xref: /freebsd/sys/powerpc/booke/pmap.c (revision a3cf0ef5a295c885c895fabfd56470c0d1db322d)
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/types.h>
55 #include <sys/param.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/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/smp.h>
67 #include <sys/vmmeter.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_pager.h>
78 #include <vm/uma.h>
79 
80 #include <machine/cpu.h>
81 #include <machine/pcb.h>
82 #include <machine/platform.h>
83 
84 #include <machine/tlb.h>
85 #include <machine/spr.h>
86 #include <machine/vmparam.h>
87 #include <machine/md_var.h>
88 #include <machine/mmuvar.h>
89 #include <machine/pmap.h>
90 #include <machine/pte.h>
91 
92 #include "mmu_if.h"
93 
94 #define DEBUG
95 #undef DEBUG
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 #include "opt_sched.h"
106 #ifndef SCHED_4BSD
107 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
108 #endif
109 extern struct mtx sched_lock;
110 
111 extern int dumpsys_minidump;
112 
113 extern unsigned char _etext[];
114 extern unsigned char _end[];
115 
116 /* Kernel physical load address. */
117 extern uint32_t kernload;
118 vm_offset_t kernstart;
119 vm_size_t kernsize;
120 
121 /* Message buffer and tables. */
122 static vm_offset_t data_start;
123 static vm_size_t data_end;
124 
125 /* Phys/avail memory regions. */
126 static struct mem_region *availmem_regions;
127 static int availmem_regions_sz;
128 static struct mem_region *physmem_regions;
129 static int physmem_regions_sz;
130 
131 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
132 static vm_offset_t zero_page_va;
133 static struct mtx zero_page_mutex;
134 
135 static struct mtx tlbivax_mutex;
136 
137 /*
138  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
139  * by idle thred only, no lock required.
140  */
141 static vm_offset_t zero_page_idle_va;
142 
143 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
144 static vm_offset_t copy_page_src_va;
145 static vm_offset_t copy_page_dst_va;
146 static struct mtx copy_page_mutex;
147 
148 /**************************************************************************/
149 /* PMAP */
150 /**************************************************************************/
151 
152 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
153     vm_prot_t, boolean_t);
154 
155 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
156 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
157 
158 /*
159  * If user pmap is processed with mmu_booke_remove and the resident count
160  * drops to 0, there are no more pages to remove, so we need not continue.
161  */
162 #define PMAP_REMOVE_DONE(pmap) \
163 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
164 
165 extern void tid_flush(tlbtid_t);
166 
167 /**************************************************************************/
168 /* TLB and TID handling */
169 /**************************************************************************/
170 
171 /* Translation ID busy table */
172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 
174 /*
175  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
176  * core revisions and should be read from h/w registers during early config.
177  */
178 uint32_t tlb0_entries;
179 uint32_t tlb0_ways;
180 uint32_t tlb0_entries_per_way;
181 
182 #define TLB0_ENTRIES		(tlb0_entries)
183 #define TLB0_WAYS		(tlb0_ways)
184 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
185 
186 #define TLB1_ENTRIES 16
187 
188 /* In-ram copy of the TLB1 */
189 static tlb_entry_t tlb1[TLB1_ENTRIES];
190 
191 /* Next free entry in the TLB1 */
192 static unsigned int tlb1_idx;
193 
194 static tlbtid_t tid_alloc(struct pmap *);
195 
196 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
197 
198 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
199 static void tlb1_write_entry(unsigned int);
200 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
201 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_offset_t, vm_size_t);
202 
203 static vm_size_t tsize2size(unsigned int);
204 static unsigned int size2tsize(vm_size_t);
205 static unsigned int ilog2(unsigned int);
206 
207 static void set_mas4_defaults(void);
208 
209 static inline void tlb0_flush_entry(vm_offset_t);
210 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
211 
212 /**************************************************************************/
213 /* Page table management */
214 /**************************************************************************/
215 
216 /* Data for the pv entry allocation mechanism */
217 static uma_zone_t pvzone;
218 static struct vm_object pvzone_obj;
219 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
220 
221 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
222 
223 #ifndef PMAP_SHPGPERPROC
224 #define PMAP_SHPGPERPROC	200
225 #endif
226 
227 static void ptbl_init(void);
228 static struct ptbl_buf *ptbl_buf_alloc(void);
229 static void ptbl_buf_free(struct ptbl_buf *);
230 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
231 
232 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
233 static void ptbl_free(mmu_t, pmap_t, unsigned int);
234 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
235 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
236 
237 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
238 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
239 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
240 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
241 
242 static pv_entry_t pv_alloc(void);
243 static void pv_free(pv_entry_t);
244 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
245 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
246 
247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
248 #define PTBL_BUFS		(128 * 16)
249 
250 struct ptbl_buf {
251 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
252 	vm_offset_t kva;		/* va of mapping */
253 };
254 
255 /* ptbl free list and a lock used for access synchronization. */
256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
257 static struct mtx ptbl_buf_freelist_lock;
258 
259 /* Base address of kva space allocated fot ptbl bufs. */
260 static vm_offset_t ptbl_buf_pool_vabase;
261 
262 /* Pointer to ptbl_buf structures. */
263 static struct ptbl_buf *ptbl_bufs;
264 
265 void pmap_bootstrap_ap(volatile uint32_t *);
266 
267 /*
268  * Kernel MMU interface
269  */
270 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
271 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
272 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
273 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
274     vm_size_t, vm_offset_t);
275 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
276 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
277     vm_prot_t, boolean_t);
278 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
279     vm_page_t, vm_prot_t);
280 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
281     vm_prot_t);
282 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
283 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
284     vm_prot_t);
285 static void		mmu_booke_init(mmu_t);
286 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
287 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
288 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
289 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
290 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
291     int);
292 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
293     vm_paddr_t *);
294 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
295     vm_object_t, vm_pindex_t, vm_size_t);
296 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
297 static void		mmu_booke_page_init(mmu_t, vm_page_t);
298 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
299 static void		mmu_booke_pinit(mmu_t, pmap_t);
300 static void		mmu_booke_pinit0(mmu_t, pmap_t);
301 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
302     vm_prot_t);
303 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
304 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
305 static void		mmu_booke_release(mmu_t, pmap_t);
306 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
307 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
308 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
309 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
310 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
311 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
312 static void		mmu_booke_activate(mmu_t, struct thread *);
313 static void		mmu_booke_deactivate(mmu_t, struct thread *);
314 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
315 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
316 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
317 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
318 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
319 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
320 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
321 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
322     vm_size_t);
323 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
324     vm_size_t, vm_size_t *);
325 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
326     vm_size_t, vm_offset_t);
327 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
328 
329 static mmu_method_t mmu_booke_methods[] = {
330 	/* pmap dispatcher interface */
331 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
332 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
333 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
334 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
335 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
336 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
337 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
338 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
339 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
340 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
341 	MMUMETHOD(mmu_init,		mmu_booke_init),
342 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
343 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
344 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
345 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
346 	MMUMETHOD(mmu_map,		mmu_booke_map),
347 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
348 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
349 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
350 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
351 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
352 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
353 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
354 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
355 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
356 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
357 	MMUMETHOD(mmu_release,		mmu_booke_release),
358 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
359 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
360 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
361 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
362 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
363 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
364 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
365 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
366 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
367 
368 	/* Internal interfaces */
369 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
370 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
371 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
372 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
373 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
374 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
375 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
376 
377 	/* dumpsys() support */
378 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
379 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
380 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
381 
382 	{ 0, 0 }
383 };
384 
385 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
386 
387 static inline void
388 tlb_miss_lock(void)
389 {
390 #ifdef SMP
391 	struct pcpu *pc;
392 
393 	if (!smp_started)
394 		return;
395 
396 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
397 		if (pc != pcpup) {
398 
399 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
400 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
401 
402 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
403 			    ("tlb_miss_lock: tried to lock self"));
404 
405 			tlb_lock(pc->pc_booke_tlb_lock);
406 
407 			CTR1(KTR_PMAP, "%s: locked", __func__);
408 		}
409 	}
410 #endif
411 }
412 
413 static inline void
414 tlb_miss_unlock(void)
415 {
416 #ifdef SMP
417 	struct pcpu *pc;
418 
419 	if (!smp_started)
420 		return;
421 
422 	SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
423 		if (pc != pcpup) {
424 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
425 			    __func__, pc->pc_cpuid);
426 
427 			tlb_unlock(pc->pc_booke_tlb_lock);
428 
429 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
430 		}
431 	}
432 #endif
433 }
434 
435 /* Return number of entries in TLB0. */
436 static __inline void
437 tlb0_get_tlbconf(void)
438 {
439 	uint32_t tlb0_cfg;
440 
441 	tlb0_cfg = mfspr(SPR_TLB0CFG);
442 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
443 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
444 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
445 }
446 
447 /* Initialize pool of kva ptbl buffers. */
448 static void
449 ptbl_init(void)
450 {
451 	int i;
452 
453 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
454 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
455 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
456 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
457 
458 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
459 	TAILQ_INIT(&ptbl_buf_freelist);
460 
461 	for (i = 0; i < PTBL_BUFS; i++) {
462 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
463 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
464 	}
465 }
466 
467 /* Get a ptbl_buf from the freelist. */
468 static struct ptbl_buf *
469 ptbl_buf_alloc(void)
470 {
471 	struct ptbl_buf *buf;
472 
473 	mtx_lock(&ptbl_buf_freelist_lock);
474 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
475 	if (buf != NULL)
476 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
477 	mtx_unlock(&ptbl_buf_freelist_lock);
478 
479 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
480 
481 	return (buf);
482 }
483 
484 /* Return ptbl buff to free pool. */
485 static void
486 ptbl_buf_free(struct ptbl_buf *buf)
487 {
488 
489 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
490 
491 	mtx_lock(&ptbl_buf_freelist_lock);
492 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
493 	mtx_unlock(&ptbl_buf_freelist_lock);
494 }
495 
496 /*
497  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
498  */
499 static void
500 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
501 {
502 	struct ptbl_buf *pbuf;
503 
504 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
505 
506 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
507 
508 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
509 		if (pbuf->kva == (vm_offset_t)ptbl) {
510 			/* Remove from pmap ptbl buf list. */
511 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
512 
513 			/* Free corresponding ptbl buf. */
514 			ptbl_buf_free(pbuf);
515 			break;
516 		}
517 }
518 
519 /* Allocate page table. */
520 static pte_t *
521 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
522 {
523 	vm_page_t mtbl[PTBL_PAGES];
524 	vm_page_t m;
525 	struct ptbl_buf *pbuf;
526 	unsigned int pidx;
527 	pte_t *ptbl;
528 	int i;
529 
530 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
531 	    (pmap == kernel_pmap), pdir_idx);
532 
533 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
534 	    ("ptbl_alloc: invalid pdir_idx"));
535 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
536 	    ("pte_alloc: valid ptbl entry exists!"));
537 
538 	pbuf = ptbl_buf_alloc();
539 	if (pbuf == NULL)
540 		panic("pte_alloc: couldn't alloc kernel virtual memory");
541 
542 	ptbl = (pte_t *)pbuf->kva;
543 
544 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
545 
546 	/* Allocate ptbl pages, this will sleep! */
547 	for (i = 0; i < PTBL_PAGES; i++) {
548 		pidx = (PTBL_PAGES * pdir_idx) + i;
549 		while ((m = vm_page_alloc(NULL, pidx,
550 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
551 
552 			PMAP_UNLOCK(pmap);
553 			vm_page_unlock_queues();
554 			VM_WAIT;
555 			vm_page_lock_queues();
556 			PMAP_LOCK(pmap);
557 		}
558 		mtbl[i] = m;
559 	}
560 
561 	/* Map allocated pages into kernel_pmap. */
562 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
563 
564 	/* Zero whole ptbl. */
565 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
566 
567 	/* Add pbuf to the pmap ptbl bufs list. */
568 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
569 
570 	return (ptbl);
571 }
572 
573 /* Free ptbl pages and invalidate pdir entry. */
574 static void
575 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
576 {
577 	pte_t *ptbl;
578 	vm_paddr_t pa;
579 	vm_offset_t va;
580 	vm_page_t m;
581 	int i;
582 
583 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
584 	    (pmap == kernel_pmap), pdir_idx);
585 
586 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
587 	    ("ptbl_free: invalid pdir_idx"));
588 
589 	ptbl = pmap->pm_pdir[pdir_idx];
590 
591 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
592 
593 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
594 
595 	/*
596 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
597 	 * don't attempt to look up the page tables we are releasing.
