xref: /freebsd/lib/libkvm/kvm_minidump_powerpc64_hpt.c (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
1 /*-
2  * Copyright (c) 2006 Peter Wemm
3  * Copyright (c) 2019 Leandro Lupori
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * From: FreeBSD: src/lib/libkvm/kvm_minidump_riscv.c
27  */
28 
29 #include <sys/param.h>
30 #include <vm/vm.h>
31 
32 #include <kvm.h>
33 
34 #include <limits.h>
35 #include <stdint.h>
36 #include <stdlib.h>
37 #include <string.h>
38 #include <unistd.h>
39 
40 #include "../../sys/powerpc/include/minidump.h"
41 #include "kvm_private.h"
42 #include "kvm_powerpc64.h"
43 
44 /*
45  * PowerPC64 HPT machine dependent routines for kvm and minidumps.
46  *
47  * Address Translation parameters:
48  *
49  * b = 12 (SLB base page size: 4 KB)
50  * b = 24 (SLB base page size: 16 MB)
51  * p = 12 (page size: 4 KB)
52  * p = 24 (page size: 16 MB)
53  * s = 28 (segment size: 256 MB)
54  */
55 
56 /* Large (huge) page params */
57 #define	LP_PAGE_SHIFT		24
58 #define	LP_PAGE_SIZE		(1ULL << LP_PAGE_SHIFT)
59 #define	LP_PAGE_MASK		0x00ffffffULL
60 
61 /* SLB */
62 
63 #define	SEGMENT_LENGTH		0x10000000ULL
64 
65 #define	round_seg(x)		roundup2((uint64_t)(x), SEGMENT_LENGTH)
66 
67 /* Virtual real-mode VSID in LPARs */
68 #define	VSID_VRMA		0x1ffffffULL
69 
70 #define	SLBV_L			0x0000000000000100ULL /* Large page selector */
71 #define	SLBV_CLASS		0x0000000000000080ULL /* Class selector */
72 #define	SLBV_LP_MASK		0x0000000000000030ULL
73 #define	SLBV_VSID_MASK		0x3ffffffffffff000ULL /* Virtual SegID mask */
74 #define	SLBV_VSID_SHIFT		12
75 
76 #define	SLBE_B_MASK		0x0000000006000000ULL
77 #define	SLBE_B_256MB		0x0000000000000000ULL
78 #define	SLBE_VALID		0x0000000008000000ULL /* SLB entry valid */
79 #define	SLBE_INDEX_MASK		0x0000000000000fffULL /* SLB index mask */
80 #define	SLBE_ESID_MASK		0xfffffffff0000000ULL /* Effective SegID mask */
81 #define	SLBE_ESID_SHIFT		28
82 
83 /* PTE */
84 
85 #define	LPTEH_VSID_SHIFT	12
86 #define	LPTEH_AVPN_MASK		0xffffffffffffff80ULL
87 #define	LPTEH_B_MASK		0xc000000000000000ULL
88 #define	LPTEH_B_256MB		0x0000000000000000ULL
89 #define	LPTEH_BIG		0x0000000000000004ULL	/* 4KB/16MB page */
90 #define	LPTEH_HID		0x0000000000000002ULL
91 #define	LPTEH_VALID		0x0000000000000001ULL
92 
93 #define	LPTEL_RPGN		0xfffffffffffff000ULL
94 #define	LPTEL_LP_MASK		0x00000000000ff000ULL
95 #define	LPTEL_NOEXEC		0x0000000000000004ULL
96 
97 /* Supervisor        (U: RW, S: RW) */
98 #define	LPTEL_BW		0x0000000000000002ULL
99 
100 /* Both Read Only    (U: RO, S: RO) */
101 #define	LPTEL_BR		0x0000000000000003ULL
102 
103 #define	LPTEL_RW		LPTEL_BW
104 #define	LPTEL_RO		LPTEL_BR
105 
106 /*
107  * PTE AVA field manipulation macros.
