xref: /freebsd/sys/powerpc/booke/pmap_64.c (revision aa1a8ff2d6dbc51ef058f46f3db5a8bb77967145)
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   * 64-bit pmap:
41   * Virtual address space layout:
42   * -----------------------------
43   * 0x0000_0000_0000_0000 - 0x3fff_ffff_ffff_ffff      : user process
44   * 0x4000_0000_0000_0000 - 0x7fff_ffff_ffff_ffff      : unused
45   * 0x8000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : mmio region
46   * 0xc000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : direct map
47   * 0xe000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : KVA
48   */
49 
50 #include <sys/cdefs.h>
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 #ifdef  DEBUG
101 #define debugf(fmt, args...) printf(fmt, ##args)
102 #else
103 #define debugf(fmt, args...)
104 #endif
105 
106 #define	PRI0ptrX	"016lx"
107 
108 /**************************************************************************/
109 /* PMAP */
110 /**************************************************************************/
111 
112 unsigned int kernel_pdirs;
113 static uma_zone_t ptbl_root_zone;
114 static pte_t ****kernel_ptbl_root;
115 
116 /*
117  * Base of the pmap_mapdev() region.  On 32-bit it immediately follows the
118  * userspace address range.  On On 64-bit it's far above, at (1 << 63), and
119  * ranges up to the DMAP, giving 62 bits of PA allowed.  This is far larger than
120  * the widest Book-E address bus, the e6500 has a 40-bit PA space.  This allows
121  * us to map akin to the DMAP, with addresses identical to the PA, offset by the
122  * base.
123  */
124 #define	VM_MAPDEV_BASE		0x8000000000000000
125 #define	VM_MAPDEV_PA_MAX	0x4000000000000000 /* Don't encroach on DMAP */
126 
127 static void tid_flush(tlbtid_t tid);
128 static unsigned long ilog2(unsigned long);
129 
130 /**************************************************************************/
131 /* Page table management */
132 /**************************************************************************/
133 
134 #define PMAP_ROOT_SIZE	(sizeof(pte_t****) * PG_ROOT_NENTRIES)
135 static pte_t *ptbl_alloc(pmap_t pmap, vm_offset_t va,
136     bool nosleep, bool *is_new);
137 static void ptbl_hold(pmap_t, pte_t *);
138 static int ptbl_unhold(pmap_t, vm_offset_t);
139 
140 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
141 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, bool);
142 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
143 static pte_t *pte_find(pmap_t, vm_offset_t);
144 static pte_t *pte_find_next(pmap_t, vm_offset_t *);
145 static void kernel_pte_alloc(vm_offset_t, vm_offset_t);
146 
147 /**************************************************************************/
148 /* Page table related */
149 /**************************************************************************/
150 
151 /* Allocate a page, to be used in a page table. */
152 static vm_offset_t
153 mmu_booke_alloc_page(pmap_t pmap, unsigned int idx, bool nosleep)
154 {
155 	vm_page_t	m;
156 	int		req;
157 
158 	req = VM_ALLOC_WIRED | VM_ALLOC_ZERO;
159 	while ((m = vm_page_alloc_noobj(req)) == NULL) {
160 		if (nosleep)
161 			return (0);
162 
163 		PMAP_UNLOCK(pmap);
164 		rw_wunlock(&pvh_global_lock);
165 		vm_wait(NULL);
166 		rw_wlock(&pvh_global_lock);
167 		PMAP_LOCK(pmap);
168 	}
169 	m->pindex = idx;
170 
171 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
172 }
173 
174 /* Initialize pool of kva ptbl buffers. */
175 static void
176 ptbl_init(void)
177 {
178 }
179 
180 /* Get a pointer to a PTE in a page table. */
181 static __inline pte_t *
182 pte_find(pmap_t pmap, vm_offset_t va)
183 {
184 	pte_t        ***pdir_l1;
185 	pte_t         **pdir;
