xref: /freebsd/lib/libkvm/kvm_minidump_powerpc64_hpt.c (revision 1d386b48a555f61cb7325543adbbb5c3f3407a66)
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/cdefs.h>
30 #include <sys/param.h>
31 #include <vm/vm.h>
32 
33 #include <kvm.h>
34 
35 #include <limits.h>
36 #include <stdint.h>
37 #include <stdlib.h>
38 #include <string.h>
39 #include <unistd.h>
40 
41 #include "../../sys/powerpc/include/minidump.h"
42 #include "kvm_private.h"
43 #include "kvm_powerpc64.h"
44 
45 /*
46  * PowerPC64 HPT machine dependent routines for kvm and minidumps.
47  *
48  * Address Translation parameters:
49  *
50  * b = 12 (SLB base page size: 4 KB)
51  * b = 24 (SLB base page size: 16 MB)
52  * p = 12 (page size: 4 KB)
53  * p = 24 (page size: 16 MB)
54  * s = 28 (segment size: 256 MB)
55  */
56 
57 /* Large (huge) page params */
58 #define	LP_PAGE_SHIFT		24
59 #define	LP_PAGE_SIZE		(1ULL << LP_PAGE_SHIFT)
60 #define	LP_PAGE_MASK		0x00ffffffULL
61 
62 /* SLB */
63 
64 #define	SEGMENT_LENGTH		0x10000000ULL
65 
66 #define	round_seg(x)		roundup2((uint64_t)(x), SEGMENT_LENGTH)
67 
68 /* Virtual real-mode VSID in LPARs */
69 #define	VSID_VRMA		0x1ffffffULL
70 
71 #define	SLBV_L			0x0000000000000100ULL /* Large page selector */
72 #define	SLBV_CLASS		0x0000000000000080ULL /* Class selector */
73 #define	SLBV_LP_MASK		0x0000000000000030ULL
74 #define	SLBV_VSID_MASK		0x3ffffffffffff000ULL /* Virtual SegID mask */
75 #define	SLBV_VSID_SHIFT		12
76 
77 #define	SLBE_B_MASK		0x0000000006000000ULL
78 #define	SLBE_B_256MB		0x0000000000000000ULL
79 #define	SLBE_VALID		0x0000000008000000ULL /* SLB entry valid */
80 #define	SLBE_INDEX_MASK		0x0000000000000fffULL /* SLB index mask */
81 #define	SLBE_ESID_MASK		0xfffffffff0000000ULL /* Effective SegID mask */
82 #define	SLBE_ESID_SHIFT		28
83 
84 /* PTE */
85 
86 #define	LPTEH_VSID_SHIFT	12
87 #define	LPTEH_AVPN_MASK		0xffffffffffffff80ULL
88 #define	LPTEH_B_MASK		0xc000000000000000ULL
89 #define	LPTEH_B_256MB		0x0000000000000000ULL
90 #define	LPTEH_BIG		0x0000000000000004ULL	/* 4KB/16MB page */
91 #define	LPTEH_HID		0x0000000000000002ULL
92 #define	LPTEH_VALID		0x0000000000000001ULL
93 
94 #define	LPTEL_RPGN		0xfffffffffffff000ULL
95 #define	LPTEL_LP_MASK		0x00000000000ff000ULL
96 #define	LPTEL_NOEXEC		0x0000000000000004ULL
97 
98 /* Supervisor        (U: RW, S: RW) */
99 #define	LPTEL_BW		0x0000000000000002ULL
100 
101 /* Both Read Only    (U: RO, S: RO) */
102 #define	LPTEL_BR		0x0000000000000003ULL
103 
104 #define	LPTEL_RW		LPTEL_BW
105 #define	LPTEL_RO		LPTEL_BR
106 
107 /*
108  * PTE AVA field manipulation macros.
