xref: /freebsd/sys/powerpc/booke/pmap_32.c (revision 1edb7116f450c1a1793f2fd25f6bdc16735ef888)
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 static unsigned long ilog2(unsigned long);
120 
121 /**************************************************************************/
122 /* Page table management */
123 /**************************************************************************/
124 
125 #define PMAP_ROOT_SIZE	(sizeof(pte_t**) * PDIR_NENTRIES)
126 static void ptbl_init(void);
127 static struct ptbl_buf *ptbl_buf_alloc(void);
128 static void ptbl_buf_free(struct ptbl_buf *);
129 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
130 
131 static pte_t *ptbl_alloc(pmap_t, unsigned int, bool);
132 static void ptbl_free(pmap_t, unsigned int);
133 static void ptbl_hold(pmap_t, unsigned int);
134 static int ptbl_unhold(pmap_t, unsigned int);
135 
136 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
137 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, bool);
138 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
139 static pte_t *pte_find(pmap_t, vm_offset_t);
140 
141 struct ptbl_buf {
142 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
143 	vm_offset_t kva;		/* va of mapping */
144 };
145 
146 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
147 #define PTBL_BUFS		(128 * 16)
148 
149 /* ptbl free list and a lock used for access synchronization. */
150 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
151 static struct mtx ptbl_buf_freelist_lock;
152 
153 /* Base address of kva space allocated fot ptbl bufs. */
154 static vm_offset_t ptbl_buf_pool_vabase;
155 
156 /* Pointer to ptbl_buf structures. */
157 static struct ptbl_buf *ptbl_bufs;
158 
159 /**************************************************************************/
160 /* Page table related */
161 /**************************************************************************/
162 
163 /* Initialize pool of kva ptbl buffers. */
164 static void
165 ptbl_init(void)
166 {
167 	int i;
168 
169 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
170 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
171 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
172 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
173 
174 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
175 	TAILQ_INIT(&ptbl_buf_freelist);
176 
177 	for (i = 0; i < PTBL_BUFS; i++) {
178 		ptbl_bufs[i].kva =
179 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
180 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
181 	}
182 }
183 
184 /* Get a ptbl_buf from the freelist. */
185 static struct ptbl_buf *
186 ptbl_buf_alloc(void)
187 {
188 	struct ptbl_buf *buf;
189 
190 	mtx_lock(&ptbl_buf_freelist_lock);
191 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
192 	if (buf != NULL)
193 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
194 	mtx_unlock(&ptbl_buf_freelist_lock);
195 
196 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
197 
198 	return (buf);
199 }
200 
201 /* Return ptbl buff to free pool. */
202 static void
203 ptbl_buf_free(struct ptbl_buf *buf)
204 {
205 
206 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
207 
208 	mtx_lock(&ptbl_buf_freelist_lock);
209 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
210 	mtx_unlock(&ptbl_buf_freelist_lock);
211 }
212 
213 /*
214  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
215  */
216 static void
217 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
218 {
219 	struct ptbl_buf *pbuf;
220 
221 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
222 
223 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
224 
225 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
226 		if (pbuf->kva == (vm_offset_t)ptbl) {
227 			/* Remove from pmap ptbl buf list. */
228 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
229 
230 			/* Free corresponding ptbl buf. */
231 			ptbl_buf_free(pbuf);
232 			break;
233 		}
234 }
235 
236 /* Allocate page table. */
237 static pte_t *
