
1737 	 * Align kernel start and end address (kernel image).
1738 	 * Note that kernel end does not necessarily relate to kernsize.
1739 	 * kernsize is the size of the kernel that is actually mapped.
1740 	 */
1741 	kernstart = trunc_page(start);
1742 	data_start = round_page(kernelend);
1743 	data_end = data_start;
1744 
1745 	/*
1746 	 * Addresses of preloaded modules (like file systems) use
1747 	 * physical addresses. Make sure we relocate those into
1748 	 * virtual addresses.
1749 	 */
1750 	preload_addr_relocate = kernstart - kernload;
1751 
1752 	/* Allocate the dynamic per-cpu area. */
1753 	dpcpu = (void *)data_end;
1754 	data_end += DPCPU_SIZE;
1755 
1756 	/* Allocate space for the message buffer. */
1757 	msgbufp = (struct msgbuf *)data_end;
1758 	data_end += msgbufsize;
1759 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1760 	    (uintptr_t)msgbufp, data_end);
1761 
1762 	data_end = round_page(data_end);
1763 
1764 	/* Allocate space for ptbl_bufs. */
1765 	ptbl_bufs = (struct ptbl_buf *)data_end;
1766 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1767 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1768 	    (uintptr_t)ptbl_bufs, data_end);
1769 
1770 	data_end = round_page(data_end);
1771 
1772 	/* Allocate PTE tables for kernel KVA. */
1773 	kernel_pdir = data_end;
1774 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
1775 	    PDIR_SIZE);
1776 #ifdef __powerpc64__
1777 	kernel_pdirs = howmany(kernel_ptbls, PDIR_NENTRIES);
1778 	data_end += kernel_pdirs * PDIR_PAGES * PAGE_SIZE;
1779 #endif
1780 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1781 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1782 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
1783 	    kernel_pdir, data_end);
1784 
1785 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
1786 	if (data_end - kernstart > kernsize) {
1787 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1788 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1789 	}
1790 	data_end = kernstart + kernsize;
1791 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
1792 
1793 	/*
1794 	 * Clear the structures - note we can only do it safely after the
1795 	 * possible additional TLB1 translations are in place (above) so that
1796 	 * all range up to the currently calculated 'data_end' is covered.
1797 	 */
1798 	dpcpu_init(dpcpu, 0);
1799 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1800 #ifdef __powerpc64__
1801 	memset((void *)kernel_pdir, 0,
1802 	    kernel_pdirs * PDIR_PAGES * PAGE_SIZE +
1803 	    kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1804 #else
1805 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1806 #endif
1807 
1808 	/*******************************************************/
1809 	/* Set the start and end of kva. */
1810 	/*******************************************************/
1811 	virtual_avail = round_page(data_end);
1812 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1813 
1814 	/* Allocate KVA space for page zero/copy operations. */
1815 	zero_page_va = virtual_avail;
1816 	virtual_avail += PAGE_SIZE;
1817 	copy_page_src_va = virtual_avail;
1818 	virtual_avail += PAGE_SIZE;
1819 	copy_page_dst_va = virtual_avail;
1820 	virtual_avail += PAGE_SIZE;
1821 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1822 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1823 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1824 
1825 	/* Initialize page zero/copy mutexes. */
1826 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1827 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1828 
1829 	/* Allocate KVA space for ptbl bufs. */
1830 	ptbl_buf_pool_vabase = virtual_avail;
1831 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1832 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1833 	    ptbl_buf_pool_vabase, virtual_avail);
1834 
1835 	/* Calculate corresponding physical addresses for the kernel region. */
1836 	phys_kernelend = kernload + kernsize;
1837 	debugf("kernel image and allocated data:\n");
1838 	debugf(" kernload    = 0x%09llx\n", (uint64_t)kernload);
1839 	debugf(" kernstart   = 0x%08x\n", kernstart);
1840 	debugf(" kernsize    = 0x%08x\n", kernsize);
1841 
1842 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1843 		panic("mmu_booke_bootstrap: phys_avail too small");
1844 
1845 	/*
1846 	 * Remove kernel physical address range from avail regions list. Page
1847 	 * align all regions.  Non-page aligned memory isn't very interesting
1848 	 * to us.  Also, sort the entries for ascending addresses.
1849 	 */
1850 
1851 	/* Retrieve phys/avail mem regions */
1852 	mem_regions(&physmem_regions, &physmem_regions_sz,
1853 	    &availmem_regions, &availmem_regions_sz);
1854 	sz = 0;
1855 	cnt = availmem_regions_sz;
1856 	debugf("processing avail regions:\n");
1857 	for (mp = availmem_regions; mp->mr_size; mp++) {
1858 		s = mp->mr_start;
1859 		e = mp->mr_start + mp->mr_size;
1860 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
1861 		/* Check whether this region holds all of the kernel. */
1862 		if (s < kernload && e > phys_kernelend) {
1863 			availmem_regions[cnt].mr_start = phys_kernelend;
1864 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1865 			e = kernload;
1866 		}
1867 		/* Look whether this regions starts within the kernel. */
1868 		if (s >= kernload && s < phys_kernelend) {
1869 			if (e <= phys_kernelend)
1870 				goto empty;
1871 			s = phys_kernelend;
1872 		}
1873 		/* Now look whether this region ends within the kernel. */
1874 		if (e > kernload && e <= phys_kernelend) {
1875 			if (s >= kernload)
1876 				goto empty;
1877 			e = kernload;
1878 		}
1879 		/* Now page align the start and size of the region. */
1880 		s = round_page(s);
1881 		e = trunc_page(e);
1882 		if (e < s)
1883 			e = s;
1884 		sz = e - s;
1885 		debugf("%09jx-%09jx = %jx\n",
1886 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
1887 
1888 		/* Check whether some memory is left here. */
1889 		if (sz == 0) {
1890 		empty:
1891 			memmove(mp, mp + 1,
1892 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1893 			cnt--;
1894 			mp--;
1895 			continue;
1896 		}
1897 
1898 		/* Do an insertion sort. */
1899 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1900 			if (s < mp1->mr_start)
1901 				break;
1902 		if (mp1 < mp) {
1903 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1904 			mp1->mr_start = s;
1905 			mp1->mr_size = sz;
1906 		} else {
1907 			mp->mr_start = s;
1908 			mp->mr_size = sz;
1909 		}
1910 	}
1911 	availmem_regions_sz = cnt;
1912 
1913 	/*******************************************************/
1914 	/* Steal physical memory for kernel stack from the end */
1915 	/* of the first avail region                           */
1916 	/*******************************************************/
1917 	kstack0_sz = kstack_pages * PAGE_SIZE;
1918 	kstack0_phys = availmem_regions[0].mr_start +
1919 	    availmem_regions[0].mr_size;
1920 	kstack0_phys -= kstack0_sz;
1921 	availmem_regions[0].mr_size -= kstack0_sz;
1922 
1923 	/*******************************************************/
1924 	/* Fill in phys_avail table, based on availmem_regions */
1925 	/*******************************************************/
1926 	phys_avail_count = 0;
1927 	physsz = 0;
1928 	hwphyssz = 0;
1929 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1930 
1931 	debugf("fill in phys_avail:\n");
1932 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1933 
1934 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
1935 		    (uintmax_t)availmem_regions[i].mr_start,
1936 		    (uintmax_t)availmem_regions[i].mr_start +
1937 		        availmem_regions[i].mr_size,
1938 		    (uintmax_t)availmem_regions[i].mr_size);
1939 
1940 		if (hwphyssz != 0 &&
1941 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1942 			debugf(" hw.physmem adjust\n");
1943 			if (physsz < hwphyssz) {
1944 				phys_avail[j] = availmem_regions[i].mr_start;
1945 				phys_avail[j + 1] =
1946 				    availmem_regions[i].mr_start +
1947 				    hwphyssz - physsz;
1948 				physsz = hwphyssz;
1949 				phys_avail_count++;
1950 			}
1951 			break;
1952 		}
1953 
1954 		phys_avail[j] = availmem_regions[i].mr_start;
1955 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1956 		    availmem_regions[i].mr_size;
1957 		phys_avail_count++;
1958 		physsz += availmem_regions[i].mr_size;
1959 	}
1960 	physmem = btoc(physsz);
1961 
1962 	/* Calculate the last available physical address. */
1963 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1964 		;
1965 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1966 
1967 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1968 	debugf("phys_avail_count = %d\n", phys_avail_count);
1969 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
1970 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
1971 
1972 	/*******************************************************/
1973 	/* Initialize (statically allocated) kernel pmap. */
1974 	/*******************************************************/
1975 	PMAP_LOCK_INIT(kernel_pmap);
1976 #ifndef __powerpc64__
1977 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1978 #endif
1979 
1980 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
1981 	kernel_pte_alloc(virtual_avail, kernstart, kernel_pdir);
1982 	for (i = 0; i < MAXCPU; i++) {
1983 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1984 
1985 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1986 		tidbusy[i][TID_KERNEL] = kernel_pmap;
1987 	}
1988 
1989 	/* Mark kernel_pmap active on all CPUs */
1990 	CPU_FILL(&kernel_pmap->pm_active);
1991 
1992  	/*
1993 	 * Initialize the global pv list lock.
1994 	 */
1995 	rw_init(&pvh_global_lock, "pmap pv global");
1996 
1997 	/*******************************************************/
1998 	/* Final setup */
1999 	/*******************************************************/
2000 
2001 	/* Enter kstack0 into kernel map, provide guard page */
2002 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
2003 	thread0.td_kstack = kstack0;
2004 	thread0.td_kstack_pages = kstack_pages;
2005 
2006 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
2007 	debugf("kstack0_phys at 0x%09llx - 0x%09llx\n",
2008 	    kstack0_phys, kstack0_phys + kstack0_sz);
2009 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
2010 	    kstack0, kstack0 + kstack0_sz);
2011 
2012 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
2013 	for (i = 0; i < kstack_pages; i++) {
2014 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
2015 		kstack0 += PAGE_SIZE;
2016 		kstack0_phys += PAGE_SIZE;
2017 	}
2018 
2019 	pmap_bootstrapped = 1;
2020 
2021 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
2022 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
2023 
2024 	debugf("mmu_booke_bootstrap: exit\n");
2025 }
2026 
2027 #ifdef SMP
2028  void
2029 tlb1_ap_prep(void)
2030 {
2031 	tlb_entry_t *e, tmp;
2032 	unsigned int i;
2033 
2034 	/* Prepare TLB1 image for AP processors */
2035 	e = __boot_tlb1;
2036 	for (i = 0; i < TLB1_ENTRIES; i++) {
2037 		tlb1_read_entry(&tmp, i);
2038 
2039 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
2040 			memcpy(e++, &tmp, sizeof(tmp));
2041 	}
2042 }
2043 
2044 void
2045 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
2046 {
2047 	int i;
2048 
2049 	/*
2050 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
2051 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
2052 	 * created by tlb1_ap_prep(), so use these values directly to
2053 	 * (re)program AP's TLB1 hardware.