109  *
110  * AVA[0:54] = PTEH[2:56]
111  * AVA[VSID] = AVA[0:49] = PTEH[2:51]
112  * AVA[PAGE] = AVA[50:54] = PTEH[52:56]
113  */
114 #define	PTEH_AVA_VSID_MASK	0x3ffffffffffff000UL
115 #define	PTEH_AVA_VSID_SHIFT	12
116 #define	PTEH_AVA_VSID(p) \
117 	(((p) & PTEH_AVA_VSID_MASK) >> PTEH_AVA_VSID_SHIFT)
118 
119 #define	PTEH_AVA_PAGE_MASK	0x0000000000000f80UL
120 #define	PTEH_AVA_PAGE_SHIFT	7
121 #define	PTEH_AVA_PAGE(p) \
122 	(((p) & PTEH_AVA_PAGE_MASK) >> PTEH_AVA_PAGE_SHIFT)
123 
124 /* Masks to obtain the Physical Address from PTE low 64-bit word. */
125 #define	PTEL_PA_MASK		0x0ffffffffffff000UL
126 #define	PTEL_LP_PA_MASK		0x0fffffffff000000UL
127 
128 #define	PTE_HASH_MASK		0x0000007fffffffffUL
129 
130 /*
131  * Number of AVA/VA page bits to shift right, in order to leave only the
132  * ones that should be considered.
133  *
134  * q = MIN(54, 77-b) (PowerISA v2.07B, 5.7.7.3)
135  * n = q + 1 - 50 (VSID size in bits)
136  * s(ava) = 5 - n
137  * s(va) = (28 - b) - n
138  *
139  * q: bit number of lower limit of VA/AVA bits to compare
140  * n: number of AVA/VA page bits to compare
141  * s: shift amount
142  * 28 - b: VA page size in bits
143  */
144 #define	AVA_PAGE_SHIFT(b)	(5 - (MIN(54, 77-(b)) + 1 - 50))
145 #define	VA_PAGE_SHIFT(b)	(28 - (b) - (MIN(54, 77-(b)) + 1 - 50))
146 
147 /* Kernel ESID -> VSID mapping */
148 #define	KERNEL_VSID_BIT	0x0000001000000000UL /* Bit set in all kernel VSIDs */
149 #define	KERNEL_VSID(esid) ((((((uint64_t)esid << 8) | ((uint64_t)esid >> 28)) \
150 				* 0x13bbUL) & (KERNEL_VSID_BIT - 1)) | \
151 				KERNEL_VSID_BIT)
152 
153 /* Types */
154 
155 typedef uint64_t	ppc64_physaddr_t;
156 
157 typedef struct {
158 	uint64_t slbv;
159 	uint64_t slbe;
160 } ppc64_slb_entry_t;
161 
162 typedef struct {
163 	uint64_t pte_hi;
164 	uint64_t pte_lo;
165 } ppc64_pt_entry_t;
166 
167 struct hpt_data {
168 	ppc64_slb_entry_t *slbs;
169 	uint32_t slbsize;
170 };
171 
172 
173 static void
174 slb_fill(ppc64_slb_entry_t *slb, uint64_t ea, uint64_t i)
175 {
176 	uint64_t esid;
177 
178 	esid = ea >> SLBE_ESID_SHIFT;
179 	slb->slbv = KERNEL_VSID(esid) << SLBV_VSID_SHIFT;
180 	slb->slbe = (esid << SLBE_ESID_SHIFT) | SLBE_VALID | i;
181 }
182 
183 static int
184 slb_init(kvm_t *kd)
185 {
186 	struct minidumphdr *hdr;
187 	struct hpt_data *data;
188 	ppc64_slb_entry_t *slb;
189 	uint32_t slbsize;
190 	uint64_t ea, i, maxmem;
191 
192 	hdr = &kd->vmst->hdr;
193 	data = PPC64_MMU_DATA(kd);
194 
195 	/* Alloc SLBs */
196 	maxmem = hdr->bitmapsize * 8 * PPC64_PAGE_SIZE;
197 	slbsize = round_seg(hdr->kernend + 1 - hdr->kernbase + maxmem) /
198 	    SEGMENT_LENGTH * sizeof(ppc64_slb_entry_t);
199 	data->slbs = _kvm_malloc(kd, slbsize);
200 	if (data->slbs == NULL) {
201 		_kvm_err(kd, kd->program, "cannot allocate slbs");
202 		return (-1);
203 	}
204 	data->slbsize = slbsize;
205 
206 	dprintf("%s: maxmem=0x%jx, segs=%jd, slbsize=0x%jx\n",
207 	    __func__, (uintmax_t)maxmem,
208 	    (uintmax_t)slbsize / sizeof(ppc64_slb_entry_t), (uintmax_t)slbsize);
209 
210 	/*
211 	 * Generate needed SLB entries.
212 	 *
213 	 * When translating addresses from EA to VA to PA, the needed SLB
214 	 * entry could be generated on the fly, but this is not the case
215 	 * for the walk_pages method, that needs to search the SLB entry
216 	 * by VSID, in order to find out the EA from a PTE.