108  *
109  * AVA[0:54] = PTEH[2:56]
110  * AVA[VSID] = AVA[0:49] = PTEH[2:51]
111  * AVA[PAGE] = AVA[50:54] = PTEH[52:56]
112  */
113 #define	PTEH_AVA_VSID_MASK	0x3ffffffffffff000UL
114 #define	PTEH_AVA_VSID_SHIFT	12
115 #define	PTEH_AVA_VSID(p) \
116 	(((p) & PTEH_AVA_VSID_MASK) >> PTEH_AVA_VSID_SHIFT)
117 
118 #define	PTEH_AVA_PAGE_MASK	0x0000000000000f80UL
119 #define	PTEH_AVA_PAGE_SHIFT	7
120 #define	PTEH_AVA_PAGE(p) \
121 	(((p) & PTEH_AVA_PAGE_MASK) >> PTEH_AVA_PAGE_SHIFT)
122 
123 /* Masks to obtain the Physical Address from PTE low 64-bit word. */
124 #define	PTEL_PA_MASK		0x0ffffffffffff000UL
125 #define	PTEL_LP_PA_MASK		0x0fffffffff000000UL
126 
127 #define	PTE_HASH_MASK		0x0000007fffffffffUL
128 
129 /*
130  * Number of AVA/VA page bits to shift right, in order to leave only the
131  * ones that should be considered.
132  *
133  * q = MIN(54, 77-b) (PowerISA v2.07B, 5.7.7.3)
134  * n = q + 1 - 50 (VSID size in bits)
135  * s(ava) = 5 - n
136  * s(va) = (28 - b) - n
137  *
138  * q: bit number of lower limit of VA/AVA bits to compare
139  * n: number of AVA/VA page bits to compare
140  * s: shift amount
141  * 28 - b: VA page size in bits
142  */
143 #define	AVA_PAGE_SHIFT(b)	(5 - (MIN(54, 77-(b)) + 1 - 50))
144 #define	VA_PAGE_SHIFT(b)	(28 - (b) - (MIN(54, 77-(b)) + 1 - 50))
145 
146 /* Kernel ESID -> VSID mapping */
147 #define	KERNEL_VSID_BIT	0x0000001000000000UL /* Bit set in all kernel VSIDs */
148 #define	KERNEL_VSID(esid) ((((((uint64_t)esid << 8) | ((uint64_t)esid >> 28)) \
149 				* 0x13bbUL) & (KERNEL_VSID_BIT - 1)) | \
150 				KERNEL_VSID_BIT)
151 
152 /* Types */
153 
154 typedef uint64_t	ppc64_physaddr_t;
155 
156 typedef struct {
157 	uint64_t slbv;
158 	uint64_t slbe;
159 } ppc64_slb_entry_t;
160 
161 typedef struct {
162 	uint64_t pte_hi;
163 	uint64_t pte_lo;
164 } ppc64_pt_entry_t;
165 
166 struct hpt_data {
167 	ppc64_slb_entry_t *slbs;
168 	uint32_t slbsize;
169 };
170 
171 
172 static void
173 slb_fill(ppc64_slb_entry_t *slb, uint64_t ea, uint64_t i)
174 {
175 	uint64_t esid;
176 
177 	esid = ea >> SLBE_ESID_SHIFT;
178 	slb->slbv = KERNEL_VSID(esid) << SLBV_VSID_SHIFT;
179 	slb->slbe = (esid << SLBE_ESID_SHIFT) | SLBE_VALID | i;
180 }
181 
182 static int
183 slb_init(kvm_t *kd)
184 {
185 	struct minidumphdr *hdr;
186 	struct hpt_data *data;
187 	ppc64_slb_entry_t *slb;
188 	uint32_t slbsize;
189 	uint64_t ea, i, maxmem;
190 
191 	hdr = &kd->vmst->hdr;
192 	data = PPC64_MMU_DATA(kd);
193 
194 	/* Alloc SLBs */
195 	maxmem = hdr->bitmapsize * 8 * PPC64_PAGE_SIZE;
196 	slbsize = round_seg(hdr->kernend + 1 - hdr->kernbase + maxmem) /
197 	    SEGMENT_LENGTH * sizeof(ppc64_slb_entry_t);
198 	data->slbs = _kvm_malloc(kd, slbsize);
199 	if (data->slbs == NULL) {
200 		_kvm_err(kd, kd->program, "cannot allocate slbs");
201 		return (-1);
202 	}
203 	data->slbsize = slbsize;
204 
205 	dprintf("%s: maxmem=0x%jx, segs=%jd, slbsize=0x%jx\n",
206 	    __func__, (uintmax_t)maxmem,
207 	    (uintmax_t)slbsize / sizeof(ppc64_slb_entry_t), (uintmax_t)slbsize);
208 
209 	/*
210 	 * Generate needed SLB entries.
