xref: /freebsd/sys/powerpc/booke/pmap_32.c (revision cab6a39d7b343596a5823e65c0f7b426551ec22d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
50 
51 #include "opt_ddb.h"
52 #include "opt_kstack_pages.h"
53 
54 #include <sys/param.h>
55 #include <sys/conf.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
66 #include <sys/lock.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
70 #include <sys/smp.h>
71 #include <sys/vmmeter.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vm_phys.h>
83 #include <vm/vm_pagequeue.h>
84 #include <vm/uma.h>
85 
86 #include <machine/_inttypes.h>
87 #include <machine/cpu.h>
88 #include <machine/pcb.h>
89 #include <machine/platform.h>
90 
91 #include <machine/tlb.h>
92 #include <machine/spr.h>
93 #include <machine/md_var.h>
94 #include <machine/mmuvar.h>
95 #include <machine/pmap.h>
96 #include <machine/pte.h>
97 
98 #include <ddb/ddb.h>
99 
100 #define	PRI0ptrX	"08x"
101 
102 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
103 static vm_offset_t zero_page_va;
104 static struct mtx zero_page_mutex;
105 
106 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
107 static vm_offset_t copy_page_src_va;
108 static vm_offset_t copy_page_dst_va;
109 static struct mtx copy_page_mutex;
110 
111 static vm_offset_t kernel_ptbl_root;
112 static unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
113 
114 /**************************************************************************/
115 /* PMAP */
116 /**************************************************************************/
117 
118 #define	VM_MAPDEV_BASE	((vm_offset_t)VM_MAXUSER_ADDRESS + PAGE_SIZE)
119 
120 static void tid_flush(tlbtid_t tid);
121 static unsigned long ilog2(unsigned long);
122 
123 /**************************************************************************/
124 /* Page table management */
125 /**************************************************************************/
126 
127 #define PMAP_ROOT_SIZE	(sizeof(pte_t**) * PDIR_NENTRIES)
128 static void ptbl_init(void);
129 static struct ptbl_buf *ptbl_buf_alloc(void);
130 static void ptbl_buf_free(struct ptbl_buf *);
131 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
132 
133 static pte_t *ptbl_alloc(pmap_t, unsigned int, boolean_t);
134 static void ptbl_free(pmap_t, unsigned int);
135 static void ptbl_hold(pmap_t, unsigned int);
136 static int ptbl_unhold(pmap_t, unsigned int);
137 
138 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
139 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
140 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
141 static pte_t *pte_find(pmap_t, vm_offset_t);
142 
143 struct ptbl_buf {
144 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
145 	vm_offset_t kva;		/* va of mapping */
146 };
147 
148 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
149 #define PTBL_BUFS		(128 * 16)
150 
151 /* ptbl free list and a lock used for access synchronization. */
152 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
153 static struct mtx ptbl_buf_freelist_lock;
154 
155 /* Base address of kva space allocated fot ptbl bufs. */
156 static vm_offset_t ptbl_buf_pool_vabase;
157 
158 /* Pointer to ptbl_buf structures. */
159 static struct ptbl_buf *ptbl_bufs;
160 
161 /**************************************************************************/
162 /* Page table related */
163 /**************************************************************************/
164 
165 /* Initialize pool of kva ptbl buffers. */
166 static void
167 ptbl_init(void)
168 {
169 	int i;
170 
171 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
172 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
173 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
174 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
175 
176 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
177 	TAILQ_INIT(&ptbl_buf_freelist);
178 
179 	for (i = 0; i < PTBL_BUFS; i++) {
180 		ptbl_bufs[i].kva =
181 		    ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
182 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
183 	}
184 }
185 
186 /* Get a ptbl_buf from the freelist. */
187 static struct ptbl_buf *
188 ptbl_buf_alloc(void)
189 {
190 	struct ptbl_buf *buf;
191 
192 	mtx_lock(&ptbl_buf_freelist_lock);
193 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