238 ptbl_alloc(pmap_t pmap, unsigned int pdir_idx, bool nosleep)
239 {
240 	vm_page_t mtbl[PTBL_PAGES];
241 	vm_page_t m;
242 	struct ptbl_buf *pbuf;
243 	unsigned int pidx;
244 	pte_t *ptbl;
245 	int i, j;
246 
247 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
248 	    (pmap == kernel_pmap), pdir_idx);
249 
250 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
251 	    ("ptbl_alloc: invalid pdir_idx"));
252 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
253 	    ("pte_alloc: valid ptbl entry exists!"));
254 
255 	pbuf = ptbl_buf_alloc();
256 	if (pbuf == NULL)
257 		panic("pte_alloc: couldn't alloc kernel virtual memory");
258 
259 	ptbl = (pte_t *)pbuf->kva;
260 
261 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
262 
263 	for (i = 0; i < PTBL_PAGES; i++) {
264 		pidx = (PTBL_PAGES * pdir_idx) + i;
265 		while ((m = vm_page_alloc_noobj(VM_ALLOC_WIRED)) == NULL) {
266 			if (nosleep) {
267 				ptbl_free_pmap_ptbl(pmap, ptbl);
268 				for (j = 0; j < i; j++)
269 					vm_page_free(mtbl[j]);
270 				vm_wire_sub(i);
271 				return (NULL);
272 			}
273 			PMAP_UNLOCK(pmap);
274 			rw_wunlock(&pvh_global_lock);
275 			vm_wait(NULL);
276 			rw_wlock(&pvh_global_lock);
277 			PMAP_LOCK(pmap);
278 		}
279 		m->pindex = pidx;
280 		mtbl[i] = m;
281 	}
282 
283 	/* Map allocated pages into kernel_pmap. */
284 	mmu_booke_qenter((vm_offset_t)ptbl, mtbl, PTBL_PAGES);
285 
286 	/* Zero whole ptbl. */
287 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
288 
289 	/* Add pbuf to the pmap ptbl bufs list. */
290 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
291 
292 	return (ptbl);
293 }
294 
295 /* Free ptbl pages and invalidate pdir entry. */
296 static void
297 ptbl_free(pmap_t pmap, unsigned int pdir_idx)
298 {
299 	pte_t *ptbl;
300 	vm_paddr_t pa;
301 	vm_offset_t va;
302 	vm_page_t m;
303 	int i;
304 
305 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
306 	    (pmap == kernel_pmap), pdir_idx);
307 
308 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
309 	    ("ptbl_free: invalid pdir_idx"));
310 
311 	ptbl = pmap->pm_pdir[pdir_idx];
312 
313 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
314 
315 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
316 
317 	/*
318 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
319 	 * don't attempt to look up the page tables we are releasing.
320 	 */
321 	mtx_lock_spin(&tlbivax_mutex);
322 	tlb_miss_lock();
323 
324 	pmap->pm_pdir[pdir_idx] = NULL;
325 
326 	tlb_miss_unlock();
327 	mtx_unlock_spin(&tlbivax_mutex);
328 
329 	for (i = 0; i < PTBL_PAGES; i++) {
330 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
331 		pa = pte_vatopa(kernel_pmap, va);
332 		m = PHYS_TO_VM_PAGE(pa);
333 		vm_page_free_zero(m);
334 		vm_wire_sub(1);
335 		mmu_booke_kremove(va);
336 	}
337 
338 	ptbl_free_pmap_ptbl(pmap, ptbl);
339 }
340 
341 /*
342  * Decrement ptbl pages hold count and attempt to free ptbl pages.
343  * Called when removing pte entry from ptbl.
344  *
345  * Return 1 if ptbl pages were freed.
346  */
347 static int
348 ptbl_unhold(pmap_t pmap, unsigned int pdir_idx)
349 {
350 	pte_t *ptbl;
351 	vm_paddr_t pa;
352 	vm_page_t m;
353 	int i;
354 
355 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
356 	    (pmap == kernel_pmap), pdir_idx);
357 
358 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
359 	    ("ptbl_unhold: invalid pdir_idx"));
360 	KASSERT((pmap != kernel_pmap),
361 	    ("ptbl_unhold: unholding kernel ptbl!"));
362 
363 	ptbl = pmap->pm_pdir[pdir_idx];
364 
365 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
366 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
367 	    ("ptbl_unhold: non kva ptbl"));
368 
369 	/* decrement hold count */
370 	for (i = 0; i < PTBL_PAGES; i++) {
371 		pa = pte_vatopa(kernel_pmap,
372 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
373 		m = PHYS_TO_VM_PAGE(pa);
374 		m->ref_count--;
375 	}
376 
377 	/*
378 	 * Free ptbl pages if there are no pte etries in this ptbl.