217 	 */
218 
219 	/* VM area */
220 	for (ea = hdr->kernbase, i = 0, slb = data->slbs;
221 	    ea < hdr->kernend; ea += SEGMENT_LENGTH, i++, slb++)
222 		slb_fill(slb, ea, i);
223 
224 	/* DMAP area */
225 	for (ea = hdr->dmapbase;
226 	    ea < MIN(hdr->dmapend, hdr->dmapbase + maxmem);
227 	    ea += SEGMENT_LENGTH, i++, slb++) {
228 		slb_fill(slb, ea, i);
229 		if (hdr->hw_direct_map)
230 			slb->slbv |= SLBV_L;
231 	}
232 
233 	return (0);
234 }
235 
236 static void
237 ppc64mmu_hpt_cleanup(kvm_t *kd)
238 {
239 	struct hpt_data *data;
240 
241 	if (kd->vmst == NULL)
242 		return;
243 
244 	data = PPC64_MMU_DATA(kd);
245 	free(data->slbs);
246 	free(data);
247 	PPC64_MMU_DATA(kd) = NULL;
248 }
249 
250 static int
251 ppc64mmu_hpt_init(kvm_t *kd)
252 {
253 	struct hpt_data *data;
254 
255 	/* Alloc MMU data */
256 	data = _kvm_malloc(kd, sizeof(*data));
257 	if (data == NULL) {
258 		_kvm_err(kd, kd->program, "cannot allocate MMU data");
259 		return (-1);
260 	}
261 	data->slbs = NULL;
262 	PPC64_MMU_DATA(kd) = data;
263 
264 	if (slb_init(kd) == -1)
265 		goto failed;
266 
267 	return (0);
268 
269 failed:
270 	ppc64mmu_hpt_cleanup(kd);
271 	return (-1);
272 }
273 
274 static ppc64_slb_entry_t *
275 slb_search(kvm_t *kd, kvaddr_t ea)
276 {
277 	struct hpt_data *data;
278 	ppc64_slb_entry_t *slb;
279 	int i, n;
280 
281 	data = PPC64_MMU_DATA(kd);
282 	slb = data->slbs;
283 	n = data->slbsize / sizeof(ppc64_slb_entry_t);
284 
285 	/* SLB search */
286 	for (i = 0; i < n; i++, slb++) {
287 		if ((slb->slbe & SLBE_VALID) == 0)
288 			continue;
289 
290 		/* Compare 36-bit ESID of EA with segment one (64-s) */
291 		if ((slb->slbe & SLBE_ESID_MASK) != (ea & SLBE_ESID_MASK))
292 			continue;
293 
294 		/* Match found */
295 		dprintf("SEG#%02d: slbv=0x%016jx, slbe=0x%016jx\n",
296 		    i, (uintmax_t)slb->slbv, (uintmax_t)slb->slbe);
297 		break;
298 	}
299 
300 	/* SLB not found */
301 	if (i == n) {
302 		_kvm_err(kd, kd->program, "%s: segment not found for EA 0x%jx",
303 		    __func__, (uintmax_t)ea);
304 		return (NULL);
305 	}
306 	return (slb);
307 }
308 
309 static ppc64_pt_entry_t
310 pte_get(kvm_t *kd, u_long ptex)
311 {
312 	ppc64_pt_entry_t pte, *p;
313 
314 	p = _kvm_pmap_get(kd, ptex, sizeof(pte));
315 	pte.pte_hi = be64toh(p->pte_hi);
316 	pte.pte_lo = be64toh(p->pte_lo);
317 	return (pte);
318 }
319 
320 static int
321 pte_search(kvm_t *kd, ppc64_slb_entry_t *slb, uint64_t hid, kvaddr_t ea,
322     ppc64_pt_entry_t *p)
323 {
324 	uint64_t hash, hmask;
325 	uint64_t pteg, ptex;
326 	uint64_t va_vsid, va_page;
327 	int b;
328 	int ava_pg_shift, va_pg_shift;
329 	ppc64_pt_entry_t pte;
330 
331 	/*
332 	 * Get VA:
333 	 *
334 	 * va(78) = va_vsid(50) || va_page(s-b) || offset(b)
335 	 *
336 	 * va_vsid: 50-bit VSID (78-s)
337 	 * va_page: (s-b)-bit VA page
338 	 */
339 	b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
340 	va_vsid = (slb->slbv & SLBV_VSID_MASK) >> SLBV_VSID_SHIFT;
341 	va_page = (ea & ~SLBE_ESID_MASK) >> b;
342 
343 	dprintf("%s: hid=0x%jx, ea=0x%016jx, b=%d, va_vsid=0x%010jx, "
344 	    "va_page=0x%04jx\n",
345 	    __func__, (uintmax_t)hid, (uintmax_t)ea, b,
346 	    (uintmax_t)va_vsid, (uintmax_t)va_page);