194 	if (buf != NULL)
195 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
196 	mtx_unlock(&ptbl_buf_freelist_lock);
197 
198 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
199 
200 	return (buf);
201 }
202 
203 /* Return ptbl buff to free pool. */
204 static void
205 ptbl_buf_free(struct ptbl_buf *buf)
206 {
207 
208 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
209 
210 	mtx_lock(&ptbl_buf_freelist_lock);
211 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
212 	mtx_unlock(&ptbl_buf_freelist_lock);
213 }
214 
215 /*
216  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
217  */
218 static void
219 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
220 {
221 	struct ptbl_buf *pbuf;
222 
223 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
224 
225 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
226 
227 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
228 		if (pbuf->kva == (vm_offset_t)ptbl) {
229 			/* Remove from pmap ptbl buf list. */
230 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
231 
232 			/* Free corresponding ptbl buf. */
233 			ptbl_buf_free(pbuf);
234 			break;
235 		}
236 }
237 
238 /* Allocate page table. */
239 static pte_t *
240 ptbl_alloc(pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
241 {
242 	vm_page_t mtbl[PTBL_PAGES];
243 	vm_page_t m;
244 	struct ptbl_buf *pbuf;
245 	unsigned int pidx;
246 	pte_t *ptbl;
247 	int i, j;
248 
249 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
250 	    (pmap == kernel_pmap), pdir_idx);
251 
252 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
253 	    ("ptbl_alloc: invalid pdir_idx"));
254 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
255 	    ("pte_alloc: valid ptbl entry exists!"));
256 
257 	pbuf = ptbl_buf_alloc();
258 	if (pbuf == NULL)
259 		panic("pte_alloc: couldn't alloc kernel virtual memory");
260 
261 	ptbl = (pte_t *)pbuf->kva;
262 
263 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
264 
265 	for (i = 0; i < PTBL_PAGES; i++) {
266 		pidx = (PTBL_PAGES * pdir_idx) + i;
267 		while ((m = vm_page_alloc(NULL, pidx,
268 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
269 			if (nosleep) {
270 				ptbl_free_pmap_ptbl(pmap, ptbl);
271 				for (j = 0; j < i; j++)
272 					vm_page_free(mtbl[j]);
273 				vm_wire_sub(i);
274 				return (NULL);
275 			}
276 			PMAP_UNLOCK(pmap);
277 			rw_wunlock(&pvh_global_lock);
278 			vm_wait(NULL);
279 			rw_wlock(&pvh_global_lock);
280 			PMAP_LOCK(pmap);
281 		}
282 		mtbl[i] = m;
283 	}
284 
285 	/* Map allocated pages into kernel_pmap. */
286 	mmu_booke_qenter((vm_offset_t)ptbl, mtbl, PTBL_PAGES);
287 
288 	/* Zero whole ptbl. */
289 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
290 
291 	/* Add pbuf to the pmap ptbl bufs list. */
292 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
293 
294 	return (ptbl);
295 }
296 
297 /* Free ptbl pages and invalidate pdir entry. */
298 static void
299 ptbl_free(pmap_t pmap, unsigned int pdir_idx)
300 {
301 	pte_t *ptbl;
302 	vm_paddr_t pa;
303 	vm_offset_t va;
304 	vm_page_t m;
305 	int i;
306 
307 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
308 	    (pmap == kernel_pmap), pdir_idx);
309 
310 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
311 	    ("ptbl_free: invalid pdir_idx"));
312 
313 	ptbl = pmap->pm_pdir[pdir_idx];
314 
315 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
316 
317 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
318 
319 	/*
320 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
321 	 * don't attempt to look up the page tables we are releasing.
322 	 */
323 	mtx_lock_spin(&tlbivax_mutex);
324 	tlb_miss_lock();
325 
326 	pmap->pm_pdir[pdir_idx] = NULL;
327 
328 	tlb_miss_unlock();
329 	mtx_unlock_spin(&tlbivax_mutex);
330 
331 	for (i = 0; i < PTBL_PAGES; i++) {
332 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
333 		pa = pte_vatopa(kernel_pmap, va);
334 		m = PHYS_TO_VM_PAGE(pa);
335 		vm_page_free_zero(m);
336 		vm_wire_sub(1);
337 		mmu_booke_kremove(va);
338 	}
339 
340 	ptbl_free_pmap_ptbl(pmap, ptbl);
341 }
342 
343 /*
344  * Decrement ptbl pages hold count and attempt to free ptbl pages.
345  * Called when removing pte entry from ptbl.
346  *
347  * Return 1 if ptbl pages were freed.
