xref: /freebsd/lib/libkvm/kvm_minidump_amd64.c (revision d4eeb02986980bf33dd56c41ceb9fc5f180c0d47)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2006 Peter Wemm
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 /*
32  * AMD64 machine dependent routines for kvm and minidumps.
33  */
34 
35 #include <sys/param.h>
36 #include <sys/endian.h>
37 #include <stdint.h>
38 #include <stdlib.h>
39 #include <string.h>
40 #include <unistd.h>
41 #include <vm/vm.h>
42 #include <kvm.h>
43 
44 #include "../../sys/amd64/include/minidump.h"
45 
46 #include <limits.h>
47 
48 #include "kvm_private.h"
49 #include "kvm_amd64.h"
50 
51 #define	amd64_round_page(x)	roundup2((kvaddr_t)(x), AMD64_PAGE_SIZE)
52 #define	VM_IS_V1(vm)		(vm->hdr.version == 1)
53 #define	VA_OFF(vm, va)		\
54 	(VM_IS_V1(vm) ? ((va) & (AMD64_PAGE_SIZE - 1)) : ((va) & AMD64_PAGE_MASK))
55 
56 struct vmstate {
57 	struct minidumphdr hdr;
58 };
59 
60 static vm_prot_t
61 _amd64_entry_to_prot(uint64_t entry)
62 {
63 	vm_prot_t prot = VM_PROT_READ;
64 
65 	if ((entry & AMD64_PG_RW) != 0)
66 		prot |= VM_PROT_WRITE;
67 	if ((entry & AMD64_PG_NX) == 0)
68 		prot |= VM_PROT_EXECUTE;
69 	return prot;
70 }
71 
72 /*
73  * Version 2 minidumps use page directory entries, while version 1 use page
74  * table entries.
75  */
76 
77 static amd64_pde_t
78 _amd64_pde_get(kvm_t *kd, u_long pdeindex)
79 {
80 	amd64_pde_t *pde = _kvm_pmap_get(kd, pdeindex, sizeof(*pde));
81 
82 	return le64toh(*pde);
83 }
84 
85 static amd64_pte_t
86 _amd64_pte_get(kvm_t *kd, u_long pteindex)
87 {
88 	amd64_pte_t *pte = _kvm_pmap_get(kd, pteindex, sizeof(*pte));
89 
90 	return le64toh(*pte);
91 }
92 
93 /* Get the first page table entry for a given page directory index. */
94 static amd64_pte_t *
95 _amd64_pde_first_pte(kvm_t *kd, u_long pdeindex)
96 {
97 	u_long *pa;
98 
99 	pa = _kvm_pmap_get(kd, pdeindex, sizeof(amd64_pde_t));
100 	if (pa == NULL)
101 		return NULL;
102 	return _kvm_map_get(kd, *pa & AMD64_PG_FRAME, AMD64_PAGE_SIZE);
103 }
104 
105 static int
106 _amd64_minidump_probe(kvm_t *kd)
107 {
108 
109 	return (_kvm_probe_elf_kernel(kd, ELFCLASS64, EM_X86_64) &&
110 	    _kvm_is_minidump(kd));
111 }
112 
113 static void
114 _amd64_minidump_freevtop(kvm_t *kd)
115 {
116 	struct vmstate *vm = kd->vmst;
117 
118 	free(vm);
119 	kd->vmst = NULL;
120 }
121 
122 static int
123 _amd64_minidump_initvtop(kvm_t *kd)
124 {
125 	struct vmstate *vmst;
126 	off_t off, dump_avail_off, sparse_off;
127 
128 	vmst = _kvm_malloc(kd, sizeof(*vmst));
129 	if (vmst == NULL) {
130 		_kvm_err(kd, kd->program, "cannot allocate vm");
131 		return (-1);
132 	}
133 	kd->vmst = vmst;
134 	if (pread(kd->pmfd, &vmst->hdr, sizeof(vmst->hdr), 0) !=
135 	    sizeof(vmst->hdr)) {
136 		_kvm_err(kd, kd->program, "cannot read dump header");
137 		return (-1);
138 	}
139 	if (strncmp(MINIDUMP_MAGIC, vmst->hdr.magic, sizeof(vmst->hdr.magic)) != 0) {
140 		_kvm_err(kd, kd->program, "not a minidump for this platform");
141 		return (-1);
142 	}
143 
144 	/*
145 	 * NB: amd64 minidump header is binary compatible between version 1
146 	 * and version 2; version 3 adds the dumpavailsize field
147 	 */
148 	vmst->hdr.version = le32toh(vmst->hdr.version);
149 	if (vmst->hdr.version > MINIDUMP_VERSION || vmst->hdr.version < 1) {
150 		_kvm_err(kd, kd->program, "wrong minidump version. expected %d got %d",
151 		    MINIDUMP_VERSION, vmst->hdr.version);
152 		return (-1);
153 	}
154 	vmst->hdr.msgbufsize = le32toh(vmst->hdr.msgbufsize);
155 	vmst->hdr.bitmapsize = le32toh(vmst->hdr.bitmapsize);
156 	vmst->hdr.pmapsize = le32toh(vmst->hdr.pmapsize);
157 	vmst->hdr.kernbase = le64toh(vmst->hdr.kernbase);
158 	vmst->hdr.dmapbase = le64toh(vmst->hdr.dmapbase);
159 	vmst->hdr.dmapend = le64toh(vmst->hdr.dmapend);
160 	vmst->hdr.dumpavailsize = vmst->hdr.version == MINIDUMP_VERSION ?
