xref: /freebsd/sys/powerpc/booke/pmap.c (revision 2e5b60079b7d8c3ca68f1390cd90f305e651f8d3)
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/conf.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
66 #include <sys/lock.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
70 #include <sys/smp.h>
71 #include <sys/vmmeter.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
82 #include <vm/uma.h>
83 
84 #include <machine/cpu.h>
85 #include <machine/pcb.h>
86 #include <machine/platform.h>
87 
88 #include <machine/tlb.h>
89 #include <machine/spr.h>
90 #include <machine/md_var.h>
91 #include <machine/mmuvar.h>
92 #include <machine/pmap.h>
93 #include <machine/pte.h>
94 
95 #include "mmu_if.h"
96 
97 #ifdef  DEBUG
98 #define debugf(fmt, args...) printf(fmt, ##args)
99 #else
100 #define debugf(fmt, args...)
101 #endif
102 
103 #define TODO			panic("%s: not implemented", __func__);
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 int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, u_int flags, int8_t psind);
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, boolean_t);
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 int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_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_clear_modify(mmu_t, vm_page_t);
270 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
271     vm_size_t, vm_offset_t);
272 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
273 static void		mmu_booke_copy_pages(mmu_t, vm_page_t *,
274     vm_offset_t, vm_page_t *, vm_offset_t, int);
275 static int		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
276     vm_prot_t, u_int flags, int8_t psind);
277 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
278     vm_page_t, vm_prot_t);
279 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
280     vm_prot_t);
281 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
282 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
283     vm_prot_t);
284 static void		mmu_booke_init(mmu_t);
285 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
286 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
287 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
288 static int		mmu_booke_ts_referenced(mmu_t, vm_page_t);
289 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
290     int);
291 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
292     vm_paddr_t *);
293 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
294     vm_object_t, vm_pindex_t, vm_size_t);
295 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
296 static void		mmu_booke_page_init(mmu_t, vm_page_t);
297 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
298 static void		mmu_booke_pinit(mmu_t, pmap_t);
299 static void		mmu_booke_pinit0(mmu_t, pmap_t);
300 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
301     vm_prot_t);
302 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
303 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
304 static void		mmu_booke_release(mmu_t, pmap_t);
305 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
306 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
307 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
308 static void		mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_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 void		mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
326     void **);
327 static void		mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
328     void *);
329 static void		mmu_booke_scan_init(mmu_t);
330 
331 static mmu_method_t mmu_booke_methods[] = {
332 	/* pmap dispatcher interface */
333 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
334 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
335 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
336 	MMUMETHOD(mmu_copy_pages,	mmu_booke_copy_pages),
337 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
338 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
339 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
340 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
341 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
342 	MMUMETHOD(mmu_init,		mmu_booke_init),
343 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
344 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
345 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
346 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
347 	MMUMETHOD(mmu_map,		mmu_booke_map),
348 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
349 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
350 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
351 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
352 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
353 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
354 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
355 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
356 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
357 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
358 	MMUMETHOD(mmu_release,		mmu_booke_release),
359 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
360 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
361 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
362 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
363 	MMUMETHOD(mmu_unwire,		mmu_booke_unwire),
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_init,	mmu_booke_scan_init),
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, boolean_t nosleep)
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, j;
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 			PMAP_UNLOCK(pmap);
592 			rw_wunlock(&pvh_global_lock);
593 			if (nosleep) {
594 				ptbl_free_pmap_ptbl(pmap, ptbl);
595 				for (j = 0; j < i; j++)
596 					vm_page_free(mtbl[j]);
597 				atomic_subtract_int(&vm_cnt.v_wire_count, i);
598 				return (NULL);
599 			}
600 			VM_WAIT;
601 			rw_wlock(&pvh_global_lock);
602 			PMAP_LOCK(pmap);
603 		}
604 		mtbl[i] = m;
605 	}
606 
607 	/* Map allocated pages into kernel_pmap. */
608 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
609 
610 	/* Zero whole ptbl. */
611 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
612 
613 	/* Add pbuf to the pmap ptbl bufs list. */
614 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
615 
616 	return (ptbl);
617 }
618 
619 /* Free ptbl pages and invalidate pdir entry. */
620 static void
621 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
622 {
623 	pte_t *ptbl;
624 	vm_paddr_t pa;
625 	vm_offset_t va;
626 	vm_page_t m;
627 	int i;
628 
629 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
630 	    (pmap == kernel_pmap), pdir_idx);
631 
632 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
633 	    ("ptbl_free: invalid pdir_idx"));
634 
635 	ptbl = pmap->pm_pdir[pdir_idx];
636 
637 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
638 
639 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
640 
641 	/*
642 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
643 	 * don't attempt to look up the page tables we are releasing.
644 	 */
645 	mtx_lock_spin(&tlbivax_mutex);
646 	tlb_miss_lock();
647 
648 	pmap->pm_pdir[pdir_idx] = NULL;
649 
650 	tlb_miss_unlock();
651 	mtx_unlock_spin(&tlbivax_mutex);
652 
653 	for (i = 0; i < PTBL_PAGES; i++) {
654 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
655 		pa = pte_vatopa(mmu, kernel_pmap, va);
656 		m = PHYS_TO_VM_PAGE(pa);
657 		vm_page_free_zero(m);
658 		atomic_subtract_int(&vm_cnt.v_wire_count, 1);
659 		mmu_booke_kremove(mmu, va);
660 	}
661 
662 	ptbl_free_pmap_ptbl(pmap, ptbl);
663 }
664 
665 /*
666  * Decrement ptbl pages hold count and attempt to free ptbl pages.
667  * Called when removing pte entry from ptbl.
668  *
669  * Return 1 if ptbl pages were freed.
670  */
671 static int
672 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
673 {
674 	pte_t *ptbl;
675 	vm_paddr_t pa;
676 	vm_page_t m;
677 	int i;
678 
679 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
680 	    (pmap == kernel_pmap), pdir_idx);
681 
682 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
683 	    ("ptbl_unhold: invalid pdir_idx"));
684 	KASSERT((pmap != kernel_pmap),
685 	    ("ptbl_unhold: unholding kernel ptbl!"));
686 
687 	ptbl = pmap->pm_pdir[pdir_idx];
688 
689 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
690 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
691 	    ("ptbl_unhold: non kva ptbl"));
692 
693 	/* decrement hold count */
694 	for (i = 0; i < PTBL_PAGES; i++) {
695 		pa = pte_vatopa(mmu, kernel_pmap,
696 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
697 		m = PHYS_TO_VM_PAGE(pa);
698 		m->wire_count--;
699 	}
700 
701 	/*
702 	 * Free ptbl pages if there are no pte etries in this ptbl.
703 	 * wire_count has the same value for all ptbl pages, so check the last
704 	 * page.
705 	 */
706 	if (m->wire_count == 0) {
707 		ptbl_free(mmu, pmap, pdir_idx);
708 
709 		//debugf("ptbl_unhold: e (freed ptbl)\n");
710 		return (1);
711 	}
712 
713 	return (0);
714 }
715 
716 /*
717  * Increment hold count for ptbl pages. This routine is used when a new pte
718  * entry is being inserted into the ptbl.
719  */
720 static void
721 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
722 {
723 	vm_paddr_t pa;
724 	pte_t *ptbl;
725 	vm_page_t m;
726 	int i;
727 
728 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
729 	    pdir_idx);
730 
731 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
732 	    ("ptbl_hold: invalid pdir_idx"));
733 	KASSERT((pmap != kernel_pmap),
734 	    ("ptbl_hold: holding kernel ptbl!"));
735 
736 	ptbl = pmap->pm_pdir[pdir_idx];
737 
738 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
739 
740 	for (i = 0; i < PTBL_PAGES; i++) {
741 		pa = pte_vatopa(mmu, kernel_pmap,
742 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
743 		m = PHYS_TO_VM_PAGE(pa);
744 		m->wire_count++;
745 	}
746 }
747 
748 /* Allocate pv_entry structure. */
749 pv_entry_t
750 pv_alloc(void)
751 {
752 	pv_entry_t pv;
753 
754 	pv_entry_count++;
755 	if (pv_entry_count > pv_entry_high_water)
756 		pagedaemon_wakeup();
757 	pv = uma_zalloc(pvzone, M_NOWAIT);
758 
759 	return (pv);
760 }
761 
762 /* Free pv_entry structure. */
763 static __inline void
764 pv_free(pv_entry_t pve)
765 {
766 
767 	pv_entry_count--;
768 	uma_zfree(pvzone, pve);
769 }
770 
771 
772 /* Allocate and initialize pv_entry structure. */
773 static void
774 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
775 {
776 	pv_entry_t pve;
777 
778 	//int su = (pmap == kernel_pmap);
779 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
780 	//	(u_int32_t)pmap, va, (u_int32_t)m);
781 
782 	pve = pv_alloc();
783 	if (pve == NULL)
784 		panic("pv_insert: no pv entries!");
785 
786 	pve->pv_pmap = pmap;
787 	pve->pv_va = va;
788 
789 	/* add to pv_list */
790 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
791 	rw_assert(&pvh_global_lock, RA_WLOCKED);
792 
793 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
794 
795 	//debugf("pv_insert: e\n");
796 }
797 
798 /* Destroy pv entry. */
799 static void
800 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
801 {
802 	pv_entry_t pve;
803 
804 	//int su = (pmap == kernel_pmap);
805 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
806 
807 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
808 	rw_assert(&pvh_global_lock, RA_WLOCKED);
809 
810 	/* find pv entry */
811 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
812 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
813 			/* remove from pv_list */
814 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
815 			if (TAILQ_EMPTY(&m->md.pv_list))
816 				vm_page_aflag_clear(m, PGA_WRITEABLE);
817 
818 			/* free pv entry struct */
819 			pv_free(pve);
820 			break;
821 		}
822 	}
823 
824 	//debugf("pv_remove: e\n");
825 }
826 
827 /*
828  * Clean pte entry, try to free page table page if requested.
829  *
830  * Return 1 if ptbl pages were freed, otherwise return 0.