2054 	 *
2055 	 * Start at index 1 because index 0 has the kernel map.
2056 	 */
2057 	for (i = 1; i < TLB1_ENTRIES; i++) {
2058 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
2059 			tlb1_write_entry(&__boot_tlb1[i], i);
2060 	}
2061 
2062 	set_mas4_defaults();
2063 }
2064 #endif
2065 
2066 static void
2067 booke_pmap_init_qpages(void)
2068 {
2069 	struct pcpu *pc;
2070 	int i;
2071 
2072 	CPU_FOREACH(i) {
2073 		pc = pcpu_find(i);
2074 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
2075 		if (pc->pc_qmap_addr == 0)
2076 			panic("pmap_init_qpages: unable to allocate KVA");
2077 	}
2078 }
2079 
2080 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
2081 
2082 /*
2083  * Get the physical page address for the given pmap/virtual address.
2084  */
2085 static vm_paddr_t
2086 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
2087 {
2088 	vm_paddr_t pa;
2089 
2090 	PMAP_LOCK(pmap);
2091 	pa = pte_vatopa(mmu, pmap, va);
2092 	PMAP_UNLOCK(pmap);
2093 
2094 	return (pa);
2095 }
2096 
2097 /*
2098  * Extract the physical page address associated with the given
2099  * kernel virtual address.
2100  */
2101 static vm_paddr_t
2102 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
2103 {
2104 	tlb_entry_t e;
2105 	vm_paddr_t p = 0;
2106 	int i;
2107 
2108 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
2109 		p = pte_vatopa(mmu, kernel_pmap, va);
2110 
2111 	if (p == 0) {
2112 		/* Check TLB1 mappings */
2113 		for (i = 0; i < TLB1_ENTRIES; i++) {
2114 			tlb1_read_entry(&e, i);
2115 			if (!(e.mas1 & MAS1_VALID))
2116 				continue;
2117 			if (va >= e.virt && va < e.virt + e.size)
2118 				return (e.phys + (va - e.virt));
2119 		}
2120 	}
2121 
2122 	return (p);
2123 }
2124 
2125 /*
2126  * Initialize the pmap module.
2127  * Called by vm_init, to initialize any structures that the pmap
2128  * system needs to map virtual memory.
2129  */
2130 static void
2131 mmu_booke_init(mmu_t mmu)
2132 {
2133 	int shpgperproc = PMAP_SHPGPERPROC;
2134 
2135 	/*
2136 	 * Initialize the address space (zone) for the pv entries.  Set a
2137 	 * high water mark so that the system can recover from excessive
2138 	 * numbers of pv entries.
2139 	 */
2140 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
2141 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
2142 
2143 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
2144 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
2145 
2146 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
2147 	pv_entry_high_water = 9 * (pv_entry_max / 10);
2148 
2149 	uma_zone_reserve_kva(pvzone, pv_entry_max);
2150 
2151 	/* Pre-fill pvzone with initial number of pv entries. */
2152 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
2153 
2154 	/* Initialize ptbl allocation. */
2155 	ptbl_init();
2156 }
2157 
2158 /*
2159  * Map a list of wired pages into kernel virtual address space.  This is
2160  * intended for temporary mappings which do not need page modification or
2161  * references recorded.  Existing mappings in the region are overwritten.
2162  */
2163 static void
2164 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
2165 {
2166 	vm_offset_t va;
2167 
2168 	va = sva;
2169 	while (count-- > 0) {
2170 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
2171 		va += PAGE_SIZE;
2172 		m++;
2173 	}
2174 }
2175 
2176 /*
2177  * Remove page mappings from kernel virtual address space.  Intended for
2178  * temporary mappings entered by mmu_booke_qenter.
2179  */
2180 static void
2181 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
2182 {
2183 	vm_offset_t va;
2184 
2185 	va = sva;
2186 	while (count-- > 0) {
2187 		mmu_booke_kremove(mmu, va);
2188 		va += PAGE_SIZE;
2189 	}
2190 }
2191 
2192 /*
2193  * Map a wired page into kernel virtual address space.
2194  */
2195 static void
2196 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
2197 {
2198 
2199 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
2200 }
2201 
2202 static void
2203 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
2204 {
2205 	uint32_t flags;
2206 	pte_t *pte;
2207 
2208 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2209 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
2210 
2211 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
2212 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
2213 	flags |= PTE_PS_4KB;
2214 
2215 	pte = pte_find(mmu, kernel_pmap, va);
2216 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
2217 
2218 	mtx_lock_spin(&tlbivax_mutex);
2219 	tlb_miss_lock();
2220 
2221 	if (PTE_ISVALID(pte)) {
2222 
2223 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
2224 
2225 		/* Flush entry from TLB0 */
2226 		tlb0_flush_entry(va);
2227 	}
2228 
2229 	*pte = PTE_RPN_FROM_PA(pa) | flags;
2230 
2231 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
2232 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
2233 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
2234 
2235 	/* Flush the real memory from the instruction cache. */
2236 	if ((flags & (PTE_I | PTE_G)) == 0)
2237 		__syncicache((void *)va, PAGE_SIZE);
2238 
2239 	tlb_miss_unlock();
2240 	mtx_unlock_spin(&tlbivax_mutex);
2241 }
2242 
2243 /*
2244  * Remove a page from kernel page table.
2245  */
2246 static void
2247 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
2248 {
2249 	pte_t *pte;
2250 
2251 	CTR2(KTR_PMAP,"%s: s (va = 0x%08x)\n", __func__, va);
2252 
2253 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
2254 	    (va <= VM_MAX_KERNEL_ADDRESS)),
2255 	    ("mmu_booke_kremove: invalid va"));
2256 
2257 	pte = pte_find(mmu, kernel_pmap, va);
2258 
2259 	if (!PTE_ISVALID(pte)) {
2260 
2261 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
2262 
2263 		return;
2264 	}
2265 
2266 	mtx_lock_spin(&tlbivax_mutex);
2267 	tlb_miss_lock();
2268 
2269 	/* Invalidate entry in TLB0, update PTE. */
2270 	tlb0_flush_entry(va);
2271 	*pte = 0;
2272 
2273 	tlb_miss_unlock();
2274 	mtx_unlock_spin(&tlbivax_mutex);
2275 }
2276 
2277 /*
2278  * Provide a kernel pointer corresponding to a given userland pointer.
2279  * The returned pointer is valid until the next time this function is
2280  * called in this thread. This is used internally in copyin/copyout.
2281  */
2282 int
2283 mmu_booke_map_user_ptr(mmu_t mmu, pmap_t pm, volatile const void *uaddr,
2284     void **kaddr, size_t ulen, size_t *klen)
2285 {
2286 
2287 	if ((uintptr_t)uaddr + ulen > VM_MAXUSER_ADDRESS + PAGE_SIZE)
2288 		return (EFAULT);
2289 
2290 	*kaddr = (void *)(uintptr_t)uaddr;
2291 	if (klen)
2292 		*klen = ulen;
2293 
2294 	return (0);
2295 }
2296 
2297 /*
2298  * Figure out where a given kernel pointer (usually in a fault) points
2299  * to from the VM's perspective, potentially remapping into userland's
2300  * address space.
2301  */
2302 static int
2303 mmu_booke_decode_kernel_ptr(mmu_t mmu, vm_offset_t addr, int *is_user,
2304     vm_offset_t *decoded_addr)
2305 {
2306 
2307 	if (addr < VM_MAXUSER_ADDRESS)
2308 		*is_user = 1;
2309 	else
2310 		*is_user = 0;
2311 
2312 	*decoded_addr = addr;
2313 	return (0);
2314 }
2315 
2316 /*
2317  * Initialize pmap associated with process 0.
2318  */
2319 static void
2320 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
2321 {
2322 
2323 	PMAP_LOCK_INIT(pmap);
2324 	mmu_booke_pinit(mmu, pmap);
2325 	PCPU_SET(curpmap, pmap);
2326 }
2327 
2328 /*
2329  * Initialize a preallocated and zeroed pmap structure,
2330  * such as one in a vmspace structure.
2331  */
2332 static void
2333 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
2334 {
2335 	int i;
2336 
2337 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
2338 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
2339 
2340 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
2341 
2342 	for (i = 0; i < MAXCPU; i++)
2343 		pmap->pm_tid[i] = TID_NONE;
2344 	CPU_ZERO(&kernel_pmap->pm_active);
2345 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
2346 #ifdef __powerpc64__
2347 	bzero(&pmap->pm_pp2d, sizeof(pte_t **) * PP2D_NENTRIES);
2348 	TAILQ_INIT(&pmap->pm_pdir_list);
2349 #else
2350 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
2351 #endif
2352 	TAILQ_INIT(&pmap->pm_ptbl_list);
2353 }
2354 
2355 /*
2356  * Release any resources held by the given physical map.
2357  * Called when a pmap initialized by mmu_booke_pinit is being released.
2358  * Should only be called if the map contains no valid mappings.
2359  */
2360 static void
2361 mmu_booke_release(mmu_t mmu, pmap_t pmap)
2362 {
2363 
2364 	KASSERT(pmap->pm_stats.resident_count == 0,
2365 	    ("pmap_release: pmap resident count %ld != 0",
2366 	    pmap->pm_stats.resident_count));
2367 }
2368 
2369 /*
2370  * Insert the given physical page at the specified virtual address in the
2371  * target physical map with the protection requested. If specified the page
2372  * will be wired down.