211 	 *
212 	 * When translating addresses from EA to VA to PA, the needed SLB
213 	 * entry could be generated on the fly, but this is not the case
214 	 * for the walk_pages method, that needs to search the SLB entry
215 	 * by VSID, in order to find out the EA from a PTE.
216 	 */
217 
218 	/* VM area */
219 	for (ea = hdr->kernbase, i = 0, slb = data->slbs;
220 	    ea < hdr->kernend; ea += SEGMENT_LENGTH, i++, slb++)
221 		slb_fill(slb, ea, i);
222 
223 	/* DMAP area */
224 	for (ea = hdr->dmapbase;
225 	    ea < MIN(hdr->dmapend, hdr->dmapbase + maxmem);
226 	    ea += SEGMENT_LENGTH, i++, slb++) {
227 		slb_fill(slb, ea, i);
228 		if (hdr->hw_direct_map)
229 			slb->slbv |= SLBV_L;
230 	}
231 
232 	return (0);
233 }
234 
235 static void
236 ppc64mmu_hpt_cleanup(kvm_t *kd)
237 {
238 	struct hpt_data *data;
239 
240 	if (kd->vmst == NULL)
241 		return;
242 
243 	data = PPC64_MMU_DATA(kd);
244 	free(data->slbs);
245 	free(data);
246 	PPC64_MMU_DATA(kd) = NULL;
247 }
248 
249 static int
250 ppc64mmu_hpt_init(kvm_t *kd)
251 {
252 	struct hpt_data *data;
253 
254 	/* Alloc MMU data */
255 	data = _kvm_malloc(kd, sizeof(*data));
256 	if (data == NULL) {
257 		_kvm_err(kd, kd->program, "cannot allocate MMU data");
258 		return (-1);
259 	}
260 	data->slbs = NULL;
261 	PPC64_MMU_DATA(kd) = data;
262 
263 	if (slb_init(kd) == -1)
264 		goto failed;
265 
266 	return (0);
267 
268 failed:
269 	ppc64mmu_hpt_cleanup(kd);
270 	return (-1);
271 }
272 
273 static ppc64_slb_entry_t *
274 slb_search(kvm_t *kd, kvaddr_t ea)
275 {
276 	struct hpt_data *data;
277 	ppc64_slb_entry_t *slb;
278 	int i, n;
279 
280 	data = PPC64_MMU_DATA(kd);
281 	slb = data->slbs;
282 	n = data->slbsize / sizeof(ppc64_slb_entry_t);
283 
284 	/* SLB search */
285 	for (i = 0; i < n; i++, slb++) {
286 		if ((slb->slbe & SLBE_VALID) == 0)
287 			continue;
288 
289 		/* Compare 36-bit ESID of EA with segment one (64-s) */
290 		if ((slb->slbe & SLBE_ESID_MASK) != (ea & SLBE_ESID_MASK))
291 			continue;
292 
293 		/* Match found */
294 		dprintf("SEG#%02d: slbv=0x%016jx, slbe=0x%016jx\n",
295 		    i, (uintmax_t)slb->slbv, (uintmax_t)slb->slbe);
296 		break;
297 	}
298 
299 	/* SLB not found */
300 	if (i == n) {
301 		_kvm_err(kd, kd->program, "%s: segment not found for EA 0x%jx",
302 		    __func__, (uintmax_t)ea);
303 		return (NULL);
304 	}
305 	return (slb);
306 }
307 
308 static ppc64_pt_entry_t
309 pte_get(kvm_t *kd, u_long ptex)
310 {
311 	ppc64_pt_entry_t pte, *p;
312 
313 	p = _kvm_pmap_get(kd, ptex, sizeof(pte));
314 	pte.pte_hi = be64toh(p->pte_hi);
315 	pte.pte_lo = be64toh(p->pte_lo);
316 	return (pte);
317 }
318 
319 static int
320 pte_search(kvm_t *kd, ppc64_slb_entry_t *slb, uint64_t hid, kvaddr_t ea,
321     ppc64_pt_entry_t *p)
322 {
323 	uint64_t hash, hmask;
324 	uint64_t pteg, ptex;
325 	uint64_t va_vsid, va_page;
326 	int b;
327 	int ava_pg_shift, va_pg_shift;
328 	ppc64_pt_entry_t pte;
329 
330 	/*
331 	 * Get VA:
332 	 *
333 	 * va(78) = va_vsid(50) || va_page(s-b) || offset(b)
334 	 *
335 	 * va_vsid: 50-bit VSID (78-s)
336 	 * va_page: (s-b)-bit VA page
337 	 */
338 	b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
339 	va_vsid = (slb->slbv & SLBV_VSID_MASK) >> SLBV_VSID_SHIFT;
340 	va_page = (ea & ~SLBE_ESID_MASK) >> b;
341 
342 	dprintf("%s: hid=0x%jx, ea=0x%016jx, b=%d, va_vsid=0x%010jx, "