831  */
832 static int
833 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
834 {
835 	unsigned int pdir_idx = PDIR_IDX(va);
836 	unsigned int ptbl_idx = PTBL_IDX(va);
837 	vm_page_t m;
838 	pte_t *ptbl;
839 	pte_t *pte;
840 
841 	//int su = (pmap == kernel_pmap);
842 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
843 	//		su, (u_int32_t)pmap, va, flags);
844 
845 	ptbl = pmap->pm_pdir[pdir_idx];
846 	KASSERT(ptbl, ("pte_remove: null ptbl"));
847 
848 	pte = &ptbl[ptbl_idx];
849 
850 	if (pte == NULL || !PTE_ISVALID(pte))
851 		return (0);
852 
853 	if (PTE_ISWIRED(pte))
854 		pmap->pm_stats.wired_count--;
855 
856 	/* Handle managed entry. */
857 	if (PTE_ISMANAGED(pte)) {
858 		/* Get vm_page_t for mapped pte. */
859 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
860 
861 		if (PTE_ISMODIFIED(pte))
862 			vm_page_dirty(m);
863 
864 		if (PTE_ISREFERENCED(pte))
865 			vm_page_aflag_set(m, PGA_REFERENCED);
866 
867 		pv_remove(pmap, va, m);
868 	}
869 
870 	mtx_lock_spin(&tlbivax_mutex);
871 	tlb_miss_lock();
872 
873 	tlb0_flush_entry(va);
874 	pte->flags = 0;
875 	pte->rpn = 0;
876 
877 	tlb_miss_unlock();
878 	mtx_unlock_spin(&tlbivax_mutex);
879 
880 	pmap->pm_stats.resident_count--;
881 
882 	if (flags & PTBL_UNHOLD) {
883 		//debugf("pte_remove: e (unhold)\n");
884 		return (ptbl_unhold(mmu, pmap, pdir_idx));
885 	}
886 
887 	//debugf("pte_remove: e\n");
888 	return (0);
889 }
890 
891 /*
892  * Insert PTE for a given page and virtual address.
893  */
894 static int
895 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
896     boolean_t nosleep)
897 {
898 	unsigned int pdir_idx = PDIR_IDX(va);
899 	unsigned int ptbl_idx = PTBL_IDX(va);
900 	pte_t *ptbl, *pte;
901 
902 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
903 	    pmap == kernel_pmap, pmap, va);
904 
905 	/* Get the page table pointer. */
906 	ptbl = pmap->pm_pdir[pdir_idx];
907 
908 	if (ptbl == NULL) {
909 		/* Allocate page table pages. */
910 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
911 		if (ptbl == NULL) {
912 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
913 			return (ENOMEM);
914 		}
915 	} else {
916 		/*
917 		 * Check if there is valid mapping for requested
918 		 * va, if there is, remove it.
919 		 */
920 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
921 		if (PTE_ISVALID(pte)) {
922 			pte_remove(mmu, pmap, va, PTBL_HOLD);
923 		} else {
924 			/*
925 			 * pte is not used, increment hold count
926 			 * for ptbl pages.
927 			 */
928 			if (pmap != kernel_pmap)
929 				ptbl_hold(mmu, pmap, pdir_idx);
930 		}
931 	}
932 
933 	/*
934 	 * Insert pv_entry into pv_list for mapped page if part of managed
935 	 * memory.
936 	 */
937 	if ((m->oflags & VPO_UNMANAGED) == 0) {
938 		flags |= PTE_MANAGED;
939 
940 		/* Create and insert pv entry. */
941 		pv_insert(pmap, va, m);
942 	}
943 
944 	pmap->pm_stats.resident_count++;
945 
946 	mtx_lock_spin(&tlbivax_mutex);
947 	tlb_miss_lock();
948 
949 	tlb0_flush_entry(va);
950 	if (pmap->pm_pdir[pdir_idx] == NULL) {
951 		/*
952 		 * If we just allocated a new page table, hook it in
953 		 * the pdir.
954 		 */
955 		pmap->pm_pdir[pdir_idx] = ptbl;
956 	}
957 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
958 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
959 	pte->flags |= (PTE_VALID | flags);
960 
961 	tlb_miss_unlock();
962 	mtx_unlock_spin(&tlbivax_mutex);
963 	return (0);
964 }
965 
966 /* Return the pa for the given pmap/va. */
967 static vm_paddr_t
968 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
969 {
970 	vm_paddr_t pa = 0;
971 	pte_t *pte;
972 
973 	pte = pte_find(mmu, pmap, va);
974 	if ((pte != NULL) && PTE_ISVALID(pte))
975 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
976 	return (pa);
977 }
978 
979 /* Get a pointer to a PTE in a page table. */
980 static pte_t *
981 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
982 {
983 	unsigned int pdir_idx = PDIR_IDX(va);
984 	unsigned int ptbl_idx = PTBL_IDX(va);
985 
986 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
987 
988 	if (pmap->pm_pdir[pdir_idx])
989 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
990 
991 	return (NULL);
992 }
993 
994 /**************************************************************************/
995 /* PMAP related */
996 /**************************************************************************/
997 
998 /*
999  * This is called during booke_init, before the system is really initialized.
1000  */
1001 static void
1002 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1003 {
1004 	vm_offset_t phys_kernelend;
1005 	struct mem_region *mp, *mp1;
1006 	int cnt, i, j;
1007 	u_int s, e, sz;
1008 	u_int phys_avail_count;
1009 	vm_size_t physsz, hwphyssz, kstack0_sz;
1010 	vm_offset_t kernel_pdir, kstack0, va;
1011 	vm_paddr_t kstack0_phys;
1012 	void *dpcpu;
1013 	pte_t *pte;
1014 
1015 	debugf("mmu_booke_bootstrap: entered\n");
1016 
1017 	/* Initialize invalidation mutex */
1018 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1019 
1020 	/* Read TLB0 size and associativity. */
1021 	tlb0_get_tlbconf();
1022 
1023 	/*
1024 	 * Align kernel start and end address (kernel image).
1025 	 * Note that kernel end does not necessarily relate to kernsize.
1026 	 * kernsize is the size of the kernel that is actually mapped.
1027 	 * Also note that "start - 1" is deliberate. With SMP, the
1028 	 * entry point is exactly a page from the actual load address.
1029 	 * As such, trunc_page() has no effect and we're off by a page.
1030 	 * Since we always have the ELF header between the load address
1031 	 * and the entry point, we can safely subtract 1 to compensate.
1032 	 */
1033 	kernstart = trunc_page(start - 1);
1034 	data_start = round_page(kernelend);
1035 	data_end = data_start;
1036 
1037 	/*
1038 	 * Addresses of preloaded modules (like file systems) use
1039 	 * physical addresses. Make sure we relocate those into
1040 	 * virtual addresses.
1041 	 */
1042 	preload_addr_relocate = kernstart - kernload;
1043 
1044 	/* Allocate the dynamic per-cpu area. */
1045 	dpcpu = (void *)data_end;
1046 	data_end += DPCPU_SIZE;
1047 
1048 	/* Allocate space for the message buffer. */
1049 	msgbufp = (struct msgbuf *)data_end;
1050 	data_end += msgbufsize;
1051 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1052 	    data_end);
1053 
1054 	data_end = round_page(data_end);
1055 
1056 	/* Allocate space for ptbl_bufs. */
1057 	ptbl_bufs = (struct ptbl_buf *)data_end;
1058 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1059 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1060 	    data_end);
1061 
1062 	data_end = round_page(data_end);
1063 
1064 	/* Allocate PTE tables for kernel KVA. */
1065 	kernel_pdir = data_end;
1066 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1067 	    PDIR_SIZE - 1) / PDIR_SIZE;
1068 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1069 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1070 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1071 
1072 	debugf(" data_end: 0x%08x\n", data_end);
1073 	if (data_end - kernstart > kernsize) {
1074 		kernsize += tlb1_mapin_region(kernstart + kernsize,
1075 		    kernload + kernsize, (data_end - kernstart) - kernsize);
1076 	}
1077 	data_end = kernstart + kernsize;
1078 	debugf(" updated data_end: 0x%08x\n", data_end);
1079 
1080 	/*
1081 	 * Clear the structures - note we can only do it safely after the
1082 	 * possible additional TLB1 translations are in place (above) so that
1083 	 * all range up to the currently calculated 'data_end' is covered.
1084 	 */
1085 	dpcpu_init(dpcpu, 0);
1086 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1087 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1088 
1089 	/*******************************************************/
1090 	/* Set the start and end of kva. */
1091 	/*******************************************************/
1092 	virtual_avail = round_page(data_end);
1093 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1094 
1095 	/* Allocate KVA space for page zero/copy operations. */
1096 	zero_page_va = virtual_avail;
1097 	virtual_avail += PAGE_SIZE;
1098 	zero_page_idle_va = virtual_avail;
1099 	virtual_avail += PAGE_SIZE;
1100 	copy_page_src_va = virtual_avail;
1101 	virtual_avail += PAGE_SIZE;
1102 	copy_page_dst_va = virtual_avail;
1103 	virtual_avail += PAGE_SIZE;
1104 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1105 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1106 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1107 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1108 
1109 	/* Initialize page zero/copy mutexes. */
1110 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1111 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1112 
1113 	/* Allocate KVA space for ptbl bufs. */
1114 	ptbl_buf_pool_vabase = virtual_avail;
1115 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1116 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1117 	    ptbl_buf_pool_vabase, virtual_avail);
1118 
1119 	/* Calculate corresponding physical addresses for the kernel region. */
1120 	phys_kernelend = kernload + kernsize;
1121 	debugf("kernel image and allocated data:\n");
1122 	debugf(" kernload    = 0x%08x\n", kernload);
1123 	debugf(" kernstart   = 0x%08x\n", kernstart);
1124 	debugf(" kernsize    = 0x%08x\n", kernsize);
1125 
1126 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1127 		panic("mmu_booke_bootstrap: phys_avail too small");
1128 
1129 	/*
1130 	 * Remove kernel physical address range from avail regions list. Page
1131 	 * align all regions.  Non-page aligned memory isn't very interesting
1132 	 * to us.  Also, sort the entries for ascending addresses.