2373  */
2374 static int
2375 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2376     vm_prot_t prot, u_int flags, int8_t psind)
2377 {
2378 	int error;
2379 
2380 	rw_wlock(&pvh_global_lock);
2381 	PMAP_LOCK(pmap);
2382 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
2383 	PMAP_UNLOCK(pmap);
2384 	rw_wunlock(&pvh_global_lock);
2385 	return (error);
2386 }
2387 
2388 static int
2389 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2390     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
2391 {
2392 	pte_t *pte;
2393 	vm_paddr_t pa;
2394 	uint32_t flags;
2395 	int error, su, sync;
2396 
2397 	pa = VM_PAGE_TO_PHYS(m);
2398 	su = (pmap == kernel_pmap);
2399 	sync = 0;
2400 
2401 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
2402 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
2403 	//		(u_int32_t)pmap, su, pmap->pm_tid,
2404 	//		(u_int32_t)m, va, pa, prot, flags);
2405 
2406 	if (su) {
2407 		KASSERT(((va >= virtual_avail) &&
2408 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2409 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
2410 	} else {
2411 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2412 		    ("mmu_booke_enter_locked: user pmap, non user va"));
2413 	}
2414 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
2415 		VM_OBJECT_ASSERT_LOCKED(m->object);
2416 
2417 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
2418 
2419 	/*
2420 	 * If there is an existing mapping, and the physical address has not
2421 	 * changed, must be protection or wiring change.
2422 	 */
2423 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
2424 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
2425 
2426 		/*
2427 		 * Before actually updating pte->flags we calculate and
2428 		 * prepare its new value in a helper var.
2429 		 */
2430 		flags = *pte;
2431 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
2432 
2433 		/* Wiring change, just update stats. */
2434 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
2435 			if (!PTE_ISWIRED(pte)) {
2436 				flags |= PTE_WIRED;
2437 				pmap->pm_stats.wired_count++;
2438 			}
2439 		} else {
2440 			if (PTE_ISWIRED(pte)) {
2441 				flags &= ~PTE_WIRED;
2442 				pmap->pm_stats.wired_count--;
2443 			}
2444 		}
2445 
2446 		if (prot & VM_PROT_WRITE) {
2447 			/* Add write permissions. */
2448 			flags |= PTE_SW;
2449 			if (!su)
2450 				flags |= PTE_UW;
2451 
2452 			if ((flags & PTE_MANAGED) != 0)
2453 				vm_page_aflag_set(m, PGA_WRITEABLE);
2454 		} else {
2455 			/* Handle modified pages, sense modify status. */
2456 
2457 			/*
2458 			 * The PTE_MODIFIED flag could be set by underlying
2459 			 * TLB misses since we last read it (above), possibly
2460 			 * other CPUs could update it so we check in the PTE
2461 			 * directly rather than rely on that saved local flags
2462 			 * copy.
2463 			 */
2464 			if (PTE_ISMODIFIED(pte))
2465 				vm_page_dirty(m);
2466 		}
2467 
2468 		if (prot & VM_PROT_EXECUTE) {
2469 			flags |= PTE_SX;
2470 			if (!su)
2471 				flags |= PTE_UX;
2472 
2473 			/*
2474 			 * Check existing flags for execute permissions: if we
2475 			 * are turning execute permissions on, icache should
2476 			 * be flushed.
2477 			 */
2478 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
2479 				sync++;
2480 		}
2481 
2482 		flags &= ~PTE_REFERENCED;
2483 
2484 		/*
2485 		 * The new flags value is all calculated -- only now actually
2486 		 * update the PTE.
2487 		 */
2488 		mtx_lock_spin(&tlbivax_mutex);
2489 		tlb_miss_lock();
2490 
2491 		tlb0_flush_entry(va);
2492 		*pte &= ~PTE_FLAGS_MASK;
2493 		*pte |= flags;
2494 
2495 		tlb_miss_unlock();
2496 		mtx_unlock_spin(&tlbivax_mutex);
2497 
2498 	} else {
2499 		/*
2500 		 * If there is an existing mapping, but it's for a different
2501 		 * physical address, pte_enter() will delete the old mapping.
2502 		 */
2503 		//if ((pte != NULL) && PTE_ISVALID(pte))
2504 		//	debugf("mmu_booke_enter_locked: replace\n");
2505 		//else
2506 		//	debugf("mmu_booke_enter_locked: new\n");
2507 
2508 		/* Now set up the flags and install the new mapping. */
2509 		flags = (PTE_SR | PTE_VALID);
2510 		flags |= PTE_M;
2511 
2512 		if (!su)
2513 			flags |= PTE_UR;
2514 
2515 		if (prot & VM_PROT_WRITE) {
2516 			flags |= PTE_SW;
2517 			if (!su)
2518 				flags |= PTE_UW;
2519 
2520 			if ((m->oflags & VPO_UNMANAGED) == 0)
2521 				vm_page_aflag_set(m, PGA_WRITEABLE);
2522 		}
2523 
2524 		if (prot & VM_PROT_EXECUTE) {
2525 			flags |= PTE_SX;
2526 			if (!su)
2527 				flags |= PTE_UX;
2528 		}
2529 
2530 		/* If its wired update stats. */
2531 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
2532 			flags |= PTE_WIRED;
2533 
2534 		error = pte_enter(mmu, pmap, m, va, flags,
2535 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
2536 		if (error != 0)
2537 			return (KERN_RESOURCE_SHORTAGE);
2538 
2539 		if ((flags & PMAP_ENTER_WIRED) != 0)
2540 			pmap->pm_stats.wired_count++;
2541 
2542 		/* Flush the real memory from the instruction cache. */
2543 		if (prot & VM_PROT_EXECUTE)
2544 			sync++;
2545 	}
2546 
2547 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
2548 		__syncicache((void *)va, PAGE_SIZE);
2549 		sync = 0;
2550 	}
2551 
2552 	return (KERN_SUCCESS);
2553 }
2554 
2555 /*
2556  * Maps a sequence of resident pages belonging to the same object.
2557  * The sequence begins with the given page m_start.  This page is
2558  * mapped at the given virtual address start.  Each subsequent page is
2559  * mapped at a virtual address that is offset from start by the same
2560  * amount as the page is offset from m_start within the object.  The
2561  * last page in the sequence is the page with the largest offset from
2562  * m_start that can be mapped at a virtual address less than the given
2563  * virtual address end.  Not every virtual page between start and end
2564  * is mapped; only those for which a resident page exists with the
2565  * corresponding offset from m_start are mapped.
2566  */
2567 static void
2568 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
2569     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
2570 {
2571 	vm_page_t m;
2572 	vm_pindex_t diff, psize;
2573 
2574 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
2575 
2576 	psize = atop(end - start);
2577 	m = m_start;
2578 	rw_wlock(&pvh_global_lock);
2579 	PMAP_LOCK(pmap);
2580 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
2581 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
2582 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
2583 		    PMAP_ENTER_NOSLEEP, 0);
2584 		m = TAILQ_NEXT(m, listq);
2585 	}
2586 	rw_wunlock(&pvh_global_lock);
2587 	PMAP_UNLOCK(pmap);
2588 }
2589 
2590 static void
2591 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
2592     vm_prot_t prot)
2593 {
2594 
2595 	rw_wlock(&pvh_global_lock);
2596 	PMAP_LOCK(pmap);
2597 	mmu_booke_enter_locked(mmu, pmap, va, m,
2598 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
2599 	    0);
2600 	rw_wunlock(&pvh_global_lock);
2601 	PMAP_UNLOCK(pmap);
2602 }
2603 
2604 /*
2605  * Remove the given range of addresses from the specified map.
2606  *
2607  * It is assumed that the start and end are properly rounded to the page size.
2608  */
2609 static void
2610 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
2611 {
2612 	pte_t *pte;
2613 	uint8_t hold_flag;
2614 
2615 	int su = (pmap == kernel_pmap);
2616 
2617 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
2618 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
2619 
2620 	if (su) {
2621 		KASSERT(((va >= virtual_avail) &&
2622 		    (va <= VM_MAX_KERNEL_ADDRESS)),
2623 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
2624 	} else {
2625 		KASSERT((va <= VM_MAXUSER_ADDRESS),
2626 		    ("mmu_booke_remove: user pmap, non user va"));
2627 	}
2628 
2629 	if (PMAP_REMOVE_DONE(pmap)) {
2630 		//debugf("mmu_booke_remove: e (empty)\n");
2631 		return;
2632 	}
2633 
2634 	hold_flag = PTBL_HOLD_FLAG(pmap);
2635 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
2636 
2637 	rw_wlock(&pvh_global_lock);
2638 	PMAP_LOCK(pmap);
2639 	for (; va < endva; va += PAGE_SIZE) {
2640 		pte = pte_find(mmu, pmap, va);
2641 		if ((pte != NULL) && PTE_ISVALID(pte))
2642 			pte_remove(mmu, pmap, va, hold_flag);
2643 	}
2644 	PMAP_UNLOCK(pmap);
2645 	rw_wunlock(&pvh_global_lock);
2646 
2647 	//debugf("mmu_booke_remove: e\n");
2648 }
2649 
2650 /*
2651  * Remove physical page from all pmaps in which it resides.
2652  */
2653 static void
2654 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
2655 {
2656 	pv_entry_t pv, pvn;
2657 	uint8_t hold_flag;
2658 
2659 	rw_wlock(&pvh_global_lock);
2660 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
2661 		pvn = TAILQ_NEXT(pv, pv_link);
2662 
2663 		PMAP_LOCK(pv->pv_pmap);
2664 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
2665 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
2666 		PMAP_UNLOCK(pv->pv_pmap);
2667 	}
2668 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2669 	rw_wunlock(&pvh_global_lock);
2670 }
2671 
2672 /*
2673  * Map a range of physical addresses into kernel virtual address space.
2674  */
2675 static vm_offset_t
2676 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
2677     vm_paddr_t pa_end, int prot)
2678 {
2679 	vm_offset_t sva = *virt;
2680 	vm_offset_t va = sva;
2681 
2682 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
2683 	//		sva, pa_start, pa_end);
2684 
2685 	while (pa_start < pa_end) {
2686 		mmu_booke_kenter(mmu, va, pa_start);
2687 		va += PAGE_SIZE;
2688 		pa_start += PAGE_SIZE;
2689 	}
2690 	*virt = va;
2691 
2692 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
2693 	return (sva);
2694 }
2695 
2696 /*
2697  * The pmap must be activated before it's address space can be accessed in any
2698  * way.