379 	 * ref_count has the same value for all ptbl pages, so check the last
380 	 * page.
381 	 */
382 	if (m->ref_count == 0) {
383 		ptbl_free(pmap, pdir_idx);
384 
385 		//debugf("ptbl_unhold: e (freed ptbl)\n");
386 		return (1);
387 	}
388 
389 	return (0);
390 }
391 
392 /*
393  * Increment hold count for ptbl pages. This routine is used when a new pte
394  * entry is being inserted into the ptbl.
395  */
396 static void
397 ptbl_hold(pmap_t pmap, unsigned int pdir_idx)
398 {
399 	vm_paddr_t pa;
400 	pte_t *ptbl;
401 	vm_page_t m;
402 	int i;
403 
404 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
405 	    pdir_idx);
406 
407 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
408 	    ("ptbl_hold: invalid pdir_idx"));
409 	KASSERT((pmap != kernel_pmap),
410 	    ("ptbl_hold: holding kernel ptbl!"));
411 
412 	ptbl = pmap->pm_pdir[pdir_idx];
413 
414 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
415 
416 	for (i = 0; i < PTBL_PAGES; i++) {
417 		pa = pte_vatopa(kernel_pmap,
418 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
419 		m = PHYS_TO_VM_PAGE(pa);
420 		m->ref_count++;
421 	}
422 }
423 
424 /*
425  * Clean pte entry, try to free page table page if requested.
426  *
427  * Return 1 if ptbl pages were freed, otherwise return 0.
428  */
429 static int
430 pte_remove(pmap_t pmap, vm_offset_t va, uint8_t flags)
431 {
432 	unsigned int pdir_idx = PDIR_IDX(va);
433 	unsigned int ptbl_idx = PTBL_IDX(va);
434 	vm_page_t m;
435 	pte_t *ptbl;
436 	pte_t *pte;
437 
438 	//int su = (pmap == kernel_pmap);
439 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
440 	//		su, (u_int32_t)pmap, va, flags);
441 
442 	ptbl = pmap->pm_pdir[pdir_idx];
443 	KASSERT(ptbl, ("pte_remove: null ptbl"));
444 
445 	pte = &ptbl[ptbl_idx];
446 
447 	if (pte == NULL || !PTE_ISVALID(pte))
448 		return (0);
449 
450 	if (PTE_ISWIRED(pte))
451 		pmap->pm_stats.wired_count--;
452 
453 	/* Get vm_page_t for mapped pte. */
454 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
455 
456 	/* Handle managed entry. */
457 	if (PTE_ISMANAGED(pte)) {
458 		if (PTE_ISMODIFIED(pte))
459 			vm_page_dirty(m);
460 
461 		if (PTE_ISREFERENCED(pte))
462 			vm_page_aflag_set(m, PGA_REFERENCED);
463 
464 		pv_remove(pmap, va, m);
465 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
466 		/*
467 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
468 		 * used.  This is needed by the NCSW support code for fast
469 		 * VA<->PA translation.
470 		 */
471 		pv_remove(pmap, va, m);
472 		if (TAILQ_EMPTY(&m->md.pv_list))
473 			m->md.pv_tracked = false;
474 	}
475 
476 	mtx_lock_spin(&tlbivax_mutex);
477 	tlb_miss_lock();
478 
479 	tlb0_flush_entry(va);
480 	*pte = 0;
481 
482 	tlb_miss_unlock();
483 	mtx_unlock_spin(&tlbivax_mutex);
484 
485 	pmap->pm_stats.resident_count--;
486 
487 	if (flags & PTBL_UNHOLD) {
488 		//debugf("pte_remove: e (unhold)\n");
489 		return (ptbl_unhold(pmap, pdir_idx));
490 	}
491 
492 	//debugf("pte_remove: e\n");
493 	return (0);
494 }
495 
496 /*
497  * Insert PTE for a given page and virtual address.