1133 	 */
1134 
1135 	/* Retrieve phys/avail mem regions */
1136 	mem_regions(&physmem_regions, &physmem_regions_sz,
1137 	    &availmem_regions, &availmem_regions_sz);
1138 	sz = 0;
1139 	cnt = availmem_regions_sz;
1140 	debugf("processing avail regions:\n");
1141 	for (mp = availmem_regions; mp->mr_size; mp++) {
1142 		s = mp->mr_start;
1143 		e = mp->mr_start + mp->mr_size;
1144 		debugf(" %08x-%08x -> ", s, e);
1145 		/* Check whether this region holds all of the kernel. */
1146 		if (s < kernload && e > phys_kernelend) {
1147 			availmem_regions[cnt].mr_start = phys_kernelend;
1148 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1149 			e = kernload;
1150 		}
1151 		/* Look whether this regions starts within the kernel. */
1152 		if (s >= kernload && s < phys_kernelend) {
1153 			if (e <= phys_kernelend)
1154 				goto empty;
1155 			s = phys_kernelend;
1156 		}
1157 		/* Now look whether this region ends within the kernel. */
1158 		if (e > kernload && e <= phys_kernelend) {
1159 			if (s >= kernload)
1160 				goto empty;
1161 			e = kernload;
1162 		}
1163 		/* Now page align the start and size of the region. */
1164 		s = round_page(s);
1165 		e = trunc_page(e);
1166 		if (e < s)
1167 			e = s;
1168 		sz = e - s;
1169 		debugf("%08x-%08x = %x\n", s, e, sz);
1170 
1171 		/* Check whether some memory is left here. */
1172 		if (sz == 0) {
1173 		empty:
1174 			memmove(mp, mp + 1,
1175 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1176 			cnt--;
1177 			mp--;
1178 			continue;
1179 		}
1180 
1181 		/* Do an insertion sort. */
1182 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1183 			if (s < mp1->mr_start)
1184 				break;
1185 		if (mp1 < mp) {
1186 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1187 			mp1->mr_start = s;
1188 			mp1->mr_size = sz;
1189 		} else {
1190 			mp->mr_start = s;
1191 			mp->mr_size = sz;
1192 		}
1193 	}
1194 	availmem_regions_sz = cnt;
1195 
1196 	/*******************************************************/
1197 	/* Steal physical memory for kernel stack from the end */
1198 	/* of the first avail region                           */
1199 	/*******************************************************/
1200 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1201 	kstack0_phys = availmem_regions[0].mr_start +
1202 	    availmem_regions[0].mr_size;
1203 	kstack0_phys -= kstack0_sz;
1204 	availmem_regions[0].mr_size -= kstack0_sz;
1205 
1206 	/*******************************************************/
1207 	/* Fill in phys_avail table, based on availmem_regions */
1208 	/*******************************************************/
1209 	phys_avail_count = 0;
1210 	physsz = 0;
1211 	hwphyssz = 0;
1212 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1213 
1214 	debugf("fill in phys_avail:\n");
1215 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1216 
1217 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1218 		    availmem_regions[i].mr_start,
1219 		    availmem_regions[i].mr_start +
1220 		        availmem_regions[i].mr_size,
1221 		    availmem_regions[i].mr_size);
1222 
1223 		if (hwphyssz != 0 &&
1224 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1225 			debugf(" hw.physmem adjust\n");
1226 			if (physsz < hwphyssz) {
1227 				phys_avail[j] = availmem_regions[i].mr_start;
1228 				phys_avail[j + 1] =
1229 				    availmem_regions[i].mr_start +
1230 				    hwphyssz - physsz;
1231 				physsz = hwphyssz;
1232 				phys_avail_count++;
1233 			}
1234 			break;
1235 		}
1236 
1237 		phys_avail[j] = availmem_regions[i].mr_start;
1238 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1239 		    availmem_regions[i].mr_size;
1240 		phys_avail_count++;
1241 		physsz += availmem_regions[i].mr_size;
1242 	}
1243 	physmem = btoc(physsz);
1244 
1245 	/* Calculate the last available physical address. */
1246 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1247 		;
1248 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1249 
1250 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1251 	debugf("phys_avail_count = %d\n", phys_avail_count);
1252 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1253 	    physmem);
1254 
1255 	/*******************************************************/
1256 	/* Initialize (statically allocated) kernel pmap. */
1257 	/*******************************************************/
1258 	PMAP_LOCK_INIT(kernel_pmap);
1259 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1260 
1261 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1262 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1263 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1264 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1265 
1266 	/* Initialize kernel pdir */
1267 	for (i = 0; i < kernel_ptbls; i++)
1268 		kernel_pmap->pm_pdir[kptbl_min + i] =
1269 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1270 
1271 	for (i = 0; i < MAXCPU; i++) {
1272 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1273 
1274 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1275 		tidbusy[i][0] = kernel_pmap;
1276 	}
1277 
1278 	/*
1279 	 * Fill in PTEs covering kernel code and data. They are not required
1280 	 * for address translation, as this area is covered by static TLB1
1281 	 * entries, but for pte_vatopa() to work correctly with kernel area
1282 	 * addresses.
1283 	 */
1284 	for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1285 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1286 		pte->rpn = kernload + (va - kernstart);
1287 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1288 		    PTE_VALID;
1289 	}
1290 	/* Mark kernel_pmap active on all CPUs */
1291 	CPU_FILL(&kernel_pmap->pm_active);
1292 
1293  	/*
1294 	 * Initialize the global pv list lock.
1295 	 */
1296 	rw_init(&pvh_global_lock, "pmap pv global");
1297 
1298 	/*******************************************************/
1299 	/* Final setup */
1300 	/*******************************************************/
1301 
1302 	/* Enter kstack0 into kernel map, provide guard page */
1303 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1304 	thread0.td_kstack = kstack0;
1305 	thread0.td_kstack_pages = KSTACK_PAGES;
1306 
1307 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1308 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1309 	    kstack0_phys, kstack0_phys + kstack0_sz);
1310 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1311 
1312 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1313 	for (i = 0; i < KSTACK_PAGES; i++) {
1314 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1315 		kstack0 += PAGE_SIZE;
1316 		kstack0_phys += PAGE_SIZE;
1317 	}
1318 
1319 	debugf("virtual_avail = %08x\n", virtual_avail);
1320 	debugf("virtual_end   = %08x\n", virtual_end);
1321 
1322 	debugf("mmu_booke_bootstrap: exit\n");
1323 }
1324 
1325 void
1326 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1327 {
1328 	int i;
1329 
1330 	/*
1331 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1332 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1333 	 * these values directly to (re)program AP's TLB1 hardware.
1334 	 */
1335 	for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1336 		/* Skip invalid entries */
1337 		if (!(tlb1[i].mas1 & MAS1_VALID))
1338 			continue;
1339 
1340 		tlb1_write_entry(i);
1341 	}
1342 
1343 	set_mas4_defaults();
1344 }
1345 
1346 /*
1347  * Get the physical page address for the given pmap/virtual address.
1348  */
1349 static vm_paddr_t
1350 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1351 {
1352 	vm_paddr_t pa;
1353 
1354 	PMAP_LOCK(pmap);
1355 	pa = pte_vatopa(mmu, pmap, va);
1356 	PMAP_UNLOCK(pmap);
1357 
1358 	return (pa);
1359 }
1360 
1361 /*
1362  * Extract the physical page address associated with the given
1363  * kernel virtual address.
1364  */
1365 static vm_paddr_t
1366 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1367 {
1368 	int i;
1369 
1370 	/* Check TLB1 mappings */
1371 	for (i = 0; i < tlb1_idx; i++) {
1372 		if (!(tlb1[i].mas1 & MAS1_VALID))
1373 			continue;
1374 		if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1375 			return (tlb1[i].phys + (va - tlb1[i].virt));
1376 	}
1377 
1378 	return (pte_vatopa(mmu, kernel_pmap, va));
1379 }
1380 
1381 /*
1382  * Initialize the pmap module.
1383  * Called by vm_init, to initialize any structures that the pmap
1384  * system needs to map virtual memory.
1385  */
1386 static void
1387 mmu_booke_init(mmu_t mmu)
1388 {
1389 	int shpgperproc = PMAP_SHPGPERPROC;
1390 
1391 	/*
1392 	 * Initialize the address space (zone) for the pv entries.  Set a
1393 	 * high water mark so that the system can recover from excessive
1394 	 * numbers of pv entries.
1395 	 */
1396 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1397 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1398 
1399 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1400 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1401 
1402 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1403 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1404 
1405 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1406 
1407 	/* Pre-fill pvzone with initial number of pv entries. */
1408 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1409 
1410 	/* Initialize ptbl allocation. */
1411 	ptbl_init();
1412 }
1413 
1414 /*
1415  * Map a list of wired pages into kernel virtual address space.  This is
1416  * intended for temporary mappings which do not need page modification or
1417  * references recorded.  Existing mappings in the region are overwritten.
1418  */
1419 static void
1420 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1421 {
1422 	vm_offset_t va;
1423 
1424 	va = sva;
1425 	while (count-- > 0) {
1426 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1427 		va += PAGE_SIZE;
1428 		m++;
1429 	}
1430 }
1431 
1432 /*
1433  * Remove page mappings from kernel virtual address space.  Intended for
1434  * temporary mappings entered by mmu_booke_qenter.
1435  */
1436 static void
1437 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1438 {
1439 	vm_offset_t va;
1440 
1441 	va = sva;
1442 	while (count-- > 0) {
1443 		mmu_booke_kremove(mmu, va);
1444 		va += PAGE_SIZE;
1445 	}
1446 }
1447 
1448 /*
1449  * Map a wired page into kernel virtual address space.
1450  */
1451 static void
1452 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1453 {
1454 
1455 	mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1456 }
1457 
1458 static void
1459 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1460 {
1461 	unsigned int pdir_idx = PDIR_IDX(va);
1462 	unsigned int ptbl_idx = PTBL_IDX(va);
1463 	uint32_t flags;
1464 	pte_t *pte;
1465 
1466 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1467 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1468 
1469 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1470 	flags |= tlb_calc_wimg(pa, ma);
1471 
1472 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1473 
1474 	mtx_lock_spin(&tlbivax_mutex);
1475 	tlb_miss_lock();
1476 
1477 	if (PTE_ISVALID(pte)) {
1478 
1479 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1480 
1481 		/* Flush entry from TLB0 */
1482 		tlb0_flush_entry(va);
1483 	}
1484 
1485 	pte->rpn = pa & ~PTE_PA_MASK;
1486 	pte->flags = flags;
1487 
1488 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1489 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1490 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1491 
1492 	/* Flush the real memory from the instruction cache. */
1493 	if ((flags & (PTE_I | PTE_G)) == 0) {
1494 		__syncicache((void *)va, PAGE_SIZE);
1495 	}
1496 
1497 	tlb_miss_unlock();
1498 	mtx_unlock_spin(&tlbivax_mutex);
1499 }
1500 
1501 /*
1502  * Remove a page from kernel page table.
1503  */
1504 static void
1505 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1506 {
1507 	unsigned int pdir_idx = PDIR_IDX(va);
1508 	unsigned int ptbl_idx = PTBL_IDX(va);
1509 	pte_t *pte;
1510 
1511 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1512 
1513 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1514 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1515 	    ("mmu_booke_kremove: invalid va"));
1516 
1517 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1518 
1519 	if (!PTE_ISVALID(pte)) {
1520 
1521 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1522 
1523 		return;
1524 	}
1525 
1526 	mtx_lock_spin(&tlbivax_mutex);
1527 	tlb_miss_lock();
1528 
1529 	/* Invalidate entry in TLB0, update PTE. */
1530 	tlb0_flush_entry(va);
1531 	pte->flags = 0;
1532 	pte->rpn = 0;
1533 
1534 	tlb_miss_unlock();
1535 	mtx_unlock_spin(&tlbivax_mutex);
1536 }
1537 
1538 /*
1539  * Initialize pmap associated with process 0.
