xref: /freebsd/sys/powerpc/booke/pmap_32.c (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (C) 2020 Justin Hibbits
5  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
6  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
21  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
23  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
24  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
25  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
26  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
27  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  *
29  * Some hw specific parts of this pmap were derived or influenced
30  * by NetBSD's ibm4xx pmap module. More generic code is shared with
31  * a few other pmap modules from the FreeBSD tree.
32  */
33 
34  /*
35   * VM layout notes:
36   *
37   * Kernel and user threads run within one common virtual address space
38   * defined by AS=0.
39   *
40   * 32-bit pmap:
41   * Virtual address space layout:
42   * -----------------------------
43   * 0x0000_0000 - 0x7fff_ffff	: user process
44   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
45   * 0xc000_0000 - 0xffff_efff	: KVA
46   */
47 
48 #include <sys/cdefs.h>
49 #include "opt_ddb.h"
50 #include "opt_kstack_pages.h"
51 
52 #include <sys/param.h>
53 #include <sys/conf.h>
54 #include <sys/malloc.h>
55 #include <sys/ktr.h>
56 #include <sys/proc.h>
57 #include <sys/user.h>
58 #include <sys/queue.h>
59 #include <sys/systm.h>
60 #include <sys/kernel.h>
61 #include <sys/kerneldump.h>
62 #include <sys/linker.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/rwlock.h>
67 #include <sys/sched.h>
68 #include <sys/smp.h>
69 #include <sys/vmmeter.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_phys.h>
81 #include <vm/vm_pagequeue.h>
82 #include <vm/uma.h>
83 
84 #include <machine/_inttypes.h>
85 #include <machine/cpu.h>
86 #include <machine/pcb.h>
87 #include <machine/platform.h>
88 
89 #include <machine/tlb.h>
90 #include <machine/spr.h>
91 #include <machine/md_var.h>
92 #include <machine/mmuvar.h>
93 #include <machine/pmap.h>
94 #include <machine/pte.h>
95 
96 #include <ddb/ddb.h>
97 
98 #define	PRI0ptrX	"08x"
99 
100 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
101 static vm_offset_t zero_page_va;
102 static struct mtx zero_page_mutex;
103 
104 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
105 static vm_offset_t copy_page_src_va;
106 static vm_offset_t copy_page_dst_va;
107 static struct mtx copy_page_mutex;
108 
109 static vm_offset_t kernel_ptbl_root;
110 static unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
111 
112 /**************************************************************************/
113 /* PMAP */
114 /**************************************************************************/
115 
116 #define	VM_MAPDEV_BASE	((vm_offset_t)VM_MAXUSER_ADDRESS + PAGE_SIZE)
117 
118 static void tid_flush(tlbtid_t tid);
119 
120 /**************************************************************************/
121 /* Page table management */
122 /**************************************************************************/
123 
124 #define PMAP_ROOT_SIZE	(sizeof(pte_t**) * PDIR_NENTRIES)
125 static void ptbl_init(void);
126 static struct ptbl_buf *ptbl_buf_alloc(void);
127 static void ptbl_buf_free(struct ptbl_buf *);
128 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
129 
130 static pte_t *ptbl_alloc(pmap_t, unsigned int, bool);
131 static void ptbl_free(pmap_t, unsigned int);
132 static void ptbl_hold(pmap_t, unsigned int);
133 static int ptbl_unhold(pmap_t, unsigned int);
134 
135 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
136 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, bool);
137 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
138 static pte_t *pte_find(pmap_t, vm_offset_t);
139 
140 struct ptbl_buf {
141 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
142 	vm_offset_t kva;		/* va of mapping */
143 };
144 
145 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
146 #define PTBL_BUFS		(128 * 16)
147 
148 /* ptbl free list and a lock used for access synchronization. */
149 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
150 static struct mtx ptbl_buf_freelist_lock;
151 
152 /* Base address of kva space allocated fot ptbl bufs. */
153 static vm_offset_t ptbl_buf_pool_vabase;
154 
155 /* Pointer to ptbl_buf structures. */
156 static struct ptbl_buf *ptbl_bufs;
157 
158 /**************************************************************************/
159 /* Page table related */