186 	pte_t          *ptbl;
187 
188 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
189 
190 	pdir_l1 = pmap->pm_root[PG_ROOT_IDX(va)];
191 	if (pdir_l1 == NULL)
192 		return (NULL);
193 	pdir = pdir_l1[PDIR_L1_IDX(va)];
194 	if (pdir == NULL)
195 		return (NULL);
196 	ptbl = pdir[PDIR_IDX(va)];
197 
198 	return ((ptbl != NULL) ? &ptbl[PTBL_IDX(va)] : NULL);
199 }
200 
201 /* Get a pointer to a PTE in a page table, or the next closest (greater) one. */
202 static __inline pte_t *
203 pte_find_next(pmap_t pmap, vm_offset_t *pva)
204 {
205 	vm_offset_t	va;
206 	pte_t	    ****pm_root;
207 	pte_t	       *pte;
208 	unsigned long	i, j, k, l;
209 
210 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
211 
212 	va = *pva;
213 	i = PG_ROOT_IDX(va);
214 	j = PDIR_L1_IDX(va);
215 	k = PDIR_IDX(va);
216 	l = PTBL_IDX(va);
217 	pm_root = pmap->pm_root;
218 
219 	/* truncate the VA for later. */
220 	va &= ~((1UL << (PG_ROOT_H + 1)) - 1);
221 	for (; i < PG_ROOT_NENTRIES; i++, j = 0, k = 0, l = 0) {
222 		if (pm_root[i] == 0)
223 			continue;
224 		for (; j < PDIR_L1_NENTRIES; j++, k = 0, l = 0) {
225 			if (pm_root[i][j] == 0)
226 				continue;
227 			for (; k < PDIR_NENTRIES; k++, l = 0) {
228 				if (pm_root[i][j][k] == NULL)
229 					continue;
230 				for (; l < PTBL_NENTRIES; l++) {
231 					pte = &pm_root[i][j][k][l];
232 					if (!PTE_ISVALID(pte))
233 						continue;
234 					*pva = va + PG_ROOT_SIZE * i +
235 					    PDIR_L1_SIZE * j +
236 					    PDIR_SIZE * k +
237 					    PAGE_SIZE * l;
238 					return (pte);
239 				}
240 			}
241 		}
242 	}
243 	return (NULL);
244 }
245 
246 static bool
247 unhold_free_page(pmap_t pmap, vm_page_t m)
248 {
249 
250 	if (vm_page_unwire_noq(m)) {
251 		vm_page_free_zero(m);
252 		return (true);
253 	}
254 
255 	return (false);
256 }
257 
258 static vm_offset_t
259 get_pgtbl_page(pmap_t pmap, vm_offset_t *ptr_tbl, uint32_t index,
260     bool nosleep, bool hold_parent, bool *isnew)
261 {
262 	vm_offset_t	page;
263 	vm_page_t	m;
264 
265 	page = ptr_tbl[index];
266 	KASSERT(page != 0 || pmap != kernel_pmap,
267 	    ("NULL page table page found in kernel pmap!"));
268 	if (page == 0) {
269 		page = mmu_booke_alloc_page(pmap, index, nosleep);
270 		if (ptr_tbl[index] == 0) {
271 			*isnew = true;
272 			ptr_tbl[index] = page;
273 			if (hold_parent) {
274 				m = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)ptr_tbl));
275 				m->ref_count++;
276 			}
277 			return (page);
278 		}
279 		m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS(page));
280 		page = ptr_tbl[index];
281 		vm_page_unwire_noq(m);
282 		vm_page_free_zero(m);
283 	}
284 
285 	*isnew = false;
286 
287 	return (page);
288 }
289 
290 /* Allocate page table. */
291 static pte_t*
292 ptbl_alloc(pmap_t pmap, vm_offset_t va, bool nosleep, bool *is_new)
293 {
294 	unsigned int	pg_root_idx = PG_ROOT_IDX(va);
295 	unsigned int	pdir_l1_idx = PDIR_L1_IDX(va);
296 	unsigned int	pdir_idx = PDIR_IDX(va);
297 	vm_offset_t	pdir_l1, pdir, ptbl;
298 
299 	/* When holding a parent, no need to hold the root index pages. */
300 	pdir_l1 = get_pgtbl_page(pmap, (vm_offset_t *)pmap->pm_root,
301 	    pg_root_idx, nosleep, false, is_new);
302 	if (pdir_l1 == 0)
303 		return (NULL);
304 	pdir = get_pgtbl_page(pmap, (vm_offset_t *)pdir_l1, pdir_l1_idx,
305 	    nosleep, !*is_new, is_new);
306 	if (pdir == 0)
307 		return (NULL);
308 	ptbl = get_pgtbl_page(pmap, (vm_offset_t *)pdir, pdir_idx,
309 	    nosleep, !*is_new, is_new);
310 
311 	return ((pte_t *)ptbl);
312 }
313 
314 /*
315  * Decrement ptbl pages hold count and attempt to free ptbl pages. Called
316  * when removing pte entry from ptbl.