1540  */
1541 static void
1542 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1543 {
1544 
1545 	PMAP_LOCK_INIT(pmap);
1546 	mmu_booke_pinit(mmu, pmap);
1547 	PCPU_SET(curpmap, pmap);
1548 }
1549 
1550 /*
1551  * Initialize a preallocated and zeroed pmap structure,
1552  * such as one in a vmspace structure.
1553  */
1554 static void
1555 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1556 {
1557 	int i;
1558 
1559 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1560 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1561 
1562 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1563 
1564 	for (i = 0; i < MAXCPU; i++)
1565 		pmap->pm_tid[i] = TID_NONE;
1566 	CPU_ZERO(&kernel_pmap->pm_active);
1567 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1568 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1569 	TAILQ_INIT(&pmap->pm_ptbl_list);
1570 }
1571 
1572 /*
1573  * Release any resources held by the given physical map.
1574  * Called when a pmap initialized by mmu_booke_pinit is being released.
1575  * Should only be called if the map contains no valid mappings.
1576  */
1577 static void
1578 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1579 {
1580 
1581 	KASSERT(pmap->pm_stats.resident_count == 0,
1582 	    ("pmap_release: pmap resident count %ld != 0",
1583 	    pmap->pm_stats.resident_count));
1584 }
1585 
1586 /*
1587  * Insert the given physical page at the specified virtual address in the
1588  * target physical map with the protection requested. If specified the page
1589  * will be wired down.
1590  */
1591 static int
1592 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1593     vm_prot_t prot, u_int flags, int8_t psind)
1594 {
1595 	int error;
1596 
1597 	rw_wlock(&pvh_global_lock);
1598 	PMAP_LOCK(pmap);
1599 	error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1600 	rw_wunlock(&pvh_global_lock);
1601 	PMAP_UNLOCK(pmap);
1602 	return (error);
1603 }
1604 
1605 static int
1606 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1607     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1608 {
1609 	pte_t *pte;
1610 	vm_paddr_t pa;
1611 	uint32_t flags;
1612 	int error, su, sync;
1613 
1614 	pa = VM_PAGE_TO_PHYS(m);
1615 	su = (pmap == kernel_pmap);
1616 	sync = 0;
1617 
1618 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1619 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1620 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1621 	//		(u_int32_t)m, va, pa, prot, flags);
1622 
1623 	if (su) {
1624 		KASSERT(((va >= virtual_avail) &&
1625 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1626 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1627 	} else {
1628 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1629 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1630 	}
1631 	if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1632 		VM_OBJECT_ASSERT_LOCKED(m->object);
1633 
1634 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1635 
1636 	/*
1637 	 * If there is an existing mapping, and the physical address has not
1638 	 * changed, must be protection or wiring change.
1639 	 */
1640 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1641 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1642 
1643 		/*
1644 		 * Before actually updating pte->flags we calculate and
1645 		 * prepare its new value in a helper var.
1646 		 */
1647 		flags = pte->flags;
1648 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1649 
1650 		/* Wiring change, just update stats. */
1651 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1652 			if (!PTE_ISWIRED(pte)) {
1653 				flags |= PTE_WIRED;
1654 				pmap->pm_stats.wired_count++;
1655 			}
1656 		} else {
1657 			if (PTE_ISWIRED(pte)) {
1658 				flags &= ~PTE_WIRED;
1659 				pmap->pm_stats.wired_count--;
1660 			}
1661 		}
1662 
1663 		if (prot & VM_PROT_WRITE) {
1664 			/* Add write permissions. */
1665 			flags |= PTE_SW;
1666 			if (!su)
1667 				flags |= PTE_UW;
1668 
1669 			if ((flags & PTE_MANAGED) != 0)
1670 				vm_page_aflag_set(m, PGA_WRITEABLE);
1671 		} else {
1672 			/* Handle modified pages, sense modify status. */
1673 
1674 			/*
1675 			 * The PTE_MODIFIED flag could be set by underlying
1676 			 * TLB misses since we last read it (above), possibly
1677 			 * other CPUs could update it so we check in the PTE
1678 			 * directly rather than rely on that saved local flags
1679 			 * copy.
1680 			 */
1681 			if (PTE_ISMODIFIED(pte))
1682 				vm_page_dirty(m);
1683 		}
1684 
1685 		if (prot & VM_PROT_EXECUTE) {
1686 			flags |= PTE_SX;
1687 			if (!su)
1688 				flags |= PTE_UX;
1689 
1690 			/*
1691 			 * Check existing flags for execute permissions: if we
1692 			 * are turning execute permissions on, icache should
1693 			 * be flushed.
1694 			 */
1695 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1696 				sync++;
1697 		}
1698 
1699 		flags &= ~PTE_REFERENCED;
1700 
1701 		/*
1702 		 * The new flags value is all calculated -- only now actually
1703 		 * update the PTE.
1704 		 */
1705 		mtx_lock_spin(&tlbivax_mutex);
1706 		tlb_miss_lock();
1707 
1708 		tlb0_flush_entry(va);
1709 		pte->flags = flags;
1710 
1711 		tlb_miss_unlock();
1712 		mtx_unlock_spin(&tlbivax_mutex);
1713 
1714 	} else {
1715 		/*
1716 		 * If there is an existing mapping, but it's for a different
1717 		 * physical address, pte_enter() will delete the old mapping.
1718 		 */
1719 		//if ((pte != NULL) && PTE_ISVALID(pte))
1720 		//	debugf("mmu_booke_enter_locked: replace\n");
1721 		//else
1722 		//	debugf("mmu_booke_enter_locked: new\n");
1723 
1724 		/* Now set up the flags and install the new mapping. */
1725 		flags = (PTE_SR | PTE_VALID);
1726 		flags |= PTE_M;
1727 
1728 		if (!su)
1729 			flags |= PTE_UR;
1730 
1731 		if (prot & VM_PROT_WRITE) {
1732 			flags |= PTE_SW;
1733 			if (!su)
1734 				flags |= PTE_UW;
1735 
1736 			if ((m->oflags & VPO_UNMANAGED) == 0)
1737 				vm_page_aflag_set(m, PGA_WRITEABLE);
1738 		}
1739 
1740 		if (prot & VM_PROT_EXECUTE) {
1741 			flags |= PTE_SX;
1742 			if (!su)
1743 				flags |= PTE_UX;
1744 		}
1745 
1746 		/* If its wired update stats. */
1747 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1748 			flags |= PTE_WIRED;
1749 
1750 		error = pte_enter(mmu, pmap, m, va, flags,
1751 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1752 		if (error != 0)
1753 			return (KERN_RESOURCE_SHORTAGE);
1754 
1755 		if ((flags & PMAP_ENTER_WIRED) != 0)
1756 			pmap->pm_stats.wired_count++;
1757 
1758 		/* Flush the real memory from the instruction cache. */
1759 		if (prot & VM_PROT_EXECUTE)
1760 			sync++;
1761 	}
1762 
1763 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1764 		__syncicache((void *)va, PAGE_SIZE);
1765 		sync = 0;
1766 	}
1767 
1768 	return (KERN_SUCCESS);
1769 }
1770 
1771 /*
1772  * Maps a sequence of resident pages belonging to the same object.
1773  * The sequence begins with the given page m_start.  This page is
1774  * mapped at the given virtual address start.  Each subsequent page is
1775  * mapped at a virtual address that is offset from start by the same
1776  * amount as the page is offset from m_start within the object.  The
1777  * last page in the sequence is the page with the largest offset from
1778  * m_start that can be mapped at a virtual address less than the given
1779  * virtual address end.  Not every virtual page between start and end
1780  * is mapped; only those for which a resident page exists with the
1781  * corresponding offset from m_start are mapped.
1782  */
1783 static void
1784 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1785     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1786 {
1787 	vm_page_t m;
1788 	vm_pindex_t diff, psize;
1789 
1790 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1791 
1792 	psize = atop(end - start);
1793 	m = m_start;
1794 	rw_wlock(&pvh_global_lock);
1795 	PMAP_LOCK(pmap);
1796 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1797 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1798 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1799 		    PMAP_ENTER_NOSLEEP, 0);
1800 		m = TAILQ_NEXT(m, listq);
1801 	}
1802 	rw_wunlock(&pvh_global_lock);
1803 	PMAP_UNLOCK(pmap);
1804 }
1805 
1806 static void
1807 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1808     vm_prot_t prot)
1809 {
1810 
1811 	rw_wlock(&pvh_global_lock);
1812 	PMAP_LOCK(pmap);
1813 	mmu_booke_enter_locked(mmu, pmap, va, m,
1814 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1815 	    0);
1816 	rw_wunlock(&pvh_global_lock);
1817 	PMAP_UNLOCK(pmap);
1818 }
1819 
1820 /*
1821  * Remove the given range of addresses from the specified map.
1822  *
1823  * It is assumed that the start and end are properly rounded to the page size.
1824  */
1825 static void
1826 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1827 {
1828 	pte_t *pte;
1829 	uint8_t hold_flag;
1830 
1831 	int su = (pmap == kernel_pmap);
1832 
1833 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1834 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1835 
1836 	if (su) {
1837 		KASSERT(((va >= virtual_avail) &&
1838 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1839 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1840 	} else {
1841 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1842 		    ("mmu_booke_remove: user pmap, non user va"));
1843 	}
1844 
1845 	if (PMAP_REMOVE_DONE(pmap)) {
1846 		//debugf("mmu_booke_remove: e (empty)\n");
1847 		return;
1848 	}
1849 
1850 	hold_flag = PTBL_HOLD_FLAG(pmap);
1851 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1852 
1853 	rw_wlock(&pvh_global_lock);
1854 	PMAP_LOCK(pmap);
1855 	for (; va < endva; va += PAGE_SIZE) {
1856 		pte = pte_find(mmu, pmap, va);
1857 		if ((pte != NULL) && PTE_ISVALID(pte))
1858 			pte_remove(mmu, pmap, va, hold_flag);
1859 	}
1860 	PMAP_UNLOCK(pmap);
1861 	rw_wunlock(&pvh_global_lock);
1862 
1863 	//debugf("mmu_booke_remove: e\n");
1864 }
1865 
1866 /*
1867  * Remove physical page from all pmaps in which it resides.