2699  */
2700 static void
2701 mmu_booke_activate(mmu_t mmu, struct thread *td)
2702 {
2703 	pmap_t pmap;
2704 	u_int cpuid;
2705 
2706 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2707 
2708 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
2709 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2710 
2711 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
2712 
2713 	sched_pin();
2714 
2715 	cpuid = PCPU_GET(cpuid);
2716 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
2717 	PCPU_SET(curpmap, pmap);
2718 
2719 	if (pmap->pm_tid[cpuid] == TID_NONE)
2720 		tid_alloc(pmap);
2721 
2722 	/* Load PID0 register with pmap tid value. */
2723 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
2724 	__asm __volatile("isync");
2725 
2726 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
2727 
2728 	sched_unpin();
2729 
2730 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
2731 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
2732 }
2733 
2734 /*
2735  * Deactivate the specified process's address space.
2736  */
2737 static void
2738 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
2739 {
2740 	pmap_t pmap;
2741 
2742 	pmap = &td->td_proc->p_vmspace->vm_pmap;
2743 
2744 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
2745 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
2746 
2747 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
2748 
2749 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
2750 	PCPU_SET(curpmap, NULL);
2751 }
2752 
2753 /*
2754  * Copy the range specified by src_addr/len
2755  * from the source map to the range dst_addr/len
2756  * in the destination map.
2757  *
2758  * This routine is only advisory and need not do anything.
2759  */
2760 static void
2761 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
2762     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
2763 {
2764 
2765 }
2766 
2767 /*
2768  * Set the physical protection on the specified range of this map as requested.
2769  */
2770 static void
2771 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2772     vm_prot_t prot)
2773 {
2774 	vm_offset_t va;
2775 	vm_page_t m;
2776 	pte_t *pte;
2777 
2778 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2779 		mmu_booke_remove(mmu, pmap, sva, eva);
2780 		return;
2781 	}
2782 
2783 	if (prot & VM_PROT_WRITE)
2784 		return;
2785 
2786 	PMAP_LOCK(pmap);
2787 	for (va = sva; va < eva; va += PAGE_SIZE) {
2788 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2789 			if (PTE_ISVALID(pte)) {
2790 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2791 
2792 				mtx_lock_spin(&tlbivax_mutex);
2793 				tlb_miss_lock();
2794 
2795 				/* Handle modified pages. */
2796 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2797 					vm_page_dirty(m);
2798 
2799 				tlb0_flush_entry(va);
2800 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2801 
2802 				tlb_miss_unlock();
2803 				mtx_unlock_spin(&tlbivax_mutex);
2804 			}
2805 		}
2806 	}
2807 	PMAP_UNLOCK(pmap);
2808 }
2809 
2810 /*
2811  * Clear the write and modified bits in each of the given page's mappings.
2812  */
2813 static void
2814 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2815 {
2816 	pv_entry_t pv;
2817 	pte_t *pte;
2818 
2819 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2820 	    ("mmu_booke_remove_write: page %p is not managed", m));
2821 
2822 	/*
2823 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2824 	 * set by another thread while the object is locked.  Thus,
2825 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2826 	 */
2827 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2828 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2829 		return;
2830 	rw_wlock(&pvh_global_lock);
2831 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2832 		PMAP_LOCK(pv->pv_pmap);
2833 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2834 			if (PTE_ISVALID(pte)) {
2835 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2836 
2837 				mtx_lock_spin(&tlbivax_mutex);
2838 				tlb_miss_lock();
2839 
2840 				/* Handle modified pages. */
2841 				if (PTE_ISMODIFIED(pte))
2842 					vm_page_dirty(m);
2843 
2844 				/* Flush mapping from TLB0. */
2845 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2846 
2847 				tlb_miss_unlock();
2848 				mtx_unlock_spin(&tlbivax_mutex);
2849 			}
2850 		}
2851 		PMAP_UNLOCK(pv->pv_pmap);
2852 	}
2853 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2854 	rw_wunlock(&pvh_global_lock);
2855 }
2856 
2857 static void
2858 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2859 {
2860 	pte_t *pte;
2861 	pmap_t pmap;
2862 	vm_page_t m;
2863 	vm_offset_t addr;
2864 	vm_paddr_t pa = 0;
2865 	int active, valid;
2866 
2867 	va = trunc_page(va);
2868 	sz = round_page(sz);
2869 
2870 	rw_wlock(&pvh_global_lock);
2871 	pmap = PCPU_GET(curpmap);
2872 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2873 	while (sz > 0) {
2874 		PMAP_LOCK(pm);
2875 		pte = pte_find(mmu, pm, va);
2876 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2877 		if (valid)
2878 			pa = PTE_PA(pte);
2879 		PMAP_UNLOCK(pm);
2880 		if (valid) {
2881 			if (!active) {
2882 				/* Create a mapping in the active pmap. */
2883 				addr = 0;
2884 				m = PHYS_TO_VM_PAGE(pa);
2885 				PMAP_LOCK(pmap);
2886 				pte_enter(mmu, pmap, m, addr,
2887 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2888 				__syncicache((void *)addr, PAGE_SIZE);
2889 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2890 				PMAP_UNLOCK(pmap);
2891 			} else
2892 				__syncicache((void *)va, PAGE_SIZE);
2893 		}
2894 		va += PAGE_SIZE;
2895 		sz -= PAGE_SIZE;
2896 	}
2897 	rw_wunlock(&pvh_global_lock);
2898 }
2899 
2900 /*
2901  * Atomically extract and hold the physical page with the given
2902  * pmap and virtual address pair if that mapping permits the given
2903  * protection.
2904  */
2905 static vm_page_t
2906 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2907     vm_prot_t prot)
2908 {
2909 	pte_t *pte;
2910 	vm_page_t m;
2911 	uint32_t pte_wbit;
2912 	vm_paddr_t pa;
2913 
2914 	m = NULL;
2915 	pa = 0;
2916 	PMAP_LOCK(pmap);
2917 retry:
2918 	pte = pte_find(mmu, pmap, va);
2919 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2920 		if (pmap == kernel_pmap)
2921 			pte_wbit = PTE_SW;
2922 		else
2923 			pte_wbit = PTE_UW;
2924 
2925 		if ((*pte & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2926 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2927 				goto retry;
2928 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2929 			vm_page_hold(m);
2930 		}
2931 	}
2932 
2933 	PA_UNLOCK_COND(pa);
2934 	PMAP_UNLOCK(pmap);
2935 	return (m);
2936 }
2937 
2938 /*
2939  * Initialize a vm_page's machine-dependent fields.
2940  */
2941 static void
2942 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2943 {
2944 
2945 	m->md.pv_tracked = 0;
2946 	TAILQ_INIT(&m->md.pv_list);
2947 }
2948 
2949 /*
2950  * mmu_booke_zero_page_area zeros the specified hardware page by
2951  * mapping it into virtual memory and using bzero to clear
2952  * its contents.
2953  *
2954  * off and size must reside within a single page.
2955  */
2956 static void
2957 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2958 {
2959 	vm_offset_t va;
2960 
2961 	/* XXX KASSERT off and size are within a single page? */
2962 
2963 	mtx_lock(&zero_page_mutex);
2964 	va = zero_page_va;
2965 
2966 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2967 	bzero((caddr_t)va + off, size);
2968 	mmu_booke_kremove(mmu, va);
2969 
2970 	mtx_unlock(&zero_page_mutex);
2971 }
2972 
2973 /*
2974  * mmu_booke_zero_page zeros the specified hardware page.
2975  */
2976 static void
2977 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2978 {
2979 	vm_offset_t off, va;
2980 
2981 	mtx_lock(&zero_page_mutex);
2982 	va = zero_page_va;
2983 
2984 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2985 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
2986 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
2987 	mmu_booke_kremove(mmu, va);
2988 
2989 	mtx_unlock(&zero_page_mutex);
2990 }
2991 
2992 /*
2993  * mmu_booke_copy_page copies the specified (machine independent) page by
2994  * mapping the page into virtual memory and using memcopy to copy the page,
2995  * one machine dependent page at a time.
2996  */
2997 static void
2998 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2999 {
3000 	vm_offset_t sva, dva;
3001 
3002 	sva = copy_page_src_va;
3003 	dva = copy_page_dst_va;
3004 
3005 	mtx_lock(&copy_page_mutex);
3006 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
3007 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
3008 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
3009 	mmu_booke_kremove(mmu, dva);
3010 	mmu_booke_kremove(mmu, sva);
3011 	mtx_unlock(&copy_page_mutex);
3012 }
3013 
3014 static inline void
3015 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
3016     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
3017 {
3018 	void *a_cp, *b_cp;
3019 	vm_offset_t a_pg_offset, b_pg_offset;
3020 	int cnt;
3021 
3022 	mtx_lock(&copy_page_mutex);
3023 	while (xfersize > 0) {
3024 		a_pg_offset = a_offset & PAGE_MASK;
3025 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
3026 		mmu_booke_kenter(mmu, copy_page_src_va,
3027 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
3028 		a_cp = (char *)copy_page_src_va + a_pg_offset;
3029 		b_pg_offset = b_offset & PAGE_MASK;
3030 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
3031 		mmu_booke_kenter(mmu, copy_page_dst_va,
3032 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
3033 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
3034 		bcopy(a_cp, b_cp, cnt);
3035 		mmu_booke_kremove(mmu, copy_page_dst_va);
3036 		mmu_booke_kremove(mmu, copy_page_src_va);
3037 		a_offset += cnt;
3038 		b_offset += cnt;
3039 		xfersize -= cnt;
3040 	}
3041 	mtx_unlock(&copy_page_mutex);
3042 }
3043 
3044 static vm_offset_t
3045 mmu_booke_quick_enter_page(mmu_t mmu, vm_page_t m)
3046 {
3047 	vm_paddr_t paddr;
3048 	vm_offset_t qaddr;
3049 	uint32_t flags;
3050 	pte_t *pte;
3051 
3052 	paddr = VM_PAGE_TO_PHYS(m);
3053 
3054 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
3055 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
3056 	flags |= PTE_PS_4KB;
3057 
3058 	critical_enter();
3059 	qaddr = PCPU_GET(qmap_addr);
3060 
3061 	pte = pte_find(mmu, kernel_pmap, qaddr);
3062 
3063 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
3064 
3065 	/*
3066 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
3067  	 * not be present in other TLBs.  Is there a better instruction
3068 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
3069 	 */
3070 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
3071 	__asm __volatile("isync; msync");
3072 
3073 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
3074 
3075 	/* Flush the real memory from the instruction cache. */
3076 	if ((flags & (PTE_I | PTE_G)) == 0)
3077 		__syncicache((void *)qaddr, PAGE_SIZE);
3078 
3079 	return (qaddr);
3080 }
3081 
3082 static void
3083 mmu_booke_quick_remove_page(mmu_t mmu, vm_offset_t addr)
3084 {
3085 	pte_t *pte;
3086 
3087 	pte = pte_find(mmu, kernel_pmap, addr);
3088 
3089 	KASSERT(PCPU_GET(qmap_addr) == addr,
3090 	    ("mmu_booke_quick_remove_page: invalid address"));
3091 	KASSERT(*pte != 0,
3092 	    ("mmu_booke_quick_remove_page: PTE not in use"));
3093 
3094 	*pte = 0;
3095 	critical_exit();
3096 }
3097 
3098 /*
3099  * Return whether or not the specified physical page was modified
3100  * in any of physical maps.