317  *
318  * Return 1 if ptbl pages were freed.
319  */
320 static int
321 ptbl_unhold(pmap_t pmap, vm_offset_t va)
322 {
323 	pte_t          *ptbl;
324 	vm_page_t	m;
325 	u_int		pg_root_idx;
326 	pte_t        ***pdir_l1;
327 	u_int		pdir_l1_idx;
328 	pte_t         **pdir;
329 	u_int		pdir_idx;
330 
331 	pg_root_idx = PG_ROOT_IDX(va);
332 	pdir_l1_idx = PDIR_L1_IDX(va);
333 	pdir_idx = PDIR_IDX(va);
334 
335 	KASSERT((pmap != kernel_pmap),
336 		("ptbl_unhold: unholding kernel ptbl!"));
337 
338 	pdir_l1 = pmap->pm_root[pg_root_idx];
339 	pdir = pdir_l1[pdir_l1_idx];
340 	ptbl = pdir[pdir_idx];
341 
342 	/* decrement hold count */
343 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
344 
345 	if (!unhold_free_page(pmap, m))
346 		return (0);
347 
348 	pdir[pdir_idx] = NULL;
349 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) pdir));
350 
351 	if (!unhold_free_page(pmap, m))
352 		return (1);
353 
354 	pdir_l1[pdir_l1_idx] = NULL;
355 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) pdir_l1));
356 
357 	if (!unhold_free_page(pmap, m))
358 		return (1);
359 	pmap->pm_root[pg_root_idx] = NULL;
360 
361 	return (1);
362 }
363 
364 /*
365  * Increment hold count for ptbl pages. This routine is used when new pte
366  * entry is being inserted into ptbl.
367  */
368 static void
369 ptbl_hold(pmap_t pmap, pte_t *ptbl)
370 {
371 	vm_page_t	m;
372 
373 	KASSERT((pmap != kernel_pmap),
374 		("ptbl_hold: holding kernel ptbl!"));
375 
376 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t) ptbl));
377 	m->ref_count++;
378 }
379 
380 /*
381  * Clean pte entry, try to free page table page if requested.
382  *
383  * Return 1 if ptbl pages were freed, otherwise return 0.
384  */
385 static int
386 pte_remove(pmap_t pmap, vm_offset_t va, u_int8_t flags)
387 {
388 	vm_page_t	m;
389 	pte_t          *pte;
390 
391 	pte = pte_find(pmap, va);
392 	KASSERT(pte != NULL, ("%s: NULL pte for va %#jx, pmap %p",
393 	    __func__, (uintmax_t)va, pmap));
394 
395 	if (!PTE_ISVALID(pte))
396 		return (0);
397 
398 	/* Get vm_page_t for mapped pte. */
399 	m = PHYS_TO_VM_PAGE(PTE_PA(pte));
400 
401 	if (PTE_ISWIRED(pte))
402 		pmap->pm_stats.wired_count--;
403 
404 	/* Handle managed entry. */
405 	if (PTE_ISMANAGED(pte)) {
406 		/* Handle modified pages. */
407 		if (PTE_ISMODIFIED(pte))
408 			vm_page_dirty(m);
409 
410 		/* Referenced pages. */
411 		if (PTE_ISREFERENCED(pte))
412 			vm_page_aflag_set(m, PGA_REFERENCED);
413 
414 		/* Remove pv_entry from pv_list. */
415 		pv_remove(pmap, va, m);
416 	} else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
417 		pv_remove(pmap, va, m);
418 		if (TAILQ_EMPTY(&m->md.pv_list))
419 			m->md.pv_tracked = false;
420 	}
421 	mtx_lock_spin(&tlbivax_mutex);
422 	tlb_miss_lock();
423 
424 	tlb0_flush_entry(va);
425 	*pte = 0;
426 
427 	tlb_miss_unlock();
428 	mtx_unlock_spin(&tlbivax_mutex);
429 
430 	pmap->pm_stats.resident_count--;
431 
432 	if (flags & PTBL_UNHOLD) {
433 		return (ptbl_unhold(pmap, va));
434 	}
435 	return (0);
436 }
437 
438 /*
439  * Insert PTE for a given page and virtual address.