347 
348 	/*
349 	 * Get hash:
350 	 *
351 	 * Primary hash: va_vsid(11:49) ^ va_page(s-b)
352 	 * Secondary hash: ~primary_hash
353 	 */
354 	hash = (va_vsid & PTE_HASH_MASK) ^ va_page;
355 	if (hid)
356 		hash = ~hash & PTE_HASH_MASK;
357 
358 	/*
359 	 * Get PTEG:
360 	 *
361 	 * pteg = (hash(0:38) & hmask) << 3
362 	 *
363 	 * hmask (hash mask): mask generated from HTABSIZE || 11*0b1
364 	 * hmask = number_of_ptegs - 1
365 	 */
366 	hmask = kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) - 1;
367 	pteg = (hash & hmask) << 3;
368 
369 	ava_pg_shift = AVA_PAGE_SHIFT(b);
370 	va_pg_shift = VA_PAGE_SHIFT(b);
371 
372 	dprintf("%s: hash=0x%010jx, hmask=0x%010jx, (hash & hmask)=0x%010jx, "
373 	    "pteg=0x%011jx, ava_pg_shift=%d, va_pg_shift=%d\n",
374 	    __func__, (uintmax_t)hash, (uintmax_t)hmask,
375 	    (uintmax_t)(hash & hmask), (uintmax_t)pteg,
376 	    ava_pg_shift, va_pg_shift);
377 
378 	/* Search PTEG */
379 	for (ptex = pteg; ptex < pteg + 8; ptex++) {
380 		pte = pte_get(kd, ptex);
381 
382 		/* Check H, V and B */
383 		if ((pte.pte_hi & LPTEH_HID) != hid ||
384 		    (pte.pte_hi & LPTEH_VALID) == 0 ||
385 		    (pte.pte_hi & LPTEH_B_MASK) != LPTEH_B_256MB)
386 			continue;
387 
388 		/* Compare AVA with VA */
389 		if (PTEH_AVA_VSID(pte.pte_hi) != va_vsid ||
390 		    (PTEH_AVA_PAGE(pte.pte_hi) >> ava_pg_shift) !=
391 		    (va_page >> va_pg_shift))
392 			continue;
393 
394 		/*
395 		 * Check if PTE[L] matches SLBV[L].
396 		 *
397 		 * Note: this check ignores PTE[LP], as does the kernel.
398 		 */
399 		if (b == PPC64_PAGE_SHIFT) {
400 			if (pte.pte_hi & LPTEH_BIG)
401 				continue;
402 		} else if ((pte.pte_hi & LPTEH_BIG) == 0)
403 			continue;
404 
405 		/* Match found */
406 		dprintf("%s: PTE found: ptex=0x%jx, pteh=0x%016jx, "
407 		    "ptel=0x%016jx\n",
408 		    __func__, (uintmax_t)ptex, (uintmax_t)pte.pte_hi,
409 		    (uintmax_t)pte.pte_lo);
410 		break;
411 	}
412 
413 	/* Not found? */
414 	if (ptex == pteg + 8) {
415 		/* Try secondary hash */
416 		if (hid == 0)
417 			return (pte_search(kd, slb, LPTEH_HID, ea, p));
418 		else {
419 			_kvm_err(kd, kd->program,
420 			    "%s: pte not found", __func__);
421 			return (-1);
422 		}
423 	}
424 
425 	/* PTE found */
426 	*p = pte;
427 	return (0);
428 }
429 
430 static int
431 pte_lookup(kvm_t *kd, kvaddr_t ea, ppc64_pt_entry_t *pte)
432 {
433 	ppc64_slb_entry_t *slb;
434 
435 	/* First, find SLB */
436 	if ((slb = slb_search(kd, ea)) == NULL)
437 		return (-1);
438 
439 	/* Next, find PTE */
440 	return (pte_search(kd, slb, 0, ea, pte));
441 }
442 
443 static int
444 ppc64mmu_hpt_kvatop(kvm_t *kd, kvaddr_t va, off_t *pa)
445 {
446 	struct minidumphdr *hdr;
447 	struct vmstate *vm;
448 	ppc64_pt_entry_t pte;
449 	ppc64_physaddr_t pgoff, pgpa;
450 	off_t ptoff;
451 	int err;
452 
453 	vm = kd->vmst;
454 	hdr = &vm->hdr;
455 	pgoff = va & PPC64_PAGE_MASK;
456 
457 	dprintf("%s: va=0x%016jx\n", __func__, (uintmax_t)va);
458 
459 	/*
460 	 * A common use case of libkvm is to first find a symbol address
461 	 * from the kernel image and then use kvatop to translate it and
462 	 * to be able to fetch its corresponding data.