3101  */
3102 static boolean_t
3103 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
3104 {
3105 	pte_t *pte;
3106 	pv_entry_t pv;
3107 	boolean_t rv;
3108 
3109 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3110 	    ("mmu_booke_is_modified: page %p is not managed", m));
3111 	rv = FALSE;
3112 
3113 	/*
3114 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
3115 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
3116 	 * is clear, no PTEs can be modified.
3117 	 */
3118 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3119 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
3120 		return (rv);
3121 	rw_wlock(&pvh_global_lock);
3122 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3123 		PMAP_LOCK(pv->pv_pmap);
3124 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3125 		    PTE_ISVALID(pte)) {
3126 			if (PTE_ISMODIFIED(pte))
3127 				rv = TRUE;
3128 		}
3129 		PMAP_UNLOCK(pv->pv_pmap);
3130 		if (rv)
3131 			break;
3132 	}
3133 	rw_wunlock(&pvh_global_lock);
3134 	return (rv);
3135 }
3136 
3137 /*
3138  * Return whether or not the specified virtual address is eligible
3139  * for prefault.
3140  */
3141 static boolean_t
3142 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
3143 {
3144 
3145 	return (FALSE);
3146 }
3147 
3148 /*
3149  * Return whether or not the specified physical page was referenced
3150  * in any physical maps.
3151  */
3152 static boolean_t
3153 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
3154 {
3155 	pte_t *pte;
3156 	pv_entry_t pv;
3157 	boolean_t rv;
3158 
3159 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3160 	    ("mmu_booke_is_referenced: page %p is not managed", m));
3161 	rv = FALSE;
3162 	rw_wlock(&pvh_global_lock);
3163 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3164 		PMAP_LOCK(pv->pv_pmap);
3165 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3166 		    PTE_ISVALID(pte)) {
3167 			if (PTE_ISREFERENCED(pte))
3168 				rv = TRUE;
3169 		}
3170 		PMAP_UNLOCK(pv->pv_pmap);
3171 		if (rv)
3172 			break;
3173 	}
3174 	rw_wunlock(&pvh_global_lock);
3175 	return (rv);
3176 }
3177 
3178 /*
3179  * Clear the modify bits on the specified physical page.
3180  */
3181 static void
3182 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
3183 {
3184 	pte_t *pte;
3185 	pv_entry_t pv;
3186 
3187 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3188 	    ("mmu_booke_clear_modify: page %p is not managed", m));
3189 	VM_OBJECT_ASSERT_WLOCKED(m->object);
3190 	KASSERT(!vm_page_xbusied(m),
3191 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
3192 
3193 	/*
3194 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
3195 	 * If the object containing the page is locked and the page is not
3196 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
3197 	 */
3198 	if ((m->aflags & PGA_WRITEABLE) == 0)
3199 		return;
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 			mtx_lock_spin(&tlbivax_mutex);
3206 			tlb_miss_lock();
3207 
3208 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
3209 				tlb0_flush_entry(pv->pv_va);
3210 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
3211 				    PTE_REFERENCED);
3212 			}
3213 
3214 			tlb_miss_unlock();
3215 			mtx_unlock_spin(&tlbivax_mutex);
3216 		}
3217 		PMAP_UNLOCK(pv->pv_pmap);
3218 	}
3219 	rw_wunlock(&pvh_global_lock);
3220 }
3221 
3222 /*
3223  * Return a count of reference bits for a page, clearing those bits.
3224  * It is not necessary for every reference bit to be cleared, but it
3225  * is necessary that 0 only be returned when there are truly no
3226  * reference bits set.
3227  *
3228  * As an optimization, update the page's dirty field if a modified bit is
3229  * found while counting reference bits.  This opportunistic update can be
3230  * performed at low cost and can eliminate the need for some future calls
3231  * to pmap_is_modified().  However, since this function stops after
3232  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
3233  * dirty pages.  Those dirty pages will only be detected by a future call
3234  * to pmap_is_modified().
3235  */
3236 static int
3237 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
3238 {
3239 	pte_t *pte;
3240 	pv_entry_t pv;
3241 	int count;
3242 
3243 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3244 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
3245 	count = 0;
3246 	rw_wlock(&pvh_global_lock);
3247 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3248 		PMAP_LOCK(pv->pv_pmap);
3249 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
3250 		    PTE_ISVALID(pte)) {
3251 			if (PTE_ISMODIFIED(pte))
3252 				vm_page_dirty(m);
3253 			if (PTE_ISREFERENCED(pte)) {
3254 				mtx_lock_spin(&tlbivax_mutex);
3255 				tlb_miss_lock();
3256 
3257 				tlb0_flush_entry(pv->pv_va);
3258 				*pte &= ~PTE_REFERENCED;
3259 
3260 				tlb_miss_unlock();
3261 				mtx_unlock_spin(&tlbivax_mutex);
3262 
3263 				if (++count >= PMAP_TS_REFERENCED_MAX) {
3264 					PMAP_UNLOCK(pv->pv_pmap);
3265 					break;
3266 				}
3267 			}
3268 		}
3269 		PMAP_UNLOCK(pv->pv_pmap);
3270 	}
3271 	rw_wunlock(&pvh_global_lock);
3272 	return (count);
3273 }
3274 
3275 /*
3276  * Clear the wired attribute from the mappings for the specified range of
3277  * addresses in the given pmap.  Every valid mapping within that range must
3278  * have the wired attribute set.  In contrast, invalid mappings cannot have
3279  * the wired attribute set, so they are ignored.
3280  *
3281  * The wired attribute of the page table entry is not a hardware feature, so
3282  * there is no need to invalidate any TLB entries.
3283  */
3284 static void
3285 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
3286 {
3287 	vm_offset_t va;
3288 	pte_t *pte;
3289 
3290 	PMAP_LOCK(pmap);
3291 	for (va = sva; va < eva; va += PAGE_SIZE) {
3292 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
3293 		    PTE_ISVALID(pte)) {
3294 			if (!PTE_ISWIRED(pte))
3295 				panic("mmu_booke_unwire: pte %p isn't wired",
3296 				    pte);
3297 			*pte &= ~PTE_WIRED;
3298 			pmap->pm_stats.wired_count--;
3299 		}
3300 	}
3301 	PMAP_UNLOCK(pmap);
3302 
3303 }
3304 
3305 /*
3306  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
3307  * page.  This count may be changed upwards or downwards in the future; it is
3308  * only necessary that true be returned for a small subset of pmaps for proper
3309  * page aging.
3310  */
3311 static boolean_t
3312 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
3313 {
3314 	pv_entry_t pv;
3315 	int loops;
3316 	boolean_t rv;
3317 
3318 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
3319 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
3320 	loops = 0;
3321 	rv = FALSE;
3322 	rw_wlock(&pvh_global_lock);
3323 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3324 		if (pv->pv_pmap == pmap) {
3325 			rv = TRUE;
3326 			break;
3327 		}
3328 		if (++loops >= 16)
3329 			break;
3330 	}
3331 	rw_wunlock(&pvh_global_lock);
3332 	return (rv);
3333 }
3334 
3335 /*
3336  * Return the number of managed mappings to the given physical page that are
3337  * wired.
3338  */
3339 static int
3340 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
3341 {
3342 	pv_entry_t pv;
3343 	pte_t *pte;
3344 	int count = 0;
3345 
3346 	if ((m->oflags & VPO_UNMANAGED) != 0)
3347 		return (count);
3348 	rw_wlock(&pvh_global_lock);
3349 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
3350 		PMAP_LOCK(pv->pv_pmap);
3351 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
3352 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
3353 				count++;
3354 		PMAP_UNLOCK(pv->pv_pmap);
3355 	}
3356 	rw_wunlock(&pvh_global_lock);
3357 	return (count);
3358 }
3359 
3360 static int
3361 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3362 {
3363 	int i;
3364 	vm_offset_t va;
3365 
3366 	/*
3367 	 * This currently does not work for entries that
3368 	 * overlap TLB1 entries.