1868  */
1869 static void
1870 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1871 {
1872 	pv_entry_t pv, pvn;
1873 	uint8_t hold_flag;
1874 
1875 	rw_wlock(&pvh_global_lock);
1876 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1877 		pvn = TAILQ_NEXT(pv, pv_link);
1878 
1879 		PMAP_LOCK(pv->pv_pmap);
1880 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1881 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1882 		PMAP_UNLOCK(pv->pv_pmap);
1883 	}
1884 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1885 	rw_wunlock(&pvh_global_lock);
1886 }
1887 
1888 /*
1889  * Map a range of physical addresses into kernel virtual address space.
1890  */
1891 static vm_offset_t
1892 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1893     vm_paddr_t pa_end, int prot)
1894 {
1895 	vm_offset_t sva = *virt;
1896 	vm_offset_t va = sva;
1897 
1898 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1899 	//		sva, pa_start, pa_end);
1900 
1901 	while (pa_start < pa_end) {
1902 		mmu_booke_kenter(mmu, va, pa_start);
1903 		va += PAGE_SIZE;
1904 		pa_start += PAGE_SIZE;
1905 	}
1906 	*virt = va;
1907 
1908 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1909 	return (sva);
1910 }
1911 
1912 /*
1913  * The pmap must be activated before it's address space can be accessed in any
1914  * way.
1915  */
1916 static void
1917 mmu_booke_activate(mmu_t mmu, struct thread *td)
1918 {
1919 	pmap_t pmap;
1920 	u_int cpuid;
1921 
1922 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1923 
1924 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1925 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1926 
1927 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1928 
1929 	sched_pin();
1930 
1931 	cpuid = PCPU_GET(cpuid);
1932 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1933 	PCPU_SET(curpmap, pmap);
1934 
1935 	if (pmap->pm_tid[cpuid] == TID_NONE)
1936 		tid_alloc(pmap);
1937 
1938 	/* Load PID0 register with pmap tid value. */
1939 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1940 	__asm __volatile("isync");
1941 
1942 	sched_unpin();
1943 
1944 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1945 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1946 }
1947 
1948 /*
1949  * Deactivate the specified process's address space.
1950  */
1951 static void
1952 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1953 {
1954 	pmap_t pmap;
1955 
1956 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1957 
1958 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1959 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1960 
1961 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1962 	PCPU_SET(curpmap, NULL);
1963 }
1964 
1965 /*
1966  * Copy the range specified by src_addr/len
1967  * from the source map to the range dst_addr/len
1968  * in the destination map.
1969  *
1970  * This routine is only advisory and need not do anything.
1971  */
1972 static void
1973 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1974     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1975 {
1976 
1977 }
1978 
1979 /*
1980  * Set the physical protection on the specified range of this map as requested.
1981  */
1982 static void
1983 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1984     vm_prot_t prot)
1985 {
1986 	vm_offset_t va;
1987 	vm_page_t m;
1988 	pte_t *pte;
1989 
1990 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1991 		mmu_booke_remove(mmu, pmap, sva, eva);
1992 		return;
1993 	}
1994 
1995 	if (prot & VM_PROT_WRITE)
1996 		return;
1997 
1998 	PMAP_LOCK(pmap);
1999 	for (va = sva; va < eva; va += PAGE_SIZE) {
2000 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2001 			if (PTE_ISVALID(pte)) {
2002 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2003 
2004 				mtx_lock_spin(&tlbivax_mutex);
2005 				tlb_miss_lock();
2006 
2007 				/* Handle modified pages. */
2008 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2009 					vm_page_dirty(m);
2010 
2011 				tlb0_flush_entry(va);
2012 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2013 
2014 				tlb_miss_unlock();
2015 				mtx_unlock_spin(&tlbivax_mutex);
2016 			}
2017 		}
2018 	}
2019 	PMAP_UNLOCK(pmap);
2020 }
2021 
2022 /*
2023  * Clear the write and modified bits in each of the given page's mappings.
2024  */
2025 static void
2026 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2027 {
2028 	pv_entry_t pv;
2029 	pte_t *pte;
2030 
2031 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2032 	    ("mmu_booke_remove_write: page %p is not managed", m));
2033 
2034 	/*
2035 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2036 	 * set by another thread while the object is locked.  Thus,
2037 	 * if PGA_WRITEABLE is clear, no page table entries need updating.
2038 	 */
2039 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2040 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2041 		return;
2042 	rw_wlock(&pvh_global_lock);
2043 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2044 		PMAP_LOCK(pv->pv_pmap);
2045 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2046 			if (PTE_ISVALID(pte)) {
2047 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2048 
2049 				mtx_lock_spin(&tlbivax_mutex);
2050 				tlb_miss_lock();
2051 
2052 				/* Handle modified pages. */
2053 				if (PTE_ISMODIFIED(pte))
2054 					vm_page_dirty(m);
2055 
2056 				/* Flush mapping from TLB0. */
2057 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2058 
2059 				tlb_miss_unlock();
2060 				mtx_unlock_spin(&tlbivax_mutex);
2061 			}
2062 		}
2063 		PMAP_UNLOCK(pv->pv_pmap);
2064 	}
2065 	vm_page_aflag_clear(m, PGA_WRITEABLE);
2066 	rw_wunlock(&pvh_global_lock);
2067 }
2068 
2069 static void
2070 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2071 {
2072 	pte_t *pte;
2073 	pmap_t pmap;
2074 	vm_page_t m;
2075 	vm_offset_t addr;
2076 	vm_paddr_t pa = 0;
2077 	int active, valid;
2078 
2079 	va = trunc_page(va);
2080 	sz = round_page(sz);
2081 
2082 	rw_wlock(&pvh_global_lock);
2083 	pmap = PCPU_GET(curpmap);
2084 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2085 	while (sz > 0) {
2086 		PMAP_LOCK(pm);
2087 		pte = pte_find(mmu, pm, va);
2088 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2089 		if (valid)
2090 			pa = PTE_PA(pte);
2091 		PMAP_UNLOCK(pm);
2092 		if (valid) {
2093 			if (!active) {
2094 				/* Create a mapping in the active pmap. */
2095 				addr = 0;
2096 				m = PHYS_TO_VM_PAGE(pa);
2097 				PMAP_LOCK(pmap);
2098 				pte_enter(mmu, pmap, m, addr,
2099 				    PTE_SR | PTE_VALID | PTE_UR, FALSE);
2100 				__syncicache((void *)addr, PAGE_SIZE);
2101 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2102 				PMAP_UNLOCK(pmap);
2103 			} else
2104 				__syncicache((void *)va, PAGE_SIZE);
2105 		}
2106 		va += PAGE_SIZE;
2107 		sz -= PAGE_SIZE;
2108 	}
2109 	rw_wunlock(&pvh_global_lock);
2110 }
2111 
2112 /*
2113  * Atomically extract and hold the physical page with the given
2114  * pmap and virtual address pair if that mapping permits the given
2115  * protection.
2116  */
2117 static vm_page_t
2118 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2119     vm_prot_t prot)
2120 {
2121 	pte_t *pte;
2122 	vm_page_t m;
2123 	uint32_t pte_wbit;
2124 	vm_paddr_t pa;
2125 
2126 	m = NULL;
2127 	pa = 0;
2128 	PMAP_LOCK(pmap);
2129 retry:
2130 	pte = pte_find(mmu, pmap, va);
2131 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2132 		if (pmap == kernel_pmap)
2133 			pte_wbit = PTE_SW;
2134 		else
2135 			pte_wbit = PTE_UW;
2136 
2137 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2138 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2139 				goto retry;
2140 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2141 			vm_page_hold(m);
2142 		}
2143 	}
2144 
2145 	PA_UNLOCK_COND(pa);
2146 	PMAP_UNLOCK(pmap);
2147 	return (m);
2148 }
2149 
2150 /*
2151  * Initialize a vm_page's machine-dependent fields.
2152  */
2153 static void
2154 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2155 {
2156 
2157 	TAILQ_INIT(&m->md.pv_list);
2158 }
2159 
2160 /*
2161  * mmu_booke_zero_page_area zeros the specified hardware page by
2162  * mapping it into virtual memory and using bzero to clear
2163  * its contents.
2164  *
2165  * off and size must reside within a single page.
2166  */
2167 static void
2168 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2169 {
2170 	vm_offset_t va;
2171 
2172 	/* XXX KASSERT off and size are within a single page? */
2173 
2174 	mtx_lock(&zero_page_mutex);
2175 	va = zero_page_va;
2176 
2177 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2178 	bzero((caddr_t)va + off, size);
2179 	mmu_booke_kremove(mmu, va);
2180 
2181 	mtx_unlock(&zero_page_mutex);
2182 }
2183 
2184 /*
2185  * mmu_booke_zero_page zeros the specified hardware page.
2186  */
2187 static void
2188 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2189 {
2190 
2191 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2192 }
2193 
2194 /*
2195  * mmu_booke_copy_page copies the specified (machine independent) page by
2196  * mapping the page into virtual memory and using memcopy to copy the page,
2197  * one machine dependent page at a time.
2198  */
2199 static void
2200 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2201 {
2202 	vm_offset_t sva, dva;
2203 
2204 	sva = copy_page_src_va;
2205 	dva = copy_page_dst_va;
2206 
2207 	mtx_lock(&copy_page_mutex);
2208 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2209 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2210 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2211 	mmu_booke_kremove(mmu, dva);
2212 	mmu_booke_kremove(mmu, sva);
2213 	mtx_unlock(&copy_page_mutex);
2214 }
2215 
2216 static inline void
2217 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2218     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2219 {
2220 	void *a_cp, *b_cp;
2221 	vm_offset_t a_pg_offset, b_pg_offset;
2222 	int cnt;
2223 
2224 	mtx_lock(&copy_page_mutex);
2225 	while (xfersize > 0) {
2226 		a_pg_offset = a_offset & PAGE_MASK;
2227 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2228 		mmu_booke_kenter(mmu, copy_page_src_va,
2229 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2230 		a_cp = (char *)copy_page_src_va + a_pg_offset;
2231 		b_pg_offset = b_offset & PAGE_MASK;
2232 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2233 		mmu_booke_kenter(mmu, copy_page_dst_va,
2234 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2235 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
2236 		bcopy(a_cp, b_cp, cnt);
2237 		mmu_booke_kremove(mmu, copy_page_dst_va);
2238 		mmu_booke_kremove(mmu, copy_page_src_va);
2239 		a_offset += cnt;
2240 		b_offset += cnt;
2241 		xfersize -= cnt;
2242 	}
2243 	mtx_unlock(&copy_page_mutex);
2244 }
2245 
2246 /*
2247  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2248  * into virtual memory and using bzero to clear its contents. This is intended
2249  * to be called from the vm_pagezero process only and outside of Giant. No
2250  * lock is required.