348  */
349 static int
350 ptbl_unhold(pmap_t pmap, unsigned int pdir_idx)
351 {
352 	pte_t *ptbl;
353 	vm_paddr_t pa;
354 	vm_page_t m;
355 	int i;
356 
357 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
358 	    (pmap == kernel_pmap), pdir_idx);
359 
360 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
361 	    ("ptbl_unhold: invalid pdir_idx"));
362 	KASSERT((pmap != kernel_pmap),
363 	    ("ptbl_unhold: unholding kernel ptbl!"));
364 
365 	ptbl = pmap->pm_pdir[pdir_idx];
366 
367 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
368 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
369 	    ("ptbl_unhold: non kva ptbl"));
370 
371 	/* decrement hold count */
372 	for (i = 0; i < PTBL_PAGES; i++) {
373 		pa = pte_vatopa(kernel_pmap,
374 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
375 		m = PHYS_TO_VM_PAGE(pa);
376 		m->ref_count--;
377 	}
378 
379 	/*
380 	 * Free ptbl pages if there are no pte etries in this ptbl.
381 	 * ref_count has the same value for all ptbl pages, so check the last
382 	 * page.
383 	 */
384 	if (m->ref_count == 0) {
385 		ptbl_free(pmap, pdir_idx);
386 
387 		//debugf("ptbl_unhold: e (freed ptbl)\n");
388 		return (1);
389 	}
390 
391 	return (0);
392 }
393 
394 /*
395  * Increment hold count for ptbl pages. This routine is used when a new pte
396  * entry is being inserted into the ptbl.
397  */
398 static void
399 ptbl_hold(pmap_t pmap, unsigned int pdir_idx)
400 {
401 	vm_paddr_t pa;
402 	pte_t *ptbl;
403 	vm_page_t m;
404 	int i;
405 
406 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
407 	    pdir_idx);
408 
409 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
410 	    ("ptbl_hold: invalid pdir_idx"));
411 	KASSERT((pmap != kernel_pmap),
412 	    ("ptbl_hold: holding kernel ptbl!"));
413 
414 	ptbl = pmap->pm_pdir[pdir_idx];
415 
416 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
417 
418 	for (i = 0; i < PTBL_PAGES; i++) {
419 		pa = pte_vatopa(kernel_pmap,
420 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
421 		m = PHYS_TO_VM_PAGE(pa);
422 		m->ref_count++;
423 	}
424 }
425 
426 /*
427  * Clean pte entry, try to free page table page if requested.
428  *
429  * Return 1 if ptbl pages were freed, otherwise return 0.
430  */
431 static int
432 pte_remove(pmap_t pmap, vm_offset_t va, uint8_t flags)
433 {
434 	unsigned int pdir_idx = PDIR_IDX(va);
435 	unsigned int ptbl_idx = PTBL_IDX(va);
436 	vm_page_t m;
437 	pte_t *ptbl;
438 	pte_t *pte;
439 
440 	//int su = (pmap == kernel_pmap);
441 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
442 	//		su, (u_int32_t)pmap, va, flags);
443 
444 	ptbl = pmap->pm_pdir[pdir_idx];
445 	KASSERT(ptbl, ("pte_remove: null ptbl"));
446 
447 	pte = &ptbl[ptbl_idx];
448 
449 	if (pte == NULL || !PTE_ISVALID(pte))
450 		return (0);
451 
452 	if (PTE_ISWIRED(pte))
453 		pmap->pm_stats.wired_count--;
454 
455 	/* Get vm_page_t for mapped pte. */
456 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
457 
458 	/* Handle managed entry. */
459 	if (PTE_ISMANAGED(pte)) {
460 		if (PTE_ISMODIFIED(pte))
461 			vm_page_dirty(m);
462 
463 		if (PTE_ISREFERENCED(pte))
464 			vm_page_aflag_set(m, PGA_REFERENCED);
465 
466 		pv_remove(pmap, va, m);
467 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
468 		/*
469 		 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
470 		 * used.  This is needed by the NCSW support code for fast
471 		 * VA<->PA translation.