598 	 */
599 	mtx_lock_spin(&tlbivax_mutex);
600 	tlb_miss_lock();
601 
602 	pmap->pm_pdir[pdir_idx] = NULL;
603 
604 	tlb_miss_unlock();
605 	mtx_unlock_spin(&tlbivax_mutex);
606 
607 	for (i = 0; i < PTBL_PAGES; i++) {
608 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
609 		pa = pte_vatopa(mmu, kernel_pmap, va);
610 		m = PHYS_TO_VM_PAGE(pa);
611 		vm_page_free_zero(m);
612 		atomic_subtract_int(&cnt.v_wire_count, 1);
613 		mmu_booke_kremove(mmu, va);
614 	}
615 
616 	ptbl_free_pmap_ptbl(pmap, ptbl);
617 }
618 
619 /*
620  * Decrement ptbl pages hold count and attempt to free ptbl pages.
621  * Called when removing pte entry from ptbl.
622  *
623  * Return 1 if ptbl pages were freed.
624  */
625 static int
626 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
627 {
628 	pte_t *ptbl;
629 	vm_paddr_t pa;
630 	vm_page_t m;
631 	int i;
632 
633 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
634 	    (pmap == kernel_pmap), pdir_idx);
635 
636 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
637 	    ("ptbl_unhold: invalid pdir_idx"));
638 	KASSERT((pmap != kernel_pmap),
639 	    ("ptbl_unhold: unholding kernel ptbl!"));
640 
641 	ptbl = pmap->pm_pdir[pdir_idx];
642 
643 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
644 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
645 	    ("ptbl_unhold: non kva ptbl"));
646 
647 	/* decrement hold count */
648 	for (i = 0; i < PTBL_PAGES; i++) {
649 		pa = pte_vatopa(mmu, kernel_pmap,
650 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
651 		m = PHYS_TO_VM_PAGE(pa);
652 		m->wire_count--;
653 	}
654 
655 	/*
656 	 * Free ptbl pages if there are no pte etries in this ptbl.
657 	 * wire_count has the same value for all ptbl pages, so check the last
658 	 * page.
659 	 */
660 	if (m->wire_count == 0) {
661 		ptbl_free(mmu, pmap, pdir_idx);
662 
663 		//debugf("ptbl_unhold: e (freed ptbl)\n");
664 		return (1);
665 	}
666 
667 	return (0);
668 }
669 
670 /*
671  * Increment hold count for ptbl pages. This routine is used when a new pte
672  * entry is being inserted into the ptbl.
673  */
674 static void
675 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
676 {
677 	vm_paddr_t pa;
678 	pte_t *ptbl;
679 	vm_page_t m;
680 	int i;
681 
682 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
683 	    pdir_idx);
684 
685 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
686 	    ("ptbl_hold: invalid pdir_idx"));
687 	KASSERT((pmap != kernel_pmap),
688 	    ("ptbl_hold: holding kernel ptbl!"));
689 
690 	ptbl = pmap->pm_pdir[pdir_idx];
691 
692 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
693 
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 /* Allocate pv_entry structure. */
703 pv_entry_t
704 pv_alloc(void)
705 {
706 	pv_entry_t pv;
707 
708 	pv_entry_count++;
709 	if (pv_entry_count > pv_entry_high_water)
710 		pagedaemon_wakeup();
711 	pv = uma_zalloc(pvzone, M_NOWAIT);
712 
713 	return (pv);
714 }
715 
716 /* Free pv_entry structure. */
717 static __inline void
718 pv_free(pv_entry_t pve)
719 {
720 
721 	pv_entry_count--;
722 	uma_zfree(pvzone, pve);
723 }
724 
725 
726 /* Allocate and initialize pv_entry structure. */
727 static void
728 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
729 {
730 	pv_entry_t pve;
731 
732 	//int su = (pmap == kernel_pmap);
733 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
734 	//	(u_int32_t)pmap, va, (u_int32_t)m);
735 
736 	pve = pv_alloc();
737 	if (pve == NULL)
738 		panic("pv_insert: no pv entries!");
739 
740 	pve->pv_pmap = pmap;
741 	pve->pv_va = va;
742 
743 	/* add to pv_list */
744 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
745 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
746 
747 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
748 
749 	//debugf("pv_insert: e\n");
750 }
751 
752 /* Destroy pv entry. */
753 static void
754 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
755 {
756 	pv_entry_t pve;
757 
758 	//int su = (pmap == kernel_pmap);
759 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
760 
761 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
762 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
763 
764 	/* find pv entry */
765 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
766 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
767 			/* remove from pv_list */
768 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
769 			if (TAILQ_EMPTY(&m->md.pv_list))
770 				vm_page_flag_clear(m, PG_WRITEABLE);
771 
772 			/* free pv entry struct */
773 			pv_free(pve);
774 			break;
775 		}
776 	}
777 
778 	//debugf("pv_remove: e\n");
779 }
780 
781 /*
782  * Clean pte entry, try to free page table page if requested.
783  *
784  * Return 1 if ptbl pages were freed, otherwise return 0.
785  */
786 static int
787 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
788 {
789 	unsigned int pdir_idx = PDIR_IDX(va);
790 	unsigned int ptbl_idx = PTBL_IDX(va);
791 	vm_page_t m;
792 	pte_t *ptbl;
793 	pte_t *pte;
794 
795 	//int su = (pmap == kernel_pmap);
796 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
797 	//		su, (u_int32_t)pmap, va, flags);
798 
799 	ptbl = pmap->pm_pdir[pdir_idx];
800 	KASSERT(ptbl, ("pte_remove: null ptbl"));
801 
802 	pte = &ptbl[ptbl_idx];
803 
804 	if (pte == NULL || !PTE_ISVALID(pte))
805 		return (0);
806 
807 	if (PTE_ISWIRED(pte))
808 		pmap->pm_stats.wired_count--;
809 
810 	/* Handle managed entry. */
811 	if (PTE_ISMANAGED(pte)) {
812 		/* Get vm_page_t for mapped pte. */
813 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
814 
815 		if (PTE_ISMODIFIED(pte))
816 			vm_page_dirty(m);
817 
818 		if (PTE_ISREFERENCED(pte))
819 			vm_page_flag_set(m, PG_REFERENCED);
820 
821 		pv_remove(pmap, va, m);
822 	}
823 
824 	mtx_lock_spin(&tlbivax_mutex);
825 	tlb_miss_lock();
826 
827 	tlb0_flush_entry(va);
828 	pte->flags = 0;
829 	pte->rpn = 0;
830 
831 	tlb_miss_unlock();
832 	mtx_unlock_spin(&tlbivax_mutex);
833 
834 	pmap->pm_stats.resident_count--;
835 
836 	if (flags & PTBL_UNHOLD) {
837 		//debugf("pte_remove: e (unhold)\n");
838 		return (ptbl_unhold(mmu, pmap, pdir_idx));
839 	}
840 
841 	//debugf("pte_remove: e\n");
842 	return (0);
843 }
844 
845 /*
846  * Insert PTE for a given page and virtual address.
847  */
848 static void
849 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
850 {
851 	unsigned int pdir_idx = PDIR_IDX(va);
852 	unsigned int ptbl_idx = PTBL_IDX(va);
853 	pte_t *ptbl, *pte;
854 
855 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
856 	    pmap == kernel_pmap, pmap, va);
857 
858 	/* Get the page table pointer. */
859 	ptbl = pmap->pm_pdir[pdir_idx];
860 
861 	if (ptbl == NULL) {
862 		/* Allocate page table pages. */
863 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
864 	} else {
865 		/*
866 		 * Check if there is valid mapping for requested
867 		 * va, if there is, remove it.
868 		 */
869 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
870 		if (PTE_ISVALID(pte)) {
871 			pte_remove(mmu, pmap, va, PTBL_HOLD);
872 		} else {
873 			/*
874 			 * pte is not used, increment hold count
875 			 * for ptbl pages.
876 			 */
877 			if (pmap != kernel_pmap)
878 				ptbl_hold(mmu, pmap, pdir_idx);
879 		}
880 	}
881 
882 	/*
883 	 * Insert pv_entry into pv_list for mapped page if part of managed
884 	 * memory.
885 	 */
886         if ((m->flags & PG_FICTITIOUS) == 0) {
887 		if ((m->flags & PG_UNMANAGED) == 0) {
888 			flags |= PTE_MANAGED;
889 
890 			/* Create and insert pv entry. */
891 			pv_insert(pmap, va, m);
892 		}
893 	}
894 
895 	pmap->pm_stats.resident_count++;
896 
897 	mtx_lock_spin(&tlbivax_mutex);
898 	tlb_miss_lock();
899 
900 	tlb0_flush_entry(va);
901 	if (pmap->pm_pdir[pdir_idx] == NULL) {
902 		/*
903 		 * If we just allocated a new page table, hook it in
904 		 * the pdir.
905 		 */
906 		pmap->pm_pdir[pdir_idx] = ptbl;
907 	}
908 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
909 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
910 	pte->flags |= (PTE_VALID | flags);
911 
912 	tlb_miss_unlock();
913 	mtx_unlock_spin(&tlbivax_mutex);
914 }
915 
916 /* Return the pa for the given pmap/va. */
917 static vm_paddr_t
918 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
919 {
920 	vm_paddr_t pa = 0;
921 	pte_t *pte;
922 
923 	pte = pte_find(mmu, pmap, va);
924 	if ((pte != NULL) && PTE_ISVALID(pte))
925 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
926 	return (pa);
927 }
928 
929 /* Get a pointer to a PTE in a page table. */
930 static pte_t *
931 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
932 {
933 	unsigned int pdir_idx = PDIR_IDX(va);
934 	unsigned int ptbl_idx = PTBL_IDX(va);
935 
936 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
937 
938 	if (pmap->pm_pdir[pdir_idx])
939 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
940 
941 	return (NULL);
942 }
943 
944 /**************************************************************************/
945 /* PMAP related */
946 /**************************************************************************/
947 
948 /*
949  * This is called during e500_init, before the system is really initialized.