2251  */
2252 static void
2253 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2254 {
2255 	vm_offset_t va;
2256 
2257 	va = zero_page_idle_va;
2258 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2259 	bzero((caddr_t)va, PAGE_SIZE);
2260 	mmu_booke_kremove(mmu, va);
2261 }
2262 
2263 /*
2264  * Return whether or not the specified physical page was modified
2265  * in any of physical maps.
2266  */
2267 static boolean_t
2268 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2269 {
2270 	pte_t *pte;
2271 	pv_entry_t pv;
2272 	boolean_t rv;
2273 
2274 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2275 	    ("mmu_booke_is_modified: page %p is not managed", m));
2276 	rv = FALSE;
2277 
2278 	/*
2279 	 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2280 	 * concurrently set while the object is locked.  Thus, if PGA_WRITEABLE
2281 	 * is clear, no PTEs can be modified.
2282 	 */
2283 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2284 	if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2285 		return (rv);
2286 	rw_wlock(&pvh_global_lock);
2287 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2288 		PMAP_LOCK(pv->pv_pmap);
2289 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2290 		    PTE_ISVALID(pte)) {
2291 			if (PTE_ISMODIFIED(pte))
2292 				rv = TRUE;
2293 		}
2294 		PMAP_UNLOCK(pv->pv_pmap);
2295 		if (rv)
2296 			break;
2297 	}
2298 	rw_wunlock(&pvh_global_lock);
2299 	return (rv);
2300 }
2301 
2302 /*
2303  * Return whether or not the specified virtual address is eligible
2304  * for prefault.
2305  */
2306 static boolean_t
2307 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2308 {
2309 
2310 	return (FALSE);
2311 }
2312 
2313 /*
2314  * Return whether or not the specified physical page was referenced
2315  * in any physical maps.
2316  */
2317 static boolean_t
2318 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2319 {
2320 	pte_t *pte;
2321 	pv_entry_t pv;
2322 	boolean_t rv;
2323 
2324 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2325 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2326 	rv = FALSE;
2327 	rw_wlock(&pvh_global_lock);
2328 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2329 		PMAP_LOCK(pv->pv_pmap);
2330 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2331 		    PTE_ISVALID(pte)) {
2332 			if (PTE_ISREFERENCED(pte))
2333 				rv = TRUE;
2334 		}
2335 		PMAP_UNLOCK(pv->pv_pmap);
2336 		if (rv)
2337 			break;
2338 	}
2339 	rw_wunlock(&pvh_global_lock);
2340 	return (rv);
2341 }
2342 
2343 /*
2344  * Clear the modify bits on the specified physical page.
2345  */
2346 static void
2347 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2348 {
2349 	pte_t *pte;
2350 	pv_entry_t pv;
2351 
2352 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2353 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2354 	VM_OBJECT_ASSERT_WLOCKED(m->object);
2355 	KASSERT(!vm_page_xbusied(m),
2356 	    ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2357 
2358 	/*
2359 	 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2360 	 * If the object containing the page is locked and the page is not
2361 	 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2362 	 */
2363 	if ((m->aflags & PGA_WRITEABLE) == 0)
2364 		return;
2365 	rw_wlock(&pvh_global_lock);
2366 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2367 		PMAP_LOCK(pv->pv_pmap);
2368 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2369 		    PTE_ISVALID(pte)) {
2370 			mtx_lock_spin(&tlbivax_mutex);
2371 			tlb_miss_lock();
2372 
2373 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2374 				tlb0_flush_entry(pv->pv_va);
2375 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2376 				    PTE_REFERENCED);
2377 			}
2378 
2379 			tlb_miss_unlock();
2380 			mtx_unlock_spin(&tlbivax_mutex);
2381 		}
2382 		PMAP_UNLOCK(pv->pv_pmap);
2383 	}
2384 	rw_wunlock(&pvh_global_lock);
2385 }
2386 
2387 /*
2388  * Return a count of reference bits for a page, clearing those bits.
2389  * It is not necessary for every reference bit to be cleared, but it
2390  * is necessary that 0 only be returned when there are truly no
2391  * reference bits set.
2392  *
2393  * XXX: The exact number of bits to check and clear is a matter that
2394  * should be tested and standardized at some point in the future for
2395  * optimal aging of shared pages.
2396  */
2397 static int
2398 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2399 {
2400 	pte_t *pte;
2401 	pv_entry_t pv;
2402 	int count;
2403 
2404 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2405 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2406 	count = 0;
2407 	rw_wlock(&pvh_global_lock);
2408 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2409 		PMAP_LOCK(pv->pv_pmap);
2410 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2411 		    PTE_ISVALID(pte)) {
2412 			if (PTE_ISREFERENCED(pte)) {
2413 				mtx_lock_spin(&tlbivax_mutex);
2414 				tlb_miss_lock();
2415 
2416 				tlb0_flush_entry(pv->pv_va);
2417 				pte->flags &= ~PTE_REFERENCED;
2418 
2419 				tlb_miss_unlock();
2420 				mtx_unlock_spin(&tlbivax_mutex);
2421 
2422 				if (++count > 4) {
2423 					PMAP_UNLOCK(pv->pv_pmap);
2424 					break;
2425 				}
2426 			}
2427 		}
2428 		PMAP_UNLOCK(pv->pv_pmap);
2429 	}
2430 	rw_wunlock(&pvh_global_lock);
2431 	return (count);
2432 }
2433 
2434 /*
2435  * Clear the wired attribute from the mappings for the specified range of
2436  * addresses in the given pmap.  Every valid mapping within that range must
2437  * have the wired attribute set.  In contrast, invalid mappings cannot have
2438  * the wired attribute set, so they are ignored.
2439  *
2440  * The wired attribute of the page table entry is not a hardware feature, so
2441  * there is no need to invalidate any TLB entries.
2442  */
2443 static void
2444 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2445 {
2446 	vm_offset_t va;
2447 	pte_t *pte;
2448 
2449 	PMAP_LOCK(pmap);
2450 	for (va = sva; va < eva; va += PAGE_SIZE) {
2451 		if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2452 		    PTE_ISVALID(pte)) {
2453 			if (!PTE_ISWIRED(pte))
2454 				panic("mmu_booke_unwire: pte %p isn't wired",
2455 				    pte);
2456 			pte->flags &= ~PTE_WIRED;
2457 			pmap->pm_stats.wired_count--;
2458 		}
2459 	}
2460 	PMAP_UNLOCK(pmap);
2461 
2462 }
2463 
2464 /*
2465  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2466  * page.  This count may be changed upwards or downwards in the future; it is
2467  * only necessary that true be returned for a small subset of pmaps for proper
2468  * page aging.
2469  */
2470 static boolean_t
2471 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2472 {
2473 	pv_entry_t pv;
2474 	int loops;
2475 	boolean_t rv;
2476 
2477 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2478 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2479 	loops = 0;
2480 	rv = FALSE;
2481 	rw_wlock(&pvh_global_lock);
2482 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2483 		if (pv->pv_pmap == pmap) {
2484 			rv = TRUE;
2485 			break;
2486 		}
2487 		if (++loops >= 16)
2488 			break;
2489 	}
2490 	rw_wunlock(&pvh_global_lock);
2491 	return (rv);
2492 }
2493 
2494 /*
2495  * Return the number of managed mappings to the given physical page that are
2496  * wired.
2497  */
2498 static int
2499 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2500 {
2501 	pv_entry_t pv;
2502 	pte_t *pte;
2503 	int count = 0;
2504 
2505 	if ((m->oflags & VPO_UNMANAGED) != 0)
2506 		return (count);
2507 	rw_wlock(&pvh_global_lock);
2508 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2509 		PMAP_LOCK(pv->pv_pmap);
2510 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2511 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2512 				count++;
2513 		PMAP_UNLOCK(pv->pv_pmap);
2514 	}
2515 	rw_wunlock(&pvh_global_lock);
2516 	return (count);
2517 }
2518 
2519 static int
2520 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2521 {
2522 	int i;
2523 	vm_offset_t va;
2524 
2525 	/*
2526 	 * This currently does not work for entries that
2527 	 * overlap TLB1 entries.
2528 	 */
2529 	for (i = 0; i < tlb1_idx; i ++) {
2530 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2531 			return (0);
2532 	}
2533 
2534 	return (EFAULT);
2535 }
2536 
2537 void
2538 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2539 {
2540 	vm_paddr_t ppa;
2541 	vm_offset_t ofs;
2542 	vm_size_t gran;
2543 
2544 	/* Minidumps are based on virtual memory addresses. */
2545 	if (do_minidump) {
2546 		*va = (void *)pa;
2547 		return;
2548 	}
2549 
2550 	/* Raw physical memory dumps don't have a virtual address. */
2551 	/* We always map a 256MB page at 256M. */
2552 	gran = 256 * 1024 * 1024;
2553 	ppa = pa & ~(gran - 1);
2554 	ofs = pa - ppa;
2555 	*va = (void *)gran;
2556 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2557 
2558 	if (sz > (gran - ofs))
2559 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2560 		    _TLB_ENTRY_IO);
2561 }
2562 
2563 void
2564 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2565 {
2566 	vm_paddr_t ppa;
2567 	vm_offset_t ofs;
2568 	vm_size_t gran;
2569 
2570 	/* Minidumps are based on virtual memory addresses. */
2571 	/* Nothing to do... */
2572 	if (do_minidump)
2573 		return;
2574 
2575 	/* Raw physical memory dumps don't have a virtual address. */
2576 	tlb1_idx--;
2577 	tlb1[tlb1_idx].mas1 = 0;
2578 	tlb1[tlb1_idx].mas2 = 0;
2579 	tlb1[tlb1_idx].mas3 = 0;
2580 	tlb1_write_entry(tlb1_idx);
2581 
2582 	gran = 256 * 1024 * 1024;
2583 	ppa = pa & ~(gran - 1);
2584 	ofs = pa - ppa;
2585 	if (sz > (gran - ofs)) {
2586 		tlb1_idx--;
2587 		tlb1[tlb1_idx].mas1 = 0;
2588 		tlb1[tlb1_idx].mas2 = 0;
2589 		tlb1[tlb1_idx].mas3 = 0;
2590 		tlb1_write_entry(tlb1_idx);
2591 	}
2592 }
2593 
2594 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2595 
2596 void
2597 mmu_booke_scan_init(mmu_t mmu)
2598 {
2599 	vm_offset_t va;
2600 	pte_t *pte;
2601 	int i;
2602 
2603 	if (!do_minidump) {
2604 		/* Initialize phys. segments for dumpsys(). */
2605 		memset(&dump_map, 0, sizeof(dump_map));
2606 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2607 		    &availmem_regions_sz);
2608 		for (i = 0; i < physmem_regions_sz; i++) {
2609 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2610 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2611 		}
2612 		return;
2613 	}
2614 
2615 	/* Virtual segments for minidumps: */
2616 	memset(&dump_map, 0, sizeof(dump_map));
2617 
2618 	/* 1st: kernel .data and .bss. */
2619 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2620 	dump_map[0].pa_size =
2621 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2622 
2623 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2624 	dump_map[1].pa_start = data_start;
2625 	dump_map[1].pa_size = data_end - data_start;
2626 
2627 	/* 3rd: kernel VM. */
2628 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2629 	/* Find start of next chunk (from va). */
2630 	while (va < virtual_end) {
2631 		/* Don't dump the buffer cache. */
2632 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2633 			va = kmi.buffer_eva;
2634 			continue;
2635 		}
2636 		pte = pte_find(mmu, kernel_pmap, va);
2637 		if (pte != NULL && PTE_ISVALID(pte))
2638 			break;
2639 		va += PAGE_SIZE;
2640 	}
2641 	if (va < virtual_end) {
2642 		dump_map[2].pa_start = va;
2643 		va += PAGE_SIZE;
2644 		/* Find last page in chunk. */
2645 		while (va < virtual_end) {
2646 			/* Don't run into the buffer cache. */
2647 			if (va == kmi.buffer_sva)
2648 				break;
2649 			pte = pte_find(mmu, kernel_pmap, va);
2650 			if (pte == NULL || !PTE_ISVALID(pte))
2651 				break;
2652 			va += PAGE_SIZE;
2653 		}
2654 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2655 	}
2656 }
2657 
2658 /*
2659  * Map a set of physical memory pages into the kernel virtual address space.