160 /**************************************************************************/
161 
162 /* Initialize pool of kva ptbl buffers. */
163 static void
164 ptbl_init(void)
165 {
166 	int i;
167 
168 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
169 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
170 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
171 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
172 
173 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
174 	TAILQ_INIT(&ptbl_buf_freelist);
175 
176 	for (i = 0; i < PTBL_BUFS; i++) {
177 		ptbl_bufs[i].kva =
178 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
179 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
180 	}
181 }
182 
183 /* Get a ptbl_buf from the freelist. */
184 static struct ptbl_buf *
185 ptbl_buf_alloc(void)
186 {
187 	struct ptbl_buf *buf;
188 
189 	mtx_lock(&ptbl_buf_freelist_lock);
190 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
191 	if (buf != NULL)
192 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
193 	mtx_unlock(&ptbl_buf_freelist_lock);
194 
195 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
196 
197 	return (buf);
198 }
199 
200 /* Return ptbl buff to free pool. */
201 static void
202 ptbl_buf_free(struct ptbl_buf *buf)
203 {
204 
205 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
206 
207 	mtx_lock(&ptbl_buf_freelist_lock);
208 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
209 	mtx_unlock(&ptbl_buf_freelist_lock);
210 }
211 
212 /*
213  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
214  */
215 static void
216 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
217 {
218 	struct ptbl_buf *pbuf;
219 
220 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
221 
222 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
223 
224 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
225 		if (pbuf->kva == (vm_offset_t)ptbl) {
226 			/* Remove from pmap ptbl buf list. */
227 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
228 
229 			/* Free corresponding ptbl buf. */
230 			ptbl_buf_free(pbuf);
231 			break;
232 		}
233 }
234 
235 /* Allocate page table. */
236 static pte_t *
237 ptbl_alloc(pmap_t pmap, unsigned int pdir_idx, bool nosleep)
238 {
239 	vm_page_t mtbl[PTBL_PAGES];
240 	vm_page_t m;
241 	struct ptbl_buf *pbuf;
242 	unsigned int pidx;
243 	pte_t *ptbl;
244 	int i, j;
245 
246 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
247 	    (pmap == kernel_pmap), pdir_idx);
248 
249 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
250 	    ("ptbl_alloc: invalid pdir_idx"));
251 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
252 	    ("pte_alloc: valid ptbl entry exists!"));
253 
254 	pbuf = ptbl_buf_alloc();
255 	if (pbuf == NULL)
256 		panic("pte_alloc: couldn't alloc kernel virtual memory");
257 
258 	ptbl = (pte_t *)pbuf->kva;
259 
260 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
261 
262 	for (i = 0; i < PTBL_PAGES; i++) {
263 		pidx = (PTBL_PAGES * pdir_idx) + i;
264 		while ((m = vm_page_alloc_noobj(VM_ALLOC_WIRED)) == NULL) {
265 			if (nosleep) {
266 				ptbl_free_pmap_ptbl(pmap, ptbl);
267 				for (j = 0; j < i; j++)
268 					vm_page_free(mtbl[j]);
269 				vm_wire_sub(i);
270 				return (NULL);
271 			}
272 			PMAP_UNLOCK(pmap);
273 			rw_wunlock(&pvh_global_lock);
274 			vm_wait(NULL);
275 			rw_wlock(&pvh_global_lock);
276 			PMAP_LOCK(pmap);
277 		}
278 		m->pindex = pidx;
279 		mtbl[i] = m;
280 	}
281 
282 	/* Map allocated pages into kernel_pmap. */
283 	mmu_booke_qenter((vm_offset_t)ptbl, mtbl, PTBL_PAGES);
284 
285 	/* Zero whole ptbl. */
286 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
287 
288 	/* Add pbuf to the pmap ptbl bufs list. */
289 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
290 
291 	return (ptbl);
292 }
293 
294 /* Free ptbl pages and invalidate pdir entry. */
295 static void
296 ptbl_free(pmap_t pmap, unsigned int pdir_idx)
297 {
298 	pte_t *ptbl;
299 	vm_paddr_t pa;
300 	vm_offset_t va;
301 	vm_page_t m;
302 	int i;
303 
304 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
305 	    (pmap == kernel_pmap), pdir_idx);
306 
307 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
308 	    ("ptbl_free: invalid pdir_idx"));
309 
310 	ptbl = pmap->pm_pdir[pdir_idx];
311 
312 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
313 
314 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
315 
316 	/*
317 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
318 	 * don't attempt to look up the page tables we are releasing.