950  */
951 static void
952 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
953 {
954 	vm_offset_t phys_kernelend;
955 	struct mem_region *mp, *mp1;
956 	int cnt, i, j;
957 	u_int s, e, sz;
958 	u_int phys_avail_count;
959 	vm_size_t physsz, hwphyssz, kstack0_sz;
960 	vm_offset_t kernel_pdir, kstack0, va;
961 	vm_paddr_t kstack0_phys;
962 	void *dpcpu;
963 	pte_t *pte;
964 
965 	debugf("mmu_booke_bootstrap: entered\n");
966 
967 	/* Initialize invalidation mutex */
968 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
969 
970 	/* Read TLB0 size and associativity. */
971 	tlb0_get_tlbconf();
972 
973 	/* Align kernel start and end address (kernel image). */
974 	kernstart = trunc_page(start);
975 	data_start = round_page(kernelend);
976 	kernsize = data_start - kernstart;
977 
978 	data_end = data_start;
979 
980 	/* Allocate space for the message buffer. */
981 	msgbufp = (struct msgbuf *)data_end;
982 	data_end += MSGBUF_SIZE;
983 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
984 	    data_end);
985 
986 	data_end = round_page(data_end);
987 
988 	/* Allocate the dynamic per-cpu area. */
989 	dpcpu = (void *)data_end;
990 	data_end += DPCPU_SIZE;
991 	dpcpu_init(dpcpu, 0);
992 
993 	/* Allocate space for ptbl_bufs. */
994 	ptbl_bufs = (struct ptbl_buf *)data_end;
995 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
996 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
997 	    data_end);
998 
999 	data_end = round_page(data_end);
1000 
1001 	/* Allocate PTE tables for kernel KVA. */
1002 	kernel_pdir = data_end;
1003 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1004 	    PDIR_SIZE - 1) / PDIR_SIZE;
1005 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1006 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1007 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1008 
1009 	debugf(" data_end: 0x%08x\n", data_end);
1010 	if (data_end - kernstart > 0x1000000) {
1011 		data_end = (data_end + 0x3fffff) & ~0x3fffff;
1012 		tlb1_mapin_region(kernstart + 0x1000000,
1013 		    kernload + 0x1000000, data_end - kernstart - 0x1000000);
1014 	} else
1015 		data_end = (data_end + 0xffffff) & ~0xffffff;
1016 
1017 	debugf(" updated data_end: 0x%08x\n", data_end);
1018 
1019 	kernsize += data_end - data_start;
1020 
1021 	/*
1022 	 * Clear the structures - note we can only do it safely after the
1023 	 * possible additional TLB1 translations are in place (above) so that
1024 	 * all range up to the currently calculated 'data_end' is covered.
1025 	 */
1026 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1027 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1028 
1029 	/*******************************************************/
1030 	/* Set the start and end of kva. */
1031 	/*******************************************************/
1032 	virtual_avail = round_page(data_end);
1033 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1034 
1035 	/* Allocate KVA space for page zero/copy operations. */
1036 	zero_page_va = virtual_avail;
1037 	virtual_avail += PAGE_SIZE;
1038 	zero_page_idle_va = virtual_avail;
1039 	virtual_avail += PAGE_SIZE;
1040 	copy_page_src_va = virtual_avail;
1041 	virtual_avail += PAGE_SIZE;
1042 	copy_page_dst_va = virtual_avail;
1043 	virtual_avail += PAGE_SIZE;
1044 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1045 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1046 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1047 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1048 
1049 	/* Initialize page zero/copy mutexes. */
1050 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1051 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1052 
1053 	/* Allocate KVA space for ptbl bufs. */
1054 	ptbl_buf_pool_vabase = virtual_avail;
1055 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1056 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1057 	    ptbl_buf_pool_vabase, virtual_avail);
1058 
1059 	/* Calculate corresponding physical addresses for the kernel region. */
1060 	phys_kernelend = kernload + kernsize;
1061 	debugf("kernel image and allocated data:\n");
1062 	debugf(" kernload    = 0x%08x\n", kernload);
1063 	debugf(" kernstart   = 0x%08x\n", kernstart);
1064 	debugf(" kernsize    = 0x%08x\n", kernsize);
1065 
1066 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1067 		panic("mmu_booke_bootstrap: phys_avail too small");
1068 
1069 	/*
1070 	 * Remove kernel physical address range from avail regions list. Page
1071 	 * align all regions.  Non-page aligned memory isn't very interesting
1072 	 * to us.  Also, sort the entries for ascending addresses.
1073 	 */
1074 
1075 	/* Retrieve phys/avail mem regions */
1076 	mem_regions(&physmem_regions, &physmem_regions_sz,
1077 	    &availmem_regions, &availmem_regions_sz);
1078 	sz = 0;
1079 	cnt = availmem_regions_sz;
1080 	debugf("processing avail regions:\n");
1081 	for (mp = availmem_regions; mp->mr_size; mp++) {
1082 		s = mp->mr_start;
1083 		e = mp->mr_start + mp->mr_size;
1084 		debugf(" %08x-%08x -> ", s, e);
1085 		/* Check whether this region holds all of the kernel. */
1086 		if (s < kernload && e > phys_kernelend) {
1087 			availmem_regions[cnt].mr_start = phys_kernelend;
1088 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1089 			e = kernload;
1090 		}
1091 		/* Look whether this regions starts within the kernel. */
1092 		if (s >= kernload && s < phys_kernelend) {
1093 			if (e <= phys_kernelend)
1094 				goto empty;
1095 			s = phys_kernelend;
1096 		}
1097 		/* Now look whether this region ends within the kernel. */
1098 		if (e > kernload && e <= phys_kernelend) {
1099 			if (s >= kernload)
1100 				goto empty;
1101 			e = kernload;
1102 		}
1103 		/* Now page align the start and size of the region. */
1104 		s = round_page(s);
1105 		e = trunc_page(e);
1106 		if (e < s)
1107 			e = s;
1108 		sz = e - s;
1109 		debugf("%08x-%08x = %x\n", s, e, sz);
1110 
1111 		/* Check whether some memory is left here. */
1112 		if (sz == 0) {
1113 		empty:
1114 			memmove(mp, mp + 1,
1115 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1116 			cnt--;
1117 			mp--;
1118 			continue;
1119 		}
1120 
1121 		/* Do an insertion sort. */
1122 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1123 			if (s < mp1->mr_start)
1124 				break;
1125 		if (mp1 < mp) {
1126 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1127 			mp1->mr_start = s;
1128 			mp1->mr_size = sz;
1129 		} else {
1130 			mp->mr_start = s;
1131 			mp->mr_size = sz;
1132 		}
1133 	}
1134 	availmem_regions_sz = cnt;
1135 
1136 	/*******************************************************/
1137 	/* Steal physical memory for kernel stack from the end */
1138 	/* of the first avail region                           */
1139 	/*******************************************************/
1140 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1141 	kstack0_phys = availmem_regions[0].mr_start +
1142 	    availmem_regions[0].mr_size;
1143 	kstack0_phys -= kstack0_sz;
1144 	availmem_regions[0].mr_size -= kstack0_sz;
1145 
1146 	/*******************************************************/
1147 	/* Fill in phys_avail table, based on availmem_regions */
1148 	/*******************************************************/
1149 	phys_avail_count = 0;
1150 	physsz = 0;
1151 	hwphyssz = 0;
1152 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1153 
1154 	debugf("fill in phys_avail:\n");
1155 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1156 
1157 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1158 		    availmem_regions[i].mr_start,
1159 		    availmem_regions[i].mr_start +
1160 		        availmem_regions[i].mr_size,
1161 		    availmem_regions[i].mr_size);
1162 
1163 		if (hwphyssz != 0 &&
1164 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1165 			debugf(" hw.physmem adjust\n");
1166 			if (physsz < hwphyssz) {
1167 				phys_avail[j] = availmem_regions[i].mr_start;
1168 				phys_avail[j + 1] =
1169 				    availmem_regions[i].mr_start +
1170 				    hwphyssz - physsz;
1171 				physsz = hwphyssz;
1172 				phys_avail_count++;
1173 			}
1174 			break;
1175 		}
1176 
1177 		phys_avail[j] = availmem_regions[i].mr_start;
1178 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1179 		    availmem_regions[i].mr_size;
1180 		phys_avail_count++;
1181 		physsz += availmem_regions[i].mr_size;
1182 	}
1183 	physmem = btoc(physsz);
1184 
1185 	/* Calculate the last available physical address. */
1186 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1187 		;
1188 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1189 
1190 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1191 	debugf("phys_avail_count = %d\n", phys_avail_count);
1192 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1193 	    physmem);
1194 
1195 	/*******************************************************/
1196 	/* Initialize (statically allocated) kernel pmap. */
1197 	/*******************************************************/
1198 	PMAP_LOCK_INIT(kernel_pmap);
1199 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1200 
1201 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1202 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1203 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1204 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1205 
1206 	/* Initialize kernel pdir */
1207 	for (i = 0; i < kernel_ptbls; i++)
1208 		kernel_pmap->pm_pdir[kptbl_min + i] =
1209 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1210 
1211 	for (i = 0; i < MAXCPU; i++) {
1212 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1213 
1214 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1215 		tidbusy[i][0] = kernel_pmap;
1216 	}
1217 
1218 	/*
1219 	 * Fill in PTEs covering kernel code and data. They are not required
1220 	 * for address translation, as this area is covered by static TLB1
1221 	 * entries, but for pte_vatopa() to work correctly with kernel area
1222 	 * addresses.
1223 	 */
1224 	for (va = KERNBASE; va < data_end; va += PAGE_SIZE) {
1225 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1226 		pte->rpn = kernload + (va - KERNBASE);
1227 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1228 		    PTE_VALID;
1229 	}
1230 	/* Mark kernel_pmap active on all CPUs */
1231 	kernel_pmap->pm_active = ~0;
1232 
1233 	/*******************************************************/
1234 	/* Final setup */
1235 	/*******************************************************/
1236 
1237 	/* Enter kstack0 into kernel map, provide guard page */
1238 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1239 	thread0.td_kstack = kstack0;
1240 	thread0.td_kstack_pages = KSTACK_PAGES;
1241 
1242 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1243 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1244 	    kstack0_phys, kstack0_phys + kstack0_sz);
1245 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1246 
1247 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1248 	for (i = 0; i < KSTACK_PAGES; i++) {
1249 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1250 		kstack0 += PAGE_SIZE;
1251 		kstack0_phys += PAGE_SIZE;
1252 	}
1253 
1254 	debugf("virtual_avail = %08x\n", virtual_avail);
1255 	debugf("virtual_end   = %08x\n", virtual_end);
1256 
1257 	debugf("mmu_booke_bootstrap: exit\n");
1258 }
1259 
1260 void
1261 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1262 {
1263 	int i;
1264 
1265 	/*
1266 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1267 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1268 	 * these values directly to (re)program AP's TLB1 hardware.
