xref: /freebsd/sys/powerpc/booke/pmap.c (revision d4ae33f0721c1b170fe37d97e395228ffcfb3f80)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/param.h>
55 #include <sys/malloc.h>
56 #include <sys/ktr.h>
57 #include <sys/proc.h>
58 #include <sys/user.h>
59 #include <sys/queue.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/linker.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/rwlock.h>
67 #include <sys/sched.h>
68 #include <sys/smp.h>
69 #include <sys/vmmeter.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_pager.h>
80 #include <vm/uma.h>
81 
82 #include <machine/cpu.h>
83 #include <machine/pcb.h>
84 #include <machine/platform.h>
85 
86 #include <machine/tlb.h>
87 #include <machine/spr.h>
88 #include <machine/md_var.h>
89 #include <machine/mmuvar.h>
90 #include <machine/pmap.h>
91 #include <machine/pte.h>
92 
93 #include "mmu_if.h"
94 
95 #ifdef  DEBUG
96 #define debugf(fmt, args...) printf(fmt, ##args)
97 #else
98 #define debugf(fmt, args...)
99 #endif
100 
101 #define TODO			panic("%s: not implemented", __func__);
102 
103 extern int dumpsys_minidump;
104 
105 extern unsigned char _etext[];
106 extern unsigned char _end[];
107 
108 extern uint32_t *bootinfo;
109 
110 #ifdef SMP
111 extern uint32_t bp_ntlb1s;
112 #endif
113 
114 vm_paddr_t kernload;
115 vm_offset_t kernstart;
116 vm_size_t kernsize;
117 
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
121 
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
127 
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
131 
132 static struct mtx tlbivax_mutex;
133 
134 /*
135  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136  * by idle thred only, no lock required.
137  */
138 static vm_offset_t zero_page_idle_va;
139 
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
144 
145 /**************************************************************************/
146 /* PMAP */
147 /**************************************************************************/
148 
149 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, boolean_t);
151 
152 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
154 
155 /*
156  * If user pmap is processed with mmu_booke_remove and the resident count
157  * drops to 0, there are no more pages to remove, so we need not continue.
158  */
159 #define PMAP_REMOVE_DONE(pmap) \
160 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
161 
162 extern void tid_flush(tlbtid_t);
163 
164 /**************************************************************************/
165 /* TLB and TID handling */
166 /**************************************************************************/
167 
168 /* Translation ID busy table */
169 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
170 
171 /*
172  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
173  * core revisions and should be read from h/w registers during early config.
174  */
175 uint32_t tlb0_entries;
176 uint32_t tlb0_ways;
177 uint32_t tlb0_entries_per_way;
178 
179 #define TLB0_ENTRIES		(tlb0_entries)
180 #define TLB0_WAYS		(tlb0_ways)
181 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
182 
183 #define TLB1_ENTRIES 16
184 
185 /* In-ram copy of the TLB1 */
186 static tlb_entry_t tlb1[TLB1_ENTRIES];
187 
188 /* Next free entry in the TLB1 */
189 static unsigned int tlb1_idx;
190 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
191 
192 static tlbtid_t tid_alloc(struct pmap *);
193 
194 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
195 
196 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
197 static void tlb1_write_entry(unsigned int);
198 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
199 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
200 
201 static vm_size_t tsize2size(unsigned int);
202 static unsigned int size2tsize(vm_size_t);
203 static unsigned int ilog2(unsigned int);
204 
205 static void set_mas4_defaults(void);
206 
207 static inline void tlb0_flush_entry(vm_offset_t);
208 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
209 
210 /**************************************************************************/
211 /* Page table management */
212 /**************************************************************************/
213 
214 static struct rwlock_padalign pvh_global_lock;
215 
216 /* Data for the pv entry allocation mechanism */
217 static uma_zone_t pvzone;
218 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
219 
220 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
221 
222 #ifndef PMAP_SHPGPERPROC
223 #define PMAP_SHPGPERPROC	200
224 #endif
225 
226 static void ptbl_init(void);
227 static struct ptbl_buf *ptbl_buf_alloc(void);
228 static void ptbl_buf_free(struct ptbl_buf *);
229 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
230 
231 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
232 static void ptbl_free(mmu_t, pmap_t, unsigned int);
233 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
234 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
235 
236 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
237 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
238 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
239 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
240 
241 static pv_entry_t pv_alloc(void);
242 static void pv_free(pv_entry_t);
243 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
244 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
245 
246 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
247 #define PTBL_BUFS		(128 * 16)
248 
249 struct ptbl_buf {
250 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
251 	vm_offset_t kva;		/* va of mapping */
252 };
253 
254 /* ptbl free list and a lock used for access synchronization. */
255 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
256 static struct mtx ptbl_buf_freelist_lock;
257 
258 /* Base address of kva space allocated fot ptbl bufs. */
259 static vm_offset_t ptbl_buf_pool_vabase;
260 
261 /* Pointer to ptbl_buf structures. */
262 static struct ptbl_buf *ptbl_bufs;
263 
264 void pmap_bootstrap_ap(volatile uint32_t *);
265 
266 /*
267  * Kernel MMU interface
268  */
269 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
270 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
271 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
272     vm_size_t, vm_offset_t);
273 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
274 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
275     vm_offset_t, vm_page_t *, vm_offset_t, int);
276 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
277     vm_prot_t, boolean_t);
278 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
279     vm_page_t, vm_prot_t);
280 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t);
282 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
283 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
284     vm_prot_t);
285 static void		mmu_booke_init(mmu_t);
286 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
287 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
288 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
289 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
290 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
291     int);
292 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
293     vm_paddr_t *);
294 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
295     vm_object_t, vm_pindex_t, vm_size_t);
296 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
297 static void		mmu_booke_page_init(mmu_t, vm_page_t);
298 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
299 static void		mmu_booke_pinit(mmu_t, pmap_t);
300 static void		mmu_booke_pinit0(mmu_t, pmap_t);
301 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
302     vm_prot_t);
303 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
304 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
305 static void		mmu_booke_release(mmu_t, pmap_t);
306 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
308 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
309 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
310 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
311 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
312 static void		mmu_booke_activate(mmu_t, struct thread *);
313 static void		mmu_booke_deactivate(mmu_t, struct thread *);
314 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
315 static void		*mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
316 static void		*mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
317 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
318 static vm_paddr_t	mmu_booke_kextract(mmu_t, vm_offset_t);
319 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
320 static void		mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
321 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
322 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
323 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
324     vm_size_t);
325 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
326     vm_size_t, vm_size_t *);
327 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
328     vm_size_t, vm_offset_t);
329 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
330 
331 static mmu_method_t mmu_booke_methods[] = {
332 	/* pmap dispatcher interface */
333 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
334 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
335 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
336 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
337 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
338 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
339 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
340 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
341 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
342 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
343 	MMUMETHOD(mmu_init,		mmu_booke_init),
344 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
345 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
346 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
347 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
348 	MMUMETHOD(mmu_map,		mmu_booke_map),
349 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
350 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
351 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
352 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
353 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
354 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
355 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
356 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
357 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
358 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
359 	MMUMETHOD(mmu_release,		mmu_booke_release),
360 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
361 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
362 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
363 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
364 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
365 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
366 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
367 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
368 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
369 
370 	/* Internal interfaces */
371 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
372 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
373 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
374 	MMUMETHOD(mmu_mapdev_attr,	mmu_booke_mapdev_attr),
375 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
376 	MMUMETHOD(mmu_kenter_attr,	mmu_booke_kenter_attr),
377 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
378 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
379 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
380 
381 	/* dumpsys() support */
382 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
383 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
384 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
385 
386 	{ 0, 0 }
387 };
388 
389 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
390 
391 static __inline uint32_t
392 tlb_calc_wimg(vm_offset_t pa, vm_memattr_t ma)
393 {
394 	uint32_t attrib;
395 	int i;
396 
397 	if (ma != VM_MEMATTR_DEFAULT) {
398 		switch (ma) {
399 		case VM_MEMATTR_UNCACHEABLE:
400 			return (PTE_I | PTE_G);
401 		case VM_MEMATTR_WRITE_COMBINING:
402 		case VM_MEMATTR_WRITE_BACK:
403 		case VM_MEMATTR_PREFETCHABLE:
404 			return (PTE_I);
405 		case VM_MEMATTR_WRITE_THROUGH:
406 			return (PTE_W | PTE_M);
407 		}
408 	}
409 
410 	/*
411 	 * Assume the page is cache inhibited and access is guarded unless
412 	 * it's in our available memory array.
413 	 */
414 	attrib = _TLB_ENTRY_IO;
415 	for (i = 0; i < physmem_regions_sz; i++) {
416 		if ((pa >= physmem_regions[i].mr_start) &&
417 		    (pa < (physmem_regions[i].mr_start +
418 		     physmem_regions[i].mr_size))) {
419 			attrib = _TLB_ENTRY_MEM;
420 			break;
421 		}
422 	}
423 
424 	return (attrib);
425 }
426 
427 static inline void
428 tlb_miss_lock(void)
429 {
430 #ifdef SMP
431 	struct pcpu *pc;
432 
433 	if (!smp_started)
434 		return;
435 
436 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
437 		if (pc != pcpup) {
438 
439 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
440 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
441 
442 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
443 			    ("tlb_miss_lock: tried to lock self"));
444 
445 			tlb_lock(pc->pc_booke_tlb_lock);
446 
447 			CTR1(KTR_PMAP, "%s: locked", __func__);
448 		}
449 	}
450 #endif
451 }
452 
453 static inline void
454 tlb_miss_unlock(void)
455 {
456 #ifdef SMP
457 	struct pcpu *pc;
458 
459 	if (!smp_started)
460 		return;
461 
462 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
463 		if (pc != pcpup) {
464 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
465 			    __func__, pc->pc_cpuid);
466 
467 			tlb_unlock(pc->pc_booke_tlb_lock);
468 
469 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
470 		}
471 	}
472 #endif
473 }
474 
475 /* Return number of entries in TLB0. */
476 static __inline void
477 tlb0_get_tlbconf(void)
478 {
479 	uint32_t tlb0_cfg;
480 
481 	tlb0_cfg = mfspr(SPR_TLB0CFG);
482 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
483 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
484 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
485 }
486 
487 /* Initialize pool of kva ptbl buffers. */
488 static void
489 ptbl_init(void)
490 {
491 	int i;
492 
493 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
494 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
495 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
496 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
497 
498 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
499 	TAILQ_INIT(&ptbl_buf_freelist);
500 
501 	for (i = 0; i < PTBL_BUFS; i++) {
502 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
503 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
504 	}
505 }
506 
507 /* Get a ptbl_buf from the freelist. */
508 static struct ptbl_buf *
509 ptbl_buf_alloc(void)
510 {
511 	struct ptbl_buf *buf;
512 
513 	mtx_lock(&ptbl_buf_freelist_lock);
514 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
515 	if (buf != NULL)
516 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
517 	mtx_unlock(&ptbl_buf_freelist_lock);
518 
519 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
520 
521 	return (buf);
522 }
523 
524 /* Return ptbl buff to free pool. */
525 static void
526 ptbl_buf_free(struct ptbl_buf *buf)
527 {
528 
529 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
530 
531 	mtx_lock(&ptbl_buf_freelist_lock);
532 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
533 	mtx_unlock(&ptbl_buf_freelist_lock);
534 }
535 
536 /*
537  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
538  */
539 static void
540 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
541 {
542 	struct ptbl_buf *pbuf;
543 
544 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
545 
546 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
547 
548 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
549 		if (pbuf->kva == (vm_offset_t)ptbl) {
550 			/* Remove from pmap ptbl buf list. */
551 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
552 
553 			/* Free corresponding ptbl buf. */
554 			ptbl_buf_free(pbuf);
555 			break;
556 		}
557 }
558 
559 /* Allocate page table. */
560 static pte_t *
561 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
562 {
563 	vm_page_t mtbl[PTBL_PAGES];
564 	vm_page_t m;
565 	struct ptbl_buf *pbuf;
566 	unsigned int pidx;
567 	pte_t *ptbl;
568 	int i;
569 
570 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
571 	    (pmap == kernel_pmap), pdir_idx);
572 
573 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
574 	    ("ptbl_alloc: invalid pdir_idx"));
575 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
576 	    ("pte_alloc: valid ptbl entry exists!"));
577 
578 	pbuf = ptbl_buf_alloc();
579 	if (pbuf == NULL)
580 		panic("pte_alloc: couldn't alloc kernel virtual memory");
581 
582 	ptbl = (pte_t *)pbuf->kva;
583 
584 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
585 
586 	/* Allocate ptbl pages, this will sleep! */
587 	for (i = 0; i < PTBL_PAGES; i++) {
588 		pidx = (PTBL_PAGES * pdir_idx) + i;
589 		while ((m = vm_page_alloc(NULL, pidx,
590 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
591 
592 			PMAP_UNLOCK(pmap);
593 			rw_wunlock(&pvh_global_lock);
594 			VM_WAIT;
595 			rw_wlock(&pvh_global_lock);
596 			PMAP_LOCK(pmap);
597 		}
598 		mtbl[i] = m;
599 	}
600 
601 	/* Map allocated pages into kernel_pmap. */
602 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
603 
604 	/* Zero whole ptbl. */
605 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
606 
607 	/* Add pbuf to the pmap ptbl bufs list. */
608 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
609 
610 	return (ptbl);
611 }
612 
613 /* Free ptbl pages and invalidate pdir entry. */
614 static void
615 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
616 {
617 	pte_t *ptbl;
618 	vm_paddr_t pa;
619 	vm_offset_t va;
620 	vm_page_t m;
621 	int i;
622 
623 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
624 	    (pmap == kernel_pmap), pdir_idx);
625 
626 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
627 	    ("ptbl_free: invalid pdir_idx"));
628 
629 	ptbl = pmap->pm_pdir[pdir_idx];
630 
631 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
632 
633 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
634 
635 	/*
636 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
637 	 * don't attempt to look up the page tables we are releasing.