319 	 */
320 	mtx_lock_spin(&tlbivax_mutex);
321 	tlb_miss_lock();
322 
323 	pmap->pm_pdir[pdir_idx] = NULL;
324 
325 	tlb_miss_unlock();
326 	mtx_unlock_spin(&tlbivax_mutex);
327 
328 	for (i = 0; i < PTBL_PAGES; i++) {
329 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
330 		pa = pte_vatopa(kernel_pmap, va);
331 		m = PHYS_TO_VM_PAGE(pa);
332 		vm_page_free_zero(m);
333 		vm_wire_sub(1);
334 		mmu_booke_kremove(va);
335 	}
336 
337 	ptbl_free_pmap_ptbl(pmap, ptbl);
338 }
339 
340 /*
341  * Decrement ptbl pages hold count and attempt to free ptbl pages.
342  * Called when removing pte entry from ptbl.
343  *
344  * Return 1 if ptbl pages were freed.
345  */
346 static int
347 ptbl_unhold(pmap_t pmap, unsigned int pdir_idx)
348 {
349 	pte_t *ptbl;
350 	vm_paddr_t pa;
351 	vm_page_t m;
352 	int i;
353 
354 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
355 	    (pmap == kernel_pmap), pdir_idx);
356 
357 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
358 	    ("ptbl_unhold: invalid pdir_idx"));
359 	KASSERT((pmap != kernel_pmap),
360 	    ("ptbl_unhold: unholding kernel ptbl!"));
361 
362 	ptbl = pmap->pm_pdir[pdir_idx];
363 
364 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
365 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
366 	    ("ptbl_unhold: non kva ptbl"));
367 
368 	/* decrement hold count */
369 	for (i = 0; i < PTBL_PAGES; i++) {
370 		pa = pte_vatopa(kernel_pmap,
371 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
372 		m = PHYS_TO_VM_PAGE(pa);
373 		m->ref_count--;
374 	}
375 
376 	/*
377 	 * Free ptbl pages if there are no pte etries in this ptbl.
378 	 * ref_count has the same value for all ptbl pages, so check the last
379 	 * page.
380 	 */
381 	if (m->ref_count == 0) {
382 		ptbl_free(pmap, pdir_idx);
383 
384 		//debugf("ptbl_unhold: e (freed ptbl)\n");
385 		return (1);
386 	}
387 
388 	return (0);
389 }
390 
391 /*
392  * Increment hold count for ptbl pages. This routine is used when a new pte
393  * entry is being inserted into the ptbl.
394  */
395 static void
396 ptbl_hold(pmap_t pmap, unsigned int pdir_idx)
397 {
398 	vm_paddr_t pa;
399 	pte_t *ptbl;
400 	vm_page_t m;
401 	int i;
402 
403 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
404 	    pdir_idx);
405 
406 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
407 	    ("ptbl_hold: invalid pdir_idx"));
408 	KASSERT((pmap != kernel_pmap),
409 	    ("ptbl_hold: holding kernel ptbl!"));
410 
411 	ptbl = pmap->pm_pdir[pdir_idx];
412 
413 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
414 
415 	for (i = 0; i < PTBL_PAGES; i++) {
416 		pa = pte_vatopa(kernel_pmap,
417 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
418 		m = PHYS_TO_VM_PAGE(pa);
419 		m->ref_count++;
420 	}
421 }
422 
423 /*
424  * Clean pte entry, try to free page table page if requested.
425  *
426  * Return 1 if ptbl pages were freed, otherwise return 0.
427  */
428 static int
429 pte_remove(pmap_t pmap, vm_offset_t va, uint8_t flags)
430 {
431 	unsigned int pdir_idx = PDIR_IDX(va);
432 	unsigned int ptbl_idx = PTBL_IDX(va);
433 	vm_page_t m;
434 	pte_t *ptbl;
435 	pte_t *pte;
436 
437 	//int su = (pmap == kernel_pmap);
438 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
439 	//		su, (u_int32_t)pmap, va, flags);
440 
441 	ptbl = pmap->pm_pdir[pdir_idx];
442 	KASSERT(ptbl, ("pte_remove: null ptbl"));
443 
444 	pte = &ptbl[ptbl_idx];
445 
446 	if (pte == NULL || !PTE_ISVALID(pte))
447 		return (0);
448 
449 	if (PTE_ISWIRED(pte))
450 		pmap->pm_stats.wired_count--;
451 
452 	/* Get vm_page_t for mapped pte. */
453 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
454 
455 	/* Handle managed entry. */
456 	if (PTE_ISMANAGED(pte)) {
457 		if (PTE_ISMODIFIED(pte))
458 			vm_page_dirty(m);
459 
460 		if (PTE_ISREFERENCED(pte))
461 			vm_page_aflag_set(m, PGA_REFERENCED);
462 
463 		pv_remove(pmap, va, m);
464 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
465 		/*
466 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
467 		 * used.  This is needed by the NCSW support code for fast
468 		 * VA<->PA translation.