343 	    "va_page=0x%04jx\n",
344 	    __func__, (uintmax_t)hid, (uintmax_t)ea, b,
345 	    (uintmax_t)va_vsid, (uintmax_t)va_page);
346 
347 	/*
348 	 * Get hash:
349 	 *
350 	 * Primary hash: va_vsid(11:49) ^ va_page(s-b)
351 	 * Secondary hash: ~primary_hash
352 	 */
353 	hash = (va_vsid & PTE_HASH_MASK) ^ va_page;
354 	if (hid)
355 		hash = ~hash & PTE_HASH_MASK;
356 
357 	/*
358 	 * Get PTEG:
359 	 *
360 	 * pteg = (hash(0:38) & hmask) << 3
361 	 *
362 	 * hmask (hash mask): mask generated from HTABSIZE || 11*0b1
363 	 * hmask = number_of_ptegs - 1
364 	 */
365 	hmask = kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) - 1;
366 	pteg = (hash & hmask) << 3;
367 
368 	ava_pg_shift = AVA_PAGE_SHIFT(b);
369 	va_pg_shift = VA_PAGE_SHIFT(b);
370 
371 	dprintf("%s: hash=0x%010jx, hmask=0x%010jx, (hash & hmask)=0x%010jx, "
372 	    "pteg=0x%011jx, ava_pg_shift=%d, va_pg_shift=%d\n",
373 	    __func__, (uintmax_t)hash, (uintmax_t)hmask,
374 	    (uintmax_t)(hash & hmask), (uintmax_t)pteg,
375 	    ava_pg_shift, va_pg_shift);
376 
377 	/* Search PTEG */
378 	for (ptex = pteg; ptex < pteg + 8; ptex++) {
379 		pte = pte_get(kd, ptex);
380 
381 		/* Check H, V and B */
382 		if ((pte.pte_hi & LPTEH_HID) != hid ||
383 		    (pte.pte_hi & LPTEH_VALID) == 0 ||
384 		    (pte.pte_hi & LPTEH_B_MASK) != LPTEH_B_256MB)
385 			continue;
386 
387 		/* Compare AVA with VA */
388 		if (PTEH_AVA_VSID(pte.pte_hi) != va_vsid ||
389 		    (PTEH_AVA_PAGE(pte.pte_hi) >> ava_pg_shift) !=
390 		    (va_page >> va_pg_shift))
391 			continue;
392 
393 		/*
394 		 * Check if PTE[L] matches SLBV[L].
395 		 *
396 		 * Note: this check ignores PTE[LP], as does the kernel.
397 		 */
398 		if (b == PPC64_PAGE_SHIFT) {
399 			if (pte.pte_hi & LPTEH_BIG)
400 				continue;
401 		} else if ((pte.pte_hi & LPTEH_BIG) == 0)
402 			continue;
403 
404 		/* Match found */
405 		dprintf("%s: PTE found: ptex=0x%jx, pteh=0x%016jx, "
406 		    "ptel=0x%016jx\n",
407 		    __func__, (uintmax_t)ptex, (uintmax_t)pte.pte_hi,
408 		    (uintmax_t)pte.pte_lo);
409 		break;
410 	}
411 
412 	/* Not found? */
413 	if (ptex == pteg + 8) {
414 		/* Try secondary hash */
415 		if (hid == 0)
416 			return (pte_search(kd, slb, LPTEH_HID, ea, p));
417 		else {
418 			_kvm_err(kd, kd->program,
419 			    "%s: pte not found", __func__);
420 			return (-1);
421 		}
422 	}
423 
424 	/* PTE found */
425 	*p = pte;
426 	return (0);
427 }
428 
429 static int
430 pte_lookup(kvm_t *kd, kvaddr_t ea, ppc64_pt_entry_t *pte)
431 {
432 	ppc64_slb_entry_t *slb;
433 
434 	/* First, find SLB */
435 	if ((slb = slb_search(kd, ea)) == NULL)
436 		return (-1);
437 
438 	/* Next, find PTE */
439 	return (pte_search(kd, slb, 0, ea, pte));
440 }
441 
442 static int
443 ppc64mmu_hpt_kvatop(kvm_t *kd, kvaddr_t va, off_t *pa)
444 {
445 	struct minidumphdr *hdr;
446 	struct vmstate *vm;
447 	ppc64_pt_entry_t pte;
448 	ppc64_physaddr_t pgoff, pgpa;
449 	off_t ptoff;
450 	int err;
451 
452 	vm = kd->vmst;
453 	hdr = &vm->hdr;
454 	pgoff = va & PPC64_PAGE_MASK;
455 
456 	dprintf("%s: va=0x%016jx\n", __func__, (uintmax_t)va);
457 
458 	/*
459 	 * A common use case of libkvm is to first find a symbol address
460 	 * from the kernel image and then use kvatop to translate it and
461 	 * to be able to fetch its corresponding data.