3369 	 */
3370 	for (i = 0; i < TLB1_ENTRIES; i ++) {
3371 		if (tlb1_iomapped(i, pa, size, &va) == 0)
3372 			return (0);
3373 	}
3374 
3375 	return (EFAULT);
3376 }
3377 
3378 void
3379 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
3380 {
3381 	vm_paddr_t ppa;
3382 	vm_offset_t ofs;
3383 	vm_size_t gran;
3384 
3385 	/* Minidumps are based on virtual memory addresses. */
3386 	if (do_minidump) {
3387 		*va = (void *)(vm_offset_t)pa;
3388 		return;
3389 	}
3390 
3391 	/* Raw physical memory dumps don't have a virtual address. */
3392 	/* We always map a 256MB page at 256M. */
3393 	gran = 256 * 1024 * 1024;
3394 	ppa = rounddown2(pa, gran);
3395 	ofs = pa - ppa;
3396 	*va = (void *)gran;
3397 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
3398 
3399 	if (sz > (gran - ofs))
3400 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
3401 		    _TLB_ENTRY_IO);
3402 }
3403 
3404 void
3405 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
3406 {
3407 	vm_paddr_t ppa;
3408 	vm_offset_t ofs;
3409 	vm_size_t gran;
3410 	tlb_entry_t e;
3411 	int i;
3412 
3413 	/* Minidumps are based on virtual memory addresses. */
3414 	/* Nothing to do... */
3415 	if (do_minidump)
3416 		return;
3417 
3418 	for (i = 0; i < TLB1_ENTRIES; i++) {
3419 		tlb1_read_entry(&e, i);
3420 		if (!(e.mas1 & MAS1_VALID))
3421 			break;
3422 	}
3423 
3424 	/* Raw physical memory dumps don't have a virtual address. */
3425 	i--;
3426 	e.mas1 = 0;
3427 	e.mas2 = 0;
3428 	e.mas3 = 0;
3429 	tlb1_write_entry(&e, i);
3430 
3431 	gran = 256 * 1024 * 1024;
3432 	ppa = rounddown2(pa, gran);
3433 	ofs = pa - ppa;
3434 	if (sz > (gran - ofs)) {
3435 		i--;
3436 		e.mas1 = 0;
3437 		e.mas2 = 0;
3438 		e.mas3 = 0;
3439 		tlb1_write_entry(&e, i);
3440 	}
3441 }
3442 
3443 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
3444 
3445 void
3446 mmu_booke_scan_init(mmu_t mmu)
3447 {
3448 	vm_offset_t va;
3449 	pte_t *pte;
3450 	int i;
3451 
3452 	if (!do_minidump) {
3453 		/* Initialize phys. segments for dumpsys(). */
3454 		memset(&dump_map, 0, sizeof(dump_map));
3455 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
3456 		    &availmem_regions_sz);
3457 		for (i = 0; i < physmem_regions_sz; i++) {
3458 			dump_map[i].pa_start = physmem_regions[i].mr_start;
3459 			dump_map[i].pa_size = physmem_regions[i].mr_size;
3460 		}
3461 		return;
3462 	}
3463 
3464 	/* Virtual segments for minidumps: */
3465 	memset(&dump_map, 0, sizeof(dump_map));
3466 
3467 	/* 1st: kernel .data and .bss. */
3468 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
3469 	dump_map[0].pa_size =
3470 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
3471 
3472 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
3473 	dump_map[1].pa_start = data_start;
3474 	dump_map[1].pa_size = data_end - data_start;
3475 
3476 	/* 3rd: kernel VM. */
3477 	va = dump_map[1].pa_start + dump_map[1].pa_size;
3478 	/* Find start of next chunk (from va). */
3479 	while (va < virtual_end) {
3480 		/* Don't dump the buffer cache. */
3481 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
3482 			va = kmi.buffer_eva;
3483 			continue;
3484 		}
3485 		pte = pte_find(mmu, kernel_pmap, va);
3486 		if (pte != NULL && PTE_ISVALID(pte))
3487 			break;
3488 		va += PAGE_SIZE;
3489 	}
3490 	if (va < virtual_end) {
3491 		dump_map[2].pa_start = va;
3492 		va += PAGE_SIZE;
3493 		/* Find last page in chunk. */
3494 		while (va < virtual_end) {
3495 			/* Don't run into the buffer cache. */
3496 			if (va == kmi.buffer_sva)
3497 				break;
3498 			pte = pte_find(mmu, kernel_pmap, va);
3499 			if (pte == NULL || !PTE_ISVALID(pte))
3500 				break;
3501 			va += PAGE_SIZE;
3502 		}
3503 		dump_map[2].pa_size = va - dump_map[2].pa_start;
3504 	}
3505 }
3506 
3507 /*
3508  * Map a set of physical memory pages into the kernel virtual address space.
3509  * Return a pointer to where it is mapped. This routine is intended to be used
3510  * for mapping device memory, NOT real memory.
3511  */
3512 static void *
3513 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
3514 {
3515 
3516 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
3517 }
3518 
3519 static void *
3520 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
3521 {
3522 	tlb_entry_t e;
3523 	void *res;
3524 	uintptr_t va, tmpva;
3525 	vm_size_t sz;
3526 	int i;
3527 
3528 	/*
3529 	 * Check if this is premapped in TLB1. Note: this should probably also
3530 	 * check whether a sequence of TLB1 entries exist that match the
3531 	 * requirement, but now only checks the easy case.
3532 	 */
3533 	for (i = 0; i < TLB1_ENTRIES; i++) {
3534 		tlb1_read_entry(&e, i);
3535 		if (!(e.mas1 & MAS1_VALID))
3536 			continue;
3537 		if (pa >= e.phys &&
3538 		    (pa + size) <= (e.phys + e.size) &&
3539 		    (ma == VM_MEMATTR_DEFAULT ||
3540 		     tlb_calc_wimg(pa, ma) ==
3541 		      (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED))))
3542 			return (void *)(e.virt +
3543 			    (vm_offset_t)(pa - e.phys));
3544 	}
3545 
3546 	size = roundup(size, PAGE_SIZE);
3547 
3548 	/*
3549 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
3550 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
3551 	 */
3552 #ifdef SPARSE_MAPDEV
3553 	/*
3554 	 * With a sparse mapdev, align to the largest starting region.  This
3555 	 * could feasibly be optimized for a 'best-fit' alignment, but that
3556 	 * calculation could be very costly.
3557 	 * Align to the smaller of:
3558 	 * - first set bit in overlap of (pa & size mask)
3559 	 * - largest size envelope
3560 	 *
3561 	 * It's possible the device mapping may start at a PA that's not larger
3562 	 * than the size mask, so we need to offset in to maximize the TLB entry
3563 	 * range and minimize the number of used TLB entries.
3564 	 */
3565 	do {
3566 	    tmpva = tlb1_map_base;
3567 	    sz = ffsl(((1 << flsl(size-1)) - 1) & pa);
3568 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
3569 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
3570 #ifdef __powerpc64__
3571 	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
3572 #else
3573 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
3574 #endif
3575 #else
3576 #ifdef __powerpc64__
3577 	va = atomic_fetchadd_long(&tlb1_map_base, size);
3578 #else
3579 	va = atomic_fetchadd_int(&tlb1_map_base, size);
3580 #endif
3581 #endif
3582 	res = (void *)va;
3583 
3584 	do {
3585 		sz = 1 << (ilog2(size) & ~1);
3586 		/* Align size to PA */
3587 		if (pa % sz != 0) {
3588 			do {
3589 				sz >>= 2;
3590 			} while (pa % sz != 0);
3591 		}
3592 		/* Now align from there to VA */
3593 		if (va % sz != 0) {
3594 			do {
3595 				sz >>= 2;
3596 			} while (va % sz != 0);
3597 		}
3598 		if (bootverbose)
3599 			printf("Wiring VA=%lx to PA=%jx (size=%lx)\n",
3600 			    va, (uintmax_t)pa, sz);
3601 		if (tlb1_set_entry(va, pa, sz,
3602 		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
3603 			return (NULL);
3604 		size -= sz;
3605 		pa += sz;
3606 		va += sz;
3607 	} while (size > 0);
3608 
3609 	return (res);
3610 }
3611 
3612 /*
3613  * 'Unmap' a range mapped by mmu_booke_mapdev().
3614  */
3615 static void
3616 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
3617 {
3618 #ifdef SUPPORTS_SHRINKING_TLB1
3619 	vm_offset_t base, offset;
3620 
3621 	/*
3622 	 * Unmap only if this is inside kernel virtual space.
3623 	 */
3624 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
3625 		base = trunc_page(va);
3626 		offset = va & PAGE_MASK;
3627 		size = roundup(offset + size, PAGE_SIZE);
3628 		kva_free(base, size);
3629 	}
3630 #endif
3631 }
3632 
3633 /*
3634  * mmu_booke_object_init_pt preloads the ptes for a given object into the
3635  * specified pmap. This eliminates the blast of soft faults on process startup
3636  * and immediately after an mmap.
3637  */
3638 static void
3639 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3640     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
3641 {
3642 
3643 	VM_OBJECT_ASSERT_WLOCKED(object);
3644 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
3645 	    ("mmu_booke_object_init_pt: non-device object"));
3646 }
3647 
3648 /*
3649  * Perform the pmap work for mincore.
3650  */
3651 static int
3652 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
3653     vm_paddr_t *locked_pa)
3654 {
3655 
3656 	/* XXX: this should be implemented at some point */
3657 	return (0);
3658 }
3659 
3660 static int
3661 mmu_booke_change_attr(mmu_t mmu, vm_offset_t addr, vm_size_t sz,
3662     vm_memattr_t mode)
3663 {
3664 	vm_offset_t va;
3665 	pte_t *pte;
3666 	int i, j;
3667 	tlb_entry_t e;
3668 
3669 	/* Check TLB1 mappings */
3670 	for (i = 0; i < TLB1_ENTRIES; i++) {
3671 		tlb1_read_entry(&e, i);
3672 		if (!(e.mas1 & MAS1_VALID))
3673 			continue;
3674 		if (addr >= e.virt && addr < e.virt + e.size)
3675 			break;
3676 	}
3677 	if (i < TLB1_ENTRIES) {
3678 		/* Only allow full mappings to be modified for now. */
3679 		/* Validate the range. */
3680 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
3681 			tlb1_read_entry(&e, j);
3682 			if (va != e.virt || (sz - (va - addr) < e.size))
3683 				return (EINVAL);
3684 		}
3685 		for (va = addr; va < addr + sz; va += e.size, i++) {
3686 			tlb1_read_entry(&e, i);
3687 			e.mas2 &= ~MAS2_WIMGE_MASK;
3688 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
3689 
3690 			/*
3691 			 * Write it out to the TLB.  Should really re-sync with other
3692 			 * cores.
3693 			 */
3694 			tlb1_write_entry(&e, i);
3695 		}
3696 		return (0);
3697 	}
3698 
3699 	/* Not in TLB1, try through pmap */
3700 	/* First validate the range. */
3701 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3702 		pte = pte_find(mmu, kernel_pmap, va);
3703 		if (pte == NULL || !PTE_ISVALID(pte))
3704 			return (EINVAL);
3705 	}
3706 
3707 	mtx_lock_spin(&tlbivax_mutex);
3708 	tlb_miss_lock();
3709 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
3710 		pte = pte_find(mmu, kernel_pmap, va);
3711 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
3712 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
3713 		tlb0_flush_entry(va);
3714 	}
3715 	tlb_miss_unlock();
3716 	mtx_unlock_spin(&tlbivax_mutex);
3717 
3718 	return (0);
3719 }
3720 
3721 /**************************************************************************/
3722 /* TID handling */
3723 /**************************************************************************/
3724 
3725 /*
3726  * Allocate a TID. If necessary, steal one from someone else.
3727  * The new TID is flushed from the TLB before returning.