1269 	 */
1270 	for (i = 0; i < tlb1_idx; i ++) {
1271 		/* Skip invalid entries */
1272 		if (!(tlb1[i].mas1 & MAS1_VALID))
1273 			continue;
1274 
1275 		tlb1_write_entry(i);
1276 	}
1277 
1278 	set_mas4_defaults();
1279 }
1280 
1281 /*
1282  * Get the physical page address for the given pmap/virtual address.
1283  */
1284 static vm_paddr_t
1285 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1286 {
1287 	vm_paddr_t pa;
1288 
1289 	PMAP_LOCK(pmap);
1290 	pa = pte_vatopa(mmu, pmap, va);
1291 	PMAP_UNLOCK(pmap);
1292 
1293 	return (pa);
1294 }
1295 
1296 /*
1297  * Extract the physical page address associated with the given
1298  * kernel virtual address.
1299  */
1300 static vm_paddr_t
1301 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1302 {
1303 
1304 	return (pte_vatopa(mmu, kernel_pmap, va));
1305 }
1306 
1307 /*
1308  * Initialize the pmap module.
1309  * Called by vm_init, to initialize any structures that the pmap
1310  * system needs to map virtual memory.
1311  */
1312 static void
1313 mmu_booke_init(mmu_t mmu)
1314 {
1315 	int shpgperproc = PMAP_SHPGPERPROC;
1316 
1317 	/*
1318 	 * Initialize the address space (zone) for the pv entries.  Set a
1319 	 * high water mark so that the system can recover from excessive
1320 	 * numbers of pv entries.
1321 	 */
1322 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1323 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1324 
1325 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1326 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1327 
1328 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1329 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1330 
1331 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1332 
1333 	/* Pre-fill pvzone with initial number of pv entries. */
1334 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1335 
1336 	/* Initialize ptbl allocation. */
1337 	ptbl_init();
1338 }
1339 
1340 /*
1341  * Map a list of wired pages into kernel virtual address space.  This is
1342  * intended for temporary mappings which do not need page modification or
1343  * references recorded.  Existing mappings in the region are overwritten.
1344  */
1345 static void
1346 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1347 {
1348 	vm_offset_t va;
1349 
1350 	va = sva;
1351 	while (count-- > 0) {
1352 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1353 		va += PAGE_SIZE;
1354 		m++;
1355 	}
1356 }
1357 
1358 /*
1359  * Remove page mappings from kernel virtual address space.  Intended for
1360  * temporary mappings entered by mmu_booke_qenter.
1361  */
1362 static void
1363 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1364 {
1365 	vm_offset_t va;
1366 
1367 	va = sva;
1368 	while (count-- > 0) {
1369 		mmu_booke_kremove(mmu, va);
1370 		va += PAGE_SIZE;
1371 	}
1372 }
1373 
1374 /*
1375  * Map a wired page into kernel virtual address space.
1376  */
1377 static void
1378 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1379 {
1380 	unsigned int pdir_idx = PDIR_IDX(va);
1381 	unsigned int ptbl_idx = PTBL_IDX(va);
1382 	uint32_t flags;
1383 	pte_t *pte;
1384 
1385 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1386 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1387 
1388 	flags = 0;
1389 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1390 	flags |= PTE_M;
1391 
1392 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1393 
1394 	mtx_lock_spin(&tlbivax_mutex);
1395 	tlb_miss_lock();
1396 
1397 	if (PTE_ISVALID(pte)) {
1398 
1399 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1400 
1401 		/* Flush entry from TLB0 */
1402 		tlb0_flush_entry(va);
1403 	}
1404 
1405 	pte->rpn = pa & ~PTE_PA_MASK;
1406 	pte->flags = flags;
1407 
1408 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1409 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1410 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1411 
1412 	/* Flush the real memory from the instruction cache. */
1413 	if ((flags & (PTE_I | PTE_G)) == 0) {
1414 		__syncicache((void *)va, PAGE_SIZE);
1415 	}
1416 
1417 	tlb_miss_unlock();
1418 	mtx_unlock_spin(&tlbivax_mutex);
1419 }
1420 
1421 /*
1422  * Remove a page from kernel page table.
1423  */
1424 static void
1425 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1426 {
1427 	unsigned int pdir_idx = PDIR_IDX(va);
1428 	unsigned int ptbl_idx = PTBL_IDX(va);
1429 	pte_t *pte;
1430 
1431 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1432 
1433 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1434 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1435 	    ("mmu_booke_kremove: invalid va"));
1436 
1437 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1438 
1439 	if (!PTE_ISVALID(pte)) {
1440 
1441 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1442 
1443 		return;
1444 	}
1445 
1446 	mtx_lock_spin(&tlbivax_mutex);
1447 	tlb_miss_lock();
1448 
1449 	/* Invalidate entry in TLB0, update PTE. */
1450 	tlb0_flush_entry(va);
1451 	pte->flags = 0;
1452 	pte->rpn = 0;
1453 
1454 	tlb_miss_unlock();
1455 	mtx_unlock_spin(&tlbivax_mutex);
1456 }
1457 
1458 /*
1459  * Initialize pmap associated with process 0.
1460  */
1461 static void
1462 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1463 {
1464 
1465 	mmu_booke_pinit(mmu, pmap);
1466 	PCPU_SET(curpmap, pmap);
1467 }
1468 
1469 /*
1470  * Initialize a preallocated and zeroed pmap structure,
1471  * such as one in a vmspace structure.
1472  */
1473 static void
1474 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1475 {
1476 	int i;
1477 
1478 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1479 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1480 
1481 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1482 
1483 	PMAP_LOCK_INIT(pmap);
1484 	for (i = 0; i < MAXCPU; i++)
1485 		pmap->pm_tid[i] = TID_NONE;
1486 	pmap->pm_active = 0;
1487 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1488 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1489 	TAILQ_INIT(&pmap->pm_ptbl_list);
1490 }
1491 
1492 /*
1493  * Release any resources held by the given physical map.
1494  * Called when a pmap initialized by mmu_booke_pinit is being released.
1495  * Should only be called if the map contains no valid mappings.
1496  */
1497 static void
1498 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1499 {
1500 
1501 	KASSERT(pmap->pm_stats.resident_count == 0,
1502 	    ("pmap_release: pmap resident count %ld != 0",
1503 	    pmap->pm_stats.resident_count));
1504 
1505 	PMAP_LOCK_DESTROY(pmap);
1506 }
1507 
1508 /*
1509  * Insert the given physical page at the specified virtual address in the
1510  * target physical map with the protection requested. If specified the page
1511  * will be wired down.
1512  */
1513 static void
1514 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1515     vm_prot_t prot, boolean_t wired)
1516 {
1517 
1518 	vm_page_lock_queues();
1519 	PMAP_LOCK(pmap);
1520 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1521 	vm_page_unlock_queues();
1522 	PMAP_UNLOCK(pmap);
1523 }
1524 
1525 static void
1526 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1527     vm_prot_t prot, boolean_t wired)
1528 {
1529 	pte_t *pte;
1530 	vm_paddr_t pa;
1531 	uint32_t flags;
1532 	int su, sync;
1533 
1534 	pa = VM_PAGE_TO_PHYS(m);
1535 	su = (pmap == kernel_pmap);
1536 	sync = 0;
1537 
1538 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1539 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1540 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1541 	//		(u_int32_t)m, va, pa, prot, wired);
1542 
1543 	if (su) {
1544 		KASSERT(((va >= virtual_avail) &&
1545 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1546 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1547 	} else {
1548 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1549 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1550 	}
1551 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1552 	    (m->oflags & VPO_BUSY) != 0 || VM_OBJECT_LOCKED(m->object),
1553 	    ("mmu_booke_enter_locked: page %p is not busy", m));
1554 
1555 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1556 
1557 	/*
1558 	 * If there is an existing mapping, and the physical address has not
1559 	 * changed, must be protection or wiring change.
1560 	 */
1561 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1562 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1563 
1564 		/*
1565 		 * Before actually updating pte->flags we calculate and
1566 		 * prepare its new value in a helper var.
1567 		 */
1568 		flags = pte->flags;
1569 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1570 
1571 		/* Wiring change, just update stats. */
1572 		if (wired) {
1573 			if (!PTE_ISWIRED(pte)) {
1574 				flags |= PTE_WIRED;
1575 				pmap->pm_stats.wired_count++;
1576 			}
1577 		} else {
1578 			if (PTE_ISWIRED(pte)) {
1579 				flags &= ~PTE_WIRED;
1580 				pmap->pm_stats.wired_count--;
1581 			}
1582 		}
1583 
1584 		if (prot & VM_PROT_WRITE) {
1585 			/* Add write permissions. */
1586 			flags |= PTE_SW;
1587 			if (!su)
1588 				flags |= PTE_UW;
1589 
1590 			if ((flags & PTE_MANAGED) != 0)
1591 				vm_page_flag_set(m, PG_WRITEABLE);
1592 		} else {
1593 			/* Handle modified pages, sense modify status. */
1594 
1595 			/*
1596 			 * The PTE_MODIFIED flag could be set by underlying
1597 			 * TLB misses since we last read it (above), possibly
1598 			 * other CPUs could update it so we check in the PTE
1599 			 * directly rather than rely on that saved local flags
1600 			 * copy.
1601 			 */
1602 			if (PTE_ISMODIFIED(pte))
1603 				vm_page_dirty(m);
1604 		}
1605 
1606 		if (prot & VM_PROT_EXECUTE) {
1607 			flags |= PTE_SX;
1608 			if (!su)
1609 				flags |= PTE_UX;
1610 
1611 			/*
1612 			 * Check existing flags for execute permissions: if we
1613 			 * are turning execute permissions on, icache should
1614 			 * be flushed.
1615 			 */
1616 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1617 				sync++;
1618 		}
1619 
1620 		flags &= ~PTE_REFERENCED;
1621 
1622 		/*
1623 		 * The new flags value is all calculated -- only now actually
1624 		 * update the PTE.
1625 		 */
1626 		mtx_lock_spin(&tlbivax_mutex);
1627 		tlb_miss_lock();
1628 
1629 		tlb0_flush_entry(va);
1630 		pte->flags = flags;
1631 
1632 		tlb_miss_unlock();
1633 		mtx_unlock_spin(&tlbivax_mutex);
1634 
1635 	} else {
1636 		/*
1637 		 * If there is an existing mapping, but it's for a different
1638 		 * physical address, pte_enter() will delete the old mapping.