462 	 *
463 	 * The problem is that, in PowerPC64 case, the addresses of relocated
464 	 * data won't match those in the kernel image. This is handled here by
465 	 * adding the relocation offset to those addresses.
466 	 */
467 	if (va < hdr->dmapbase)
468 		va += hdr->startkernel - PPC64_KERNBASE;
469 
470 	/* Handle DMAP */
471 	if (va >= hdr->dmapbase && va <= hdr->dmapend) {
472 		pgpa = (va & ~hdr->dmapbase) & ~PPC64_PAGE_MASK;
473 		ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
474 		if (ptoff == -1) {
475 			_kvm_err(kd, kd->program, "%s: "
476 			    "direct map address 0x%jx not in minidump",
477 			    __func__, (uintmax_t)va);
478 			goto invalid;
479 		}
480 		*pa = ptoff + pgoff;
481 		return (PPC64_PAGE_SIZE - pgoff);
482 	/* Translate VA to PA */
483 	} else if (va >= hdr->kernbase) {
484 		if ((err = pte_lookup(kd, va, &pte)) == -1) {
485 			_kvm_err(kd, kd->program,
486 			    "%s: pte not valid", __func__);
487 			goto invalid;
488 		}
489 
490 		if (pte.pte_hi & LPTEH_BIG)
491 			pgpa = (pte.pte_lo & PTEL_LP_PA_MASK) |
492 			    (va & ~PPC64_PAGE_MASK & LP_PAGE_MASK);
493 		else
494 			pgpa = pte.pte_lo & PTEL_PA_MASK;
495 		dprintf("%s: pgpa=0x%016jx\n", __func__, (uintmax_t)pgpa);
496 
497 		ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
498 		if (ptoff == -1) {
499 			_kvm_err(kd, kd->program, "%s: "
500 			    "physical address 0x%jx not in minidump",
501 			    __func__, (uintmax_t)pgpa);
502 			goto invalid;
503 		}
504 		*pa = ptoff + pgoff;
505 		return (PPC64_PAGE_SIZE - pgoff);
506 	} else {
507 		_kvm_err(kd, kd->program,
508 		    "%s: virtual address 0x%jx not minidumped",
509 		    __func__, (uintmax_t)va);
510 		goto invalid;
511 	}
512 
513 invalid:
514 	_kvm_err(kd, 0, "invalid address (0x%jx)", (uintmax_t)va);
515 	return (0);
516 }
517 
518 static vm_prot_t
519 entry_to_prot(ppc64_pt_entry_t *pte)
520 {
521 	vm_prot_t prot = VM_PROT_READ;
522 
523 	if (pte->pte_lo & LPTEL_RW)
524 		prot |= VM_PROT_WRITE;
525 	if ((pte->pte_lo & LPTEL_NOEXEC) != 0)
526 		prot |= VM_PROT_EXECUTE;
527 	return (prot);
528 }
529 
530 static ppc64_slb_entry_t *
531 slb_vsid_search(kvm_t *kd, uint64_t vsid)
532 {
533 	struct hpt_data *data;
534 	ppc64_slb_entry_t *slb;
535 	int i, n;
536 
537 	data = PPC64_MMU_DATA(kd);
538 	slb = data->slbs;
539 	n = data->slbsize / sizeof(ppc64_slb_entry_t);
540 	vsid <<= SLBV_VSID_SHIFT;
541 
542 	/* SLB search */
543 	for (i = 0; i < n; i++, slb++) {
544 		/* Check if valid and compare VSID */
545 		if ((slb->slbe & SLBE_VALID) &&
546 		    (slb->slbv & SLBV_VSID_MASK) == vsid)
547 			break;
548 	}
549 
550 	/* SLB not found */
551 	if (i == n) {
552 		_kvm_err(kd, kd->program,
553 		    "%s: segment not found for VSID 0x%jx",
554 		    __func__, (uintmax_t)vsid >> SLBV_VSID_SHIFT);
555 		return (NULL);
556 	}
557 	return (slb);
558 }
559 
560 static u_long
561 get_ea(kvm_t *kd, ppc64_pt_entry_t *pte, u_long ptex)
562 {
563 	ppc64_slb_entry_t *slb;
564 	uint64_t ea, hash, vsid;
565 	int b, shift;
566 
567 	/* Find SLB */
568 	vsid = PTEH_AVA_VSID(pte->pte_hi);
569 	if ((slb = slb_vsid_search(kd, vsid)) == NULL)
570 		return (~0UL);
571 
572 	/* Get ESID part of EA */
573 	ea = slb->slbe & SLBE_ESID_MASK;
574 
575 	b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
576 
577 	/*
578 	 * If there are less than 64K PTEGs (16-bit), the upper bits of
579 	 * EA page must be obtained from PTEH's AVA.