638 	 */
639 	mtx_lock_spin(&tlbivax_mutex);
640 	tlb_miss_lock();
641 
642 	pmap->pm_pdir[pdir_idx] = NULL;
643 
644 	tlb_miss_unlock();
645 	mtx_unlock_spin(&tlbivax_mutex);
646 
647 	for (i = 0; i < PTBL_PAGES; i++) {
648 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
649 		pa = pte_vatopa(mmu, kernel_pmap, va);
650 		m = PHYS_TO_VM_PAGE(pa);
651 		vm_page_free_zero(m);
652 		atomic_subtract_int(&cnt.v_wire_count, 1);
653 		mmu_booke_kremove(mmu, va);
654 	}
655 
656 	ptbl_free_pmap_ptbl(pmap, ptbl);
657 }
658 
659 /*
660  * Decrement ptbl pages hold count and attempt to free ptbl pages.
661  * Called when removing pte entry from ptbl.
662  *
663  * Return 1 if ptbl pages were freed.
664  */
665 static int
666 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
667 {
668 	pte_t *ptbl;
669 	vm_paddr_t pa;
670 	vm_page_t m;
671 	int i;
672 
673 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
674 	    (pmap == kernel_pmap), pdir_idx);
675 
676 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
677 	    ("ptbl_unhold: invalid pdir_idx"));
678 	KASSERT((pmap != kernel_pmap),
679 	    ("ptbl_unhold: unholding kernel ptbl!"));
680 
681 	ptbl = pmap->pm_pdir[pdir_idx];
682 
683 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
684 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
685 	    ("ptbl_unhold: non kva ptbl"));
686 
687 	/* decrement hold count */
688 	for (i = 0; i < PTBL_PAGES; i++) {
689 		pa = pte_vatopa(mmu, kernel_pmap,
690 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
691 		m = PHYS_TO_VM_PAGE(pa);
692 		m->wire_count--;
693 	}
694 
695 	/*
696 	 * Free ptbl pages if there are no pte etries in this ptbl.
697 	 * wire_count has the same value for all ptbl pages, so check the last
698 	 * page.
699 	 */
700 	if (m->wire_count == 0) {
701 		ptbl_free(mmu, pmap, pdir_idx);
702 
703 		//debugf("ptbl_unhold: e (freed ptbl)\n");
704 		return (1);
705 	}
706 
707 	return (0);
708 }
709 
710 /*
711  * Increment hold count for ptbl pages. This routine is used when a new pte
712  * entry is being inserted into the ptbl.
713  */
714 static void
715 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
716 {
717 	vm_paddr_t pa;
718 	pte_t *ptbl;
719 	vm_page_t m;
720 	int i;
721 
722 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
723 	    pdir_idx);
724 
725 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
726 	    ("ptbl_hold: invalid pdir_idx"));
727 	KASSERT((pmap != kernel_pmap),
728 	    ("ptbl_hold: holding kernel ptbl!"));
729 
730 	ptbl = pmap->pm_pdir[pdir_idx];
731 
732 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
733 
734 	for (i = 0; i < PTBL_PAGES; i++) {
735 		pa = pte_vatopa(mmu, kernel_pmap,
736 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
737 		m = PHYS_TO_VM_PAGE(pa);
738 		m->wire_count++;
739 	}
740 }
741 
742 /* Allocate pv_entry structure. */
743 pv_entry_t
744 pv_alloc(void)
745 {
746 	pv_entry_t pv;
747 
748 	pv_entry_count++;
749 	if (pv_entry_count > pv_entry_high_water)
750 		pagedaemon_wakeup();
751 	pv = uma_zalloc(pvzone, M_NOWAIT);
752 
753 	return (pv);
754 }
755 
756 /* Free pv_entry structure. */
757 static __inline void
758 pv_free(pv_entry_t pve)
759 {
760 
761 	pv_entry_count--;
762 	uma_zfree(pvzone, pve);
763 }
764 
765 
766 /* Allocate and initialize pv_entry structure. */
767 static void
768 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
769 {
770 	pv_entry_t pve;
771 
772 	//int su = (pmap == kernel_pmap);
773 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
774 	//	(u_int32_t)pmap, va, (u_int32_t)m);
775 
776 	pve = pv_alloc();
777 	if (pve == NULL)
778 		panic("pv_insert: no pv entries!");
779 
780 	pve->pv_pmap = pmap;
781 	pve->pv_va = va;
782 
783 	/* add to pv_list */
784 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
785 	rw_assert(&pvh_global_lock, RA_WLOCKED);
786 
787 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
788 
789 	//debugf("pv_insert: e\n");
790 }
791 
792 /* Destroy pv entry. */
793 static void
794 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
795 {
796 	pv_entry_t pve;
797 
798 	//int su = (pmap == kernel_pmap);
799 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
800 
801 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
802 	rw_assert(&pvh_global_lock, RA_WLOCKED);
803 
804 	/* find pv entry */
805 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
806 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
807 			/* remove from pv_list */
808 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
809 			if (TAILQ_EMPTY(&m->md.pv_list))
810 				vm_page_aflag_clear(m, PGA_WRITEABLE);
811 
812 			/* free pv entry struct */
813 			pv_free(pve);
814 			break;
815 		}
816 	}
817 
818 	//debugf("pv_remove: e\n");
819 }
820 
821 /*
822  * Clean pte entry, try to free page table page if requested.
823  *
824  * Return 1 if ptbl pages were freed, otherwise return 0.
825  */
826 static int
827 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
828 {
829 	unsigned int pdir_idx = PDIR_IDX(va);
830 	unsigned int ptbl_idx = PTBL_IDX(va);
831 	vm_page_t m;
832 	pte_t *ptbl;
833 	pte_t *pte;
834 
835 	//int su = (pmap == kernel_pmap);
836 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
837 	//		su, (u_int32_t)pmap, va, flags);
838 
839 	ptbl = pmap->pm_pdir[pdir_idx];
840 	KASSERT(ptbl, ("pte_remove: null ptbl"));
841 
842 	pte = &ptbl[ptbl_idx];
843 
844 	if (pte == NULL || !PTE_ISVALID(pte))
845 		return (0);
846 
847 	if (PTE_ISWIRED(pte))
848 		pmap->pm_stats.wired_count--;
849 
850 	/* Handle managed entry. */
851 	if (PTE_ISMANAGED(pte)) {
852 		/* Get vm_page_t for mapped pte. */
853 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
854 
855 		if (PTE_ISMODIFIED(pte))
856 			vm_page_dirty(m);
857 
858 		if (PTE_ISREFERENCED(pte))
859 			vm_page_aflag_set(m, PGA_REFERENCED);
860 
861 		pv_remove(pmap, va, m);
862 	}
863 
864 	mtx_lock_spin(&tlbivax_mutex);
865 	tlb_miss_lock();
866 
867 	tlb0_flush_entry(va);
868 	pte->flags = 0;
869 	pte->rpn = 0;
870 
871 	tlb_miss_unlock();
872 	mtx_unlock_spin(&tlbivax_mutex);
873 
874 	pmap->pm_stats.resident_count--;
875 
876 	if (flags & PTBL_UNHOLD) {
877 		//debugf("pte_remove: e (unhold)\n");
878 		return (ptbl_unhold(mmu, pmap, pdir_idx));
879 	}
880 
881 	//debugf("pte_remove: e\n");
882 	return (0);
883 }
884 
885 /*
886  * Insert PTE for a given page and virtual address.
887  */
888 static void
889 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
890 {
891 	unsigned int pdir_idx = PDIR_IDX(va);
892 	unsigned int ptbl_idx = PTBL_IDX(va);
893 	pte_t *ptbl, *pte;
894 
895 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
896 	    pmap == kernel_pmap, pmap, va);
897 
898 	/* Get the page table pointer. */
899 	ptbl = pmap->pm_pdir[pdir_idx];
900 
901 	if (ptbl == NULL) {
902 		/* Allocate page table pages. */
903 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
904 	} else {
905 		/*
906 		 * Check if there is valid mapping for requested
907 		 * va, if there is, remove it.
908 		 */
909 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
910 		if (PTE_ISVALID(pte)) {
911 			pte_remove(mmu, pmap, va, PTBL_HOLD);
912 		} else {
913 			/*
914 			 * pte is not used, increment hold count
915 			 * for ptbl pages.
916 			 */
917 			if (pmap != kernel_pmap)
918 				ptbl_hold(mmu, pmap, pdir_idx);
919 		}
920 	}
921 
922 	/*
923 	 * Insert pv_entry into pv_list for mapped page if part of managed
924 	 * memory.
925 	 */
926 	if ((m->oflags & VPO_UNMANAGED) == 0) {
927 		flags |= PTE_MANAGED;
928 
929 		/* Create and insert pv entry. */
930 		pv_insert(pmap, va, m);
931 	}
932 
933 	pmap->pm_stats.resident_count++;
934 
935 	mtx_lock_spin(&tlbivax_mutex);
936 	tlb_miss_lock();
937 
938 	tlb0_flush_entry(va);
939 	if (pmap->pm_pdir[pdir_idx] == NULL) {
940 		/*
941 		 * If we just allocated a new page table, hook it in
942 		 * the pdir.