440  */
441 static int
442 pte_enter(pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
443     bool nosleep)
444 {
445 	unsigned int	ptbl_idx = PTBL_IDX(va);
446 	pte_t          *ptbl, *pte, pte_tmp;
447 	bool		is_new;
448 
449 	/* Get the page directory pointer. */
450 	ptbl = ptbl_alloc(pmap, va, nosleep, &is_new);
451 	if (ptbl == NULL) {
452 		KASSERT(nosleep, ("nosleep and NULL ptbl"));
453 		return (ENOMEM);
454 	}
455 	if (is_new) {
456 		pte = &ptbl[ptbl_idx];
457 	} else {
458 		/*
459 		 * Check if there is valid mapping for requested va, if there
460 		 * is, remove it.
461 		 */
462 		pte = &ptbl[ptbl_idx];
463 		if (PTE_ISVALID(pte)) {
464 			pte_remove(pmap, va, PTBL_HOLD);
465 		} else {
466 			/*
467 			 * pte is not used, increment hold count for ptbl
468 			 * pages.
469 			 */
470 			if (pmap != kernel_pmap)
471 				ptbl_hold(pmap, ptbl);
472 		}
473 	}
474 
475 	/*
476 	 * Insert pv_entry into pv_list for mapped page if part of managed
477 	 * memory.
478 	 */
479 	if ((m->oflags & VPO_UNMANAGED) == 0) {
480 		flags |= PTE_MANAGED;
481 
482 		/* Create and insert pv entry. */
483 		pv_insert(pmap, va, m);
484 	}
485 
486 	pmap->pm_stats.resident_count++;
487 
488 	pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
489 	pte_tmp |= (PTE_VALID | flags);
490 
491 	mtx_lock_spin(&tlbivax_mutex);
492 	tlb_miss_lock();
493 
494 	tlb0_flush_entry(va);
495 	*pte = pte_tmp;
496 
497 	tlb_miss_unlock();
498 	mtx_unlock_spin(&tlbivax_mutex);
499 
500 	return (0);
501 }
502 
503 /* Return the pa for the given pmap/va. */
504 static	vm_paddr_t
505 pte_vatopa(pmap_t pmap, vm_offset_t va)
506 {
507 	vm_paddr_t	pa = 0;
508 	pte_t          *pte;
509 
510 	pte = pte_find(pmap, va);
511 	if ((pte != NULL) && PTE_ISVALID(pte))
512 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
513 	return (pa);
514 }
515 
516 /* allocate pte entries to manage (addr & mask) to (addr & mask) + size */
517 static void
518 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr)
519 {
520 	pte_t		*pte;
521 	vm_size_t	kva_size;
522 	int		kernel_pdirs, kernel_pgtbls, pdir_l1s;
523 	vm_offset_t	va, l1_va, pdir_va, ptbl_va;
524 	int		i, j, k;
525 
526 	kva_size = VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS;
527 	kernel_pmap->pm_root = kernel_ptbl_root;
528 	pdir_l1s = howmany(kva_size, PG_ROOT_SIZE);
529 	kernel_pdirs = howmany(kva_size, PDIR_L1_SIZE);
530 	kernel_pgtbls = howmany(kva_size, PDIR_SIZE);
531 
532 	/* Initialize kernel pdir */
533 	l1_va = (vm_offset_t)kernel_ptbl_root +
534 	    round_page(PG_ROOT_NENTRIES * sizeof(pte_t ***));
535 	pdir_va = l1_va + pdir_l1s * PAGE_SIZE;
536 	ptbl_va = pdir_va + kernel_pdirs * PAGE_SIZE;
537 	if (bootverbose) {
538 		printf("ptbl_root_va: %#lx\n", (vm_offset_t)kernel_ptbl_root);
539 		printf("l1_va: %#lx (%d entries)\n", l1_va, pdir_l1s);
540 		printf("pdir_va: %#lx(%d entries)\n", pdir_va, kernel_pdirs);
541 		printf("ptbl_va: %#lx(%d entries)\n", ptbl_va, kernel_pgtbls);
542 	}
543 
544 	va = VM_MIN_KERNEL_ADDRESS;
545 	for (i = PG_ROOT_IDX(va); i < PG_ROOT_IDX(va) + pdir_l1s;
546 	    i++, l1_va += PAGE_SIZE) {
547 		kernel_pmap->pm_root[i] = (pte_t ***)l1_va;
548 		for (j = 0;
549 		    j < PDIR_L1_NENTRIES && va < VM_MAX_KERNEL_ADDRESS;
550 		    j++, pdir_va += PAGE_SIZE) {
551 			kernel_pmap->pm_root[i][j] = (pte_t **)pdir_va;
552 			for (k = 0;
553 			    k < PDIR_NENTRIES && va < VM_MAX_KERNEL_ADDRESS;
554 			    k++, va += PDIR_SIZE, ptbl_va += PAGE_SIZE)
555 				kernel_pmap->pm_root[i][j][k] = (pte_t *)ptbl_va;
556 		}
557 	}
558 	/*
559 	 * Fill in PTEs covering kernel code and data. They are not required
560 	 * for address translation, as this area is covered by static TLB1
561 	 * entries, but for pte_vatopa() to work correctly with kernel area
562 	 * addresses.