3728  */
3729 static tlbtid_t
3730 tid_alloc(pmap_t pmap)
3731 {
3732 	tlbtid_t tid;
3733 	int thiscpu;
3734 
3735 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
3736 
3737 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
3738 
3739 	thiscpu = PCPU_GET(cpuid);
3740 
3741 	tid = PCPU_GET(booke.tid_next);
3742 	if (tid > TID_MAX)
3743 		tid = TID_MIN;
3744 	PCPU_SET(booke.tid_next, tid + 1);
3745 
3746 	/* If we are stealing TID then clear the relevant pmap's field */
3747 	if (tidbusy[thiscpu][tid] != NULL) {
3748 
3749 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
3750 
3751 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
3752 
3753 		/* Flush all entries from TLB0 matching this TID. */
3754 		tid_flush(tid);
3755 	}
3756 
3757 	tidbusy[thiscpu][tid] = pmap;
3758 	pmap->pm_tid[thiscpu] = tid;
3759 	__asm __volatile("msync; isync");
3760 
3761 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
3762 	    PCPU_GET(booke.tid_next));
3763 
3764 	return (tid);
3765 }
3766 
3767 /**************************************************************************/
3768 /* TLB0 handling */
3769 /**************************************************************************/
3770 
3771 static void
3772 #ifdef __powerpc64__
3773 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
3774 #else
3775 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3776 #endif
3777     uint32_t mas7)
3778 {
3779 	int as;
3780 	char desc[3];
3781 	tlbtid_t tid;
3782 	vm_size_t size;
3783 	unsigned int tsize;
3784 
3785 	desc[2] = '\0';
3786 	if (mas1 & MAS1_VALID)
3787 		desc[0] = 'V';
3788 	else
3789 		desc[0] = ' ';
3790 
3791 	if (mas1 & MAS1_IPROT)
3792 		desc[1] = 'P';
3793 	else
3794 		desc[1] = ' ';
3795 
3796 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3797 	tid = MAS1_GETTID(mas1);
3798 
3799 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3800 	size = 0;
3801 	if (tsize)
3802 		size = tsize2size(tsize);
3803 
3804 	debugf("%3d: (%s) [AS=%d] "
3805 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
3806 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3807 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
3808 }
3809 
3810 /* Convert TLB0 va and way number to tlb0[] table index. */
3811 static inline unsigned int
3812 tlb0_tableidx(vm_offset_t va, unsigned int way)
3813 {
3814 	unsigned int idx;
3815 
3816 	idx = (way * TLB0_ENTRIES_PER_WAY);
3817 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
3818 	return (idx);
3819 }
3820 
3821 /*
3822  * Invalidate TLB0 entry.
3823  */
3824 static inline void
3825 tlb0_flush_entry(vm_offset_t va)
3826 {
3827 
3828 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
3829 
3830 	mtx_assert(&tlbivax_mutex, MA_OWNED);
3831 
3832 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
3833 	__asm __volatile("isync; msync");
3834 	__asm __volatile("tlbsync; msync");
3835 
3836 	CTR1(KTR_PMAP, "%s: e", __func__);
3837 }
3838 
3839 /* Print out contents of the MAS registers for each TLB0 entry */
3840 void
3841 tlb0_print_tlbentries(void)
3842 {
3843 	uint32_t mas0, mas1, mas3, mas7;
3844 #ifdef __powerpc64__
3845 	uint64_t mas2;
3846 #else
3847 	uint32_t mas2;
3848 #endif
3849 	int entryidx, way, idx;
3850 
3851 	debugf("TLB0 entries:\n");
3852 	for (way = 0; way < TLB0_WAYS; way ++)
3853 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3854 
3855 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3856 			mtspr(SPR_MAS0, mas0);
3857 			__asm __volatile("isync");
3858 
3859 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3860 			mtspr(SPR_MAS2, mas2);
3861 
3862 			__asm __volatile("isync; tlbre");
3863 
3864 			mas1 = mfspr(SPR_MAS1);
3865 			mas2 = mfspr(SPR_MAS2);
3866 			mas3 = mfspr(SPR_MAS3);
3867 			mas7 = mfspr(SPR_MAS7);
3868 
3869 			idx = tlb0_tableidx(mas2, way);
3870 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3871 		}
3872 }
3873 
3874 /**************************************************************************/
3875 /* TLB1 handling */
3876 /**************************************************************************/
3877 
3878 /*
3879  * TLB1 mapping notes:
3880  *
3881  * TLB1[0]	Kernel text and data.
3882  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
3883  *		windows, other devices mappings.
3884  */
3885 
3886  /*
3887  * Read an entry from given TLB1 slot.
3888  */
3889 void
3890 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
3891 {
3892 	register_t msr;
3893 	uint32_t mas0;
3894 
3895 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
3896 
3897 	msr = mfmsr();
3898 	__asm __volatile("wrteei 0");
3899 
3900 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
3901 	mtspr(SPR_MAS0, mas0);
3902 	__asm __volatile("isync; tlbre");
3903 
3904 	entry->mas1 = mfspr(SPR_MAS1);
3905 	entry->mas2 = mfspr(SPR_MAS2);
3906 	entry->mas3 = mfspr(SPR_MAS3);
3907 
3908 	switch ((mfpvr() >> 16) & 0xFFFF) {
3909 	case FSL_E500v2:
3910 	case FSL_E500mc:
3911 	case FSL_E5500:
3912 	case FSL_E6500:
3913 		entry->mas7 = mfspr(SPR_MAS7);
3914 		break;
3915 	default:
3916 		entry->mas7 = 0;
3917 		break;
3918 	}
3919 	mtmsr(msr);
3920 
3921 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
3922 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
3923 	    (entry->mas3 & MAS3_RPN);
3924 	entry->size =
3925 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
3926 }
3927 
3928 struct tlbwrite_args {
3929 	tlb_entry_t *e;
3930 	unsigned int idx;
3931 };
3932 
3933 static void
3934 tlb1_write_entry_int(void *arg)
3935 {
3936 	struct tlbwrite_args *args = arg;
3937 	uint32_t mas0;
3938 
3939 	/* Select entry */
3940 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(args->idx);
3941 
3942 	mtspr(SPR_MAS0, mas0);
3943 	__asm __volatile("isync");
3944 	mtspr(SPR_MAS1, args->e->mas1);
3945 	__asm __volatile("isync");
3946 	mtspr(SPR_MAS2, args->e->mas2);
3947 	__asm __volatile("isync");
3948 	mtspr(SPR_MAS3, args->e->mas3);
3949 	__asm __volatile("isync");
3950 	switch ((mfpvr() >> 16) & 0xFFFF) {
3951 	case FSL_E500mc:
3952 	case FSL_E5500:
3953 	case FSL_E6500:
3954 		mtspr(SPR_MAS8, 0);
3955 		__asm __volatile("isync");
3956 		/* FALLTHROUGH */
3957 	case FSL_E500v2:
3958 		mtspr(SPR_MAS7, args->e->mas7);
3959 		__asm __volatile("isync");
3960 		break;
3961 	default:
3962 		break;
3963 	}
3964 
3965 	__asm __volatile("tlbwe; isync; msync");
3966 
3967 }
3968 
3969 static void
3970 tlb1_write_entry_sync(void *arg)
3971 {
3972 	/* Empty synchronization point for smp_rendezvous(). */
3973 }
3974 
3975 /*
3976  * Write given entry to TLB1 hardware.
3977  */
3978 static void
3979 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
3980 {
3981 	struct tlbwrite_args args;
3982 
3983 	args.e = e;
3984 	args.idx = idx;
3985 
3986 #ifdef SMP
3987 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
3988 		mb();
3989 		smp_rendezvous(tlb1_write_entry_sync,
3990 		    tlb1_write_entry_int,
3991 		    tlb1_write_entry_sync, &args);
3992 	} else
3993 #endif
3994 	{
3995 		register_t msr;
3996 
3997 		msr = mfmsr();
3998 		__asm __volatile("wrteei 0");
3999 		tlb1_write_entry_int(&args);
4000 		mtmsr(msr);
4001 	}
4002 }
4003 
4004 /*
4005  * Return the largest uint value log such that 2^log <= num.
4006  */
4007 static unsigned int
4008 ilog2(unsigned int num)
4009 {
4010 	int lz;
4011 
4012 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
4013 	return (31 - lz);
4014 }
4015 
4016 /*
4017  * Convert TLB TSIZE value to mapped region size.
4018  */
4019 static vm_size_t
4020 tsize2size(unsigned int tsize)
4021 {
4022 
4023 	/*
4024 	 * size = 4^tsize KB
4025 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
4026 	 */
4027 
4028 	return ((1 << (2 * tsize)) * 1024);
4029 }
4030 
4031 /*
4032  * Convert region size (must be power of 4) to TLB TSIZE value.
4033  */
4034 static unsigned int
4035 size2tsize(vm_size_t size)
4036 {
4037 
4038 	return (ilog2(size) / 2 - 5);
4039 }
4040 
4041 /*
4042  * Register permanent kernel mapping in TLB1.
4043  *
4044  * Entries are created starting from index 0 (current free entry is
4045  * kept in tlb1_idx) and are not supposed to be invalidated.
4046  */
4047 int
4048 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
4049     uint32_t flags)
4050 {
4051 	tlb_entry_t e;
4052 	uint32_t ts, tid;
4053 	int tsize, index;
4054 
4055 	for (index = 0; index < TLB1_ENTRIES; index++) {
4056 		tlb1_read_entry(&e, index);
4057 		if ((e.mas1 & MAS1_VALID) == 0)
4058 			break;
4059 		/* Check if we're just updating the flags, and update them. */
4060 		if (e.phys == pa && e.virt == va && e.size == size) {
4061 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
4062 			tlb1_write_entry(&e, index);
4063 			return (0);
4064 		}
4065 	}
4066 	if (index >= TLB1_ENTRIES) {
4067 		printf("tlb1_set_entry: TLB1 full!\n");
4068 		return (-1);
4069 	}
4070 
4071 	/* Convert size to TSIZE */
4072 	tsize = size2tsize(size);
4073 
4074 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
4075 	/* XXX TS is hard coded to 0 for now as we only use single address space */
4076 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
4077 
4078 	e.phys = pa;
4079 	e.virt = va;
4080 	e.size = size;
4081 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
4082 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
4083 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
4084 
4085 	/* Set supervisor RWX permission bits */
4086 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
4087 	e.mas7 = (pa >> 32) & MAS7_RPN;
4088 
4089 	tlb1_write_entry(&e, index);
4090 
4091 	/*
4092 	 * XXX in general TLB1 updates should be propagated between CPUs,
4093 	 * since current design assumes to have the same TLB1 set-up on all
4094 	 * cores.
4095 	 */
4096 	return (0);
4097 }
4098 
4099 /*
4100  * Map in contiguous RAM region into the TLB1 using maximum of
4101  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
4102  *
4103  * If necessary round up last entry size and return total size
4104  * used by all allocated entries.