943 		 */
944 		pmap->pm_pdir[pdir_idx] = ptbl;
945 	}
946 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
947 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
948 	pte->flags |= (PTE_VALID | flags);
949 
950 	tlb_miss_unlock();
951 	mtx_unlock_spin(&tlbivax_mutex);
952 }
953 
954 /* Return the pa for the given pmap/va. */
955 static vm_paddr_t
956 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
957 {
958 	vm_paddr_t pa = 0;
959 	pte_t *pte;
960 
961 	pte = pte_find(mmu, pmap, va);
962 	if ((pte != NULL) && PTE_ISVALID(pte))
963 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
964 	return (pa);
965 }
966 
967 /* Get a pointer to a PTE in a page table. */
968 static pte_t *
969 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
970 {
971 	unsigned int pdir_idx = PDIR_IDX(va);
972 	unsigned int ptbl_idx = PTBL_IDX(va);
973 
974 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
975 
976 	if (pmap->pm_pdir[pdir_idx])
977 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
978 
979 	return (NULL);
980 }
981 
982 /**************************************************************************/
983 /* PMAP related */
984 /**************************************************************************/
985 
986 /*
987  * This is called during booke_init, before the system is really initialized.
988  */
989 static void
990 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
991 {
992 	vm_offset_t phys_kernelend;
993 	struct mem_region *mp, *mp1;
994 	int cnt, i, j;
995 	u_int s, e, sz;
996 	u_int phys_avail_count;
997 	vm_size_t physsz, hwphyssz, kstack0_sz;
998 	vm_offset_t kernel_pdir, kstack0, va;
999 	vm_paddr_t kstack0_phys;
1000 	void *dpcpu;
1001 	pte_t *pte;
1002 
1003 	debugf("mmu_booke_bootstrap: entered\n");
1004 
1005 	/* Initialize invalidation mutex */
1006 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1007 
1008 	/* Read TLB0 size and associativity. */
1009 	tlb0_get_tlbconf();
1010 
1011 	/*
1012 	 * Align kernel start and end address (kernel image).
1013 	 * Note that kernel end does not necessarily relate to kernsize.
1014 	 * kernsize is the size of the kernel that is actually mapped.
1015 	 * Also note that "start - 1" is deliberate. With SMP, the
1016 	 * entry point is exactly a page from the actual load address.
1017 	 * As such, trunc_page() has no effect and we're off by a page.
1018 	 * Since we always have the ELF header between the load address
1019 	 * and the entry point, we can safely subtract 1 to compensate.
1020 	 */
1021 	kernstart = trunc_page(start - 1);
1022 	data_start = round_page(kernelend);
1023 	data_end = data_start;
1024 
1025 	/*
1026 	 * Addresses of preloaded modules (like file systems) use
1027 	 * physical addresses. Make sure we relocate those into
1028 	 * virtual addresses.
1029 	 */
1030 	preload_addr_relocate = kernstart - kernload;
1031 
1032 	/* Allocate the dynamic per-cpu area. */
1033 	dpcpu = (void *)data_end;
1034 	data_end += DPCPU_SIZE;
1035 
1036 	/* Allocate space for the message buffer. */
1037 	msgbufp = (struct msgbuf *)data_end;
1038 	data_end += msgbufsize;
1039 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1040 	    data_end);
1041 
1042 	data_end = round_page(data_end);
1043 
1044 	/* Allocate space for ptbl_bufs. */
1045 	ptbl_bufs = (struct ptbl_buf *)data_end;
1046 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1047 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1048 	    data_end);
1049 
1050 	data_end = round_page(data_end);
1051 
1052 	/* Allocate PTE tables for kernel KVA. */
1053 	kernel_pdir = data_end;
1054 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1055 	    PDIR_SIZE - 1) / PDIR_SIZE;
1056 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1057 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1058 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1059 
1060 	debugf(" data_end: 0x%08x\n", data_end);
1061 	if (data_end - kernstart > kernsize) {
1062 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1063 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1064 	}
1065 	data_end = kernstart + kernsize;
1066 	debugf(" updated data_end: 0x%08x\n", data_end);
1067 
1068 	/*
1069 	 * Clear the structures - note we can only do it safely after the
1070 	 * possible additional TLB1 translations are in place (above) so that
1071 	 * all range up to the currently calculated 'data_end' is covered.
1072 	 */
1073 	dpcpu_init(dpcpu, 0);
1074 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1075 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1076 
1077 	/*******************************************************/
1078 	/* Set the start and end of kva. */
1079 	/*******************************************************/
1080 	virtual_avail = round_page(data_end);
1081 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1082 
1083 	/* Allocate KVA space for page zero/copy operations. */
1084 	zero_page_va = virtual_avail;
1085 	virtual_avail += PAGE_SIZE;
1086 	zero_page_idle_va = virtual_avail;
1087 	virtual_avail += PAGE_SIZE;
1088 	copy_page_src_va = virtual_avail;
1089 	virtual_avail += PAGE_SIZE;
1090 	copy_page_dst_va = virtual_avail;
1091 	virtual_avail += PAGE_SIZE;
1092 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1093 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1094 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1095 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1096 
1097 	/* Initialize page zero/copy mutexes. */
1098 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1099 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1100 
1101 	/* Allocate KVA space for ptbl bufs. */
1102 	ptbl_buf_pool_vabase = virtual_avail;
1103 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1104 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1105 	    ptbl_buf_pool_vabase, virtual_avail);
1106 
1107 	/* Calculate corresponding physical addresses for the kernel region. */
1108 	phys_kernelend = kernload + kernsize;
1109 	debugf("kernel image and allocated data:\n");
1110 	debugf(" kernload    = 0x%08x\n", kernload);
1111 	debugf(" kernstart   = 0x%08x\n", kernstart);
1112 	debugf(" kernsize    = 0x%08x\n", kernsize);
1113 
1114 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1115 		panic("mmu_booke_bootstrap: phys_avail too small");
1116 
1117 	/*
1118 	 * Remove kernel physical address range from avail regions list. Page
1119 	 * align all regions.  Non-page aligned memory isn't very interesting
1120 	 * to us.  Also, sort the entries for ascending addresses.
1121 	 */
1122 
1123 	/* Retrieve phys/avail mem regions */
1124 	mem_regions(&physmem_regions, &physmem_regions_sz,
1125 	    &availmem_regions, &availmem_regions_sz);
1126 	sz = 0;
1127 	cnt = availmem_regions_sz;
1128 	debugf("processing avail regions:\n");
1129 	for (mp = availmem_regions; mp->mr_size; mp++) {
1130 		s = mp->mr_start;
1131 		e = mp->mr_start + mp->mr_size;
1132 		debugf(" %08x-%08x -> ", s, e);
1133 		/* Check whether this region holds all of the kernel. */
1134 		if (s < kernload && e > phys_kernelend) {
1135 			availmem_regions[cnt].mr_start = phys_kernelend;
1136 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1137 			e = kernload;
1138 		}
1139 		/* Look whether this regions starts within the kernel. */
1140 		if (s >= kernload && s < phys_kernelend) {
1141 			if (e <= phys_kernelend)
1142 				goto empty;
1143 			s = phys_kernelend;
1144 		}
1145 		/* Now look whether this region ends within the kernel. */
1146 		if (e > kernload && e <= phys_kernelend) {
1147 			if (s >= kernload)
1148 				goto empty;
1149 			e = kernload;
1150 		}
1151 		/* Now page align the start and size of the region. */
1152 		s = round_page(s);
1153 		e = trunc_page(e);
1154 		if (e < s)
1155 			e = s;
1156 		sz = e - s;
1157 		debugf("%08x-%08x = %x\n", s, e, sz);
1158 
1159 		/* Check whether some memory is left here. */
1160 		if (sz == 0) {
1161 		empty:
1162 			memmove(mp, mp + 1,
1163 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1164 			cnt--;
1165 			mp--;
1166 			continue;
1167 		}
1168 
1169 		/* Do an insertion sort. */
1170 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1171 			if (s < mp1->mr_start)
1172 				break;
1173 		if (mp1 < mp) {
1174 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1175 			mp1->mr_start = s;
1176 			mp1->mr_size = sz;
1177 		} else {
1178 			mp->mr_start = s;
1179 			mp->mr_size = sz;
1180 		}
1181 	}
1182 	availmem_regions_sz = cnt;
1183 
1184 	/*******************************************************/
1185 	/* Steal physical memory for kernel stack from the end */
1186 	/* of the first avail region                           */
1187 	/*******************************************************/
1188 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1189 	kstack0_phys = availmem_regions[0].mr_start +
1190 	    availmem_regions[0].mr_size;
1191 	kstack0_phys -= kstack0_sz;
1192 	availmem_regions[0].mr_size -= kstack0_sz;
1193 
1194 	/*******************************************************/
1195 	/* Fill in phys_avail table, based on availmem_regions */
1196 	/*******************************************************/
1197 	phys_avail_count = 0;
1198 	physsz = 0;
1199 	hwphyssz = 0;
1200 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1201 
1202 	debugf("fill in phys_avail:\n");
1203 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1204 
1205 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1206 		    availmem_regions[i].mr_start,
1207 		    availmem_regions[i].mr_start +
1208 		        availmem_regions[i].mr_size,
1209 		    availmem_regions[i].mr_size);
1210 
1211 		if (hwphyssz != 0 &&
1212 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1213 			debugf(" hw.physmem adjust\n");
1214 			if (physsz < hwphyssz) {
1215 				phys_avail[j] = availmem_regions[i].mr_start;
1216 				phys_avail[j + 1] =
1217 				    availmem_regions[i].mr_start +
1218 				    hwphyssz - physsz;
1219 				physsz = hwphyssz;
1220 				phys_avail_count++;
1221 			}
1222 			break;
1223 		}
1224 
1225 		phys_avail[j] = availmem_regions[i].mr_start;
1226 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1227 		    availmem_regions[i].mr_size;
1228 		phys_avail_count++;
1229 		physsz += availmem_regions[i].mr_size;
1230 	}
1231 	physmem = btoc(physsz);
1232 
1233 	/* Calculate the last available physical address. */
1234 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1235 		;
1236 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1237 
1238 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1239 	debugf("phys_avail_count = %d\n", phys_avail_count);
1240 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1241 	    physmem);
1242 
1243 	/*******************************************************/
1244 	/* Initialize (statically allocated) kernel pmap. */
1245 	/*******************************************************/
1246 	PMAP_LOCK_INIT(kernel_pmap);
1247 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1248 
1249 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1250 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1251 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1252 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1253 
1254 	/* Initialize kernel pdir */
1255 	for (i = 0; i < kernel_ptbls; i++)
1256 		kernel_pmap->pm_pdir[kptbl_min + i] =
1257 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1258 
1259 	for (i = 0; i < MAXCPU; i++) {
1260 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1261 
1262 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1263 		tidbusy[i][0] = kernel_pmap;
1264 	}
1265 
1266 	/*
1267 	 * Fill in PTEs covering kernel code and data. They are not required
1268 	 * for address translation, as this area is covered by static TLB1
1269 	 * entries, but for pte_vatopa() to work correctly with kernel area
1270 	 * addresses.
1271 	 */
1272 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1273 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1274 		pte->rpn = kernload + (va - kernstart);
1275 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1276 		    PTE_VALID;
1277 	}
1278 	/* Mark kernel_pmap active on all CPUs */
1279 	CPU_FILL(&kernel_pmap->pm_active);
1280 
1281  	/*
1282 	 * Initialize the global pv list lock.