472 		 */
473 		pv_remove(pmap, va, m);
474 		if (TAILQ_EMPTY(&m->md.pv_list))
475 			m->md.pv_tracked = false;
476 	}
477 
478 	mtx_lock_spin(&tlbivax_mutex);
479 	tlb_miss_lock();
480 
481 	tlb0_flush_entry(va);
482 	*pte = 0;
483 
484 	tlb_miss_unlock();
485 	mtx_unlock_spin(&tlbivax_mutex);
486 
487 	pmap->pm_stats.resident_count--;
488 
489 	if (flags & PTBL_UNHOLD) {
490 		//debugf("pte_remove: e (unhold)\n");
491 		return (ptbl_unhold(pmap, pdir_idx));
492 	}
493 
494 	//debugf("pte_remove: e\n");
495 	return (0);
496 }
497 
498 /*
499  * Insert PTE for a given page and virtual address.
500  */
501 static int
502 pte_enter(pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
503     boolean_t nosleep)
504 {
505 	unsigned int pdir_idx = PDIR_IDX(va);
506 	unsigned int ptbl_idx = PTBL_IDX(va);
507 	pte_t *ptbl, *pte, pte_tmp;
508 
509 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
510 	    pmap == kernel_pmap, pmap, va);
511 
512 	/* Get the page table pointer. */
513 	ptbl = pmap->pm_pdir[pdir_idx];
514 
515 	if (ptbl == NULL) {
516 		/* Allocate page table pages. */
517 		ptbl = ptbl_alloc(pmap, pdir_idx, nosleep);
518 		if (ptbl == NULL) {
519 			KASSERT(nosleep, ("nosleep and NULL ptbl"));
520 			return (ENOMEM);
521 		}
522 		pmap->pm_pdir[pdir_idx] = ptbl;
523 		pte = &ptbl[ptbl_idx];
524 	} else {
525 		/*
526 		 * Check if there is valid mapping for requested
527 		 * va, if there is, remove it.
528 		 */
529 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
530 		if (PTE_ISVALID(pte)) {
531 			pte_remove(pmap, va, PTBL_HOLD);
532 		} else {
533 			/*
534 			 * pte is not used, increment hold count
535 			 * for ptbl pages.
536 			 */
537 			if (pmap != kernel_pmap)
538 				ptbl_hold(pmap, pdir_idx);
539 		}
540 	}
541 
542 	/*
543 	 * Insert pv_entry into pv_list for mapped page if part of managed
544 	 * memory.
545 	 */
546 	if ((m->oflags & VPO_UNMANAGED) == 0) {
547 		flags |= PTE_MANAGED;
548 
549 		/* Create and insert pv entry. */
550 		pv_insert(pmap, va, m);
551 	}
552 
553 	pmap->pm_stats.resident_count++;
554 
555 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
556 	pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
557 
558 	mtx_lock_spin(&tlbivax_mutex);
559 	tlb_miss_lock();
560 
561 	tlb0_flush_entry(va);
562 	*pte = pte_tmp;
563 
564 	tlb_miss_unlock();
565 	mtx_unlock_spin(&tlbivax_mutex);
566 	return (0);
567 }
568 
569 /* Return the pa for the given pmap/va. */
570 static vm_paddr_t
571 pte_vatopa(pmap_t pmap, vm_offset_t va)
572 {
573 	vm_paddr_t pa = 0;
574 	pte_t *pte;
575 
576 	pte = pte_find(pmap, va);
577 	if ((pte != NULL) && PTE_ISVALID(pte))
578 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
579 	return (pa);
580 }
581 
582 /* Get a pointer to a PTE in a page table. */
583 static pte_t *
584 pte_find(pmap_t pmap, vm_offset_t va)
585 {
586 	unsigned int pdir_idx = PDIR_IDX(va);
587 	unsigned int ptbl_idx = PTBL_IDX(va);
588 
589 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
590 
591 	if (pmap->pm_pdir[pdir_idx])
592 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
593 
594 	return (NULL);
595 }
596 
597 /* Get a pointer to a PTE in a page table, or the next closest (greater) one. */
598 static __inline pte_t *
599 pte_find_next(pmap_t pmap, vm_offset_t *pva)
600 {
601 	vm_offset_t	va;
602 	pte_t	      **pdir;
603 	pte_t	       *pte;
604 	unsigned long	i, j;
605 
606 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
607 
608 	va = *pva;
609 	i = PDIR_IDX(va);
610 	j = PTBL_IDX(va);
611 	pdir = pmap->pm_pdir;
612 	for (; i < PDIR_NENTRIES; i++, j = 0) {
613 		if (pdir[i] == NULL)
614 			continue;
615 		for (; j < PTBL_NENTRIES; j++) {
616 			pte = &pdir[i][j];
617 			if (!PTE_ISVALID(pte))
618 				continue;
619 			*pva = PDIR_SIZE * i + PAGE_SIZE * j;
620 			return (pte);
621 		}
622 	}
623 	return (NULL);
624 }
625 
626 /* Set up kernel page tables. */
627 static void