4105  */
4106 vm_size_t
4107 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
4108 {
4109 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
4110 	vm_size_t mapped, pgsz, base, mask;
4111 	int idx, nents;
4112 
4113 	/* Round up to the next 1M */
4114 	size = roundup2(size, 1 << 20);
4115 
4116 	mapped = 0;
4117 	idx = 0;
4118 	base = va;
4119 	pgsz = 64*1024*1024;
4120 	while (mapped < size) {
4121 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
4122 			while (pgsz > (size - mapped))
4123 				pgsz >>= 2;
4124 			pgs[idx++] = pgsz;
4125 			mapped += pgsz;
4126 		}
4127 
4128 		/* We under-map. Correct for this. */
4129 		if (mapped < size) {
4130 			while (pgs[idx - 1] == pgsz) {
4131 				idx--;
4132 				mapped -= pgsz;
4133 			}
4134 			/* XXX We may increase beyond out starting point. */
4135 			pgsz <<= 2;
4136 			pgs[idx++] = pgsz;
4137 			mapped += pgsz;
4138 		}
4139 	}
4140 
4141 	nents = idx;
4142 	mask = pgs[0] - 1;
4143 	/* Align address to the boundary */
4144 	if (va & mask) {
4145 		va = (va + mask) & ~mask;
4146 		pa = (pa + mask) & ~mask;
4147 	}
4148 
4149 	for (idx = 0; idx < nents; idx++) {
4150 		pgsz = pgs[idx];
4151 		debugf("%u: %llx -> %x, size=%x\n", idx, pa, va, pgsz);
4152 		tlb1_set_entry(va, pa, pgsz,
4153 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
4154 		pa += pgsz;
4155 		va += pgsz;
4156 	}
4157 
4158 	mapped = (va - base);
4159 	printf("mapped size 0x%"PRI0ptrX" (wasted space 0x%"PRIxPTR")\n",
4160 	    mapped, mapped - size);
4161 	return (mapped);
4162 }
4163 
4164 /*
4165  * TLB1 initialization routine, to be called after the very first
4166  * assembler level setup done in locore.S.
4167  */
4168 void
4169 tlb1_init()
4170 {
4171 	uint32_t mas0, mas1, mas2, mas3, mas7;
4172 	uint32_t tsz;
4173 
4174 	tlb1_get_tlbconf();
4175 
4176 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
4177 	mtspr(SPR_MAS0, mas0);
4178 	__asm __volatile("isync; tlbre");
4179 
4180 	mas1 = mfspr(SPR_MAS1);
4181 	mas2 = mfspr(SPR_MAS2);
4182 	mas3 = mfspr(SPR_MAS3);
4183 	mas7 = mfspr(SPR_MAS7);
4184 
4185 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
4186 	    (mas3 & MAS3_RPN);
4187 
4188 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4189 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
4190 
4191 	/* Setup TLB miss defaults */
4192 	set_mas4_defaults();
4193 }
4194 
4195 /*
4196  * pmap_early_io_unmap() should be used in short conjunction with
4197  * pmap_early_io_map(), as in the following snippet:
4198  *
4199  * x = pmap_early_io_map(...);
4200  * <do something with x>
4201  * pmap_early_io_unmap(x, size);
4202  *
4203  * And avoiding more allocations between.
4204  */
4205 void
4206 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
4207 {
4208 	int i;
4209 	tlb_entry_t e;
4210 	vm_size_t isize;
4211 
4212 	size = roundup(size, PAGE_SIZE);
4213 	isize = size;
4214 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
4215 		tlb1_read_entry(&e, i);
4216 		if (!(e.mas1 & MAS1_VALID))
4217 			continue;
4218 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
4219 			size -= e.size;
4220 			e.mas1 &= ~MAS1_VALID;
4221 			tlb1_write_entry(&e, i);
4222 		}
4223 	}
4224 	if (tlb1_map_base == va + isize)
4225 		tlb1_map_base -= isize;
4226 }
4227 
4228 vm_offset_t
4229 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
4230 {
4231 	vm_paddr_t pa_base;
4232 	vm_offset_t va, sz;
4233 	int i;
4234 	tlb_entry_t e;
4235 
4236 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
4237 
4238 	for (i = 0; i < TLB1_ENTRIES; i++) {
4239 		tlb1_read_entry(&e, i);
4240 		if (!(e.mas1 & MAS1_VALID))
4241 			continue;
4242 		if (pa >= e.phys && (pa + size) <=
4243 		    (e.phys + e.size))
4244 			return (e.virt + (pa - e.phys));
4245 	}
4246 
4247 	pa_base = rounddown(pa, PAGE_SIZE);
4248 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
4249 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
4250 	va = tlb1_map_base + (pa - pa_base);
4251 
4252 	do {
4253 		sz = 1 << (ilog2(size) & ~1);
4254 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
4255 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
4256 		size -= sz;
4257 		pa_base += sz;
4258 		tlb1_map_base += sz;
4259 	} while (size > 0);
4260 
4261 	return (va);
4262 }
4263 
4264 void
4265 pmap_track_page(pmap_t pmap, vm_offset_t va)
4266 {
4267 	vm_paddr_t pa;
4268 	vm_page_t page;
4269 	struct pv_entry *pve;
4270 
4271 	va = trunc_page(va);
4272 	pa = pmap_kextract(va);
4273 	page = PHYS_TO_VM_PAGE(pa);
4274 
4275 	rw_wlock(&pvh_global_lock);
4276 	PMAP_LOCK(pmap);
4277 
4278 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
4279 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
4280 			goto out;
4281 		}
4282 	}
4283 	page->md.pv_tracked = true;
4284 	pv_insert(pmap, va, page);
4285 out:
4286 	PMAP_UNLOCK(pmap);
4287 	rw_wunlock(&pvh_global_lock);
4288 }
4289 
4290 
4291 /*
4292  * Setup MAS4 defaults.
4293  * These values are loaded to MAS0-2 on a TLB miss.
4294  */
4295 static void
4296 set_mas4_defaults(void)
4297 {
4298 	uint32_t mas4;
4299 
4300 	/* Defaults: TLB0, PID0, TSIZED=4K */
4301 	mas4 = MAS4_TLBSELD0;
4302 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
4303 #ifdef SMP
4304 	mas4 |= MAS4_MD;
4305 #endif
4306 	mtspr(SPR_MAS4, mas4);
4307 	__asm __volatile("isync");
4308 }
4309 
4310 /*
4311  * Print out contents of the MAS registers for each TLB1 entry
4312  */
4313 void
4314 tlb1_print_tlbentries(void)
4315 {
4316 	uint32_t mas0, mas1, mas3, mas7;
4317 #ifdef __powerpc64__
4318 	uint64_t mas2;
4319 #else
4320 	uint32_t mas2;
4321 #endif
4322 	int i;
4323 
4324 	debugf("TLB1 entries:\n");
4325 	for (i = 0; i < TLB1_ENTRIES; i++) {
4326 
4327 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
4328 		mtspr(SPR_MAS0, mas0);
4329 
4330 		__asm __volatile("isync; tlbre");
4331 
4332 		mas1 = mfspr(SPR_MAS1);
4333 		mas2 = mfspr(SPR_MAS2);
4334 		mas3 = mfspr(SPR_MAS3);
4335 		mas7 = mfspr(SPR_MAS7);
4336 
4337 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
4338 	}
4339 }
4340 
4341 /*
4342  * Return 0 if the physical IO range is encompassed by one of the
4343  * the TLB1 entries, otherwise return related error code.
4344  */
4345 static int
4346 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
4347 {
4348 	uint32_t prot;
4349 	vm_paddr_t pa_start;
4350 	vm_paddr_t pa_end;
4351 	unsigned int entry_tsize;
4352 	vm_size_t entry_size;
4353 	tlb_entry_t e;
4354 
4355 	*va = (vm_offset_t)NULL;
4356 
4357 	tlb1_read_entry(&e, i);
4358 	/* Skip invalid entries */
4359 	if (!(e.mas1 & MAS1_VALID))
4360 		return (EINVAL);
4361 
4362 	/*
4363 	 * The entry must be cache-inhibited, guarded, and r/w
4364 	 * so it can function as an i/o page
4365 	 */
4366 	prot = e.mas2 & (MAS2_I | MAS2_G);
4367 	if (prot != (MAS2_I | MAS2_G))
4368 		return (EPERM);
4369 
4370 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
4371 	if (prot != (MAS3_SR | MAS3_SW))
4372 		return (EPERM);
4373 
4374 	/* The address should be within the entry range. */
4375 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
4376 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
4377 
4378 	entry_size = tsize2size(entry_tsize);
4379 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
4380 	    (e.mas3 & MAS3_RPN);
4381 	pa_end = pa_start + entry_size;
4382 
4383 	if ((pa < pa_start) || ((pa + size) > pa_end))
4384 		return (ERANGE);
4385 
4386 	/* Return virtual address of this mapping. */
4387 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
4388 	return (0);
4389 }
4390 
4391 /*
4392  * Invalidate all TLB0 entries which match the given TID. Note this is
4393  * dedicated for cases when invalidations should NOT be propagated to other
4394  * CPUs.
4395  */
4396 static void
4397 tid_flush(tlbtid_t tid)
4398 {
4399 	register_t msr;
4400 	uint32_t mas0, mas1, mas2;
4401 	int entry, way;
4402 
4403 
4404 	/* Don't evict kernel translations */
4405 	if (tid == TID_KERNEL)
4406 		return;
4407 
4408 	msr = mfmsr();
4409 	__asm __volatile("wrteei 0");
4410 
4411 	for (way = 0; way < TLB0_WAYS; way++)
4412 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
4413 
4414 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
4415 			mtspr(SPR_MAS0, mas0);
4416 			__asm __volatile("isync");
4417 
4418 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
4419 			mtspr(SPR_MAS2, mas2);
4420 
4421 			__asm __volatile("isync; tlbre");
4422 
4423 			mas1 = mfspr(SPR_MAS1);
4424 
4425 			if (!(mas1 & MAS1_VALID))
4426 				continue;
4427 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
4428 				continue;
4429 			mas1 &= ~MAS1_VALID;
4430 			mtspr(SPR_MAS1, mas1);
4431 			__asm __volatile("isync; tlbwe; isync; msync");
4432 		}
4433 	mtmsr(msr);
4434 }
4435