1283 	 */
1284 	rw_init(&pvh_global_lock, "pmap pv global");
1285 
1286 	/*******************************************************/
1287 	/* Final setup */
1288 	/*******************************************************/
1289 
1290 	/* Enter kstack0 into kernel map, provide guard page */
1291 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1292 	thread0.td_kstack = kstack0;
1293 	thread0.td_kstack_pages = KSTACK_PAGES;
1294 
1295 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1296 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1297 	    kstack0_phys, kstack0_phys + kstack0_sz);
1298 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1299 
1300 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1301 	for (i = 0; i < KSTACK_PAGES; i++) {
1302 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1303 		kstack0 += PAGE_SIZE;
1304 		kstack0_phys += PAGE_SIZE;
1305 	}
1306 
1307 	debugf("virtual_avail = %08x\n", virtual_avail);
1308 	debugf("virtual_end   = %08x\n", virtual_end);
1309 
1310 	debugf("mmu_booke_bootstrap: exit\n");
1311 }
1312 
1313 void
1314 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1315 {
1316 	int i;
1317 
1318 	/*
1319 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1320 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1321 	 * these values directly to (re)program AP's TLB1 hardware.
1322 	 */
1323 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1324 		/* Skip invalid entries */
1325 		if (!(tlb1[i].mas1 & MAS1_VALID))
1326 			continue;
1327 
1328 		tlb1_write_entry(i);
1329 	}
1330 
1331 	set_mas4_defaults();
1332 }
1333 
1334 /*
1335  * Get the physical page address for the given pmap/virtual address.
1336  */
1337 static vm_paddr_t
1338 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1339 {
1340 	vm_paddr_t pa;
1341 
1342 	PMAP_LOCK(pmap);
1343 	pa = pte_vatopa(mmu, pmap, va);
1344 	PMAP_UNLOCK(pmap);
1345 
1346 	return (pa);
1347 }
1348 
1349 /*
1350  * Extract the physical page address associated with the given
1351  * kernel virtual address.
1352  */
1353 static vm_paddr_t
1354 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1355 {
1356 	int i;
1357 
1358 	/* Check TLB1 mappings */
1359 	for (i = 0; i < tlb1_idx; i++) {
1360 		if (!(tlb1[i].mas1 & MAS1_VALID))
1361 			continue;
1362 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1363 			return (tlb1[i].phys + (va - tlb1[i].virt));
1364 	}
1365 
1366 	return (pte_vatopa(mmu, kernel_pmap, va));
1367 }
1368 
1369 /*
1370  * Initialize the pmap module.
1371  * Called by vm_init, to initialize any structures that the pmap
1372  * system needs to map virtual memory.
1373  */
1374 static void
1375 mmu_booke_init(mmu_t mmu)
1376 {
1377 	int shpgperproc = PMAP_SHPGPERPROC;
1378 
1379 	/*
1380 	 * Initialize the address space (zone) for the pv entries.  Set a
1381 	 * high water mark so that the system can recover from excessive
1382 	 * numbers of pv entries.
1383 	 */
1384 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1385 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1386 
1387 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1388 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1389 
1390 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1391 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1392 
1393 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1394 
1395 	/* Pre-fill pvzone with initial number of pv entries. */
1396 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1397 
1398 	/* Initialize ptbl allocation. */
1399 	ptbl_init();
1400 }
1401 
1402 /*
1403  * Map a list of wired pages into kernel virtual address space.  This is
1404  * intended for temporary mappings which do not need page modification or
1405  * references recorded.  Existing mappings in the region are overwritten.
1406  */
1407 static void
1408 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1409 {
1410 	vm_offset_t va;
1411 
1412 	va = sva;
1413 	while (count-- > 0) {
1414 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1415 		va += PAGE_SIZE;
1416 		m++;
1417 	}
1418 }
1419 
1420 /*
1421  * Remove page mappings from kernel virtual address space.  Intended for
1422  * temporary mappings entered by mmu_booke_qenter.
1423  */
1424 static void
1425 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1426 {
1427 	vm_offset_t va;
1428 
1429 	va = sva;
1430 	while (count-- > 0) {
1431 		mmu_booke_kremove(mmu, va);
1432 		va += PAGE_SIZE;
1433 	}
1434 }
1435 
1436 /*
1437  * Map a wired page into kernel virtual address space.
1438  */
1439 static void
1440 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1441 {
1442 
1443 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1444 }
1445 
1446 static void
1447 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1448 {
1449 	unsigned int pdir_idx = PDIR_IDX(va);
1450 	unsigned int ptbl_idx = PTBL_IDX(va);
1451 	uint32_t flags;
1452 	pte_t *pte;
1453 
1454 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1455 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1456 
1457 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1458 	flags |= tlb_calc_wimg(pa, ma);
1459 
1460 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1461 
1462 	mtx_lock_spin(&tlbivax_mutex);
1463 	tlb_miss_lock();
1464 
1465 	if (PTE_ISVALID(pte)) {
1466 
1467 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1468 
1469 		/* Flush entry from TLB0 */
1470 		tlb0_flush_entry(va);
1471 	}
1472 
1473 	pte->rpn = pa & ~PTE_PA_MASK;
1474 	pte->flags = flags;
1475 
1476 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1477 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1478 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1479 
1480 	/* Flush the real memory from the instruction cache. */
1481 	if ((flags & (PTE_I | PTE_G)) == 0) {
1482 		__syncicache((void *)va, PAGE_SIZE);
1483 	}
1484 
1485 	tlb_miss_unlock();
1486 	mtx_unlock_spin(&tlbivax_mutex);
1487 }
1488 
1489 /*
1490  * Remove a page from kernel page table.
1491  */
1492 static void
1493 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1494 {
1495 	unsigned int pdir_idx = PDIR_IDX(va);
1496 	unsigned int ptbl_idx = PTBL_IDX(va);
1497 	pte_t *pte;
1498 
1499 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1500 
1501 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1502 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1503 	    ("mmu_booke_kremove: invalid va"));
1504 
1505 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1506 
1507 	if (!PTE_ISVALID(pte)) {
1508 
1509 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1510 
1511 		return;
1512 	}
1513 
1514 	mtx_lock_spin(&tlbivax_mutex);
1515 	tlb_miss_lock();
1516 
1517 	/* Invalidate entry in TLB0, update PTE. */
1518 	tlb0_flush_entry(va);
1519 	pte->flags = 0;
1520 	pte->rpn = 0;
1521 
1522 	tlb_miss_unlock();
1523 	mtx_unlock_spin(&tlbivax_mutex);
1524 }
1525 
1526 /*
1527  * Initialize pmap associated with process 0.
1528  */
1529 static void
1530 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1531 {
1532 
1533 	PMAP_LOCK_INIT(pmap);
1534 	mmu_booke_pinit(mmu, pmap);
1535 	PCPU_SET(curpmap, pmap);
1536 }
1537 
1538 /*
1539  * Initialize a preallocated and zeroed pmap structure,
1540  * such as one in a vmspace structure.
1541  */
1542 static void
1543 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1544 {
1545 	int i;
1546 
1547 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1548 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1549 
1550 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1551 
1552 	for (i = 0; i < MAXCPU; i++)
1553 		pmap->pm_tid[i] = TID_NONE;
1554 	CPU_ZERO(&kernel_pmap->pm_active);
1555 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1556 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1557 	TAILQ_INIT(&pmap->pm_ptbl_list);
1558 }
1559 
1560 /*
1561  * Release any resources held by the given physical map.
1562  * Called when a pmap initialized by mmu_booke_pinit is being released.
1563  * Should only be called if the map contains no valid mappings.
1564  */
1565 static void
1566 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1567 {
1568 
1569 	KASSERT(pmap->pm_stats.resident_count == 0,
1570 	    ("pmap_release: pmap resident count %ld != 0",
1571 	    pmap->pm_stats.resident_count));
1572 }
1573 
1574 /*
1575  * Insert the given physical page at the specified virtual address in the
1576  * target physical map with the protection requested. If specified the page
1577  * will be wired down.
1578  */
1579 static void
1580 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1581     vm_prot_t prot, boolean_t wired)
1582 {
1583 
1584 	rw_wlock(&pvh_global_lock);
1585 	PMAP_LOCK(pmap);
1586 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1587 	rw_wunlock(&pvh_global_lock);
1588 	PMAP_UNLOCK(pmap);
1589 }
1590 
1591 static void
1592 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1593     vm_prot_t prot, boolean_t wired)
1594 {
1595 	pte_t *pte;
1596 	vm_paddr_t pa;
1597 	uint32_t flags;
1598 	int su, sync;
1599 
1600 	pa = VM_PAGE_TO_PHYS(m);
1601 	su = (pmap == kernel_pmap);
1602 	sync = 0;
1603 
1604 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1605 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1606 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1607 	//		(u_int32_t)m, va, pa, prot, wired);
1608 
1609 	if (su) {
1610 		KASSERT(((va >= virtual_avail) &&
1611 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1612 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1613 	} else {
1614 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1615 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1616 	}
1617 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1618 		VM_OBJECT_ASSERT_LOCKED(m->object);
1619 
1620 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1621 
1622 	/*
1623 	 * If there is an existing mapping, and the physical address has not
1624 	 * changed, must be protection or wiring change.
1625 	 */
1626 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1627 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1628 
1629 		/*
1630 		 * Before actually updating pte->flags we calculate and
1631 		 * prepare its new value in a helper var.
1632 		 */
1633 		flags = pte->flags;
1634 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1635 
1636 		/* Wiring change, just update stats. */
1637 		if (wired) {
1638 			if (!PTE_ISWIRED(pte)) {
1639 				flags |= PTE_WIRED;
1640 				pmap->pm_stats.wired_count++;
1641 			}
1642 		} else {
1643 			if (PTE_ISWIRED(pte)) {
1644 				flags &= ~PTE_WIRED;
1645 				pmap->pm_stats.wired_count--;
1646 			}
1647 		}
1648 
1649 		if (prot & VM_PROT_WRITE) {
1650 			/* Add write permissions. */
1651 			flags |= PTE_SW;
1652 			if (!su)
1653 				flags |= PTE_UW;
1654 
1655 			if ((flags & PTE_MANAGED) != 0)
1656 				vm_page_aflag_set(m, PGA_WRITEABLE);
1657 		} else {
1658 			/* Handle modified pages, sense modify status. */
1659 
1660 			/*
1661 			 * The PTE_MODIFIED flag could be set by underlying
1662 			 * TLB misses since we last read it (above), possibly
1663 			 * other CPUs could update it so we check in the PTE
1664 			 * directly rather than rely on that saved local flags
1665 			 * copy.
1666 			 */
1667 			if (PTE_ISMODIFIED(pte))
1668 				vm_page_dirty(m);
1669 		}
1670 
1671 		if (prot & VM_PROT_EXECUTE) {
1672 			flags |= PTE_SX;
1673 			if (!su)
1674 				flags |= PTE_UX;
1675 
1676 			/*
1677 			 * Check existing flags for execute permissions: if we
1678 			 * are turning execute permissions on, icache should
1679 			 * be flushed.
1680 			 */
1681 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1682 				sync++;
1683 		}
1684 
1685 		flags &= ~PTE_REFERENCED;
1686 
1687 		/*
1688 		 * The new flags value is all calculated -- only now actually
1689 		 * update the PTE.
1690 		 */
1691 		mtx_lock_spin(&tlbivax_mutex);
1692 		tlb_miss_lock();
1693 
1694 		tlb0_flush_entry(va);
1695 		pte->flags = flags;
1696 
1697 		tlb_miss_unlock();
1698 		mtx_unlock_spin(&tlbivax_mutex);
1699 
1700 	} else {
1701 		/*
1702 		 * If there is an existing mapping, but it's for a different
1703 		 * physical address, pte_enter() will delete the old mapping.