2660  * Return a pointer to where it is mapped. This routine is intended to be used
2661  * for mapping device memory, NOT real memory.
2662  */
2663 static void *
2664 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2665 {
2666 
2667 	return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2668 }
2669 
2670 static void *
2671 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2672 {
2673 	void *res;
2674 	uintptr_t va;
2675 	vm_size_t sz;
2676 	int i;
2677 
2678 	/*
2679 	 * Check if this is premapped in TLB1. Note: this should probably also
2680 	 * check whether a sequence of TLB1 entries exist that match the
2681 	 * requirement, but now only checks the easy case.
2682 	 */
2683 	if (ma == VM_MEMATTR_DEFAULT) {
2684 		for (i = 0; i < tlb1_idx; i++) {
2685 			if (!(tlb1[i].mas1 & MAS1_VALID))
2686 				continue;
2687 			if (pa >= tlb1[i].phys &&
2688 			    (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2689 				return (void *)(tlb1[i].virt +
2690 				    (pa - tlb1[i].phys));
2691 		}
2692 	}
2693 
2694 	size = roundup(size, PAGE_SIZE);
2695 
2696 	/*
2697 	 * We leave a hole for device direct mapping between the maximum user
2698 	 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2699 	 * devices are in there, just map them 1:1. If not, map them to the
2700 	 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2701 	 * addresses should be pulled from an allocator, but since we do not
2702 	 * ever free TLB1 entries, it is safe just to increment a counter.
2703 	 * Note that there isn't a lot of address space here (128 MB) and it
2704 	 * is not at all difficult to imagine running out, since that is a 4:1
2705 	 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2706 	 * mapped there.
2707 	 */
2708 	if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2709 	    (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2710 		va = pa;
2711 	else
2712 		va = atomic_fetchadd_int(&tlb1_map_base, size);
2713 	res = (void *)va;
2714 
2715 	do {
2716 		sz = 1 << (ilog2(size) & ~1);
2717 		if (bootverbose)
2718 			printf("Wiring VA=%x to PA=%x (size=%x), "
2719 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2720 		tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2721 		size -= sz;
2722 		pa += sz;
2723 		va += sz;
2724 	} while (size > 0);
2725 
2726 	return (res);
2727 }
2728 
2729 /*
2730  * 'Unmap' a range mapped by mmu_booke_mapdev().
2731  */
2732 static void
2733 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2734 {
2735 #ifdef SUPPORTS_SHRINKING_TLB1
2736 	vm_offset_t base, offset;
2737 
2738 	/*
2739 	 * Unmap only if this is inside kernel virtual space.
2740 	 */
2741 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2742 		base = trunc_page(va);
2743 		offset = va & PAGE_MASK;
2744 		size = roundup(offset + size, PAGE_SIZE);
2745 		kva_free(base, size);
2746 	}
2747 #endif
2748 }
2749 
2750 /*
2751  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2752  * specified pmap. This eliminates the blast of soft faults on process startup
2753  * and immediately after an mmap.
2754  */
2755 static void
2756 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2757     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2758 {
2759 
2760 	VM_OBJECT_ASSERT_WLOCKED(object);
2761 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2762 	    ("mmu_booke_object_init_pt: non-device object"));
2763 }
2764 
2765 /*
2766  * Perform the pmap work for mincore.
2767  */
2768 static int
2769 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2770     vm_paddr_t *locked_pa)
2771 {
2772 
2773 	/* XXX: this should be implemented at some point */
2774 	return (0);
2775 }
2776 
2777 /**************************************************************************/
2778 /* TID handling */
2779 /**************************************************************************/
2780 
2781 /*
2782  * Allocate a TID. If necessary, steal one from someone else.
2783  * The new TID is flushed from the TLB before returning.
2784  */
2785 static tlbtid_t
2786 tid_alloc(pmap_t pmap)
2787 {
2788 	tlbtid_t tid;
2789 	int thiscpu;
2790 
2791 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2792 
2793 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2794 
2795 	thiscpu = PCPU_GET(cpuid);
2796 
2797 	tid = PCPU_GET(tid_next);
2798 	if (tid > TID_MAX)
2799 		tid = TID_MIN;
2800 	PCPU_SET(tid_next, tid + 1);
2801 
2802 	/* If we are stealing TID then clear the relevant pmap's field */
2803 	if (tidbusy[thiscpu][tid] != NULL) {
2804 
2805 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2806 
2807 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2808 
2809 		/* Flush all entries from TLB0 matching this TID. */
2810 		tid_flush(tid);
2811 	}
2812 
2813 	tidbusy[thiscpu][tid] = pmap;
2814 	pmap->pm_tid[thiscpu] = tid;
2815 	__asm __volatile("msync; isync");
2816 
2817 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2818 	    PCPU_GET(tid_next));
2819 
2820 	return (tid);
2821 }
2822 
2823 /**************************************************************************/
2824 /* TLB0 handling */
2825 /**************************************************************************/
2826 
2827 static void
2828 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2829     uint32_t mas7)
2830 {
2831 	int as;
2832 	char desc[3];
2833 	tlbtid_t tid;
2834 	vm_size_t size;
2835 	unsigned int tsize;
2836 
2837 	desc[2] = '\0';
2838 	if (mas1 & MAS1_VALID)
2839 		desc[0] = 'V';
2840 	else
2841 		desc[0] = ' ';
2842 
2843 	if (mas1 & MAS1_IPROT)
2844 		desc[1] = 'P';
2845 	else
2846 		desc[1] = ' ';
2847 
2848 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2849 	tid = MAS1_GETTID(mas1);
2850 
2851 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2852 	size = 0;
2853 	if (tsize)
2854 		size = tsize2size(tsize);
2855 
2856 	debugf("%3d: (%s) [AS=%d] "
2857 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2858 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2859 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2860 }
2861 
2862 /* Convert TLB0 va and way number to tlb0[] table index. */
2863 static inline unsigned int
2864 tlb0_tableidx(vm_offset_t va, unsigned int way)
2865 {
2866 	unsigned int idx;
2867 
2868 	idx = (way * TLB0_ENTRIES_PER_WAY);
2869 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2870 	return (idx);
2871 }
2872 
2873 /*
2874  * Invalidate TLB0 entry.
2875  */
2876 static inline void
2877 tlb0_flush_entry(vm_offset_t va)
2878 {
2879 
2880 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2881 
2882 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2883 
2884 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2885 	__asm __volatile("isync; msync");
2886 	__asm __volatile("tlbsync; msync");
2887 
2888 	CTR1(KTR_PMAP, "%s: e", __func__);
2889 }
2890 
2891 /* Print out contents of the MAS registers for each TLB0 entry */
2892 void
2893 tlb0_print_tlbentries(void)
2894 {
2895 	uint32_t mas0, mas1, mas2, mas3, mas7;
2896 	int entryidx, way, idx;
2897 
2898 	debugf("TLB0 entries:\n");
2899 	for (way = 0; way < TLB0_WAYS; way ++)
2900 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2901 
2902 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2903 			mtspr(SPR_MAS0, mas0);
2904 			__asm __volatile("isync");
2905 
2906 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2907 			mtspr(SPR_MAS2, mas2);
2908 
2909 			__asm __volatile("isync; tlbre");
2910 
2911 			mas1 = mfspr(SPR_MAS1);
2912 			mas2 = mfspr(SPR_MAS2);
2913 			mas3 = mfspr(SPR_MAS3);
2914 			mas7 = mfspr(SPR_MAS7);
2915 
2916 			idx = tlb0_tableidx(mas2, way);
2917 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2918 		}
2919 }
2920 
2921 /**************************************************************************/
2922 /* TLB1 handling */
2923 /**************************************************************************/
2924 
2925 /*
2926  * TLB1 mapping notes:
2927  *
2928  * TLB1[0]	Kernel text and data.
2929  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2930  *		windows, other devices mappings.
2931  */
2932 
2933 /*
2934  * Write given entry to TLB1 hardware.
2935  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2936  */
2937 static void
2938 tlb1_write_entry(unsigned int idx)
2939 {
2940 	uint32_t mas0, mas7;
2941 
2942 	//debugf("tlb1_write_entry: s\n");
2943 
2944 	/* Clear high order RPN bits */
2945 	mas7 = 0;
2946 
2947 	/* Select entry */
2948 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2949 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2950 
2951 	mtspr(SPR_MAS0, mas0);
2952 	__asm __volatile("isync");
2953 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2954 	__asm __volatile("isync");
2955 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2956 	__asm __volatile("isync");
2957 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2958 	__asm __volatile("isync");
2959 	mtspr(SPR_MAS7, mas7);
2960 	__asm __volatile("isync; tlbwe; isync; msync");
2961 
2962 	//debugf("tlb1_write_entry: e\n");
2963 }
2964 
2965 /*
2966  * Return the largest uint value log such that 2^log <= num.