498  */
499 static int
500 pte_enter(pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
501     bool nosleep)
502 {
503 	unsigned int pdir_idx = PDIR_IDX(va);
504 	unsigned int ptbl_idx = PTBL_IDX(va);
505 	pte_t *ptbl, *pte, pte_tmp;
506 
507 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
508 	    pmap == kernel_pmap, pmap, va);
509 
510 	/* Get the page table pointer. */
511 	ptbl = pmap->pm_pdir[pdir_idx];
512 
513 	if (ptbl == NULL) {
514 		/* Allocate page table pages. */
515 		ptbl = ptbl_alloc(pmap, pdir_idx, nosleep);
516 		if (ptbl == NULL) {
517 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
518 			return (ENOMEM);
519 		}
520 		pmap->pm_pdir[pdir_idx] = ptbl;
521 		pte = &ptbl[ptbl_idx];
522 	} else {
523 		/*
524 		 * Check if there is valid mapping for requested
525 		 * va, if there is, remove it.
526 		 */
527 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
528 		if (PTE_ISVALID(pte)) {
529 			pte_remove(pmap, va, PTBL_HOLD);
530 		} else {
531 			/*
532 			 * pte is not used, increment hold count
533 			 * for ptbl pages.
534 			 */
535 			if (pmap != kernel_pmap)
536 				ptbl_hold(pmap, pdir_idx);
537 		}
538 	}
539 
540 	/*
541 	 * Insert pv_entry into pv_list for mapped page if part of managed
542 	 * memory.
543 	 */
544 	if ((m->oflags & VPO_UNMANAGED) == 0) {
545 		flags |= PTE_MANAGED;
546 
547 		/* Create and insert pv entry. */
548 		pv_insert(pmap, va, m);
549 	}
550 
551 	pmap->pm_stats.resident_count++;
552 
553 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
554 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
555 
556 	mtx_lock_spin(&tlbivax_mutex);
557 	tlb_miss_lock();
558 
559 	tlb0_flush_entry(va);
560 	*pte = pte_tmp;
561 
562 	tlb_miss_unlock();
563 	mtx_unlock_spin(&tlbivax_mutex);
564 	return (0);
565 }
566 
567 /* Return the pa for the given pmap/va. */
568 static vm_paddr_t
569 pte_vatopa(pmap_t pmap, vm_offset_t va)
570 {
571 	vm_paddr_t pa = 0;
572 	pte_t *pte;
573 
574 	pte = pte_find(pmap, va);
575 	if ((pte != NULL) && PTE_ISVALID(pte))
576 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
577 	return (pa);
578 }
579 
580 /* Get a pointer to a PTE in a page table. */
581 static pte_t *
582 pte_find(pmap_t pmap, vm_offset_t va)
583 {
584 	unsigned int pdir_idx = PDIR_IDX(va);
585 	unsigned int ptbl_idx = PTBL_IDX(va);
586 
587 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
588 
589 	if (pmap->pm_pdir[pdir_idx])
590 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
591 
592 	return (NULL);
593 }
594 
595 /* Get a pointer to a PTE in a page table, or the next closest (greater) one. */
596 static __inline pte_t *
597 pte_find_next(pmap_t pmap, vm_offset_t *pva)
598 {
599 	vm_offset_t	va;
600 	pte_t	      **pdir;
601 	pte_t	       *pte;
602 	unsigned long	i, j;
603 
604 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
605 
606 	va = *pva;
607 	i = PDIR_IDX(va);
608 	j = PTBL_IDX(va);
609 	pdir = pmap->pm_pdir;
610 	for (; i < PDIR_NENTRIES; i++, j = 0) {
611 		if (pdir[i] == NULL)
612 			continue;
613 		for (; j < PTBL_NENTRIES; j++) {
614 			pte = &pdir[i][j];
615 			if (!PTE_ISVALID(pte))
616 				continue;
617 			*pva = PDIR_SIZE * i + PAGE_SIZE * j;
618 			return (pte);
619 		}
620 	}
621 	return (NULL);
622 }
623 
624 /* Set up kernel page tables. */
625 static void
626 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr)
627 {
628 	pte_t		*pte;
629 	vm_offset_t	va;
630 	vm_offset_t	pdir_start;
631 	int		i;
632 
633 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
634 	kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
635 
636 	pdir_start = kernel_ptbl_root + PDIR_NENTRIES * sizeof(pte_t);
637 
638 	/* Initialize kernel pdir */
639 	for (i = 0; i < kernel_ptbls; i++) {
640 		kernel_pmap->pm_pdir[kptbl_min + i] =
641 		    (pte_t *)(pdir_start + (i * PAGE_SIZE * PTBL_PAGES));
642 	}
643 
644 	/*
645 	 * Fill in PTEs covering kernel code and data. They are not required
646 	 * for address translation, as this area is covered by static TLB1
647 	 * entries, but for pte_vatopa() to work correctly with kernel area
648 	 * addresses.