1704 		 */
1705 		//if ((pte != NULL) && PTE_ISVALID(pte))
1706 		//	debugf("mmu_booke_enter_locked: replace\n");
1707 		//else
1708 		//	debugf("mmu_booke_enter_locked: new\n");
1709 
1710 		/* Now set up the flags and install the new mapping. */
1711 		flags = (PTE_SR | PTE_VALID);
1712 		flags |= PTE_M;
1713 
1714 		if (!su)
1715 			flags |= PTE_UR;
1716 
1717 		if (prot & VM_PROT_WRITE) {
1718 			flags |= PTE_SW;
1719 			if (!su)
1720 				flags |= PTE_UW;
1721 
1722 			if ((m->oflags & VPO_UNMANAGED) == 0)
1723 				vm_page_aflag_set(m, PGA_WRITEABLE);
1724 		}
1725 
1726 		if (prot & VM_PROT_EXECUTE) {
1727 			flags |= PTE_SX;
1728 			if (!su)
1729 				flags |= PTE_UX;
1730 		}
1731 
1732 		/* If its wired update stats. */
1733 		if (wired) {
1734 			pmap->pm_stats.wired_count++;
1735 			flags |= PTE_WIRED;
1736 		}
1737 
1738 		pte_enter(mmu, pmap, m, va, flags);
1739 
1740 		/* Flush the real memory from the instruction cache. */
1741 		if (prot & VM_PROT_EXECUTE)
1742 			sync++;
1743 	}
1744 
1745 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1746 		__syncicache((void *)va, PAGE_SIZE);
1747 		sync = 0;
1748 	}
1749 }
1750 
1751 /*
1752  * Maps a sequence of resident pages belonging to the same object.
1753  * The sequence begins with the given page m_start.  This page is
1754  * mapped at the given virtual address start.  Each subsequent page is
1755  * mapped at a virtual address that is offset from start by the same
1756  * amount as the page is offset from m_start within the object.  The
1757  * last page in the sequence is the page with the largest offset from
1758  * m_start that can be mapped at a virtual address less than the given
1759  * virtual address end.  Not every virtual page between start and end
1760  * is mapped; only those for which a resident page exists with the
1761  * corresponding offset from m_start are mapped.
1762  */
1763 static void
1764 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1765     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1766 {
1767 	vm_page_t m;
1768 	vm_pindex_t diff, psize;
1769 
1770 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1771 
1772 	psize = atop(end - start);
1773 	m = m_start;
1774 	rw_wlock(&pvh_global_lock);
1775 	PMAP_LOCK(pmap);
1776 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1777 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1778 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1779 		m = TAILQ_NEXT(m, listq);
1780 	}
1781 	rw_wunlock(&pvh_global_lock);
1782 	PMAP_UNLOCK(pmap);
1783 }
1784 
1785 static void
1786 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1787     vm_prot_t prot)
1788 {
1789 
1790 	rw_wlock(&pvh_global_lock);
1791 	PMAP_LOCK(pmap);
1792 	mmu_booke_enter_locked(mmu, pmap, va, m,
1793 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1794 	rw_wunlock(&pvh_global_lock);
1795 	PMAP_UNLOCK(pmap);
1796 }
1797 
1798 /*
1799  * Remove the given range of addresses from the specified map.
1800  *
1801  * It is assumed that the start and end are properly rounded to the page size.
1802  */
1803 static void
1804 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1805 {
1806 	pte_t *pte;
1807 	uint8_t hold_flag;
1808 
1809 	int su = (pmap == kernel_pmap);
1810 
1811 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1812 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1813 
1814 	if (su) {
1815 		KASSERT(((va >= virtual_avail) &&
1816 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1817 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1818 	} else {
1819 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1820 		    ("mmu_booke_remove: user pmap, non user va"));
1821 	}
1822 
1823 	if (PMAP_REMOVE_DONE(pmap)) {
1824 		//debugf("mmu_booke_remove: e (empty)\n");
1825 		return;
1826 	}
1827 
1828 	hold_flag = PTBL_HOLD_FLAG(pmap);
1829 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1830 
1831 	rw_wlock(&pvh_global_lock);
1832 	PMAP_LOCK(pmap);
1833 	for (; va < endva; va += PAGE_SIZE) {
1834 		pte = pte_find(mmu, pmap, va);
1835 		if ((pte != NULL) && PTE_ISVALID(pte))
1836 			pte_remove(mmu, pmap, va, hold_flag);
1837 	}
1838 	PMAP_UNLOCK(pmap);
1839 	rw_wunlock(&pvh_global_lock);
1840 
1841 	//debugf("mmu_booke_remove: e\n");
1842 }
1843 
1844 /*
1845  * Remove physical page from all pmaps in which it resides.
1846  */
1847 static void
1848 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1849 {
1850 	pv_entry_t pv, pvn;
1851 	uint8_t hold_flag;
1852 
1853 	rw_wlock(&pvh_global_lock);
1854 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1855 		pvn = TAILQ_NEXT(pv, pv_link);
1856 
1857 		PMAP_LOCK(pv->pv_pmap);
1858 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1859 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1860 		PMAP_UNLOCK(pv->pv_pmap);
1861 	}
1862 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1863 	rw_wunlock(&pvh_global_lock);
1864 }
1865 
1866 /*
1867  * Map a range of physical addresses into kernel virtual address space.
1868  */
1869 static vm_offset_t
1870 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1871     vm_paddr_t pa_end, int prot)
1872 {
1873 	vm_offset_t sva = *virt;
1874 	vm_offset_t va = sva;
1875 
1876 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1877 	//		sva, pa_start, pa_end);
1878 
1879 	while (pa_start < pa_end) {
1880 		mmu_booke_kenter(mmu, va, pa_start);
1881 		va += PAGE_SIZE;
1882 		pa_start += PAGE_SIZE;
1883 	}
1884 	*virt = va;
1885 
1886 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1887 	return (sva);
1888 }
1889 
1890 /*
1891  * The pmap must be activated before it's address space can be accessed in any
1892  * way.
1893  */
1894 static void
1895 mmu_booke_activate(mmu_t mmu, struct thread *td)
1896 {
1897 	pmap_t pmap;
1898 	u_int cpuid;
1899 
1900 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1901 
1902 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1903 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1904 
1905 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1906 
1907 	sched_pin();
1908 
1909 	cpuid = PCPU_GET(cpuid);
1910 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1911 	PCPU_SET(curpmap, pmap);
1912 
1913 	if (pmap->pm_tid[cpuid] == TID_NONE)
1914 		tid_alloc(pmap);
1915 
1916 	/* Load PID0 register with pmap tid value. */
1917 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1918 	__asm __volatile("isync");
1919 
1920 	sched_unpin();
1921 
1922 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1923 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1924 }
1925 
1926 /*
1927  * Deactivate the specified process's address space.
1928  */
1929 static void
1930 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1931 {
1932 	pmap_t pmap;
1933 
1934 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1935 
1936 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1937 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1938 
1939 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1940 	PCPU_SET(curpmap, NULL);
1941 }
1942 
1943 /*
1944  * Copy the range specified by src_addr/len
1945  * from the source map to the range dst_addr/len
1946  * in the destination map.
1947  *
1948  * This routine is only advisory and need not do anything.
1949  */
1950 static void
1951 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1952     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1953 {
1954 
1955 }
1956 
1957 /*
1958  * Set the physical protection on the specified range of this map as requested.
1959  */
1960 static void
1961 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1962     vm_prot_t prot)
1963 {
1964 	vm_offset_t va;
1965 	vm_page_t m;
1966 	pte_t *pte;
1967 
1968 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1969 		mmu_booke_remove(mmu, pmap, sva, eva);
1970 		return;
1971 	}
1972 
1973 	if (prot & VM_PROT_WRITE)
1974 		return;
1975 
1976 	PMAP_LOCK(pmap);
1977 	for (va = sva; va < eva; va += PAGE_SIZE) {
1978 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1979 			if (PTE_ISVALID(pte)) {
1980 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1981 
1982 				mtx_lock_spin(&tlbivax_mutex);
1983 				tlb_miss_lock();
1984 
1985 				/* Handle modified pages. */
1986 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1987 					vm_page_dirty(m);
1988 
1989 				tlb0_flush_entry(va);
1990 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1991 
1992 				tlb_miss_unlock();
1993 				mtx_unlock_spin(&tlbivax_mutex);
1994 			}
1995 		}
1996 	}
1997 	PMAP_UNLOCK(pmap);
1998 }
1999 
2000 /*
2001  * Clear the write and modified bits in each of the given page's mappings.
2002  */
2003 static void
2004 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2005 {
2006 	pv_entry_t pv;
2007 	pte_t *pte;
2008 
2009 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2010 	    ("mmu_booke_remove_write: page %p is not managed", m));
2011 
2012 	/*
2013 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2014 	 * set by another thread while the object is locked.  Thus,
2015 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2016 	 */
2017 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2018 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2019 		return;
2020 	rw_wlock(&pvh_global_lock);
2021 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2022 		PMAP_LOCK(pv->pv_pmap);
2023 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2024 			if (PTE_ISVALID(pte)) {
2025 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2026 
2027 				mtx_lock_spin(&tlbivax_mutex);
2028 				tlb_miss_lock();
2029 
2030 				/* Handle modified pages. */
2031 				if (PTE_ISMODIFIED(pte))
2032 					vm_page_dirty(m);
2033 
2034 				/* Flush mapping from TLB0. */
2035 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2036 
2037 				tlb_miss_unlock();
2038 				mtx_unlock_spin(&tlbivax_mutex);
2039 			}
2040 		}
2041 		PMAP_UNLOCK(pv->pv_pmap);
2042 	}
2043 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2044 	rw_wunlock(&pvh_global_lock);
2045 }
2046 
2047 static void
2048 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2049 {
2050 	pte_t *pte;
2051 	pmap_t pmap;
2052 	vm_page_t m;
2053 	vm_offset_t addr;
2054 	vm_paddr_t pa;
2055 	int active, valid;
2056 
2057 	va = trunc_page(va);
2058 	sz = round_page(sz);
2059 
2060 	rw_wlock(&pvh_global_lock);
2061 	pmap = PCPU_GET(curpmap);
2062 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2063 	while (sz > 0) {
2064 		PMAP_LOCK(pm);
2065 		pte = pte_find(mmu, pm, va);
2066 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2067 		if (valid)
2068 			pa = PTE_PA(pte);
2069 		PMAP_UNLOCK(pm);
2070 		if (valid) {
2071 			if (!active) {
2072 				/* Create a mapping in the active pmap. */
2073 				addr = 0;
2074 				m = PHYS_TO_VM_PAGE(pa);
2075 				PMAP_LOCK(pmap);
2076 				pte_enter(mmu, pmap, m, addr,
2077 				    PTE_SR | PTE_VALID | PTE_UR);
2078 				__syncicache((void *)addr, PAGE_SIZE);
2079 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2080 				PMAP_UNLOCK(pmap);
2081 			} else
2082 				__syncicache((void *)va, PAGE_SIZE);
2083 		}
2084 		va += PAGE_SIZE;
2085 		sz -= PAGE_SIZE;
2086 	}
2087 	rw_wunlock(&pvh_global_lock);
2088 }
2089 
2090 /*
2091  * Atomically extract and hold the physical page with the given
2092  * pmap and virtual address pair if that mapping permits the given
2093  * protection.