161 	    le32toh(vmst->hdr.dumpavailsize) : 0;
162 
163 	/* Skip header and msgbuf */
164 	dump_avail_off = AMD64_PAGE_SIZE + amd64_round_page(vmst->hdr.msgbufsize);
165 
166 	/* Skip dump_avail */
167 	off = dump_avail_off + amd64_round_page(vmst->hdr.dumpavailsize);
168 
169 	sparse_off = off + amd64_round_page(vmst->hdr.bitmapsize) +
170 	    amd64_round_page(vmst->hdr.pmapsize);
171 	if (_kvm_pt_init(kd, vmst->hdr.dumpavailsize, dump_avail_off,
172 	    vmst->hdr.bitmapsize, off, sparse_off, AMD64_PAGE_SIZE) == -1) {
173 		return (-1);
174 	}
175 	off += amd64_round_page(vmst->hdr.bitmapsize);
176 
177 	if (_kvm_pmap_init(kd, vmst->hdr.pmapsize, off) == -1) {
178 		return (-1);
179 	}
180 	off += amd64_round_page(vmst->hdr.pmapsize);
181 
182 	return (0);
183 }
184 
185 static int
186 _amd64_minidump_vatop_v1(kvm_t *kd, kvaddr_t va, off_t *pa)
187 {
188 	struct vmstate *vm;
189 	amd64_physaddr_t offset;
190 	amd64_pte_t pte;
191 	kvaddr_t pteindex;
192 	amd64_physaddr_t a;
193 	off_t ofs;
194 
195 	vm = kd->vmst;
196 	offset = va & AMD64_PAGE_MASK;
197 
198 	if (va >= vm->hdr.kernbase) {
199 		pteindex = (va - vm->hdr.kernbase) >> AMD64_PAGE_SHIFT;
200 		if (pteindex >= vm->hdr.pmapsize / sizeof(pte))
201 			goto invalid;
202 		pte = _amd64_pte_get(kd, pteindex);
203 		if ((pte & AMD64_PG_V) == 0) {
204 			_kvm_err(kd, kd->program,
205 			    "_amd64_minidump_vatop_v1: pte not valid");
206 			goto invalid;
207 		}
208 		a = pte & AMD64_PG_FRAME;
209 		ofs = _kvm_pt_find(kd, a, AMD64_PAGE_SIZE);
210 		if (ofs == -1) {
211 			_kvm_err(kd, kd->program,
212 	    "_amd64_minidump_vatop_v1: physical address 0x%jx not in minidump",
213 			    (uintmax_t)a);
214 			goto invalid;
215 		}
216 		*pa = ofs + offset;
217 		return (AMD64_PAGE_SIZE - offset);
218 	} else if (va >= vm->hdr.dmapbase && va < vm->hdr.dmapend) {
219 		a = (va - vm->hdr.dmapbase) & ~AMD64_PAGE_MASK;
220 		ofs = _kvm_pt_find(kd, a, AMD64_PAGE_SIZE);
221 		if (ofs == -1) {
222 			_kvm_err(kd, kd->program,
223     "_amd64_minidump_vatop_v1: direct map address 0x%jx not in minidump",
224 			    (uintmax_t)va);
225 			goto invalid;
226 		}
227 		*pa = ofs + offset;
228 		return (AMD64_PAGE_SIZE - offset);
229 	} else {
230 		_kvm_err(kd, kd->program,
231 	    "_amd64_minidump_vatop_v1: virtual address 0x%jx not minidumped",
232 		    (uintmax_t)va);
233 		goto invalid;
234 	}
235 
236 invalid:
237 	_kvm_err(kd, 0, "invalid address (0x%jx)", (uintmax_t)va);
238 	return (0);
239 }
240 
241 static int
242 _amd64_minidump_vatop(kvm_t *kd, kvaddr_t va, off_t *pa)
243 {
244 	amd64_pte_t pt[AMD64_NPTEPG];
245 	struct vmstate *vm;
246 	amd64_physaddr_t offset;
247 	amd64_pde_t pde;
248 	amd64_pte_t pte;
249 	kvaddr_t pteindex;
250 	kvaddr_t pdeindex;
251 	amd64_physaddr_t a;
252 	off_t ofs;
253 
254 	vm = kd->vmst;
255 	offset = va & AMD64_PAGE_MASK;
256 
257 	if (va >= vm->hdr.kernbase) {