649 	 */
650 	for (va = addr; va < data_end; va += PAGE_SIZE) {
651 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
652 		powerpc_sync();
653 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
654 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
655 		    PTE_VALID | PTE_PS_4KB;
656 	}
657 }
658 
659 static vm_offset_t
660 mmu_booke_alloc_kernel_pgtables(vm_offset_t data_end)
661 {
662 	/* Allocate space for ptbl_bufs. */
663 	ptbl_bufs = (struct ptbl_buf *)data_end;
664 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
665 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
666 	    (uintptr_t)ptbl_bufs, data_end);
667 
668 	data_end = round_page(data_end);
669 
670 	kernel_ptbl_root = data_end;
671 	data_end += PDIR_NENTRIES * sizeof(pte_t*);
672 
673 	/* Allocate PTE tables for kernel KVA. */
674 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
675 	    PDIR_SIZE);
676 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
677 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
678 	debugf(" kernel pdir at %#jx end = %#jx\n",
679 	    (uintmax_t)kernel_ptbl_root, (uintmax_t)data_end);
680 
681 	return (data_end);
682 }
683 
684 /*
685  * Initialize a preallocated and zeroed pmap structure,
686  * such as one in a vmspace structure.
687  */
688 static int
689 mmu_booke_pinit(pmap_t pmap)
690 {
691 	int i;
692 
693 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
694 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
695 
696 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
697 
698 	for (i = 0; i < MAXCPU; i++)
699 		pmap->pm_tid[i] = TID_NONE;
700 	CPU_ZERO(&kernel_pmap->pm_active);
701 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
702 	pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
703 	bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
704 	TAILQ_INIT(&pmap->pm_ptbl_list);
705 
706 	return (1);
707 }
708 
709 /*
710  * Release any resources held by the given physical map.
711  * Called when a pmap initialized by mmu_booke_pinit is being released.
712  * Should only be called if the map contains no valid mappings.
713  */
714 static void
715 mmu_booke_release(pmap_t pmap)
716 {
717 
718 	KASSERT(pmap->pm_stats.resident_count == 0,
719 	    ("pmap_release: pmap resident count %ld != 0",
720 	    pmap->pm_stats.resident_count));
721 	uma_zfree(ptbl_root_zone, pmap->pm_pdir);
722 }
723 
724 static void
725 mmu_booke_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
726 {
727 	pte_t *pte;
728 	vm_paddr_t pa = 0;
729 	int sync_sz, valid;
730 	pmap_t pmap;
731 	vm_page_t m;
732 	vm_offset_t addr;
733 	int active;
734 
735 	rw_wlock(&pvh_global_lock);
736 	pmap = PCPU_GET(curpmap);
737 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
738 	while (sz > 0) {
739 		PMAP_LOCK(pm);
740 		pte = pte_find(pm, va);
741 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
742 		if (valid)
743 			pa = PTE_PA(pte);
744 		PMAP_UNLOCK(pm);
745 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
746 		sync_sz = min(sync_sz, sz);
747 		if (valid) {
748 			if (!active) {
749 				/*
750 				 * Create a mapping in the active pmap.