2094  */
2095 static vm_page_t
2096 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2097     vm_prot_t prot)
2098 {
2099 	pte_t *pte;
2100 	vm_page_t m;
2101 	uint32_t pte_wbit;
2102 	vm_paddr_t pa;
2103 
2104 	m = NULL;
2105 	pa = 0;
2106 	PMAP_LOCK(pmap);
2107 retry:
2108 	pte = pte_find(mmu, pmap, va);
2109 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2110 		if (pmap == kernel_pmap)
2111 			pte_wbit = PTE_SW;
2112 		else
2113 			pte_wbit = PTE_UW;
2114 
2115 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2116 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2117 				goto retry;
2118 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2119 			vm_page_hold(m);
2120 		}
2121 	}
2122 
2123 	PA_UNLOCK_COND(pa);
2124 	PMAP_UNLOCK(pmap);
2125 	return (m);
2126 }
2127 
2128 /*
2129  * Initialize a vm_page's machine-dependent fields.
2130  */
2131 static void
2132 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2133 {
2134 
2135 	TAILQ_INIT(&m->md.pv_list);
2136 }
2137 
2138 /*
2139  * mmu_booke_zero_page_area zeros the specified hardware page by
2140  * mapping it into virtual memory and using bzero to clear
2141  * its contents.
2142  *
2143  * off and size must reside within a single page.
2144  */
2145 static void
2146 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2147 {
2148 	vm_offset_t va;
2149 
2150 	/* XXX KASSERT off and size are within a single page? */
2151 
2152 	mtx_lock(&zero_page_mutex);
2153 	va = zero_page_va;
2154 
2155 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2156 	bzero((caddr_t)va + off, size);
2157 	mmu_booke_kremove(mmu, va);
2158 
2159 	mtx_unlock(&zero_page_mutex);
2160 }
2161 
2162 /*
2163  * mmu_booke_zero_page zeros the specified hardware page.
2164  */
2165 static void
2166 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2167 {
2168 
2169 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2170 }
2171 
2172 /*
2173  * mmu_booke_copy_page copies the specified (machine independent) page by
2174  * mapping the page into virtual memory and using memcopy to copy the page,
2175  * one machine dependent page at a time.
2176  */
2177 static void
2178 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2179 {
2180 	vm_offset_t sva, dva;
2181 
2182 	sva = copy_page_src_va;
2183 	dva = copy_page_dst_va;
2184 
2185 	mtx_lock(&copy_page_mutex);
2186 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2187 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2188 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2189 	mmu_booke_kremove(mmu, dva);
2190 	mmu_booke_kremove(mmu, sva);
2191 	mtx_unlock(&copy_page_mutex);
2192 }
2193 
2194 static inline void
2195 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2196     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2197 {
2198 	void *a_cp, *b_cp;
2199 	vm_offset_t a_pg_offset, b_pg_offset;
2200 	int cnt;
2201 
2202 	mtx_lock(&copy_page_mutex);
2203 	while (xfersize > 0) {
2204 		a_pg_offset = a_offset & PAGE_MASK;
2205 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2206 		mmu_booke_kenter(mmu, copy_page_src_va,
2207 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2208 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2209 		b_pg_offset = b_offset & PAGE_MASK;
2210 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2211 		mmu_booke_kenter(mmu, copy_page_dst_va,
2212 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2213 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2214 		bcopy(a_cp, b_cp, cnt);
2215 		mmu_booke_kremove(mmu, copy_page_dst_va);
2216 		mmu_booke_kremove(mmu, copy_page_src_va);
2217 		a_offset += cnt;
2218 		b_offset += cnt;
2219 		xfersize -= cnt;
2220 	}
2221 	mtx_unlock(&copy_page_mutex);
2222 }
2223 
2224 /*
2225  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2226  * into virtual memory and using bzero to clear its contents. This is intended
2227  * to be called from the vm_pagezero process only and outside of Giant. No
2228  * lock is required.
2229  */
2230 static void
2231 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2232 {
2233 	vm_offset_t va;
2234 
2235 	va = zero_page_idle_va;
2236 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2237 	bzero((caddr_t)va, PAGE_SIZE);
2238 	mmu_booke_kremove(mmu, va);
2239 }
2240 
2241 /*
2242  * Return whether or not the specified physical page was modified
2243  * in any of physical maps.
2244  */
2245 static boolean_t
2246 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2247 {
2248 	pte_t *pte;
2249 	pv_entry_t pv;
2250 	boolean_t rv;
2251 
2252 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2253 	    ("mmu_booke_is_modified: page %p is not managed", m));
2254 	rv = FALSE;
2255 
2256 	/*
2257 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2258 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2259 	 * is clear, no PTEs can be modified.
2260 	 */
2261 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2262 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2263 		return (rv);
2264 	rw_wlock(&pvh_global_lock);
2265 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2266 		PMAP_LOCK(pv->pv_pmap);
2267 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2268 		    PTE_ISVALID(pte)) {
2269 			if (PTE_ISMODIFIED(pte))
2270 				rv = TRUE;
2271 		}
2272 		PMAP_UNLOCK(pv->pv_pmap);
2273 		if (rv)
2274 			break;
2275 	}
2276 	rw_wunlock(&pvh_global_lock);
2277 	return (rv);
2278 }
2279 
2280 /*
2281  * Return whether or not the specified virtual address is eligible
2282  * for prefault.
2283  */
2284 static boolean_t
2285 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2286 {
2287 
2288 	return (FALSE);
2289 }
2290 
2291 /*
2292  * Return whether or not the specified physical page was referenced
2293  * in any physical maps.
2294  */
2295 static boolean_t
2296 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2297 {
2298 	pte_t *pte;
2299 	pv_entry_t pv;
2300 	boolean_t rv;
2301 
2302 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2303 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2304 	rv = FALSE;
2305 	rw_wlock(&pvh_global_lock);
2306 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2307 		PMAP_LOCK(pv->pv_pmap);
2308 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2309 		    PTE_ISVALID(pte)) {
2310 			if (PTE_ISREFERENCED(pte))
2311 				rv = TRUE;
2312 		}
2313 		PMAP_UNLOCK(pv->pv_pmap);
2314 		if (rv)
2315 			break;
2316 	}
2317 	rw_wunlock(&pvh_global_lock);
2318 	return (rv);
2319 }
2320 
2321 /*
2322  * Clear the modify bits on the specified physical page.
2323  */
2324 static void
2325 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2326 {
2327 	pte_t *pte;
2328 	pv_entry_t pv;
2329 
2330 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2331 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2332 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2333 	KASSERT(!vm_page_xbusied(m),
2334 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2335 
2336 	/*
2337 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2338 	 * If the object containing the page is locked and the page is not
2339 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2340 	 */
2341 	if ((m->aflags & PGA_WRITEABLE) == 0)
2342 		return;
2343 	rw_wlock(&pvh_global_lock);
2344 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2345 		PMAP_LOCK(pv->pv_pmap);
2346 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2347 		    PTE_ISVALID(pte)) {
2348 			mtx_lock_spin(&tlbivax_mutex);
2349 			tlb_miss_lock();
2350 
2351 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2352 				tlb0_flush_entry(pv->pv_va);
2353 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2354 				    PTE_REFERENCED);
2355 			}
2356 
2357 			tlb_miss_unlock();
2358 			mtx_unlock_spin(&tlbivax_mutex);
2359 		}
2360 		PMAP_UNLOCK(pv->pv_pmap);
2361 	}
2362 	rw_wunlock(&pvh_global_lock);
2363 }
2364 
2365 /*
2366  * Return a count of reference bits for a page, clearing those bits.
2367  * It is not necessary for every reference bit to be cleared, but it
2368  * is necessary that 0 only be returned when there are truly no
2369  * reference bits set.
2370  *
2371  * XXX: The exact number of bits to check and clear is a matter that
2372  * should be tested and standardized at some point in the future for
2373  * optimal aging of shared pages.
2374  */
2375 static int
2376 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2377 {
2378 	pte_t *pte;
2379 	pv_entry_t pv;
2380 	int count;
2381 
2382 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2383 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2384 	count = 0;
2385 	rw_wlock(&pvh_global_lock);
2386 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2387 		PMAP_LOCK(pv->pv_pmap);
2388 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2389 		    PTE_ISVALID(pte)) {
2390 			if (PTE_ISREFERENCED(pte)) {
2391 				mtx_lock_spin(&tlbivax_mutex);
2392 				tlb_miss_lock();
2393 
2394 				tlb0_flush_entry(pv->pv_va);
2395 				pte->flags &= ~PTE_REFERENCED;
2396 
2397 				tlb_miss_unlock();
2398 				mtx_unlock_spin(&tlbivax_mutex);
2399 
2400 				if (++count > 4) {
2401 					PMAP_UNLOCK(pv->pv_pmap);
2402 					break;
2403 				}
2404 			}
2405 		}
2406 		PMAP_UNLOCK(pv->pv_pmap);
2407 	}
2408 	rw_wunlock(&pvh_global_lock);
2409 	return (count);
2410 }
2411 
2412 /*
2413  * Change wiring attribute for a map/virtual-address pair.
2414  */
2415 static void
2416 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2417 {
2418 	pte_t *pte;
2419 
2420 	PMAP_LOCK(pmap);
2421 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2422 		if (wired) {
2423 			if (!PTE_ISWIRED(pte)) {
2424 				pte->flags |= PTE_WIRED;
2425 				pmap->pm_stats.wired_count++;
2426 			}
2427 		} else {
2428 			if (PTE_ISWIRED(pte)) {
2429 				pte->flags &= ~PTE_WIRED;
2430 				pmap->pm_stats.wired_count--;
2431 			}
2432 		}
2433 	}
2434 	PMAP_UNLOCK(pmap);
2435 }
2436 
2437 /*
2438  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2439  * page.  This count may be changed upwards or downwards in the future; it is
2440  * only necessary that true be returned for a small subset of pmaps for proper
2441  * page aging.
2442  */
2443 static boolean_t
2444 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2445 {
2446 	pv_entry_t pv;
2447 	int loops;
2448 	boolean_t rv;
2449 
2450 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2451 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2452 	loops = 0;
2453 	rv = FALSE;
2454 	rw_wlock(&pvh_global_lock);
2455 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2456 		if (pv->pv_pmap == pmap) {
2457 			rv = TRUE;
2458 			break;
2459 		}
2460 		if (++loops >= 16)
2461 			break;
2462 	}
2463 	rw_wunlock(&pvh_global_lock);
2464 	return (rv);
2465 }
2466 
2467 /*
2468  * Return the number of managed mappings to the given physical page that are
2469  * wired.
2470  */
2471 static int
2472 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2473 {
2474 	pv_entry_t pv;
2475 	pte_t *pte;
2476 	int count = 0;
2477 
2478 	if ((m->oflags & VPO_UNMANAGED) != 0)
2479 		return (count);
2480 	rw_wlock(&pvh_global_lock);
2481 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2482 		PMAP_LOCK(pv->pv_pmap);
2483 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2484 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2485 				count++;
2486 		PMAP_UNLOCK(pv->pv_pmap);
2487 	}
2488 	rw_wunlock(&pvh_global_lock);
2489 	return (count);
2490 }
2491 
2492 static int
2493 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2494 {
2495 	int i;
2496 	vm_offset_t va;
2497 
2498 	/*
2499 	 * This currently does not work for entries that
2500 	 * overlap TLB1 entries.