1639 		 */
1640 		//if ((pte != NULL) && PTE_ISVALID(pte))
1641 		//	debugf("mmu_booke_enter_locked: replace\n");
1642 		//else
1643 		//	debugf("mmu_booke_enter_locked: new\n");
1644 
1645 		/* Now set up the flags and install the new mapping. */
1646 		flags = (PTE_SR | PTE_VALID);
1647 		flags |= PTE_M;
1648 
1649 		if (!su)
1650 			flags |= PTE_UR;
1651 
1652 		if (prot & VM_PROT_WRITE) {
1653 			flags |= PTE_SW;
1654 			if (!su)
1655 				flags |= PTE_UW;
1656 
1657 			if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0)
1658 				vm_page_flag_set(m, PG_WRITEABLE);
1659 		}
1660 
1661 		if (prot & VM_PROT_EXECUTE) {
1662 			flags |= PTE_SX;
1663 			if (!su)
1664 				flags |= PTE_UX;
1665 		}
1666 
1667 		/* If its wired update stats. */
1668 		if (wired) {
1669 			pmap->pm_stats.wired_count++;
1670 			flags |= PTE_WIRED;
1671 		}
1672 
1673 		pte_enter(mmu, pmap, m, va, flags);
1674 
1675 		/* Flush the real memory from the instruction cache. */
1676 		if (prot & VM_PROT_EXECUTE)
1677 			sync++;
1678 	}
1679 
1680 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1681 		__syncicache((void *)va, PAGE_SIZE);
1682 		sync = 0;
1683 	}
1684 }
1685 
1686 /*
1687  * Maps a sequence of resident pages belonging to the same object.
1688  * The sequence begins with the given page m_start.  This page is
1689  * mapped at the given virtual address start.  Each subsequent page is
1690  * mapped at a virtual address that is offset from start by the same
1691  * amount as the page is offset from m_start within the object.  The
1692  * last page in the sequence is the page with the largest offset from
1693  * m_start that can be mapped at a virtual address less than the given
1694  * virtual address end.  Not every virtual page between start and end
1695  * is mapped; only those for which a resident page exists with the
1696  * corresponding offset from m_start are mapped.
1697  */
1698 static void
1699 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1700     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1701 {
1702 	vm_page_t m;
1703 	vm_pindex_t diff, psize;
1704 
1705 	psize = atop(end - start);
1706 	m = m_start;
1707 	vm_page_lock_queues();
1708 	PMAP_LOCK(pmap);
1709 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1710 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1711 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1712 		m = TAILQ_NEXT(m, listq);
1713 	}
1714 	vm_page_unlock_queues();
1715 	PMAP_UNLOCK(pmap);
1716 }
1717 
1718 static void
1719 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1720     vm_prot_t prot)
1721 {
1722 
1723 	vm_page_lock_queues();
1724 	PMAP_LOCK(pmap);
1725 	mmu_booke_enter_locked(mmu, pmap, va, m,
1726 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1727 	vm_page_unlock_queues();
1728 	PMAP_UNLOCK(pmap);
1729 }
1730 
1731 /*
1732  * Remove the given range of addresses from the specified map.
1733  *
1734  * It is assumed that the start and end are properly rounded to the page size.
1735  */
1736 static void
1737 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1738 {
1739 	pte_t *pte;
1740 	uint8_t hold_flag;
1741 
1742 	int su = (pmap == kernel_pmap);
1743 
1744 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1745 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1746 
1747 	if (su) {
1748 		KASSERT(((va >= virtual_avail) &&
1749 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1750 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1751 	} else {
1752 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1753 		    ("mmu_booke_remove: user pmap, non user va"));
1754 	}
1755 
1756 	if (PMAP_REMOVE_DONE(pmap)) {
1757 		//debugf("mmu_booke_remove: e (empty)\n");
1758 		return;
1759 	}
1760 
1761 	hold_flag = PTBL_HOLD_FLAG(pmap);
1762 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1763 
1764 	vm_page_lock_queues();
1765 	PMAP_LOCK(pmap);
1766 	for (; va < endva; va += PAGE_SIZE) {
1767 		pte = pte_find(mmu, pmap, va);
1768 		if ((pte != NULL) && PTE_ISVALID(pte))
1769 			pte_remove(mmu, pmap, va, hold_flag);
1770 	}
1771 	PMAP_UNLOCK(pmap);
1772 	vm_page_unlock_queues();
1773 
1774 	//debugf("mmu_booke_remove: e\n");
1775 }
1776 
1777 /*
1778  * Remove physical page from all pmaps in which it resides.
1779  */
1780 static void
1781 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1782 {
1783 	pv_entry_t pv, pvn;
1784 	uint8_t hold_flag;
1785 
1786 	vm_page_lock_queues();
1787 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1788 		pvn = TAILQ_NEXT(pv, pv_link);
1789 
1790 		PMAP_LOCK(pv->pv_pmap);
1791 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1792 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1793 		PMAP_UNLOCK(pv->pv_pmap);
1794 	}
1795 	vm_page_flag_clear(m, PG_WRITEABLE);
1796 	vm_page_unlock_queues();
1797 }
1798 
1799 /*
1800  * Map a range of physical addresses into kernel virtual address space.
1801  */
1802 static vm_offset_t
1803 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1804     vm_offset_t pa_end, int prot)
1805 {
1806 	vm_offset_t sva = *virt;
1807 	vm_offset_t va = sva;
1808 
1809 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1810 	//		sva, pa_start, pa_end);
1811 
1812 	while (pa_start < pa_end) {
1813 		mmu_booke_kenter(mmu, va, pa_start);
1814 		va += PAGE_SIZE;
1815 		pa_start += PAGE_SIZE;
1816 	}
1817 	*virt = va;
1818 
1819 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1820 	return (sva);
1821 }
1822 
1823 /*
1824  * The pmap must be activated before it's address space can be accessed in any
1825  * way.
1826  */
1827 static void
1828 mmu_booke_activate(mmu_t mmu, struct thread *td)
1829 {
1830 	pmap_t pmap;
1831 
1832 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1833 
1834 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1835 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1836 
1837 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1838 
1839 	mtx_lock_spin(&sched_lock);
1840 
1841 	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
1842 	PCPU_SET(curpmap, pmap);
1843 
1844 	if (pmap->pm_tid[PCPU_GET(cpuid)] == TID_NONE)
1845 		tid_alloc(pmap);
1846 
1847 	/* Load PID0 register with pmap tid value. */
1848 	mtspr(SPR_PID0, pmap->pm_tid[PCPU_GET(cpuid)]);
1849 	__asm __volatile("isync");
1850 
1851 	mtx_unlock_spin(&sched_lock);
1852 
1853 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1854 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1855 }
1856 
1857 /*
1858  * Deactivate the specified process's address space.
1859  */
1860 static void
1861 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1862 {
1863 	pmap_t pmap;
1864 
1865 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1866 
1867 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1868 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1869 
1870 	atomic_clear_int(&pmap->pm_active, PCPU_GET(cpumask));
1871 	PCPU_SET(curpmap, NULL);
1872 }
1873 
1874 /*
1875  * Copy the range specified by src_addr/len
1876  * from the source map to the range dst_addr/len
1877  * in the destination map.
1878  *
1879  * This routine is only advisory and need not do anything.
1880  */
1881 static void
1882 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1883     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1884 {
1885 
1886 }
1887 
1888 /*
1889  * Set the physical protection on the specified range of this map as requested.
1890  */
1891 static void
1892 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1893     vm_prot_t prot)
1894 {
1895 	vm_offset_t va;
1896 	vm_page_t m;
1897 	pte_t *pte;
1898 
1899 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1900 		mmu_booke_remove(mmu, pmap, sva, eva);
1901 		return;
1902 	}
1903 
1904 	if (prot & VM_PROT_WRITE)
1905 		return;
1906 
1907 	vm_page_lock_queues();
1908 	PMAP_LOCK(pmap);
1909 	for (va = sva; va < eva; va += PAGE_SIZE) {
1910 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1911 			if (PTE_ISVALID(pte)) {
1912 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1913 
1914 				mtx_lock_spin(&tlbivax_mutex);
1915 				tlb_miss_lock();
1916 
1917 				/* Handle modified pages. */
1918 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1919 					vm_page_dirty(m);
1920 
1921 				tlb0_flush_entry(va);
1922 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1923 
1924 				tlb_miss_unlock();
1925 				mtx_unlock_spin(&tlbivax_mutex);
1926 			}
1927 		}
1928 	}
1929 	PMAP_UNLOCK(pmap);
1930 	vm_page_unlock_queues();
1931 }
1932 
1933 /*
1934  * Clear the write and modified bits in each of the given page's mappings.
1935  */
1936 static void
1937 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1938 {
1939 	pv_entry_t pv;
1940 	pte_t *pte;
1941 
1942 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
1943 	    ("mmu_booke_remove_write: page %p is not managed", m));
1944 
1945 	/*
1946 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be set by
1947 	 * another thread while the object is locked.  Thus, if PG_WRITEABLE
1948 	 * is clear, no page table entries need updating.
1949 	 */
1950 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1951 	if ((m->oflags & VPO_BUSY) == 0 &&
1952 	    (m->flags & PG_WRITEABLE) == 0)
1953 		return;
1954 	vm_page_lock_queues();
1955 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1956 		PMAP_LOCK(pv->pv_pmap);
1957 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1958 			if (PTE_ISVALID(pte)) {
1959 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1960 
1961 				mtx_lock_spin(&tlbivax_mutex);
1962 				tlb_miss_lock();
1963 
1964 				/* Handle modified pages. */
1965 				if (PTE_ISMODIFIED(pte))
1966 					vm_page_dirty(m);
1967 
1968 				/* Flush mapping from TLB0. */
1969 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1970 
1971 				tlb_miss_unlock();
1972 				mtx_unlock_spin(&tlbivax_mutex);
1973 			}
1974 		}
1975 		PMAP_UNLOCK(pv->pv_pmap);
1976 	}
1977 	vm_page_flag_clear(m, PG_WRITEABLE);
1978 	vm_page_unlock_queues();
1979 }
1980 
1981 static void
1982 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1983 {
1984 	pte_t *pte;
1985 	pmap_t pmap;
1986 	vm_page_t m;
1987 	vm_offset_t addr;
1988 	vm_paddr_t pa;
1989 	int active, valid;
1990 
1991 	va = trunc_page(va);
1992 	sz = round_page(sz);
1993 
1994 	vm_page_lock_queues();
1995 	pmap = PCPU_GET(curpmap);
1996 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
1997 	while (sz > 0) {
1998 		PMAP_LOCK(pm);
1999 		pte = pte_find(mmu, pm, va);
2000 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2001 		if (valid)
2002 			pa = PTE_PA(pte);
2003 		PMAP_UNLOCK(pm);
2004 		if (valid) {
2005 			if (!active) {
2006 				/* Create a mapping in the active pmap. */
2007 				addr = 0;
2008 				m = PHYS_TO_VM_PAGE(pa);
2009 				PMAP_LOCK(pmap);
2010 				pte_enter(mmu, pmap, m, addr,
2011 				    PTE_SR | PTE_VALID | PTE_UR);
2012 				__syncicache((void *)addr, PAGE_SIZE);
2013 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2014 				PMAP_UNLOCK(pmap);
2015 			} else
2016 				__syncicache((void *)va, PAGE_SIZE);
2017 		}
2018 		va += PAGE_SIZE;
2019 		sz -= PAGE_SIZE;
2020 	}
2021 	vm_page_unlock_queues();
2022 }
2023 
2024 /*
2025  * Atomically extract and hold the physical page with the given
2026  * pmap and virtual address pair if that mapping permits the given
2027  * protection.