628 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr)
629 {
630 	pte_t		*pte;
631 	vm_offset_t	va;
632 	vm_offset_t	pdir_start;
633 	int		i;
634 
635 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
636 	kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
637 
638 	pdir_start = kernel_ptbl_root + PDIR_NENTRIES * sizeof(pte_t);
639 
640 	/* Initialize kernel pdir */
641 	for (i = 0; i < kernel_ptbls; i++) {
642 		kernel_pmap->pm_pdir[kptbl_min + i] =
643 		    (pte_t *)(pdir_start + (i * PAGE_SIZE * PTBL_PAGES));
644 	}
645 
646 	/*
647 	 * Fill in PTEs covering kernel code and data. They are not required
648 	 * for address translation, as this area is covered by static TLB1
649 	 * entries, but for pte_vatopa() to work correctly with kernel area
650 	 * addresses.
651 	 */
652 	for (va = addr; va < data_end; va += PAGE_SIZE) {
653 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
654 		powerpc_sync();
655 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
656 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
657 		    PTE_VALID | PTE_PS_4KB;
658 	}
659 }
660 
661 static vm_offset_t
662 mmu_booke_alloc_kernel_pgtables(vm_offset_t data_end)
663 {
664 	/* Allocate space for ptbl_bufs. */
665 	ptbl_bufs = (struct ptbl_buf *)data_end;
666 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
667 	debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
668 	    (uintptr_t)ptbl_bufs, data_end);
669 
670 	data_end = round_page(data_end);
671 
672 	kernel_ptbl_root = data_end;
673 	data_end += PDIR_NENTRIES * sizeof(pte_t*);
674 
675 	/* Allocate PTE tables for kernel KVA. */
676 	kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
677 	    PDIR_SIZE);
678 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
679 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
680 	debugf(" kernel pdir at %#jx end = %#jx\n",
681 	    (uintmax_t)kernel_ptbl_root, (uintmax_t)data_end);
682 
683 	return (data_end);
684 }
685 
686 /*
687  * Initialize a preallocated and zeroed pmap structure,
688  * such as one in a vmspace structure.
689  */
690 static int
691 mmu_booke_pinit(pmap_t pmap)
692 {
693 	int i;
694 
695 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
696 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
697 
698 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
699 
700 	for (i = 0; i < MAXCPU; i++)
701 		pmap->pm_tid[i] = TID_NONE;
702 	CPU_ZERO(&kernel_pmap->pm_active);
703 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
704 	pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
705 	bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
706 	TAILQ_INIT(&pmap->pm_ptbl_list);
707 
708 	return (1);
709 }
710 
711 /*
712  * Release any resources held by the given physical map.
713  * Called when a pmap initialized by mmu_booke_pinit is being released.
714  * Should only be called if the map contains no valid mappings.
715  */
716 static void
717 mmu_booke_release(pmap_t pmap)
718 {
719 
720 	KASSERT(pmap->pm_stats.resident_count == 0,
721 	    ("pmap_release: pmap resident count %ld != 0",
722 	    pmap->pm_stats.resident_count));
723 	uma_zfree(ptbl_root_zone, pmap->pm_pdir);
724 }
725 
726 static void
727 mmu_booke_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
728 {
729 	pte_t *pte;
730 	vm_paddr_t pa = 0;
731 	int sync_sz, valid;
732 	pmap_t pmap;
733 	vm_page_t m;
734 	vm_offset_t addr;
735 	int active;
736 
737 	rw_wlock(&pvh_global_lock);
738 	pmap = PCPU_GET(curpmap);
739 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
740 	while (sz > 0) {
741 		PMAP_LOCK(pm);
742 		pte = pte_find(pm, va);
743 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
744 		if (valid)
745 			pa = PTE_PA(pte);
746 		PMAP_UNLOCK(pm);
747 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
748 		sync_sz = min(sync_sz, sz);
749 		if (valid) {
750 			if (!active) {
751 				/*
752 				 * Create a mapping in the active pmap.