751 				 *
752 				 * XXX: We use the zero page here, because
753 				 * it isn't likely to be in use.
754 				 * If we ever decide to support
755 				 * security.bsd.map_at_zero on Book-E, change
756 				 * this to some other address that isn't
757 				 * normally mappable.
758 				 */
759 				addr = 0;
760 				m = PHYS_TO_VM_PAGE(pa);
761 				PMAP_LOCK(pmap);
762 				pte_enter(pmap, m, addr,
763 				    PTE_SR | PTE_VALID, false);
764 				__syncicache((void *)(addr + (va & PAGE_MASK)),
765 				    sync_sz);
766 				pte_remove(pmap, addr, PTBL_UNHOLD);
767 				PMAP_UNLOCK(pmap);
768 			} else
769 				__syncicache((void *)va, sync_sz);
770 		}
771 		va += sync_sz;
772 		sz -= sync_sz;
773 	}
774 	rw_wunlock(&pvh_global_lock);
775 }
776 
777 /*
778  * mmu_booke_zero_page_area zeros the specified hardware page by
779  * mapping it into virtual memory and using bzero to clear
780  * its contents.
781  *
782  * off and size must reside within a single page.
783  */
784 static void
785 mmu_booke_zero_page_area(vm_page_t m, int off, int size)
786 {
787 	vm_offset_t va;
788 
789 	/* XXX KASSERT off and size are within a single page? */
790 
791 	mtx_lock(&zero_page_mutex);
792 	va = zero_page_va;
793 
794 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
795 	bzero((caddr_t)va + off, size);
796 	mmu_booke_kremove(va);
797 
798 	mtx_unlock(&zero_page_mutex);
799 }
800 
801 /*
802  * mmu_booke_zero_page zeros the specified hardware page.
803  */
804 static void
805 mmu_booke_zero_page(vm_page_t m)
806 {
807 	vm_offset_t off, va;
808 
809 	va = zero_page_va;
810 	mtx_lock(&zero_page_mutex);
811 
812 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
813 
814 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
815 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
816 
817 	mmu_booke_kremove(va);
818 
819 	mtx_unlock(&zero_page_mutex);
820 }
821 
822 /*
823  * mmu_booke_copy_page copies the specified (machine independent) page by
824  * mapping the page into virtual memory and using memcopy to copy the page,
825  * one machine dependent page at a time.
826  */
827 static void
828 mmu_booke_copy_page(vm_page_t sm, vm_page_t dm)
829 {
830 	vm_offset_t sva, dva;
831 
832 	sva = copy_page_src_va;
833 	dva = copy_page_dst_va;
834 
835 	mtx_lock(&copy_page_mutex);
836 	mmu_booke_kenter(sva, VM_PAGE_TO_PHYS(sm));
837 	mmu_booke_kenter(dva, VM_PAGE_TO_PHYS(dm));
838 
839 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
840 
841 	mmu_booke_kremove(dva);
842 	mmu_booke_kremove(sva);
843 	mtx_unlock(&copy_page_mutex);
844 }
845 
846 static inline void
847 mmu_booke_copy_pages(vm_page_t *ma, vm_offset_t a_offset,
848     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
849 {
850 	void *a_cp, *b_cp;
851 	vm_offset_t a_pg_offset, b_pg_offset;
852 	int cnt;
853 
854 	mtx_lock(&copy_page_mutex);
855 	while (xfersize > 0) {
856 		a_pg_offset = a_offset & PAGE_MASK;
857 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
858 		mmu_booke_kenter(copy_page_src_va,
859 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
860 		a_cp = (char *)copy_page_src_va + a_pg_offset;
861 		b_pg_offset = b_offset & PAGE_MASK;
862 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
863 		mmu_booke_kenter(copy_page_dst_va,
864 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
865 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
866 		bcopy(a_cp, b_cp, cnt);
867 		mmu_booke_kremove(copy_page_dst_va);
868 		mmu_booke_kremove(copy_page_src_va);
869 		a_offset += cnt;
870 		b_offset += cnt;