2501 	 */
2502 	for (i = 0; i < tlb1_idx; i ++) {
2503 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2504 			return (0);
2505 	}
2506 
2507 	return (EFAULT);
2508 }
2509 
2510 vm_offset_t
2511 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2512     vm_size_t *sz)
2513 {
2514 	vm_paddr_t pa, ppa;
2515 	vm_offset_t va;
2516 	vm_size_t gran;
2517 
2518 	/* Raw physical memory dumps don't have a virtual address. */
2519 	if (md->md_vaddr == ~0UL) {
2520 		/* We always map a 256MB page at 256M. */
2521 		gran = 256 * 1024 * 1024;
2522 		pa = md->md_paddr + ofs;
2523 		ppa = pa & ~(gran - 1);
2524 		ofs = pa - ppa;
2525 		va = gran;
2526 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2527 		if (*sz > (gran - ofs))
2528 			*sz = gran - ofs;
2529 		return (va + ofs);
2530 	}
2531 
2532 	/* Minidumps are based on virtual memory addresses. */
2533 	va = md->md_vaddr + ofs;
2534 	if (va >= kernstart + kernsize) {
2535 		gran = PAGE_SIZE - (va & PAGE_MASK);
2536 		if (*sz > gran)
2537 			*sz = gran;
2538 	}
2539 	return (va);
2540 }
2541 
2542 void
2543 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2544     vm_offset_t va)
2545 {
2546 
2547 	/* Raw physical memory dumps don't have a virtual address. */
2548 	if (md->md_vaddr == ~0UL) {
2549 		tlb1_idx--;
2550 		tlb1[tlb1_idx].mas1 = 0;
2551 		tlb1[tlb1_idx].mas2 = 0;
2552 		tlb1[tlb1_idx].mas3 = 0;
2553 		tlb1_write_entry(tlb1_idx);
2554 		return;
2555 	}
2556 
2557 	/* Minidumps are based on virtual memory addresses. */
2558 	/* Nothing to do... */
2559 }
2560 
2561 struct pmap_md *
2562 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2563 {
2564 	static struct pmap_md md;
2565 	pte_t *pte;
2566 	vm_offset_t va;
2567 
2568 	if (dumpsys_minidump) {
2569 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2570 		if (prev == NULL) {
2571 			/* 1st: kernel .data and .bss. */
2572 			md.md_index = 1;
2573 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2574 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2575 			return (&md);
2576 		}
2577 		switch (prev->md_index) {
2578 		case 1:
2579 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2580 			md.md_index = 2;
2581 			md.md_vaddr = data_start;
2582 			md.md_size = data_end - data_start;
2583 			break;
2584 		case 2:
2585 			/* 3rd: kernel VM. */
2586 			va = prev->md_vaddr + prev->md_size;
2587 			/* Find start of next chunk (from va). */
2588 			while (va < virtual_end) {
2589 				/* Don't dump the buffer cache. */
2590 				if (va >= kmi.buffer_sva &&
2591 				    va < kmi.buffer_eva) {
2592 					va = kmi.buffer_eva;
2593 					continue;
2594 				}
2595 				pte = pte_find(mmu, kernel_pmap, va);
2596 				if (pte != NULL && PTE_ISVALID(pte))
2597 					break;
2598 				va += PAGE_SIZE;
2599 			}
2600 			if (va < virtual_end) {
2601 				md.md_vaddr = va;
2602 				va += PAGE_SIZE;
2603 				/* Find last page in chunk. */
2604 				while (va < virtual_end) {
2605 					/* Don't run into the buffer cache. */
2606 					if (va == kmi.buffer_sva)
2607 						break;
2608 					pte = pte_find(mmu, kernel_pmap, va);
2609 					if (pte == NULL || !PTE_ISVALID(pte))
2610 						break;
2611 					va += PAGE_SIZE;
2612 				}
2613 				md.md_size = va - md.md_vaddr;
2614 				break;
2615 			}
2616 			md.md_index = 3;
2617 			/* FALLTHROUGH */
2618 		default:
2619 			return (NULL);
2620 		}
2621 	} else { /* minidumps */
2622 		mem_regions(&physmem_regions, &physmem_regions_sz,
2623 		    &availmem_regions, &availmem_regions_sz);
2624 
2625 		if (prev == NULL) {
2626 			/* first physical chunk. */
2627 			md.md_paddr = physmem_regions[0].mr_start;
2628 			md.md_size = physmem_regions[0].mr_size;
2629 			md.md_vaddr = ~0UL;
2630 			md.md_index = 1;
2631 		} else if (md.md_index < physmem_regions_sz) {
2632 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2633 			md.md_size = physmem_regions[md.md_index].mr_size;
2634 			md.md_vaddr = ~0UL;
2635 			md.md_index++;
2636 		} else {
2637 			/* There's no next physical chunk. */
2638 			return (NULL);
2639 		}
2640 	}
2641 
2642 	return (&md);
2643 }
2644 
2645 /*
2646  * Map a set of physical memory pages into the kernel virtual address space.
2647  * Return a pointer to where it is mapped. This routine is intended to be used
2648  * for mapping device memory, NOT real memory.
2649  */
2650 static void *
2651 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2652 {
2653 
2654 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2655 }
2656 
2657 static void *
2658 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2659 {
2660 	void *res;
2661 	uintptr_t va;
2662 	vm_size_t sz;
2663 	int i;
2664 
2665 	/*
2666 	 * Check if this is premapped in TLB1. Note: this should probably also
2667 	 * check whether a sequence of TLB1 entries exist that match the
2668 	 * requirement, but now only checks the easy case.
2669 	 */
2670 	if (ma == VM_MEMATTR_DEFAULT) {
2671 		for (i = 0; i < tlb1_idx; i++) {
2672 			if (!(tlb1[i].mas1 & MAS1_VALID))
2673 				continue;
2674 			if (pa >= tlb1[i].phys &&
2675 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2676 				return (void *)(tlb1[i].virt +
2677 				    (pa - tlb1[i].phys));
2678 		}
2679 	}
2680 
2681 	size = roundup(size, PAGE_SIZE);
2682 
2683 	/*
2684 	 * We leave a hole for device direct mapping between the maximum user
2685 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2686 	 * devices are in there, just map them 1:1. If not, map them to the
2687 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2688 	 * addresses should be pulled from an allocator, but since we do not
2689 	 * ever free TLB1 entries, it is safe just to increment a counter.
2690 	 * Note that there isn't a lot of address space here (128 MB) and it
2691 	 * is not at all difficult to imagine running out, since that is a 4:1
2692 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2693 	 * mapped there.
2694 	 */
2695 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2696 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2697 		va = pa;
2698 	else
2699 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2700 	res = (void *)va;
2701 
2702 	do {
2703 		sz = 1 << (ilog2(size) & ~1);
2704 		if (bootverbose)
2705 			printf("Wiring VA=%x to PA=%x (size=%x), "
2706 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2707 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2708 		size -= sz;
2709 		pa += sz;
2710 		va += sz;
2711 	} while (size > 0);
2712 
2713 	return (res);
2714 }
2715 
2716 /*
2717  * 'Unmap' a range mapped by mmu_booke_mapdev().
2718  */
2719 static void
2720 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2721 {
2722 #ifdef SUPPORTS_SHRINKING_TLB1
2723 	vm_offset_t base, offset;
2724 
2725 	/*
2726 	 * Unmap only if this is inside kernel virtual space.
2727 	 */
2728 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2729 		base = trunc_page(va);
2730 		offset = va & PAGE_MASK;
2731 		size = roundup(offset + size, PAGE_SIZE);
2732 		kva_free(base, size);
2733 	}
2734 #endif
2735 }
2736 
2737 /*
2738  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2739  * specified pmap. This eliminates the blast of soft faults on process startup
2740  * and immediately after an mmap.
2741  */
2742 static void
2743 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2744     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2745 {
2746 
2747 	VM_OBJECT_ASSERT_WLOCKED(object);
2748 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2749 	    ("mmu_booke_object_init_pt: non-device object"));
2750 }
2751 
2752 /*
2753  * Perform the pmap work for mincore.
2754  */
2755 static int
2756 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2757     vm_paddr_t *locked_pa)
2758 {
2759 
2760 	/* XXX: this should be implemented at some point */
2761 	return (0);
2762 }
2763 
2764 /**************************************************************************/
2765 /* TID handling */
2766 /**************************************************************************/
2767 
2768 /*
2769  * Allocate a TID. If necessary, steal one from someone else.
2770  * The new TID is flushed from the TLB before returning.
2771  */
2772 static tlbtid_t
2773 tid_alloc(pmap_t pmap)
2774 {
2775 	tlbtid_t tid;
2776 	int thiscpu;
2777 
2778 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2779 
2780 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2781 
2782 	thiscpu = PCPU_GET(cpuid);
2783 
2784 	tid = PCPU_GET(tid_next);
2785 	if (tid > TID_MAX)
2786 		tid = TID_MIN;
2787 	PCPU_SET(tid_next, tid + 1);
2788 
2789 	/* If we are stealing TID then clear the relevant pmap's field */
2790 	if (tidbusy[thiscpu][tid] != NULL) {
2791 
2792 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2793 
2794 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2795 
2796 		/* Flush all entries from TLB0 matching this TID. */
2797 		tid_flush(tid);
2798 	}
2799 
2800 	tidbusy[thiscpu][tid] = pmap;
2801 	pmap->pm_tid[thiscpu] = tid;
2802 	__asm __volatile("msync; isync");
2803 
2804 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2805 	    PCPU_GET(tid_next));
2806 
2807 	return (tid);
2808 }
2809 
2810 /**************************************************************************/
2811 /* TLB0 handling */
2812 /**************************************************************************/
2813 
2814 static void
2815 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2816     uint32_t mas7)
2817 {
2818 	int as;
2819 	char desc[3];
2820 	tlbtid_t tid;
2821 	vm_size_t size;
2822 	unsigned int tsize;
2823 
2824 	desc[2] = '\0';
2825 	if (mas1 & MAS1_VALID)
2826 		desc[0] = 'V';
2827 	else
2828 		desc[0] = ' ';
2829 
2830 	if (mas1 & MAS1_IPROT)
2831 		desc[1] = 'P';
2832 	else
2833 		desc[1] = ' ';
2834 
2835 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2836 	tid = MAS1_GETTID(mas1);
2837 
2838 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2839 	size = 0;
2840 	if (tsize)
2841 		size = tsize2size(tsize);
2842 
2843 	debugf("%3d: (%s) [AS=%d] "
2844 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2845 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2846 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2847 }
2848 
2849 /* Convert TLB0 va and way number to tlb0[] table index. */
2850 static inline unsigned int
2851 tlb0_tableidx(vm_offset_t va, unsigned int way)
2852 {
2853 	unsigned int idx;
2854 
2855 	idx = (way * TLB0_ENTRIES_PER_WAY);
2856 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2857 	return (idx);
2858 }
2859 
2860 /*
2861  * Invalidate TLB0 entry.
2862  */
2863 static inline void
2864 tlb0_flush_entry(vm_offset_t va)
2865 {
2866 
2867 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2868 
2869 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2870 
2871 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2872 	__asm __volatile("isync; msync");
2873 	__asm __volatile("tlbsync; msync");
2874 
2875 	CTR1(KTR_PMAP, "%s: e", __func__);
2876 }
2877 
2878 /* Print out contents of the MAS registers for each TLB0 entry */
2879 void
2880 tlb0_print_tlbentries(void)
2881 {
2882 	uint32_t mas0, mas1, mas2, mas3, mas7;
2883 	int entryidx, way, idx;
2884 
2885 	debugf("TLB0 entries:\n");
2886 	for (way = 0; way < TLB0_WAYS; way ++)
2887 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2888 
2889 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2890 			mtspr(SPR_MAS0, mas0);
2891 			__asm __volatile("isync");
2892 
2893 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2894 			mtspr(SPR_MAS2, mas2);
2895 
2896 			__asm __volatile("isync; tlbre");
2897 
2898 			mas1 = mfspr(SPR_MAS1);
2899 			mas2 = mfspr(SPR_MAS2);
2900 			mas3 = mfspr(SPR_MAS3);
2901 			mas7 = mfspr(SPR_MAS7);
2902 
2903 			idx = tlb0_tableidx(mas2, way);
2904 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2905 		}
2906 }
2907 
2908 /**************************************************************************/
2909 /* TLB1 handling */
2910 /**************************************************************************/
2911 
2912 /*
2913  * TLB1 mapping notes:
2914  *
2915  * TLB1[0]	Kernel text and data.