580 	 */
581 	if (kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) <
582 	    0x10000U) {
583 		/*
584 		 * Add 0 to 5 EA bits, right after VSID.
585 		 * b == 12: 5 bits
586 		 * b == 24: 4 bits
587 		 */
588 		shift = AVA_PAGE_SHIFT(b);
589 		ea |= (PTEH_AVA_PAGE(pte->pte_hi) >> shift) <<
590 		    (SLBE_ESID_SHIFT - 5 + shift);
591 	}
592 
593 	/* Get VA page from hash and add to EA. */
594 	hash = (ptex & ~7) >> 3;
595 	if (pte->pte_hi & LPTEH_HID)
596 		hash = ~hash & PTE_HASH_MASK;
597 	ea |= ((hash ^ (vsid & PTE_HASH_MASK)) << b) & ~SLBE_ESID_MASK;
598 	return (ea);
599 }
600 
601 static int
602 ppc64mmu_hpt_walk_pages(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg)
603 {
604 	struct vmstate *vm;
605 	int ret;
606 	unsigned int pagesz;
607 	u_long dva, pa, va;
608 	u_long ptex, nptes;
609 	uint64_t vsid;
610 
611 	ret = 0;
612 	vm = kd->vmst;
613 	nptes = vm->hdr.pmapsize / sizeof(ppc64_pt_entry_t);
614 
615 	/* Walk through PTEs */
616 	for (ptex = 0; ptex < nptes; ptex++) {
617 		ppc64_pt_entry_t pte = pte_get(kd, ptex);
618 		if ((pte.pte_hi & LPTEH_VALID) == 0)
619 			continue;
620 
621 		/* Skip non-kernel related pages, as well as VRMA ones */
622 		vsid = PTEH_AVA_VSID(pte.pte_hi);
623 		if ((vsid & KERNEL_VSID_BIT) == 0 ||
624 		    (vsid >> PPC64_PAGE_SHIFT) == VSID_VRMA)
625 			continue;
626 
627 		/* Retrieve page's VA (EA on PPC64 terminology) */
628 		if ((va = get_ea(kd, &pte, ptex)) == ~0UL)
629 			goto out;
630 
631 		/* Get PA and page size */
632 		if (pte.pte_hi & LPTEH_BIG) {
633 			pa = pte.pte_lo & PTEL_LP_PA_MASK;
634 			pagesz = LP_PAGE_SIZE;
635 		} else {
636 			pa = pte.pte_lo & PTEL_PA_MASK;
637 			pagesz = PPC64_PAGE_SIZE;
638 		}
639 
640 		/* Get DMAP address */
641 		dva = vm->hdr.dmapbase + pa;
642 
643 		if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva,
644 		    entry_to_prot(&pte), pagesz, 0))
645 			goto out;
646 	}
647 	ret = 1;
648 
649 out:
650 	return (ret);
651 }
652 
653 
654 static struct ppc64_mmu_ops ops = {
655 	.init		= ppc64mmu_hpt_init,
656 	.cleanup	= ppc64mmu_hpt_cleanup,
657 	.kvatop		= ppc64mmu_hpt_kvatop,
658 	.walk_pages	= ppc64mmu_hpt_walk_pages,
659 };
660 struct ppc64_mmu_ops *ppc64_mmu_ops_hpt = &ops;
661