2967  */
2968 static unsigned int
2969 ilog2(unsigned int num)
2970 {
2971 	int lz;
2972 
2973 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2974 	return (31 - lz);
2975 }
2976 
2977 /*
2978  * Convert TLB TSIZE value to mapped region size.
2979  */
2980 static vm_size_t
2981 tsize2size(unsigned int tsize)
2982 {
2983 
2984 	/*
2985 	 * size = 4^tsize KB
2986 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2987 	 */
2988 
2989 	return ((1 << (2 * tsize)) * 1024);
2990 }
2991 
2992 /*
2993  * Convert region size (must be power of 4) to TLB TSIZE value.
2994  */
2995 static unsigned int
2996 size2tsize(vm_size_t size)
2997 {
2998 
2999 	return (ilog2(size) / 2 - 5);
3000 }
3001 
3002 /*
3003  * Register permanent kernel mapping in TLB1.
3004  *
3005  * Entries are created starting from index 0 (current free entry is
3006  * kept in tlb1_idx) and are not supposed to be invalidated.
3007  */
3008 static int
3009 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
3010     uint32_t flags)
3011 {
3012 	uint32_t ts, tid;
3013 	int tsize, index;
3014 
3015 	index = atomic_fetchadd_int(&tlb1_idx, 1);
3016 	if (index >= TLB1_ENTRIES) {
3017 		printf("tlb1_set_entry: TLB1 full!\n");
3018 		return (-1);
3019 	}
3020 
3021 	/* Convert size to TSIZE */
3022 	tsize = size2tsize(size);
3023 
3024 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3025 	/* XXX TS is hard coded to 0 for now as we only use single address space */
3026 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3027 
3028 	/*
3029 	 * Atomicity is preserved by the atomic increment above since nothing
3030 	 * is ever removed from tlb1.
3031 	 */
3032 
3033 	tlb1[index].phys = pa;
3034 	tlb1[index].virt = va;
3035 	tlb1[index].size = size;
3036 	tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3037 	tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3038 	tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3039 
3040 	/* Set supervisor RWX permission bits */
3041 	tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3042 
3043 	tlb1_write_entry(index);
3044 
3045 	/*
3046 	 * XXX in general TLB1 updates should be propagated between CPUs,
3047 	 * since current design assumes to have the same TLB1 set-up on all
3048 	 * cores.
3049 	 */
3050 	return (0);
3051 }
3052 
3053 /*
3054  * Map in contiguous RAM region into the TLB1 using maximum of
3055  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3056  *
3057  * If necessary round up last entry size and return total size
3058  * used by all allocated entries.
3059  */
3060 vm_size_t
3061 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3062 {
3063 	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3064 	vm_size_t mapped, pgsz, base, mask;
3065 	int idx, nents;
3066 
3067 	/* Round up to the next 1M */
3068 	size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3069 
3070 	mapped = 0;
3071 	idx = 0;
3072 	base = va;
3073 	pgsz = 64*1024*1024;
3074 	while (mapped < size) {
3075 		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3076 			while (pgsz > (size - mapped))
3077 				pgsz >>= 2;
3078 			pgs[idx++] = pgsz;
3079 			mapped += pgsz;
3080 		}
3081 
3082 		/* We under-map. Correct for this. */
3083 		if (mapped < size) {
3084 			while (pgs[idx - 1] == pgsz) {
3085 				idx--;
3086 				mapped -= pgsz;
3087 			}
3088 			/* XXX We may increase beyond out starting point. */
3089 			pgsz <<= 2;
3090 			pgs[idx++] = pgsz;
3091 			mapped += pgsz;
3092 		}
3093 	}
3094 
3095 	nents = idx;
3096 	mask = pgs[0] - 1;
3097 	/* Align address to the boundary */
3098 	if (va & mask) {
3099 		va = (va + mask) & ~mask;
3100 		pa = (pa + mask) & ~mask;
3101 	}
3102 
3103 	for (idx = 0; idx < nents; idx++) {
3104 		pgsz = pgs[idx];
3105 		debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3106 		tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3107 		pa += pgsz;
3108 		va += pgsz;
3109 	}
3110 
3111 	mapped = (va - base);
3112 	printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3113 	    mapped, mapped - size);
3114 	return (mapped);
3115 }
3116 
3117 /*
3118  * TLB1 initialization routine, to be called after the very first
3119  * assembler level setup done in locore.S.
3120  */
3121 void
3122 tlb1_init()
3123 {
3124 	uint32_t mas0, mas1, mas2, mas3;
3125 	uint32_t tsz;
3126 	u_int i;
3127 
3128 	if (bootinfo != NULL && bootinfo[0] != 1) {
3129 		tlb1_idx = *((uint16_t *)(bootinfo + 8));
3130 	} else
3131 		tlb1_idx = 1;
3132 
3133 	/* The first entry/entries are used to map the kernel. */
3134 	for (i = 0; i < tlb1_idx; i++) {
3135 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3136 		mtspr(SPR_MAS0, mas0);
3137 		__asm __volatile("isync; tlbre");
3138 
3139 		mas1 = mfspr(SPR_MAS1);
3140 		if ((mas1 & MAS1_VALID) == 0)
3141 			continue;
3142 
3143 		mas2 = mfspr(SPR_MAS2);
3144 		mas3 = mfspr(SPR_MAS3);
3145 
3146 		tlb1[i].mas1 = mas1;
3147 		tlb1[i].mas2 = mfspr(SPR_MAS2);
3148 		tlb1[i].mas3 = mas3;
3149 		tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3150 		tlb1[i].phys = mas3 & MAS3_RPN;
3151 
3152 		if (i == 0)
3153 			kernload = mas3 & MAS3_RPN;
3154 
3155 		tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3156 		tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3157 		kernsize += tlb1[i].size;
3158 	}
3159 
3160 #ifdef SMP
3161 	bp_ntlb1s = tlb1_idx;
3162 #endif
3163 
3164 	/* Purge the remaining entries */
3165 	for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3166 		tlb1_write_entry(i);
3167 
3168 	/* Setup TLB miss defaults */
3169 	set_mas4_defaults();
3170 }
3171 
3172 vm_offset_t
3173 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3174 {
3175 	vm_paddr_t pa_base;
3176 	vm_offset_t va, sz;
3177 	int i;
3178 
3179 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3180 
3181 	for (i = 0; i < tlb1_idx; i++) {
3182 		if (!(tlb1[i].mas1 & MAS1_VALID))
3183 			continue;
3184 		if (pa >= tlb1[i].phys && (pa + size) <=
3185 		    (tlb1[i].phys + tlb1[i].size))
3186 			return (tlb1[i].virt + (pa - tlb1[i].phys));
3187 	}
3188 
3189 	pa_base = trunc_page(pa);
3190 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
3191 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3192 	va = tlb1_map_base + (pa - pa_base);
3193 
3194 	do {
3195 		sz = 1 << (ilog2(size) & ~1);
3196 		tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3197 		size -= sz;
3198 		pa_base += sz;
3199 		tlb1_map_base += sz;
3200 	} while (size > 0);
3201 
3202 #ifdef SMP
3203 	bp_ntlb1s = tlb1_idx;
3204 #endif
3205 
3206 	return (va);
3207 }
3208 
3209 /*
3210  * Setup MAS4 defaults.
3211  * These values are loaded to MAS0-2 on a TLB miss.
3212  */
3213 static void
3214 set_mas4_defaults(void)
3215 {
3216 	uint32_t mas4;
3217 
3218 	/* Defaults: TLB0, PID0, TSIZED=4K */
3219 	mas4 = MAS4_TLBSELD0;
3220 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3221 #ifdef SMP
3222 	mas4 |= MAS4_MD;
3223 #endif
3224 	mtspr(SPR_MAS4, mas4);
3225 	__asm __volatile("isync");
3226 }
3227 
3228 /*
3229  * Print out contents of the MAS registers for each TLB1 entry
3230  */
3231 void
3232 tlb1_print_tlbentries(void)
3233 {
3234 	uint32_t mas0, mas1, mas2, mas3, mas7;
3235 	int i;
3236 
3237 	debugf("TLB1 entries:\n");
3238 	for (i = 0; i < TLB1_ENTRIES; i++) {
3239 
3240 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3241 		mtspr(SPR_MAS0, mas0);
3242 
3243 		__asm __volatile("isync; tlbre");
3244 
3245 		mas1 = mfspr(SPR_MAS1);
3246 		mas2 = mfspr(SPR_MAS2);
3247 		mas3 = mfspr(SPR_MAS3);
3248 		mas7 = mfspr(SPR_MAS7);
3249 
3250 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3251 	}
3252 }
3253 
3254 /*
3255  * Print out contents of the in-ram tlb1 table.
3256  */
3257 void
3258 tlb1_print_entries(void)
3259 {
3260 	int i;
3261 
3262 	debugf("tlb1[] table entries:\n");
3263 	for (i = 0; i < TLB1_ENTRIES; i++)
3264 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3265 }
3266 
3267 /*
3268  * Return 0 if the physical IO range is encompassed by one of the
3269  * the TLB1 entries, otherwise return related error code.
3270  */
3271 static int
3272 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3273 {
3274 	uint32_t prot;
3275 	vm_paddr_t pa_start;
3276 	vm_paddr_t pa_end;
3277 	unsigned int entry_tsize;
3278 	vm_size_t entry_size;
3279 
3280 	*va = (vm_offset_t)NULL;
3281 
3282 	/* Skip invalid entries */
3283 	if (!(tlb1[i].mas1 & MAS1_VALID))
3284 		return (EINVAL);
3285 
3286 	/*
3287 	 * The entry must be cache-inhibited, guarded, and r/w
3288 	 * so it can function as an i/o page
3289 	 */
3290 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3291 	if (prot != (MAS2_I | MAS2_G))
3292 		return (EPERM);
3293 
3294 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3295 	if (prot != (MAS3_SR | MAS3_SW))
3296 		return (EPERM);
3297 
3298 	/* The address should be within the entry range. */
3299 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3300 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3301 
3302 	entry_size = tsize2size(entry_tsize);
3303 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3304 	pa_end = pa_start + entry_size - 1;
3305 
3306 	if ((pa < pa_start) || ((pa + size) > pa_end))
3307 		return (ERANGE);
3308 
3309 	/* Return virtual address of this mapping. */
3310 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3311 	return (0);
3312 }
3313