753 				 *
754 				 * XXX: We use the zero page here, because
755 				 * it isn't likely to be in use.
756 				 * If we ever decide to support
757 				 * security.bsd.map_at_zero on Book-E, change
758 				 * this to some other address that isn't
759 				 * normally mappable.
760 				 */
761 				addr = 0;
762 				m = PHYS_TO_VM_PAGE(pa);
763 				PMAP_LOCK(pmap);
764 				pte_enter(pmap, m, addr,
765 				    PTE_SR | PTE_VALID, FALSE);
766 				__syncicache((void *)(addr + (va & PAGE_MASK)),
767 				    sync_sz);
768 				pte_remove(pmap, addr, PTBL_UNHOLD);
769 				PMAP_UNLOCK(pmap);
770 			} else
771 				__syncicache((void *)va, sync_sz);
772 		}
773 		va += sync_sz;
774 		sz -= sync_sz;
775 	}
776 	rw_wunlock(&pvh_global_lock);
777 }
778 
779 /*
780  * mmu_booke_zero_page_area zeros the specified hardware page by
781  * mapping it into virtual memory and using bzero to clear
782  * its contents.
783  *
784  * off and size must reside within a single page.
785  */
786 static void
787 mmu_booke_zero_page_area(vm_page_t m, int off, int size)
788 {
789 	vm_offset_t va;
790 
791 	/* XXX KASSERT off and size are within a single page? */
792 
793 	mtx_lock(&zero_page_mutex);
794 	va = zero_page_va;
795 
796 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
797 	bzero((caddr_t)va + off, size);
798 	mmu_booke_kremove(va);
799 
800 	mtx_unlock(&zero_page_mutex);
801 }
802 
803 /*
804  * mmu_booke_zero_page zeros the specified hardware page.
805  */
806 static void
807 mmu_booke_zero_page(vm_page_t m)
808 {
809 	vm_offset_t off, va;
810 
811 	va = zero_page_va;
812 	mtx_lock(&zero_page_mutex);
813 
814 	mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
815 
816 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
817 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
818 
819 	mmu_booke_kremove(va);
820 
821 	mtx_unlock(&zero_page_mutex);
822 }
823 
824 /*
825  * mmu_booke_copy_page copies the specified (machine independent) page by
826  * mapping the page into virtual memory and using memcopy to copy the page,
827  * one machine dependent page at a time.
828  */
829 static void
830 mmu_booke_copy_page(vm_page_t sm, vm_page_t dm)
831 {
832 	vm_offset_t sva, dva;
833 
834 	sva = copy_page_src_va;
835 	dva = copy_page_dst_va;
836 
837 	mtx_lock(&copy_page_mutex);
838 	mmu_booke_kenter(sva, VM_PAGE_TO_PHYS(sm));
839 	mmu_booke_kenter(dva, VM_PAGE_TO_PHYS(dm));
840 
841 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
842 
843 	mmu_booke_kremove(dva);
844 	mmu_booke_kremove(sva);
845 	mtx_unlock(&copy_page_mutex);
846 }
847 
848 static inline void
849 mmu_booke_copy_pages(vm_page_t *ma, vm_offset_t a_offset,
850     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
851 {
852 	void *a_cp, *b_cp;
853 	vm_offset_t a_pg_offset, b_pg_offset;
854 	int cnt;
855 
856 	mtx_lock(&copy_page_mutex);
857 	while (xfersize > 0) {
858 		a_pg_offset = a_offset & PAGE_MASK;
859 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
860 		mmu_booke_kenter(copy_page_src_va,
861 		    VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
862 		a_cp = (char *)copy_page_src_va + a_pg_offset;
863 		b_pg_offset = b_offset & PAGE_MASK;
864 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
865 		mmu_booke_kenter(copy_page_dst_va,
866 		    VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
867 		b_cp = (char *)copy_page_dst_va + b_pg_offset;
868 		bcopy(a_cp, b_cp, cnt);
869 		mmu_booke_kremove(copy_page_dst_va);
870 		mmu_booke_kremove(copy_page_src_va);
871 		a_offset += cnt;
872 		b_offset += cnt;
873 		xfersize -= cnt;
874 	}
875 	mtx_unlock(&copy_page_mutex);
876 }
877 
878 static vm_offset_t
879 mmu_booke_quick_enter_page(vm_page_t m)
880 {
881 	vm_paddr_t paddr;
882 	vm_offset_t qaddr;
883 	uint32_t flags;
884 	pte_t *pte;
885 
886 	paddr = VM_PAGE_TO_PHYS(m);
887 
888 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
889 	flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
890 	flags |= PTE_PS_4KB;
891 
892 	critical_enter();
893 	qaddr = PCPU_GET(qmap_addr);
894 
895 	pte = pte_find(kernel_pmap, qaddr);
896 
897 	KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
898 
899 	/*
900 	 * XXX: tlbivax is broadcast to other cores, but qaddr should
901  	 * not be present in other TLBs.  Is there a better instruction
902 	 * sequence to use? Or just forget it & use mmu_booke_kenter()...