469 		 */
470 		pv_remove(pmap, va, m);
471 		if (TAILQ_EMPTY(&m->md.pv_list))
472 			m->md.pv_tracked = false;
473 	}
474 
475 	mtx_lock_spin(&tlbivax_mutex);
476 	tlb_miss_lock();
477 
478 	tlb0_flush_entry(va);
479 	*pte = 0;
480 
481 	tlb_miss_unlock();
482 	mtx_unlock_spin(&tlbivax_mutex);
483 
484 	pmap->pm_stats.resident_count--;
485 
486 	if (flags & PTBL_UNHOLD) {
487 		//debugf("pte_remove: e (unhold)\n");
488 		return (ptbl_unhold(pmap, pdir_idx));
489 	}
490 
491 	//debugf("pte_remove: e\n");
492 	return (0);
493 }
494 
495 /*
496  * Insert PTE for a given page and virtual address.
497  */
498 static int
499 pte_enter(pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
500     bool nosleep)
501 {
502 	unsigned int pdir_idx = PDIR_IDX(va);
503 	unsigned int ptbl_idx = PTBL_IDX(va);
504 	pte_t *ptbl, *pte, pte_tmp;
505 
506 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
507 	    pmap == kernel_pmap, pmap, va);
508 
509 	/* Get the page table pointer. */
510 	ptbl = pmap->pm_pdir[pdir_idx];
511 
512 	if (ptbl == NULL) {
513 		/* Allocate page table pages. */
514 		ptbl = ptbl_alloc(pmap, pdir_idx, nosleep);
515 		if (ptbl == NULL) {
516 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
517 			return (ENOMEM);
518 		}
519 		pmap->pm_pdir[pdir_idx] = ptbl;
520 		pte = &ptbl[ptbl_idx];
521 	} else {
522 		/*
523 		 * Check if there is valid mapping for requested
524 		 * va, if there is, remove it.
525 		 */
526 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
527 		if (PTE_ISVALID(pte)) {
528 			pte_remove(pmap, va, PTBL_HOLD);
529 		} else {
530 			/*
531 			 * pte is not used, increment hold count
532 			 * for ptbl pages.
533 			 */
534 			if (pmap != kernel_pmap)
535 				ptbl_hold(pmap, pdir_idx);
536 		}
537 	}
538 
539 	/*
540 	 * Insert pv_entry into pv_list for mapped page if part of managed
541 	 * memory.