463 	 *
464 	 * The problem is that, in PowerPC64 case, the addresses of relocated
465 	 * data won't match those in the kernel image. This is handled here by
466 	 * adding the relocation offset to those addresses.
467 	 */
468 	if (va < hdr->dmapbase)
469 		va += hdr->startkernel - PPC64_KERNBASE;
470 
471 	/* Handle DMAP */
472 	if (va >= hdr->dmapbase && va <= hdr->dmapend) {
473 		pgpa = (va & ~hdr->dmapbase) & ~PPC64_PAGE_MASK;
474 		ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
475 		if (ptoff == -1) {
476 			_kvm_err(kd, kd->program, "%s: "
477 			    "direct map address 0x%jx not in minidump",
478 			    __func__, (uintmax_t)va);
479 			goto invalid;
480 		}
481 		*pa = ptoff + pgoff;
482 		return (PPC64_PAGE_SIZE - pgoff);
483 	/* Translate VA to PA */
484 	} else if (va >= hdr->kernbase) {
485 		if ((err = pte_lookup(kd, va, &pte)) == -1) {
486 			_kvm_err(kd, kd->program,
487 			    "%s: pte not valid", __func__);
488 			goto invalid;
489 		}
490 
491 		if (pte.pte_hi & LPTEH_BIG)
492 			pgpa = (pte.pte_lo & PTEL_LP_PA_MASK) |
493 			    (va & ~PPC64_PAGE_MASK & LP_PAGE_MASK);
494 		else
495 			pgpa = pte.pte_lo & PTEL_PA_MASK;
496 		dprintf("%s: pgpa=0x%016jx\n", __func__, (uintmax_t)pgpa);
497 
498 		ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
499 		if (ptoff == -1) {
500 			_kvm_err(kd, kd->program, "%s: "
501 			    "physical address 0x%jx not in minidump",
502 			    __func__, (uintmax_t)pgpa);
503 			goto invalid;
504 		}
505 		*pa = ptoff + pgoff;
506 		return (PPC64_PAGE_SIZE - pgoff);
507 	} else {
508 		_kvm_err(kd, kd->program,
509 		    "%s: virtual address 0x%jx not minidumped",
510 		    __func__, (uintmax_t)va);
511 		goto invalid;
512 	}
513 
514 invalid:
515 	_kvm_err(kd, 0, "invalid address (0x%jx)", (uintmax_t)va);
516 	return (0);
517 }
518 
519 static vm_prot_t
520 entry_to_prot(ppc64_pt_entry_t *pte)
521 {
522 	vm_prot_t prot = VM_PROT_READ;
523 
524 	if (pte->pte_lo & LPTEL_RW)
525 		prot |= VM_PROT_WRITE;
526 	if ((pte->pte_lo & LPTEL_NOEXEC) != 0)
527 		prot |= VM_PROT_EXECUTE;
528 	return (prot);
529 }
530 
531 static ppc64_slb_entry_t *
532 slb_vsid_search(kvm_t *kd, uint64_t vsid)
533 {
534 	struct hpt_data *data;
535 	ppc64_slb_entry_t *slb;
536 	int i, n;
537 
538 	data = PPC64_MMU_DATA(kd);
539 	slb = data->slbs;
540 	n = data->slbsize / sizeof(ppc64_slb_entry_t);
541 	vsid <<= SLBV_VSID_SHIFT;
542 
543 	/* SLB search */
544 	for (i = 0; i < n; i++, slb++) {
545 		/* Check if valid and compare VSID */
546 		if ((slb->slbe & SLBE_VALID) &&
547 		    (slb->slbv & SLBV_VSID_MASK) == vsid)
548 			break;
549 	}
550 
551 	/* SLB not found */
552 	if (i == n) {
553 		_kvm_err(kd, kd->program,
554 		    "%s: segment not found for VSID 0x%jx",
555 		    __func__, (uintmax_t)vsid >> SLBV_VSID_SHIFT);
556 		return (NULL);
557 	}
558 	return (slb);
559 }
560 
561 static u_long
562 get_ea(kvm_t *kd, ppc64_pt_entry_t *pte, u_long ptex)
563 {
564 	ppc64_slb_entry_t *slb;
565 	uint64_t ea, hash, vsid;
566 	int b, shift;
567 
568 	/* Find SLB */
569 	vsid = PTEH_AVA_VSID(pte->pte_hi);
570 	if ((slb = slb_vsid_search(kd, vsid)) == NULL)
571 		return (~0UL);
572 
573 	/* Get ESID part of EA */
574 	ea = slb->slbe & SLBE_ESID_MASK;
575 
576 	b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
577 
578 	/*
579 	 * If there are less than 64K PTEGs (16-bit), the upper bits of
580 	 * EA page must be obtained from PTEH's AVA.