2916  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2917  *		windows, other devices mappings.
2918  */
2919 
2920 /*
2921  * Write given entry to TLB1 hardware.
2922  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2923  */
2924 static void
2925 tlb1_write_entry(unsigned int idx)
2926 {
2927 	uint32_t mas0, mas7;
2928 
2929 	//debugf("tlb1_write_entry: s\n");
2930 
2931 	/* Clear high order RPN bits */
2932 	mas7 = 0;
2933 
2934 	/* Select entry */
2935 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2936 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2937 
2938 	mtspr(SPR_MAS0, mas0);
2939 	__asm __volatile("isync");
2940 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2941 	__asm __volatile("isync");
2942 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2943 	__asm __volatile("isync");
2944 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2945 	__asm __volatile("isync");
2946 	mtspr(SPR_MAS7, mas7);
2947 	__asm __volatile("isync; tlbwe; isync; msync");
2948 
2949 	//debugf("tlb1_write_entry: e\n");
2950 }
2951 
2952 /*
2953  * Return the largest uint value log such that 2^log <= num.
2954  */
2955 static unsigned int
2956 ilog2(unsigned int num)
2957 {
2958 	int lz;
2959 
2960 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2961 	return (31 - lz);
2962 }
2963 
2964 /*
2965  * Convert TLB TSIZE value to mapped region size.
2966  */
2967 static vm_size_t
2968 tsize2size(unsigned int tsize)
2969 {
2970 
2971 	/*
2972 	 * size = 4^tsize KB
2973 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2974 	 */
2975 
2976 	return ((1 << (2 * tsize)) * 1024);
2977 }
2978 
2979 /*
2980  * Convert region size (must be power of 4) to TLB TSIZE value.
2981  */
2982 static unsigned int
2983 size2tsize(vm_size_t size)
2984 {
2985 
2986 	return (ilog2(size) / 2 - 5);
2987 }
2988 
2989 /*
2990  * Register permanent kernel mapping in TLB1.
2991  *
2992  * Entries are created starting from index 0 (current free entry is
2993  * kept in tlb1_idx) and are not supposed to be invalidated.
2994  */
2995 static int
2996 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2997     uint32_t flags)
2998 {
2999 	uint32_t ts, tid;
3000 	int tsize, index;
3001 
3002 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3003 	if (index >= TLB1_ENTRIES) {
3004 		printf("tlb1_set_entry: TLB1 full!\n");
3005 		return (-1);
3006 	}
3007 
3008 	/* Convert size to TSIZE */
3009 	tsize = size2tsize(size);
3010 
3011 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3012 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3013 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3014 
3015 	/*
3016 	 * Atomicity is preserved by the atomic increment above since nothing
3017 	 * is ever removed from tlb1.
3018 	 */
3019 
3020 	tlb1[index].phys = pa;
3021 	tlb1[index].virt = va;
3022 	tlb1[index].size = size;
3023 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3024 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3025 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3026 
3027 	/* Set supervisor RWX permission bits */
3028 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3029 
3030 	tlb1_write_entry(index);
3031 
3032 	/*
3033 	 * XXX in general TLB1 updates should be propagated between CPUs,
3034 	 * since current design assumes to have the same TLB1 set-up on all
3035 	 * cores.
3036 	 */
3037 	return (0);
3038 }
3039 
3040 /*
3041  * Map in contiguous RAM region into the TLB1 using maximum of
3042  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3043  *
3044  * If necessary round up last entry size and return total size
3045  * used by all allocated entries.
3046  */
3047 vm_size_t
3048 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3049 {
3050 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3051 	vm_size_t mapped, pgsz, base, mask;
3052 	int idx, nents;
3053 
3054 	/* Round up to the next 1M */
3055 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3056 
3057 	mapped = 0;
3058 	idx = 0;
3059 	base = va;
3060 	pgsz = 64*1024*1024;
3061 	while (mapped < size) {
3062 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3063 			while (pgsz > (size - mapped))
3064 				pgsz >>= 2;
3065 			pgs[idx++] = pgsz;
3066 			mapped += pgsz;
3067 		}
3068 
3069 		/* We under-map. Correct for this. */
3070 		if (mapped < size) {
3071 			while (pgs[idx - 1] == pgsz) {
3072 				idx--;
3073 				mapped -= pgsz;
3074 			}
3075 			/* XXX We may increase beyond out starting point. */
3076 			pgsz <<= 2;
3077 			pgs[idx++] = pgsz;
3078 			mapped += pgsz;
3079 		}
3080 	}
3081 
3082 	nents = idx;
3083 	mask = pgs[0] - 1;
3084 	/* Align address to the boundary */
3085 	if (va & mask) {
3086 		va = (va + mask) & ~mask;
3087 		pa = (pa + mask) & ~mask;
3088 	}
3089 
3090 	for (idx = 0; idx < nents; idx++) {
3091 		pgsz = pgs[idx];
3092 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3093 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3094 		pa += pgsz;
3095 		va += pgsz;
3096 	}
3097 
3098 	mapped = (va - base);
3099 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3100 	    mapped, mapped - size);
3101 	return (mapped);
3102 }
3103 
3104 /*
3105  * TLB1 initialization routine, to be called after the very first
3106  * assembler level setup done in locore.S.
3107  */
3108 void
3109 tlb1_init()
3110 {
3111 	uint32_t mas0, mas1, mas2, mas3;
3112 	uint32_t tsz;
3113 	u_int i;
3114 
3115 	if (bootinfo != NULL && bootinfo[0] != 1) {
3116 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3117 	} else
3118 		tlb1_idx = 1;
3119 
3120 	/* The first entry/entries are used to map the kernel. */
3121 	for (i = 0; i < tlb1_idx; i++) {
3122 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3123 		mtspr(SPR_MAS0, mas0);
3124 		__asm __volatile("isync; tlbre");
3125 
3126 		mas1 = mfspr(SPR_MAS1);
3127 		if ((mas1 & MAS1_VALID) == 0)
3128 			continue;
3129 
3130 		mas2 = mfspr(SPR_MAS2);
3131 		mas3 = mfspr(SPR_MAS3);
3132 
3133 		tlb1[i].mas1 = mas1;
3134 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3135 		tlb1[i].mas3 = mas3;
3136 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3137 		tlb1[i].phys = mas3 & MAS3_RPN;
3138 
3139 		if (i == 0)
3140 			kernload = mas3 & MAS3_RPN;
3141 
3142 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3143 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3144 		kernsize += tlb1[i].size;
3145 	}
3146 
3147 #ifdef SMP
3148 	bp_ntlb1s = tlb1_idx;
3149 #endif
3150 
3151 	/* Purge the remaining entries */
3152 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3153 		tlb1_write_entry(i);
3154 
3155 	/* Setup TLB miss defaults */
3156 	set_mas4_defaults();
3157 }
3158 
3159 vm_offset_t
3160 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3161 {
3162 	vm_paddr_t pa_base;
3163 	vm_offset_t va, sz;
3164 	int i;
3165 
3166 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3167 
3168 	for (i = 0; i < tlb1_idx; i++) {
3169 		if (!(tlb1[i].mas1 & MAS1_VALID))
3170 			continue;
3171 		if (pa >= tlb1[i].phys && (pa + size) <=
3172 		    (tlb1[i].phys + tlb1[i].size))
3173 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3174 	}
3175 
3176 	pa_base = trunc_page(pa);
3177 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3178 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3179 	va = tlb1_map_base + (pa - pa_base);
3180 
3181 	do {
3182 		sz = 1 << (ilog2(size) & ~1);
3183 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3184 		size -= sz;
3185 		pa_base += sz;
3186 		tlb1_map_base += sz;
3187 	} while (size > 0);
3188 
3189 #ifdef SMP
3190 	bp_ntlb1s = tlb1_idx;
3191 #endif
3192 
3193 	return (va);
3194 }
3195 
3196 /*
3197  * Setup MAS4 defaults.
3198  * These values are loaded to MAS0-2 on a TLB miss.
3199  */
3200 static void
3201 set_mas4_defaults(void)
3202 {
3203 	uint32_t mas4;
3204 
3205 	/* Defaults: TLB0, PID0, TSIZED=4K */
3206 	mas4 = MAS4_TLBSELD0;
3207 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3208 #ifdef SMP
3209 	mas4 |= MAS4_MD;
3210 #endif
3211 	mtspr(SPR_MAS4, mas4);
3212 	__asm __volatile("isync");
3213 }
3214 
3215 /*
3216  * Print out contents of the MAS registers for each TLB1 entry
3217  */
3218 void
3219 tlb1_print_tlbentries(void)
3220 {
3221 	uint32_t mas0, mas1, mas2, mas3, mas7;
3222 	int i;
3223 
3224 	debugf("TLB1 entries:\n");
3225 	for (i = 0; i < TLB1_ENTRIES; i++) {
3226 
3227 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3228 		mtspr(SPR_MAS0, mas0);
3229 
3230 		__asm __volatile("isync; tlbre");
3231 
3232 		mas1 = mfspr(SPR_MAS1);
3233 		mas2 = mfspr(SPR_MAS2);
3234 		mas3 = mfspr(SPR_MAS3);
3235 		mas7 = mfspr(SPR_MAS7);
3236 
3237 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3238 	}
3239 }
3240 
3241 /*
3242  * Print out contents of the in-ram tlb1 table.
3243  */
3244 void
3245 tlb1_print_entries(void)
3246 {
3247 	int i;
3248 
3249 	debugf("tlb1[] table entries:\n");
3250 	for (i = 0; i < TLB1_ENTRIES; i++)
3251 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3252 }
3253 
3254 /*
3255  * Return 0 if the physical IO range is encompassed by one of the
3256  * the TLB1 entries, otherwise return related error code.
3257  */
3258 static int
3259 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3260 {
3261 	uint32_t prot;
3262 	vm_paddr_t pa_start;
3263 	vm_paddr_t pa_end;
3264 	unsigned int entry_tsize;
3265 	vm_size_t entry_size;
3266 
3267 	*va = (vm_offset_t)NULL;
3268 
3269 	/* Skip invalid entries */
3270 	if (!(tlb1[i].mas1 & MAS1_VALID))
3271 		return (EINVAL);
3272 
3273 	/*
3274 	 * The entry must be cache-inhibited, guarded, and r/w
3275 	 * so it can function as an i/o page
3276 	 */
3277 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3278 	if (prot != (MAS2_I | MAS2_G))
3279 		return (EPERM);
3280 
3281 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3282 	if (prot != (MAS3_SR | MAS3_SW))
3283 		return (EPERM);
3284 
3285 	/* The address should be within the entry range. */
3286 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3287 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3288 
3289 	entry_size = tsize2size(entry_tsize);
3290 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3291 	pa_end = pa_start + entry_size - 1;
3292 
3293 	if ((pa < pa_start) || ((pa + size) > pa_end))
3294 		return (ERANGE);
3295 
3296 	/* Return virtual address of this mapping. */
3297 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3298 	return (0);
3299 }
3300