542 	 */
543 	if ((m->oflags & VPO_UNMANAGED) == 0) {
544 		flags |= PTE_MANAGED;
545 
546 		/* Create and insert pv entry. */
547 		pv_insert(pmap, va, m);
548 	}
549 
550 	pmap->pm_stats.resident_count++;
551 
552 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
553 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
554 
555 	mtx_lock_spin(&tlbivax_mutex);
556 	tlb_miss_lock();
557 
558 	tlb0_flush_entry(va);
559 	*pte = pte_tmp;
560 
561 	tlb_miss_unlock();
562 	mtx_unlock_spin(&tlbivax_mutex);
563 	return (0);
564 }
565 
566 /* Return the pa for the given pmap/va. */
567 static vm_paddr_t
568 pte_vatopa(pmap_t pmap, vm_offset_t va)
569 {
570 	vm_paddr_t pa = 0;
571 	pte_t *pte;
572 
573 	pte = pte_find(pmap, va);
574 	if ((pte != NULL) && PTE_ISVALID(pte))
575 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
576 	return (pa);
577 }
578 
579 /* Get a pointer to a PTE in a page table. */
580 static pte_t *
581 pte_find(pmap_t pmap, vm_offset_t va)
582 {
583 	unsigned int pdir_idx = PDIR_IDX(va);
584 	unsigned int ptbl_idx = PTBL_IDX(va);
585 
586 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
587 
588 	if (pmap->pm_pdir[pdir_idx])
589 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
590 
591 	return (NULL);
592 }
593 
594 /* Get a pointer to a PTE in a page table, or the next closest (greater) one. */
595 static __inline pte_t *
596 pte_find_next(pmap_t pmap, vm_offset_t *pva)
597 {
598 	vm_offset_t	va;
599 	pte_t	      **pdir;
600 	pte_t	       *pte;
601 	unsigned long	i, j;
602 
603 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
604 
605 	va = *pva;
606 	i = PDIR_IDX(va);
607 	j = PTBL_IDX(va);
608 	pdir = pmap->pm_pdir;
609 	for (; i < PDIR_NENTRIES; i++, j = 0) {
610 		if (pdir[i] == NULL)
611 			continue;
612 		for (; j < PTBL_NENTRIES; j++) {
613 			pte = &pdir[i][j];
614 			if (!PTE_ISVALID(pte))
615 				continue;
616 			*pva = PDIR_SIZE * i + PAGE_SIZE * j;
617 			return (pte);
618 		}
619 	}
620 	return (NULL);
621 }
622 
623 /* Set up kernel page tables. */
624 static void
625 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr)
626 {
627 	pte_t		*pte;
628 	vm_offset_t	va;
629 	vm_offset_t	pdir_start;
630 	int		i;
631 
632 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
633 	kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
634 
635 	pdir_start = kernel_ptbl_root + PDIR_NENTRIES * sizeof(pte_t);
636 
637 	/* Initialize kernel pdir */
638 	for (i = 0; i < kernel_ptbls; i++) {
639 		kernel_pmap->pm_pdir[kptbl_min + i] =
640 		    (pte_t *)(pdir_start + (i * PAGE_SIZE * PTBL_PAGES));
641 	}
642 
643 	/*
644 	 * Fill in PTEs covering kernel code and data. They are not required
645 	 * for address translation, as this area is covered by static TLB1
646 	 * entries, but for pte_vatopa() to work correctly with kernel area
647 	 * addresses.
648 	 */
649 	for (va = addr; va < data_end; va += PAGE_SIZE) {
650 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
651 		powerpc_sync();
652 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
653 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
654 		    PTE_VALID | PTE_PS_4KB;
655 	}
656 }
657 
658 static vm_offset_t
659 mmu_booke_alloc_kernel_pgtables(vm_offset_t data_end)
660 {
661 	/* Allocate space for ptbl_bufs. */
662 	ptbl_bufs = (struct ptbl_buf *)data_end;
663 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
664 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
665 	    (uintptr_t)ptbl_bufs, data_end);
666 
667 	data_end = round_page(data_end);
668 
669 	kernel_ptbl_root = data_end;
670 	data_end += PDIR_NENTRIES * sizeof(pte_t*);
671 
672 	/* Allocate PTE tables for kernel KVA. */
673 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
674 	    PDIR_SIZE);
675 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
676 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
677 	debugf(" kernel pdir at %#jx end = %#jx\n",
678 	    (uintmax_t)kernel_ptbl_root, (uintmax_t)data_end);
679 
680 	return (data_end);
681 }
682 
683 /*
684  * Initialize a preallocated and zeroed pmap structure,
685  * such as one in a vmspace structure.
686  */
687 static int
688 mmu_booke_pinit(pmap_t pmap)
689 {
690 	int i;
691 
692 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
693 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
694 
695 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
696 
697 	for (i = 0; i < MAXCPU; i++)
698 		pmap->pm_tid[i] = TID_NONE;
699 	CPU_ZERO(&kernel_pmap->pm_active);
700 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
701 	pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
702 	bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
703 	TAILQ_INIT(&pmap->pm_ptbl_list);
704 
705 	return (1);
706 }
707 
708 /*
709  * Release any resources held by the given physical map.
710  * Called when a pmap initialized by mmu_booke_pinit is being released.
711  * Should only be called if the map contains no valid mappings.