2028  */
2029 static vm_page_t
2030 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2031     vm_prot_t prot)
2032 {
2033 	pte_t *pte;
2034 	vm_page_t m;
2035 	uint32_t pte_wbit;
2036 	vm_paddr_t pa;
2037 
2038 	m = NULL;
2039 	pa = 0;
2040 	PMAP_LOCK(pmap);
2041 retry:
2042 	pte = pte_find(mmu, pmap, va);
2043 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2044 		if (pmap == kernel_pmap)
2045 			pte_wbit = PTE_SW;
2046 		else
2047 			pte_wbit = PTE_UW;
2048 
2049 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2050 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2051 				goto retry;
2052 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2053 			vm_page_hold(m);
2054 		}
2055 	}
2056 
2057 	PA_UNLOCK_COND(pa);
2058 	PMAP_UNLOCK(pmap);
2059 	return (m);
2060 }
2061 
2062 /*
2063  * Initialize a vm_page's machine-dependent fields.
2064  */
2065 static void
2066 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2067 {
2068 
2069 	TAILQ_INIT(&m->md.pv_list);
2070 }
2071 
2072 /*
2073  * mmu_booke_zero_page_area zeros the specified hardware page by
2074  * mapping it into virtual memory and using bzero to clear
2075  * its contents.
2076  *
2077  * off and size must reside within a single page.
2078  */
2079 static void
2080 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2081 {
2082 	vm_offset_t va;
2083 
2084 	/* XXX KASSERT off and size are within a single page? */
2085 
2086 	mtx_lock(&zero_page_mutex);
2087 	va = zero_page_va;
2088 
2089 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2090 	bzero((caddr_t)va + off, size);
2091 	mmu_booke_kremove(mmu, va);
2092 
2093 	mtx_unlock(&zero_page_mutex);
2094 }
2095 
2096 /*
2097  * mmu_booke_zero_page zeros the specified hardware page.
2098  */
2099 static void
2100 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2101 {
2102 
2103 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2104 }
2105 
2106 /*
2107  * mmu_booke_copy_page copies the specified (machine independent) page by
2108  * mapping the page into virtual memory and using memcopy to copy the page,
2109  * one machine dependent page at a time.
2110  */
2111 static void
2112 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2113 {
2114 	vm_offset_t sva, dva;
2115 
2116 	sva = copy_page_src_va;
2117 	dva = copy_page_dst_va;
2118 
2119 	mtx_lock(&copy_page_mutex);
2120 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2121 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2122 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2123 	mmu_booke_kremove(mmu, dva);
2124 	mmu_booke_kremove(mmu, sva);
2125 	mtx_unlock(&copy_page_mutex);
2126 }
2127 
2128 /*
2129  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2130  * into virtual memory and using bzero to clear its contents. This is intended
2131  * to be called from the vm_pagezero process only and outside of Giant. No
2132  * lock is required.
2133  */
2134 static void
2135 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2136 {
2137 	vm_offset_t va;
2138 
2139 	va = zero_page_idle_va;
2140 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2141 	bzero((caddr_t)va, PAGE_SIZE);
2142 	mmu_booke_kremove(mmu, va);
2143 }
2144 
2145 /*
2146  * Return whether or not the specified physical page was modified
2147  * in any of physical maps.
2148  */
2149 static boolean_t
2150 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2151 {
2152 	pte_t *pte;
2153 	pv_entry_t pv;
2154 	boolean_t rv;
2155 
2156 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2157 	    ("mmu_booke_is_modified: page %p is not managed", m));
2158 	rv = FALSE;
2159 
2160 	/*
2161 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be
2162 	 * concurrently set while the object is locked.  Thus, if PG_WRITEABLE
2163 	 * is clear, no PTEs can be modified.
2164 	 */
2165 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2166 	if ((m->oflags & VPO_BUSY) == 0 &&
2167 	    (m->flags & PG_WRITEABLE) == 0)
2168 		return (rv);
2169 	vm_page_lock_queues();
2170 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2171 		PMAP_LOCK(pv->pv_pmap);
2172 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2173 		    PTE_ISVALID(pte)) {
2174 			if (PTE_ISMODIFIED(pte))
2175 				rv = TRUE;
2176 		}
2177 		PMAP_UNLOCK(pv->pv_pmap);
2178 		if (rv)
2179 			break;
2180 	}
2181 	vm_page_unlock_queues();
2182 	return (rv);
2183 }
2184 
2185 /*
2186  * Return whether or not the specified virtual address is eligible
2187  * for prefault.
2188  */
2189 static boolean_t
2190 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2191 {
2192 
2193 	return (FALSE);
2194 }
2195 
2196 /*
2197  * Return whether or not the specified physical page was referenced
2198  * in any physical maps.
2199  */
2200 static boolean_t
2201 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2202 {
2203 	pte_t *pte;
2204 	pv_entry_t pv;
2205 	boolean_t rv;
2206 
2207 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2208 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2209 	rv = FALSE;
2210 	vm_page_lock_queues();
2211 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2212 		PMAP_LOCK(pv->pv_pmap);
2213 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2214 		    PTE_ISVALID(pte)) {
2215 			if (PTE_ISREFERENCED(pte))
2216 				rv = TRUE;
2217 		}
2218 		PMAP_UNLOCK(pv->pv_pmap);
2219 		if (rv)
2220 			break;
2221 	}
2222 	vm_page_unlock_queues();
2223 	return (rv);
2224 }
2225 
2226 /*
2227  * Clear the modify bits on the specified physical page.
2228  */
2229 static void
2230 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2231 {
2232 	pte_t *pte;
2233 	pv_entry_t pv;
2234 
2235 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2236 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2237 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2238 	KASSERT((m->oflags & VPO_BUSY) == 0,
2239 	    ("mmu_booke_clear_modify: page %p is busy", m));
2240 
2241 	/*
2242 	 * If the page is not PG_WRITEABLE, then no PTEs can be modified.
2243 	 * If the object containing the page is locked and the page is not
2244 	 * VPO_BUSY, then PG_WRITEABLE cannot be concurrently set.
2245 	 */
2246 	if ((m->flags & PG_WRITEABLE) == 0)
2247 		return;
2248 	vm_page_lock_queues();
2249 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2250 		PMAP_LOCK(pv->pv_pmap);
2251 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2252 		    PTE_ISVALID(pte)) {
2253 			mtx_lock_spin(&tlbivax_mutex);
2254 			tlb_miss_lock();
2255 
2256 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2257 				tlb0_flush_entry(pv->pv_va);
2258 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2259 				    PTE_REFERENCED);
2260 			}
2261 
2262 			tlb_miss_unlock();
2263 			mtx_unlock_spin(&tlbivax_mutex);
2264 		}
2265 		PMAP_UNLOCK(pv->pv_pmap);
2266 	}
2267 	vm_page_unlock_queues();
2268 }
2269 
2270 /*
2271  * Return a count of reference bits for a page, clearing those bits.
2272  * It is not necessary for every reference bit to be cleared, but it
2273  * is necessary that 0 only be returned when there are truly no
2274  * reference bits set.
2275  *
2276  * XXX: The exact number of bits to check and clear is a matter that
2277  * should be tested and standardized at some point in the future for
2278  * optimal aging of shared pages.
2279  */
2280 static int
2281 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2282 {
2283 	pte_t *pte;
2284 	pv_entry_t pv;
2285 	int count;
2286 
2287 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2288 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2289 	count = 0;
2290 	vm_page_lock_queues();
2291 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2292 		PMAP_LOCK(pv->pv_pmap);
2293 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2294 		    PTE_ISVALID(pte)) {
2295 			if (PTE_ISREFERENCED(pte)) {
2296 				mtx_lock_spin(&tlbivax_mutex);
2297 				tlb_miss_lock();
2298 
2299 				tlb0_flush_entry(pv->pv_va);
2300 				pte->flags &= ~PTE_REFERENCED;
2301 
2302 				tlb_miss_unlock();
2303 				mtx_unlock_spin(&tlbivax_mutex);
2304 
2305 				if (++count > 4) {
2306 					PMAP_UNLOCK(pv->pv_pmap);
2307 					break;
2308 				}
2309 			}
2310 		}
2311 		PMAP_UNLOCK(pv->pv_pmap);
2312 	}
2313 	vm_page_unlock_queues();
2314 	return (count);
2315 }
2316 
2317 /*
2318  * Clear the reference bit on the specified physical page.
2319  */
2320 static void
2321 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2322 {
2323 	pte_t *pte;
2324 	pv_entry_t pv;
2325 
2326 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2327 	    ("mmu_booke_clear_reference: page %p is not managed", m));
2328 	vm_page_lock_queues();
2329 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2330 		PMAP_LOCK(pv->pv_pmap);
2331 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2332 		    PTE_ISVALID(pte)) {
2333 			if (PTE_ISREFERENCED(pte)) {
2334 				mtx_lock_spin(&tlbivax_mutex);
2335 				tlb_miss_lock();
2336 
2337 				tlb0_flush_entry(pv->pv_va);
2338 				pte->flags &= ~PTE_REFERENCED;
2339 
2340 				tlb_miss_unlock();
2341 				mtx_unlock_spin(&tlbivax_mutex);
2342 			}
2343 		}
2344 		PMAP_UNLOCK(pv->pv_pmap);
2345 	}
2346 	vm_page_unlock_queues();
2347 }
2348 
2349 /*
2350  * Change wiring attribute for a map/virtual-address pair.
2351  */
2352 static void
2353 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2354 {
2355 	pte_t *pte;
2356 
2357 	PMAP_LOCK(pmap);
2358 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2359 		if (wired) {
2360 			if (!PTE_ISWIRED(pte)) {
2361 				pte->flags |= PTE_WIRED;
2362 				pmap->pm_stats.wired_count++;
2363 			}
2364 		} else {
2365 			if (PTE_ISWIRED(pte)) {
2366 				pte->flags &= ~PTE_WIRED;
2367 				pmap->pm_stats.wired_count--;
2368 			}
2369 		}
2370 	}
2371 	PMAP_UNLOCK(pmap);
2372 }
2373 
2374 /*
2375  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2376  * page.  This count may be changed upwards or downwards in the future; it is
2377  * only necessary that true be returned for a small subset of pmaps for proper
2378  * page aging.
2379  */
2380 static boolean_t
2381 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2382 {
2383 	pv_entry_t pv;
2384 	int loops;
2385 	boolean_t rv;
2386 
2387 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2388 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2389 	loops = 0;
2390 	rv = FALSE;
2391 	vm_page_lock_queues();
2392 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2393 		if (pv->pv_pmap == pmap) {
2394 			rv = TRUE;
2395 			break;
2396 		}
2397 		if (++loops >= 16)
2398 			break;
2399 	}
2400 	vm_page_unlock_queues();
2401 	return (rv);
2402 }
2403 
2404 /*
2405  * Return the number of managed mappings to the given physical page that are
2406  * wired.