581 	 */
582 	if (kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) <
583 	    0x10000U) {
584 		/*
585 		 * Add 0 to 5 EA bits, right after VSID.
586 		 * b == 12: 5 bits
587 		 * b == 24: 4 bits
588 		 */
589 		shift = AVA_PAGE_SHIFT(b);
590 		ea |= (PTEH_AVA_PAGE(pte->pte_hi) >> shift) <<
591 		    (SLBE_ESID_SHIFT - 5 + shift);
592 	}
593 
594 	/* Get VA page from hash and add to EA. */
595 	hash = (ptex & ~7) >> 3;
596 	if (pte->pte_hi & LPTEH_HID)
597 		hash = ~hash & PTE_HASH_MASK;
598 	ea |= ((hash ^ (vsid & PTE_HASH_MASK)) << b) & ~SLBE_ESID_MASK;
599 	return (ea);
600 }
601 
602 static int
603 ppc64mmu_hpt_walk_pages(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg)
604 {
605 	struct vmstate *vm;
606 	int ret;
607 	unsigned int pagesz;
608 	u_long dva, pa, va;
609 	u_long ptex, nptes;
610 	uint64_t vsid;
611 
612 	ret = 0;
613 	vm = kd->vmst;
614 	nptes = vm->hdr.pmapsize / sizeof(ppc64_pt_entry_t);
615 
616 	/* Walk through PTEs */
617 	for (ptex = 0; ptex < nptes; ptex++) {
618 		ppc64_pt_entry_t pte = pte_get(kd, ptex);
619 		if ((pte.pte_hi & LPTEH_VALID) == 0)
620 			continue;
621 
622 		/* Skip non-kernel related pages, as well as VRMA ones */
623 		vsid = PTEH_AVA_VSID(pte.pte_hi);
624 		if ((vsid & KERNEL_VSID_BIT) == 0 ||
625 		    (vsid >> PPC64_PAGE_SHIFT) == VSID_VRMA)
626 			continue;
627 
628 		/* Retrieve page's VA (EA on PPC64 terminology) */
629 		if ((va = get_ea(kd, &pte, ptex)) == ~0UL)
630 			goto out;
631 
632 		/* Get PA and page size */
633 		if (pte.pte_hi & LPTEH_BIG) {
634 			pa = pte.pte_lo & PTEL_LP_PA_MASK;
635 			pagesz = LP_PAGE_SIZE;
636 		} else {
637 			pa = pte.pte_lo & PTEL_PA_MASK;
638 			pagesz = PPC64_PAGE_SIZE;
639 		}
640 
641 		/* Get DMAP address */
642 		dva = vm->hdr.dmapbase + pa;
643 
644 		if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva,
645 		    entry_to_prot(&pte), pagesz, 0))
646 			goto out;
647 	}
648 	ret = 1;
649 
650 out:
651 	return (ret);
652 }
653 
654 
655 static struct ppc64_mmu_ops ops = {
656 	.init		= ppc64mmu_hpt_init,
657 	.cleanup	= ppc64mmu_hpt_cleanup,
658 	.kvatop		= ppc64mmu_hpt_kvatop,
659 	.walk_pages	= ppc64mmu_hpt_walk_pages,
660 };
661 struct ppc64_mmu_ops *ppc64_mmu_ops_hpt = &ops;
662