712  */
713 static void
714 mmu_booke_release(pmap_t pmap)
715 {
716 
717 	KASSERT(pmap->pm_stats.resident_count == 0,
718 	    ("pmap_release: pmap resident count %ld != 0",
719 	    pmap->pm_stats.resident_count));
720 	uma_zfree(ptbl_root_zone, pmap->pm_pdir);
721 }
722 
723 static void
724 mmu_booke_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
725 {
726 	pte_t *pte;
727 	vm_paddr_t pa = 0;
728 	int sync_sz, valid;
729 	pmap_t pmap;
730 	vm_page_t m;
731 	vm_offset_t addr;
732 	int active;
733 
734 	rw_wlock(&pvh_global_lock);
735 	pmap = PCPU_GET(curpmap);
736 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
737 	while (sz > 0) {
738 		PMAP_LOCK(pm);
739 		pte = pte_find(pm, va);
740 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
741 		if (valid)
742 			pa = PTE_PA(pte);
743 		PMAP_UNLOCK(pm);
744 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
745 		sync_sz = min(sync_sz, sz);
746 		if (valid) {
747 			if (!active) {
748 				/*
749 				 * Create a mapping in the active pmap.
750 				 *
751 				 * XXX: We use the zero page here, because
752 				 * it isn't likely to be in use.
753 				 * If we ever decide to support
754 				 * security.bsd.map_at_zero on Book-E, change
755 				 * this to some other address that isn't
756 				 * normally mappable.
757 				 */
758 				addr = 0;
759 				m = PHYS_TO_VM_PAGE(pa);
760 				PMAP_LOCK(pmap);
761 				pte_enter(pmap, m, addr,
762 				    PTE_SR | PTE_VALID, false);
763 				__syncicache((void *)(addr + (va & PAGE_MASK)),
764 				    sync_sz);
765 				pte_remove(pmap, addr, PTBL_UNHOLD);
766 				PMAP_UNLOCK(pmap);
767 			} else
768 				__syncicache((void *)va, sync_sz);
769 		}
770 		va += sync_sz;
771 		sz -= sync_sz;
772 	}
773 	rw_wunlock(&pvh_global_lock);
774 }
775 
776 /*
777  * mmu_booke_zero_page_area zeros the specified hardware page by
778  * mapping it into virtual memory and using bzero to clear
779  * its contents.
780  *
781  * off and size must reside within a single page.
782  */
783 static void
784 mmu_booke_zero_page_area(vm_page_t m, int off, int size)
785 {
786 	vm_offset_t va;
787 
788 	/* XXX KASSERT off and size are within a single page? */
789 
790 	mtx_lock(&zero_page_mutex);
791 	va = zero_page_va;
792 
793 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
794 	bzero((caddr_t)va + off, size);
795 	mmu_booke_kremove(va);
796 
797 	mtx_unlock(&zero_page_mutex);
798 }
799 
800 /*
801  * mmu_booke_zero_page zeros the specified hardware page.
802  */
803 static void
804 mmu_booke_zero_page(vm_page_t m)
805 {
806 	vm_offset_t off, va;
807 
808 	va = zero_page_va;
809 	mtx_lock(&zero_page_mutex);
810 
811 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
812 
813 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
814 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
815 
816 	mmu_booke_kremove(va);
817 
818 	mtx_unlock(&zero_page_mutex);
819 }
820 
821 /*
822  * mmu_booke_copy_page copies the specified (machine independent) page by
823  * mapping the page into virtual memory and using memcopy to copy the page,
824  * one machine dependent page at a time.