2407  */
2408 static int
2409 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2410 {
2411 	pv_entry_t pv;
2412 	pte_t *pte;
2413 	int count = 0;
2414 
2415 	if ((m->flags & PG_FICTITIOUS) != 0)
2416 		return (count);
2417 	vm_page_lock_queues();
2418 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2419 		PMAP_LOCK(pv->pv_pmap);
2420 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2421 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2422 				count++;
2423 		PMAP_UNLOCK(pv->pv_pmap);
2424 	}
2425 	vm_page_unlock_queues();
2426 	return (count);
2427 }
2428 
2429 static int
2430 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2431 {
2432 	int i;
2433 	vm_offset_t va;
2434 
2435 	/*
2436 	 * This currently does not work for entries that
2437 	 * overlap TLB1 entries.
2438 	 */
2439 	for (i = 0; i < tlb1_idx; i ++) {
2440 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2441 			return (0);
2442 	}
2443 
2444 	return (EFAULT);
2445 }
2446 
2447 vm_offset_t
2448 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2449     vm_size_t *sz)
2450 {
2451 	vm_paddr_t pa, ppa;
2452 	vm_offset_t va;
2453 	vm_size_t gran;
2454 
2455 	/* Raw physical memory dumps don't have a virtual address. */
2456 	if (md->md_vaddr == ~0UL) {
2457 		/* We always map a 256MB page at 256M. */
2458 		gran = 256 * 1024 * 1024;
2459 		pa = md->md_paddr + ofs;
2460 		ppa = pa & ~(gran - 1);
2461 		ofs = pa - ppa;
2462 		va = gran;
2463 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2464 		if (*sz > (gran - ofs))
2465 			*sz = gran - ofs;
2466 		return (va + ofs);
2467 	}
2468 
2469 	/* Minidumps are based on virtual memory addresses. */
2470 	va = md->md_vaddr + ofs;
2471 	if (va >= kernstart + kernsize) {
2472 		gran = PAGE_SIZE - (va & PAGE_MASK);
2473 		if (*sz > gran)
2474 			*sz = gran;
2475 	}
2476 	return (va);
2477 }
2478 
2479 void
2480 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2481     vm_offset_t va)
2482 {
2483 
2484 	/* Raw physical memory dumps don't have a virtual address. */
2485 	if (md->md_vaddr == ~0UL) {
2486 		tlb1_idx--;
2487 		tlb1[tlb1_idx].mas1 = 0;
2488 		tlb1[tlb1_idx].mas2 = 0;
2489 		tlb1[tlb1_idx].mas3 = 0;
2490 		tlb1_write_entry(tlb1_idx);
2491 		return;
2492 	}
2493 
2494 	/* Minidumps are based on virtual memory addresses. */
2495 	/* Nothing to do... */
2496 }
2497 
2498 struct pmap_md *
2499 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2500 {
2501 	static struct pmap_md md;
2502 	pte_t *pte;
2503 	vm_offset_t va;
2504 
2505 	if (dumpsys_minidump) {
2506 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2507 		if (prev == NULL) {
2508 			/* 1st: kernel .data and .bss. */
2509 			md.md_index = 1;
2510 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2511 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2512 			return (&md);
2513 		}
2514 		switch (prev->md_index) {
2515 		case 1:
2516 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2517 			md.md_index = 2;
2518 			md.md_vaddr = data_start;
2519 			md.md_size = data_end - data_start;
2520 			break;
2521 		case 2:
2522 			/* 3rd: kernel VM. */
2523 			va = prev->md_vaddr + prev->md_size;
2524 			/* Find start of next chunk (from va). */
2525 			while (va < virtual_end) {
2526 				/* Don't dump the buffer cache. */
2527 				if (va >= kmi.buffer_sva &&
2528 				    va < kmi.buffer_eva) {
2529 					va = kmi.buffer_eva;
2530 					continue;
2531 				}
2532 				pte = pte_find(mmu, kernel_pmap, va);
2533 				if (pte != NULL && PTE_ISVALID(pte))
2534 					break;
2535 				va += PAGE_SIZE;
2536 			}
2537 			if (va < virtual_end) {
2538 				md.md_vaddr = va;
2539 				va += PAGE_SIZE;
2540 				/* Find last page in chunk. */
2541 				while (va < virtual_end) {
2542 					/* Don't run into the buffer cache. */
2543 					if (va == kmi.buffer_sva)
2544 						break;
2545 					pte = pte_find(mmu, kernel_pmap, va);
2546 					if (pte == NULL || !PTE_ISVALID(pte))
2547 						break;
2548 					va += PAGE_SIZE;
2549 				}
2550 				md.md_size = va - md.md_vaddr;
2551 				break;
2552 			}
2553 			md.md_index = 3;
2554 			/* FALLTHROUGH */
2555 		default:
2556 			return (NULL);
2557 		}
2558 	} else { /* minidumps */
2559 		mem_regions(&physmem_regions, &physmem_regions_sz,
2560 		    &availmem_regions, &availmem_regions_sz);
2561 
2562 		if (prev == NULL) {
2563 			/* first physical chunk. */
2564 			md.md_paddr = physmem_regions[0].mr_start;
2565 			md.md_size = physmem_regions[0].mr_size;
2566 			md.md_vaddr = ~0UL;
2567 			md.md_index = 1;
2568 		} else if (md.md_index < physmem_regions_sz) {
2569 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2570 			md.md_size = physmem_regions[md.md_index].mr_size;
2571 			md.md_vaddr = ~0UL;
2572 			md.md_index++;
2573 		} else {
2574 			/* There's no next physical chunk. */
2575 			return (NULL);
2576 		}
2577 	}
2578 
2579 	return (&md);
2580 }
2581 
2582 /*
2583  * Map a set of physical memory pages into the kernel virtual address space.
2584  * Return a pointer to where it is mapped. This routine is intended to be used
2585  * for mapping device memory, NOT real memory.
2586  */
2587 static void *
2588 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2589 {
2590 	void *res;
2591 	uintptr_t va;
2592 	vm_size_t sz;
2593 
2594 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2595 	res = (void *)va;
2596 
2597 	do {
2598 		sz = 1 << (ilog2(size) & ~1);
2599 		if (bootverbose)
2600 			printf("Wiring VA=%x to PA=%x (size=%x), "
2601 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2602 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2603 		size -= sz;
2604 		pa += sz;
2605 		va += sz;
2606 	} while (size > 0);
2607 
2608 	return (res);
2609 }
2610 
2611 /*
2612  * 'Unmap' a range mapped by mmu_booke_mapdev().
2613  */
2614 static void
2615 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2616 {
2617 	vm_offset_t base, offset;
2618 
2619 	/*
2620 	 * Unmap only if this is inside kernel virtual space.
2621 	 */
2622 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2623 		base = trunc_page(va);
2624 		offset = va & PAGE_MASK;
2625 		size = roundup(offset + size, PAGE_SIZE);
2626 		kmem_free(kernel_map, base, size);
2627 	}
2628 }
2629 
2630 /*
2631  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2632  * specified pmap. This eliminates the blast of soft faults on process startup
2633  * and immediately after an mmap.
2634  */
2635 static void
2636 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2637     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2638 {
2639 
2640 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2641 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2642 	    ("mmu_booke_object_init_pt: non-device object"));
2643 }
2644 
2645 /*
2646  * Perform the pmap work for mincore.
2647  */
2648 static int
2649 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2650     vm_paddr_t *locked_pa)
2651 {
2652 
2653 	TODO;
2654 	return (0);
2655 }
2656 
2657 /**************************************************************************/
2658 /* TID handling */
2659 /**************************************************************************/
2660 
2661 /*
2662  * Allocate a TID. If necessary, steal one from someone else.
2663  * The new TID is flushed from the TLB before returning.
2664  */
2665 static tlbtid_t
2666 tid_alloc(pmap_t pmap)
2667 {
2668 	tlbtid_t tid;
2669 	int thiscpu;
2670 
2671 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2672 
2673 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2674 
2675 	thiscpu = PCPU_GET(cpuid);
2676 
2677 	tid = PCPU_GET(tid_next);
2678 	if (tid > TID_MAX)
2679 		tid = TID_MIN;
2680 	PCPU_SET(tid_next, tid + 1);
2681 
2682 	/* If we are stealing TID then clear the relevant pmap's field */
2683 	if (tidbusy[thiscpu][tid] != NULL) {
2684 
2685 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2686 
2687 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2688 
2689 		/* Flush all entries from TLB0 matching this TID. */
2690 		tid_flush(tid);
2691 	}
2692 
2693 	tidbusy[thiscpu][tid] = pmap;
2694 	pmap->pm_tid[thiscpu] = tid;
2695 	__asm __volatile("msync; isync");
2696 
2697 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2698 	    PCPU_GET(tid_next));
2699 
2700 	return (tid);
2701 }
2702 
2703 /**************************************************************************/
2704 /* TLB0 handling */
2705 /**************************************************************************/
2706 
2707 static void
2708 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2709     uint32_t mas7)
2710 {
2711 	int as;
2712 	char desc[3];
2713 	tlbtid_t tid;
2714 	vm_size_t size;
2715 	unsigned int tsize;
2716 
2717 	desc[2] = '\0';
2718 	if (mas1 & MAS1_VALID)
2719 		desc[0] = 'V';
2720 	else
2721 		desc[0] = ' ';
2722 
2723 	if (mas1 & MAS1_IPROT)
2724 		desc[1] = 'P';
2725 	else
2726 		desc[1] = ' ';
2727 
2728 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2729 	tid = MAS1_GETTID(mas1);
2730 
2731 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2732 	size = 0;
2733 	if (tsize)
2734 		size = tsize2size(tsize);
2735 
2736 	debugf("%3d: (%s) [AS=%d] "
2737 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2738 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2739 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2740 }
2741 
2742 /* Convert TLB0 va and way number to tlb0[] table index. */
2743 static inline unsigned int
2744 tlb0_tableidx(vm_offset_t va, unsigned int way)
2745 {
2746 	unsigned int idx;
2747 
2748 	idx = (way * TLB0_ENTRIES_PER_WAY);
2749 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2750 	return (idx);
2751 }
2752 
2753 /*
2754  * Invalidate TLB0 entry.