258 		pdeindex = (va - vm->hdr.kernbase) >> AMD64_PDRSHIFT;
259 		if (pdeindex >= vm->hdr.pmapsize / sizeof(pde))
260 			goto invalid;
261 		pde = _amd64_pde_get(kd, pdeindex);
262 		if ((pde & AMD64_PG_V) == 0) {
263 			_kvm_err(kd, kd->program,
264 			    "_amd64_minidump_vatop: pde not valid");
265 			goto invalid;
266 		}
267 		if ((pde & AMD64_PG_PS) == 0) {
268 			a = pde & AMD64_PG_FRAME;
269 			/* TODO: Just read the single PTE */
270 			ofs = _kvm_pt_find(kd, a, AMD64_PAGE_SIZE);
271 			if (ofs == -1) {
272 				_kvm_err(kd, kd->program,
273 				    "cannot find page table entry for %ju",
274 				    (uintmax_t)a);
275 				goto invalid;
276 			}
277 			if (pread(kd->pmfd, &pt, AMD64_PAGE_SIZE, ofs) !=
278 			    AMD64_PAGE_SIZE) {
279 				_kvm_err(kd, kd->program,
280 				    "cannot read page table entry for %ju",
281 				    (uintmax_t)a);
282 				goto invalid;
283 			}
284 			pteindex = (va >> AMD64_PAGE_SHIFT) &
285 			    (AMD64_NPTEPG - 1);
286 			pte = le64toh(pt[pteindex]);
287 			if ((pte & AMD64_PG_V) == 0) {
288 				_kvm_err(kd, kd->program,
289 				    "_amd64_minidump_vatop: pte not valid");
290 				goto invalid;
291 			}
292 			a = pte & AMD64_PG_FRAME;
293 		} else {
294 			a = pde & AMD64_PG_PS_FRAME;
295 			a += (va & AMD64_PDRMASK) ^ offset;
296 		}
297 		ofs = _kvm_pt_find(kd, a, AMD64_PAGE_SIZE);
298 		if (ofs == -1) {
299 			_kvm_err(kd, kd->program,
300 	    "_amd64_minidump_vatop: physical address 0x%jx not in minidump",
301 			    (uintmax_t)a);
302 			goto invalid;
303 		}
304 		*pa = ofs + offset;
305 		return (AMD64_PAGE_SIZE - offset);
306 	} else if (va >= vm->hdr.dmapbase && va < vm->hdr.dmapend) {
307 		a = (va - vm->hdr.dmapbase) & ~AMD64_PAGE_MASK;
308 		ofs = _kvm_pt_find(kd, a, AMD64_PAGE_SIZE);
309 		if (ofs == -1) {
310 			_kvm_err(kd, kd->program,
311 	    "_amd64_minidump_vatop: direct map address 0x%jx not in minidump",
312 			    (uintmax_t)va);
313 			goto invalid;
314 		}
315 		*pa = ofs + offset;
316 		return (AMD64_PAGE_SIZE - offset);
317 	} else {
318 		_kvm_err(kd, kd->program,
319 	    "_amd64_minidump_vatop: virtual address 0x%jx not minidumped",
320 		    (uintmax_t)va);
321 		goto invalid;
322 	}
323 
324 invalid:
325 	_kvm_err(kd, 0, "invalid address (0x%jx)", (uintmax_t)va);
326 	return (0);
327 }
328 
329 static int
330 _amd64_minidump_kvatop(kvm_t *kd, kvaddr_t va, off_t *pa)
331 {
332 
333 	if (ISALIVE(kd)) {
334 		_kvm_err(kd, 0,
335 		    "_amd64_minidump_kvatop called in live kernel!");
336 		return (0);
337 	}
338 	if (((struct vmstate *)kd->vmst)->hdr.version == 1)
339 		return (_amd64_minidump_vatop_v1(kd, va, pa));
340 	else
341 		return (_amd64_minidump_vatop(kd, va, pa));
342 }
343 
344 static int
345 _amd64_minidump_walk_pages(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg)
346 {
347 	struct vmstate *vm = kd->vmst;