563 	 */
564 	for (va = addr; va < data_end; va += PAGE_SIZE) {
565 		pte = &(kernel_pmap->pm_root[PG_ROOT_IDX(va)][PDIR_L1_IDX(va)][PDIR_IDX(va)][PTBL_IDX(va)]);
566 		*pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
567 		*pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
568 		    PTE_VALID | PTE_PS_4KB;
569 	}
570 }
571 
572 static vm_offset_t
573 mmu_booke_alloc_kernel_pgtables(vm_offset_t data_end)
574 {
575 	vm_size_t kva_size = VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS;
576 	kernel_ptbl_root = (pte_t ****)data_end;
577 
578 	data_end += round_page(PG_ROOT_NENTRIES * sizeof(pte_t ***));
579 	data_end += howmany(kva_size, PG_ROOT_SIZE) * PAGE_SIZE;
580 	data_end += howmany(kva_size, PDIR_L1_SIZE) * PAGE_SIZE;
581 	data_end += howmany(kva_size, PDIR_SIZE) * PAGE_SIZE;
582 
583 	return (data_end);
584 }
585 
586 /*
587  * Initialize a preallocated and zeroed pmap structure,
588  * such as one in a vmspace structure.
589  */
590 static int
591 mmu_booke_pinit(pmap_t pmap)
592 {
593 	int i;
594 
595 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
596 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
597 
598 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
599 
600 	for (i = 0; i < MAXCPU; i++)
601 		pmap->pm_tid[i] = TID_NONE;
602 	CPU_ZERO(&kernel_pmap->pm_active);
603 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
604 	pmap->pm_root = uma_zalloc(ptbl_root_zone, M_WAITOK);
605 	bzero(pmap->pm_root, sizeof(pte_t **) * PG_ROOT_NENTRIES);
606 
607 	return (1);
608 }
609 
610 /*
611  * Release any resources held by the given physical map.
612  * Called when a pmap initialized by mmu_booke_pinit is being released.
613  * Should only be called if the map contains no valid mappings.
614  */
615 static void
616 mmu_booke_release(pmap_t pmap)
617 {
618 
619 	KASSERT(pmap->pm_stats.resident_count == 0,
620 	    ("pmap_release: pmap resident count %ld != 0",
621 	    pmap->pm_stats.resident_count));
622 #ifdef INVARIANTS
623 	/*
624 	 * Verify that all page directories are gone.
625 	 * Protects against reference count leakage.
626 	 */
627 	for (int i = 0; i < PG_ROOT_NENTRIES; i++)
628 		KASSERT(pmap->pm_root[i] == 0,
629 		    ("Index %d on root page %p is non-zero!\n", i, pmap->pm_root));
630 #endif
631 	uma_zfree(ptbl_root_zone, pmap->pm_root);
632 }
633 
634 static void
635 mmu_booke_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
636 {
637 	pte_t *pte;
638 	vm_paddr_t pa = 0;
639 	int sync_sz, valid;
640 
641 	while (sz > 0) {
642 		PMAP_LOCK(pm);
643 		pte = pte_find(pm, va);
644 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
645 		if (valid)
646 			pa = PTE_PA(pte);
647 		PMAP_UNLOCK(pm);
648 		sync_sz = PAGE_SIZE - (va & PAGE_MASK);
649 		sync_sz = min(sync_sz, sz);
650 		if (valid) {
651 			pa += (va & PAGE_MASK);
652 			__syncicache((void *)PHYS_TO_DMAP(pa), sync_sz);
653 		}
654 		va += sync_sz;
655 		sz -= sync_sz;
656 	}
657 }
658 
659 /*
660  * mmu_booke_zero_page_area zeros the specified hardware page by
661  * mapping it into virtual memory and using bzero to clear
662  * its contents.