871 		xfersize -= cnt;
872 	}
873 	mtx_unlock(&copy_page_mutex);
874 }
875 
876 static vm_offset_t
877 mmu_booke_quick_enter_page(vm_page_t m)
878 {
879 	vm_paddr_t paddr;
880 	vm_offset_t qaddr;
881 	uint32_t flags;
882 	pte_t *pte;
883 
884 	paddr = VM_PAGE_TO_PHYS(m);
885 
886 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
887 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
888 	flags |= PTE_PS_4KB;
889 
890 	critical_enter();
891 	qaddr = PCPU_GET(qmap_addr);
892 
893 	pte = pte_find(kernel_pmap, qaddr);
894 
895 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
896 
897 	/*
898 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
899  	 * not be present in other TLBs.  Is there a better instruction
900 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
901 	 */
902 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
903 	__asm __volatile("isync; msync");
904 
905 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
906 
907 	/* Flush the real memory from the instruction cache. */
908 	if ((flags & (PTE_I | PTE_G)) == 0)
909 		__syncicache((void *)qaddr, PAGE_SIZE);
910 
911 	return (qaddr);
912 }
913 
914 static void
915 mmu_booke_quick_remove_page(vm_offset_t addr)
916 {
917 	pte_t *pte;
918 
919 	pte = pte_find(kernel_pmap, addr);
920 
921 	KASSERT(PCPU_GET(qmap_addr) == addr,
922 	    ("mmu_booke_quick_remove_page: invalid address"));
923 	KASSERT(*pte != 0,
924 	    ("mmu_booke_quick_remove_page: PTE not in use"));
925 
926 	*pte = 0;
927 	critical_exit();
928 }
929 
930 /**************************************************************************/
931 /* TID handling */
932 /**************************************************************************/
933 
934 /*
935  * Return the largest uint value log such that 2^log <= num.
936  */
937 static unsigned long
938 ilog2(unsigned long num)
939 {
940 	long lz;
941 
942 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
943 	return (31 - lz);
944 }
945 
946 /*
947  * Invalidate all TLB0 entries which match the given TID. Note this is
948  * dedicated for cases when invalidations should NOT be propagated to other
949  * CPUs.
950  */
951 static void
952 tid_flush(tlbtid_t tid)
953 {
954 	register_t msr;
955 	uint32_t mas0, mas1, mas2;
956 	int entry, way;
957 
958 	/* Don't evict kernel translations */
959 	if (tid == TID_KERNEL)
960 		return;
961 
962 	msr = mfmsr();
963 	__asm __volatile("wrteei 0");
964 
965 	/*
966 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
967 	 * it for PID invalidation.
968 	 */
969 	switch ((mfpvr() >> 16) & 0xffff) {
970 	case FSL_E500mc:
971 	case FSL_E5500:
972 	case FSL_E6500:
973 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
974 		/* tlbilxpid */
975 		__asm __volatile("isync; .long 0x7c200024; isync; msync");
976 		__asm __volatile("wrtee %0" :: "r"(msr));
977 		return;
978 	}
979 
980 	for (way = 0; way < TLB0_WAYS; way++)
981 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
982 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
983 			mtspr(SPR_MAS0, mas0);
984 
985 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
986 			mtspr(SPR_MAS2, mas2);
987 
988 			__asm __volatile("isync; tlbre");
989 
990 			mas1 = mfspr(SPR_MAS1);
991 
992 			if (!(mas1 & MAS1_VALID))
993 				continue;
994 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
995 				continue;
996 			mas1 &= ~MAS1_VALID;
997 			mtspr(SPR_MAS1, mas1);
998 			__asm __volatile("isync; tlbwe; isync; msync");
999 		}
1000 	__asm __volatile("wrtee %0" :: "r"(msr));
1001 }
1002