348 	u_long npdes = vm->hdr.pmapsize / sizeof(amd64_pde_t);
349 	u_long bmindex, dva, pa, pdeindex, va;
350 	struct kvm_bitmap bm;
351 	int ret = 0;
352 	vm_prot_t prot;
353 	unsigned int pgsz = AMD64_PAGE_SIZE;
354 
355 	if (vm->hdr.version < 2)
356 		return (0);
357 
358 	if (!_kvm_bitmap_init(&bm, vm->hdr.bitmapsize, &bmindex))
359 		return (0);
360 
361 	for (pdeindex = 0; pdeindex < npdes; pdeindex++) {
362 		amd64_pde_t pde = _amd64_pde_get(kd, pdeindex);
363 		amd64_pte_t *ptes;
364 		u_long i;
365 
366 		va = vm->hdr.kernbase + (pdeindex << AMD64_PDRSHIFT);
367 		if ((pde & AMD64_PG_V) == 0)
368 			continue;
369 
370 		if ((pde & AMD64_PG_PS) != 0) {
371 			/*
372 			 * Large page.  Iterate on each 4K page section
373 			 * within this page.  This differs from 4K pages in
374 			 * that every page here uses the same PDE to
375 			 * generate permissions.
376 			 */
377 			pa = (pde & AMD64_PG_PS_FRAME) +
378 			    ((va & AMD64_PDRMASK) ^ VA_OFF(vm, va));
379 			dva = vm->hdr.dmapbase + pa;
380 			_kvm_bitmap_set(&bm, _kvm_pa_bit_id(kd, pa, AMD64_PAGE_SIZE));
381 			if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva,
382 			    _amd64_entry_to_prot(pde), AMD64_NBPDR, pgsz)) {
383 				goto out;
384 			}
385 			continue;
386 		}
387 
388 		/* 4K pages: pde references another page of entries. */
389 		ptes = _amd64_pde_first_pte(kd, pdeindex);
390 		/* Ignore page directory pages that were not dumped. */
391 		if (ptes == NULL)
392 			continue;
393 
394 		for (i = 0; i < AMD64_NPTEPG; i++) {
395 			amd64_pte_t pte = (u_long)ptes[i];
396 
397 			pa = pte & AMD64_PG_FRAME;
398 			dva = vm->hdr.dmapbase + pa;
399 			if ((pte & AMD64_PG_V) != 0) {
400 				_kvm_bitmap_set(&bm,
401 				    _kvm_pa_bit_id(kd, pa, AMD64_PAGE_SIZE));
402 				if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva,
403 				    _amd64_entry_to_prot(pte), pgsz, 0)) {
404 					goto out;
405 				}
406 			}
407 			va += AMD64_PAGE_SIZE;
408 		}
409 	}
410 
411 	while (_kvm_bitmap_next(&bm, &bmindex)) {
412 		pa = _kvm_bit_id_pa(kd, bmindex, AMD64_PAGE_SIZE);
413 		if (pa == _KVM_PA_INVALID)
414 			break;
415 		dva = vm->hdr.dmapbase + pa;
416 		if (vm->hdr.dmapend < (dva + pgsz))
417 			break;
418 		va = 0;
419 		/* amd64/pmap.c: create_pagetables(): dmap always R|W. */
420 		prot = VM_PROT_READ | VM_PROT_WRITE;
421 		if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva, prot, pgsz, 0)) {
422 			goto out;
423 		}
424 	}
425 
426 	ret = 1;
427 
428 out:
429 	_kvm_bitmap_deinit(&bm);
430 	return (ret);
431 }
432 
433 static struct kvm_arch kvm_amd64_minidump = {
434 	.ka_probe = _amd64_minidump_probe,
435 	.ka_initvtop = _amd64_minidump_initvtop,
436 	.ka_freevtop = _amd64_minidump_freevtop,
437 	.ka_kvatop = _amd64_minidump_kvatop,
438 	.ka_native = _amd64_native,
439 	.ka_walk_pages = _amd64_minidump_walk_pages,
440 };
441 
442 KVM_ARCH(kvm_amd64_minidump);
443