663  *
664  * off and size must reside within a single page.
665  */
666 static void
667 mmu_booke_zero_page_area(vm_page_t m, int off, int size)
668 {
669 	vm_offset_t va;
670 
671 	/* XXX KASSERT off and size are within a single page? */
672 
673 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
674 	bzero((caddr_t)va + off, size);
675 }
676 
677 /*
678  * mmu_booke_zero_page zeros the specified hardware page.
679  */
680 static void
681 mmu_booke_zero_page(vm_page_t m)
682 {
683 	vm_offset_t off, va;
684 
685 	va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
686 
687 	for (off = 0; off < PAGE_SIZE; off += cacheline_size)
688 		__asm __volatile("dcbz 0,%0" :: "r"(va + off));
689 }
690 
691 /*
692  * mmu_booke_copy_page copies the specified (machine independent) page by
693  * mapping the page into virtual memory and using memcopy to copy the page,
694  * one machine dependent page at a time.
695  */
696 static void
697 mmu_booke_copy_page(vm_page_t sm, vm_page_t dm)
698 {
699 	vm_offset_t sva, dva;
700 
701 	sva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(sm));
702 	dva = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dm));
703 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
704 }
705 
706 static inline void
707 mmu_booke_copy_pages(vm_page_t *ma, vm_offset_t a_offset,
708     vm_page_t *mb, vm_offset_t b_offset, int xfersize)
709 {
710 	void *a_cp, *b_cp;
711 	vm_offset_t a_pg_offset, b_pg_offset;
712 	int cnt;
713 
714 	vm_page_t pa, pb;
715 
716 	while (xfersize > 0) {
717 		a_pg_offset = a_offset & PAGE_MASK;
718 		pa = ma[a_offset >> PAGE_SHIFT];
719 		b_pg_offset = b_offset & PAGE_MASK;
720 		pb = mb[b_offset >> PAGE_SHIFT];
721 		cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
722 		cnt = min(cnt, PAGE_SIZE - b_pg_offset);
723 		a_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pa)) +
724 		    a_pg_offset);
725 		b_cp = (caddr_t)((uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pb)) +
726 		    b_pg_offset);
727 		bcopy(a_cp, b_cp, cnt);
728 		a_offset += cnt;
729 		b_offset += cnt;
730 		xfersize -= cnt;
731 	}
732 }
733 
734 static vm_offset_t
735 mmu_booke_quick_enter_page(vm_page_t m)
736 {
737 	return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
738 }
739 
740 static void
741 mmu_booke_quick_remove_page(vm_offset_t addr)
742 {
743 }
744 
745 /**************************************************************************/
746 /* TID handling */
747 /**************************************************************************/
748 
749 /*
750  * Return the largest uint value log such that 2^log <= num.
751  */
752 static unsigned long
753 ilog2(unsigned long num)
754 {
755 	long lz;
756 
757 	__asm ("cntlzd %0, %1" : "=r" (lz) : "r" (num));
758 	return (63 - lz);
759 }
760 
761 /*
762  * Invalidate all TLB0 entries which match the given TID. Note this is
763  * dedicated for cases when invalidations should NOT be propagated to other
764  * CPUs.
765  */
766 static void
767 tid_flush(tlbtid_t tid)
768 {
769 	register_t msr;
770 
771 	/* Don't evict kernel translations */
772 	if (tid == TID_KERNEL)
773 		return;
774 
775 	msr = mfmsr();
776 	__asm __volatile("wrteei 0");
777 
778 	/*
779 	 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
780 	 * it for PID invalidation.
781 	 */
782 	mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
783 	__asm __volatile("isync; .long 0x7c200024; isync; msync");
784 
785 	__asm __volatile("wrtee %0" :: "r"(msr));
786 }
787