2755  */
2756 static inline void
2757 tlb0_flush_entry(vm_offset_t va)
2758 {
2759 
2760 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2761 
2762 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2763 
2764 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2765 	__asm __volatile("isync; msync");
2766 	__asm __volatile("tlbsync; msync");
2767 
2768 	CTR1(KTR_PMAP, "%s: e", __func__);
2769 }
2770 
2771 /* Print out contents of the MAS registers for each TLB0 entry */
2772 void
2773 tlb0_print_tlbentries(void)
2774 {
2775 	uint32_t mas0, mas1, mas2, mas3, mas7;
2776 	int entryidx, way, idx;
2777 
2778 	debugf("TLB0 entries:\n");
2779 	for (way = 0; way < TLB0_WAYS; way ++)
2780 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2781 
2782 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2783 			mtspr(SPR_MAS0, mas0);
2784 			__asm __volatile("isync");
2785 
2786 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2787 			mtspr(SPR_MAS2, mas2);
2788 
2789 			__asm __volatile("isync; tlbre");
2790 
2791 			mas1 = mfspr(SPR_MAS1);
2792 			mas2 = mfspr(SPR_MAS2);
2793 			mas3 = mfspr(SPR_MAS3);
2794 			mas7 = mfspr(SPR_MAS7);
2795 
2796 			idx = tlb0_tableidx(mas2, way);
2797 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2798 		}
2799 }
2800 
2801 /**************************************************************************/
2802 /* TLB1 handling */
2803 /**************************************************************************/
2804 
2805 /*
2806  * TLB1 mapping notes:
2807  *
2808  * TLB1[0]	CCSRBAR
2809  * TLB1[1]	Kernel text and data.
2810  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2811  *		windows, other devices mappings.
2812  */
2813 
2814 /*
2815  * Write given entry to TLB1 hardware.
2816  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2817  */
2818 static void
2819 tlb1_write_entry(unsigned int idx)
2820 {
2821 	uint32_t mas0, mas7;
2822 
2823 	//debugf("tlb1_write_entry: s\n");
2824 
2825 	/* Clear high order RPN bits */
2826 	mas7 = 0;
2827 
2828 	/* Select entry */
2829 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2830 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2831 
2832 	mtspr(SPR_MAS0, mas0);
2833 	__asm __volatile("isync");
2834 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2835 	__asm __volatile("isync");
2836 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2837 	__asm __volatile("isync");
2838 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2839 	__asm __volatile("isync");
2840 	mtspr(SPR_MAS7, mas7);
2841 	__asm __volatile("isync; tlbwe; isync; msync");
2842 
2843 	//debugf("tlb1_write_entry: e\n");
2844 }
2845 
2846 /*
2847  * Return the largest uint value log such that 2^log <= num.
2848  */
2849 static unsigned int
2850 ilog2(unsigned int num)
2851 {
2852 	int lz;
2853 
2854 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2855 	return (31 - lz);
2856 }
2857 
2858 /*
2859  * Convert TLB TSIZE value to mapped region size.
2860  */
2861 static vm_size_t
2862 tsize2size(unsigned int tsize)
2863 {
2864 
2865 	/*
2866 	 * size = 4^tsize KB
2867 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2868 	 */
2869 
2870 	return ((1 << (2 * tsize)) * 1024);
2871 }
2872 
2873 /*
2874  * Convert region size (must be power of 4) to TLB TSIZE value.
2875  */
2876 static unsigned int
2877 size2tsize(vm_size_t size)
2878 {
2879 
2880 	return (ilog2(size) / 2 - 5);
2881 }
2882 
2883 /*
2884  * Register permanent kernel mapping in TLB1.
2885  *
2886  * Entries are created starting from index 0 (current free entry is
2887  * kept in tlb1_idx) and are not supposed to be invalidated.
2888  */
2889 static int
2890 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2891     uint32_t flags)
2892 {
2893 	uint32_t ts, tid;
2894 	int tsize;
2895 
2896 	if (tlb1_idx >= TLB1_ENTRIES) {
2897 		printf("tlb1_set_entry: TLB1 full!\n");
2898 		return (-1);
2899 	}
2900 
2901 	/* Convert size to TSIZE */
2902 	tsize = size2tsize(size);
2903 
2904 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2905 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2906 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2907 
2908 	/* XXX LOCK tlb1[] */
2909 
2910 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2911 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2912 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2913 
2914 	/* Set supervisor RWX permission bits */
2915 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2916 
2917 	tlb1_write_entry(tlb1_idx++);
2918 
2919 	/* XXX UNLOCK tlb1[] */
2920 
2921 	/*
2922 	 * XXX in general TLB1 updates should be propagated between CPUs,
2923 	 * since current design assumes to have the same TLB1 set-up on all
2924 	 * cores.
2925 	 */
2926 	return (0);
2927 }
2928 
2929 static int
2930 tlb1_entry_size_cmp(const void *a, const void *b)
2931 {
2932 	const vm_size_t *sza;
2933 	const vm_size_t *szb;
2934 
2935 	sza = a;
2936 	szb = b;
2937 	if (*sza > *szb)
2938 		return (-1);
2939 	else if (*sza < *szb)
2940 		return (1);
2941 	else
2942 		return (0);
2943 }
2944 
2945 /*
2946  * Map in contiguous RAM region into the TLB1 using maximum of
2947  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2948  *
2949  * If necessary round up last entry size and return total size
2950  * used by all allocated entries.
2951  */
2952 vm_size_t
2953 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2954 {
2955 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2956 	vm_size_t mapped_size, sz, esz;
2957 	unsigned int log;
2958 	int i;
2959 
2960 	CTR4(KTR_PMAP, "%s: region size = 0x%08x va = 0x%08x pa = 0x%08x",
2961 	    __func__, size, va, pa);
2962 
2963 	mapped_size = 0;
2964 	sz = size;
2965 	memset(entry_size, 0, sizeof(entry_size));
2966 
2967 	/* Calculate entry sizes. */
2968 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2969 
2970 		/* Largest region that is power of 4 and fits within size */
2971 		log = ilog2(sz) / 2;
2972 		esz = 1 << (2 * log);
2973 
2974 		/* If this is last entry cover remaining size. */
2975 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2976 			while (esz < sz)
2977 				esz = esz << 2;
2978 		}
2979 
2980 		entry_size[i] = esz;
2981 		mapped_size += esz;
2982 		if (esz < sz)
2983 			sz -= esz;
2984 		else
2985 			sz = 0;
2986 	}
2987 
2988 	/* Sort entry sizes, required to get proper entry address alignment. */
2989 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2990 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2991 
2992 	/* Load TLB1 entries. */
2993 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
2994 		esz = entry_size[i];
2995 		if (!esz)
2996 			break;
2997 
2998 		CTR5(KTR_PMAP, "%s: entry %d: sz  = 0x%08x (va = 0x%08x "
2999 		    "pa = 0x%08x)", __func__, tlb1_idx, esz, va, pa);
3000 
3001 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
3002 
3003 		va += esz;
3004 		pa += esz;
3005 	}
3006 
3007 	CTR3(KTR_PMAP, "%s: mapped size 0x%08x (wasted space 0x%08x)",
3008 	    __func__, mapped_size, mapped_size - size);
3009 
3010 	return (mapped_size);
3011 }
3012 
3013 /*
3014  * TLB1 initialization routine, to be called after the very first
3015  * assembler level setup done in locore.S.
3016  */
3017 void
3018 tlb1_init(vm_offset_t ccsrbar)
3019 {
3020 	uint32_t mas0;
3021 
3022 	/* TLB1[1] is used to map the kernel. Save that entry. */
3023 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(1);
3024 	mtspr(SPR_MAS0, mas0);
3025 	__asm __volatile("isync; tlbre");
3026 
3027 	tlb1[1].mas1 = mfspr(SPR_MAS1);
3028 	tlb1[1].mas2 = mfspr(SPR_MAS2);
3029 	tlb1[1].mas3 = mfspr(SPR_MAS3);
3030 
3031 	/* Map in CCSRBAR in TLB1[0] */
3032 	tlb1_idx = 0;
3033 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
3034 	/*
3035 	 * Set the next available TLB1 entry index. Note TLB[1] is reserved
3036 	 * for initial mapping of kernel text+data, which was set early in
3037 	 * locore, we need to skip this [busy] entry.
3038 	 */
3039 	tlb1_idx = 2;
3040 
3041 	/* Setup TLB miss defaults */
3042 	set_mas4_defaults();
3043 }
3044 
3045 /*
3046  * Setup MAS4 defaults.
3047  * These values are loaded to MAS0-2 on a TLB miss.
3048  */
3049 static void
3050 set_mas4_defaults(void)
3051 {
3052 	uint32_t mas4;
3053 
3054 	/* Defaults: TLB0, PID0, TSIZED=4K */
3055 	mas4 = MAS4_TLBSELD0;
3056 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3057 #ifdef SMP
3058 	mas4 |= MAS4_MD;
3059 #endif
3060 	mtspr(SPR_MAS4, mas4);
3061 	__asm __volatile("isync");
3062 }
3063 
3064 /*
3065  * Print out contents of the MAS registers for each TLB1 entry
3066  */
3067 void
3068 tlb1_print_tlbentries(void)
3069 {
3070 	uint32_t mas0, mas1, mas2, mas3, mas7;
3071 	int i;
3072 
3073 	debugf("TLB1 entries:\n");
3074 	for (i = 0; i < TLB1_ENTRIES; i++) {
3075 
3076 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3077 		mtspr(SPR_MAS0, mas0);
3078 
3079 		__asm __volatile("isync; tlbre");
3080 
3081 		mas1 = mfspr(SPR_MAS1);
3082 		mas2 = mfspr(SPR_MAS2);
3083 		mas3 = mfspr(SPR_MAS3);
3084 		mas7 = mfspr(SPR_MAS7);
3085 
3086 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3087 	}
3088 }
3089 
3090 /*
3091  * Print out contents of the in-ram tlb1 table.
3092  */
3093 void
3094 tlb1_print_entries(void)
3095 {
3096 	int i;
3097 
3098 	debugf("tlb1[] table entries:\n");
3099 	for (i = 0; i < TLB1_ENTRIES; i++)
3100 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3101 }
3102 
3103 /*
3104  * Return 0 if the physical IO range is encompassed by one of the
3105  * the TLB1 entries, otherwise return related error code.
3106  */
3107 static int
3108 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3109 {
3110 	uint32_t prot;
3111 	vm_paddr_t pa_start;
3112 	vm_paddr_t pa_end;
3113 	unsigned int entry_tsize;
3114 	vm_size_t entry_size;
3115 
3116 	*va = (vm_offset_t)NULL;
3117 
3118 	/* Skip invalid entries */
3119 	if (!(tlb1[i].mas1 & MAS1_VALID))
3120 		return (EINVAL);
3121 
3122 	/*
3123 	 * The entry must be cache-inhibited, guarded, and r/w
3124 	 * so it can function as an i/o page
3125 	 */
3126 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3127 	if (prot != (MAS2_I | MAS2_G))
3128 		return (EPERM);
3129 
3130 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3131 	if (prot != (MAS3_SR | MAS3_SW))
3132 		return (EPERM);
3133 
3134 	/* The address should be within the entry range. */
3135 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3136 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3137 
3138 	entry_size = tsize2size(entry_tsize);
3139 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3140 	pa_end = pa_start + entry_size - 1;
3141 
3142 	if ((pa < pa_start) || ((pa + size) > pa_end))
3143 		return (ERANGE);
3144 
3145 	/* Return virtual address of this mapping. */
3146 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3147 	return (0);
3148 }
3149