825  */
826 static void
827 mmu_booke_copy_page(vm_page_t sm, vm_page_t dm)
828 {
829 	vm_offset_t sva, dva;
830 
831 	sva = copy_page_src_va;
832 	dva = copy_page_dst_va;
833 
834 	mtx_lock(&copy_page_mutex);
835 	mmu_booke_kenter(sva, VM_PAGE_TO_PHYS(sm));
836 	mmu_booke_kenter(dva, VM_PAGE_TO_PHYS(dm));
837 
838 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
839 
840 	mmu_booke_kremove(dva);
841 	mmu_booke_kremove(sva);
842 	mtx_unlock(&copy_page_mutex);
843 }
844 
845 static inline void
846 mmu_booke_copy_pages(vm_page_t *ma, vm_offset_t a_offset,
847     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
848 {
849 	void *a_cp, *b_cp;
850 	vm_offset_t a_pg_offset, b_pg_offset;
851 	int cnt;
852 
853 	mtx_lock(&copy_page_mutex);
854 	while (xfersize > 0) {
855 		a_pg_offset = a_offset & PAGE_MASK;
856 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
857 		mmu_booke_kenter(copy_page_src_va,
858 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
859 		a_cp = (char *)copy_page_src_va + a_pg_offset;
860 		b_pg_offset = b_offset & PAGE_MASK;
861 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
862 		mmu_booke_kenter(copy_page_dst_va,
863 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
864 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
865 		bcopy(a_cp, b_cp, cnt);
866 		mmu_booke_kremove(copy_page_dst_va);
867 		mmu_booke_kremove(copy_page_src_va);
868 		a_offset += cnt;
869 		b_offset += cnt;
870 		xfersize -= cnt;
871 	}
872 	mtx_unlock(&copy_page_mutex);
873 }
874 
875 static vm_offset_t
876 mmu_booke_quick_enter_page(vm_page_t m)
877 {
878 	vm_paddr_t paddr;
879 	vm_offset_t qaddr;
880 	uint32_t flags;
881 	pte_t *pte;
882 
883 	paddr = VM_PAGE_TO_PHYS(m);
884 
885 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
886 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
887 	flags |= PTE_PS_4KB;
888 
889 	critical_enter();
890 	qaddr = PCPU_GET(qmap_addr);
891 
892 	pte = pte_find(kernel_pmap, qaddr);
893 
894 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
895 
896 	/*
897 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
898  	 * not be present in other TLBs.  Is there a better instruction
899 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
900 	 */
901 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
902 	__asm __volatile("isync; msync");
903 
904 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
905 
906 	/* Flush the real memory from the instruction cache. */
907 	if ((flags & (PTE_I | PTE_G)) == 0)
908 		__syncicache((void *)qaddr, PAGE_SIZE);
909 
910 	return (qaddr);
911 }
912 
913 static void
914 mmu_booke_quick_remove_page(vm_offset_t addr)
915 {
916 	pte_t *pte;
917 
918 	pte = pte_find(kernel_pmap, addr);
919 
920 	KASSERT(PCPU_GET(qmap_addr) == addr,
921 	    ("mmu_booke_quick_remove_page: invalid address"));
922 	KASSERT(*pte != 0,
923 	    ("mmu_booke_quick_remove_page: PTE not in use"));
924 
925 	*pte = 0;
926 	critical_exit();
927 }
928 
929 /**************************************************************************/
930 /* TID handling */
931 /**************************************************************************/
932 
933 /*
934  * Invalidate all TLB0 entries which match the given TID. Note this is
935  * dedicated for cases when invalidations should NOT be propagated to other
936  * CPUs.
937  */
938 static void
939 tid_flush(tlbtid_t tid)
940 {
941 	register_t msr;
942 	uint32_t mas0, mas1, mas2;
943 	int entry, way;
944 
945 	/* Don't evict kernel translations */
946 	if (tid == TID_KERNEL)
947 		return;
948 
949 	msr = mfmsr();
950 	__asm __volatile("wrteei 0");
951 
952 	/*
953 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
954 	 * it for PID invalidation.
955 	 */
956 	switch ((mfpvr() >> 16) & 0xffff) {
957 	case FSL_E500mc:
958 	case FSL_E5500:
959 	case FSL_E6500:
960 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
961 		/* tlbilxpid */
962 		__asm __volatile("isync; .long 0x7c200024; isync; msync");
963 		__asm __volatile("wrtee %0" :: "r"(msr));
964 		return;
965 	}
966 
967 	for (way = 0; way < TLB0_WAYS; way++)
968 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
969 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
970 			mtspr(SPR_MAS0, mas0);
971 
972 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
973 			mtspr(SPR_MAS2, mas2);
974 
975 			__asm __volatile("isync; tlbre");
976 
977 			mas1 = mfspr(SPR_MAS1);
978 
979 			if (!(mas1 & MAS1_VALID))
980 				continue;
981 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
982 				continue;
983 			mas1 &= ~MAS1_VALID;
984 			mtspr(SPR_MAS1, mas1);
985 			__asm __volatile("isync; tlbwe; isync; msync");
986 		}
987 	__asm __volatile("wrtee %0" :: "r"(msr));
988 }
989