903 	 */
904 	__asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
905 	__asm __volatile("isync; msync");
906 
907 	*pte = PTE_RPN_FROM_PA(paddr) | flags;
908 
909 	/* Flush the real memory from the instruction cache. */
910 	if ((flags & (PTE_I | PTE_G)) == 0)
911 		__syncicache((void *)qaddr, PAGE_SIZE);
912 
913 	return (qaddr);
914 }
915 
916 static void
917 mmu_booke_quick_remove_page(vm_offset_t addr)
918 {
919 	pte_t *pte;
920 
921 	pte = pte_find(kernel_pmap, addr);
922 
923 	KASSERT(PCPU_GET(qmap_addr) == addr,
924 	    ("mmu_booke_quick_remove_page: invalid address"));
925 	KASSERT(*pte != 0,
926 	    ("mmu_booke_quick_remove_page: PTE not in use"));
927 
928 	*pte = 0;
929 	critical_exit();
930 }
931 
932 /**************************************************************************/
933 /* TID handling */
934 /**************************************************************************/
935 
936 /*
937  * Return the largest uint value log such that 2^log <= num.
938  */
939 static unsigned long
940 ilog2(unsigned long num)
941 {
942 	long lz;
943 
944 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
945 	return (31 - lz);
946 }
947 
948 /*
949  * Invalidate all TLB0 entries which match the given TID. Note this is
950  * dedicated for cases when invalidations should NOT be propagated to other
951  * CPUs.
952  */
953 static void
954 tid_flush(tlbtid_t tid)
955 {
956 	register_t msr;
957 	uint32_t mas0, mas1, mas2;
958 	int entry, way;
959 
960 	/* Don't evict kernel translations */
961 	if (tid == TID_KERNEL)
962 		return;
963 
964 	msr = mfmsr();
965 	__asm __volatile("wrteei 0");
966 
967 	/*
968 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
969 	 * it for PID invalidation.
970 	 */
971 	switch ((mfpvr() >> 16) & 0xffff) {
972 	case FSL_E500mc:
973 	case FSL_E5500:
974 	case FSL_E6500:
975 		mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
976 		/* tlbilxpid */
977 		__asm __volatile("isync; .long 0x7c200024; isync; msync");
978 		__asm __volatile("wrtee %0" :: "r"(msr));
979 		return;
980 	}
981 
982 	for (way = 0; way < TLB0_WAYS; way++)
983 		for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
984 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
985 			mtspr(SPR_MAS0, mas0);
986 
987 			mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
988 			mtspr(SPR_MAS2, mas2);
989 
990 			__asm __volatile("isync; tlbre");
991 
992 			mas1 = mfspr(SPR_MAS1);
993 
994 			if (!(mas1 & MAS1_VALID))
995 				continue;
996 			if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
997 				continue;
998 			mas1 &= ~MAS1_VALID;
999 			mtspr(SPR_MAS1, mas1);
1000 			__asm __volatile("isync; tlbwe; isync; msync");
1001 		}
1002 	__asm __volatile("wrtee %0" :: "r"(msr));
1003 }
1004