minidump_machdep.c revision 330897 
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2006 Peter Wemm
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/11/sys/i386/i386/minidump_machdep.c 330897 2018-03-14 03:19:51Z eadler $");
31
32#include "opt_watchdog.h"
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/conf.h>
37#include <sys/cons.h>
38#include <sys/kernel.h>
39#include <sys/kerneldump.h>
40#include <sys/msgbuf.h>
41#include <sys/watchdog.h>
42#include <vm/vm.h>
43#include <vm/pmap.h>
44#include <machine/atomic.h>
45#include <machine/elf.h>
46#include <machine/md_var.h>
47#include <machine/vmparam.h>
48#include <machine/minidump.h>
49
50CTASSERT(sizeof(struct kerneldumpheader) == 512);
51
52/*
53 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
54 * is to protect us from metadata and to protect metadata from us.
55 */
56#define	SIZEOF_METADATA		(64*1024)
57
58#define	MD_ALIGN(x)	(((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
59#define	DEV_ALIGN(x)	roundup2((off_t)(x), DEV_BSIZE)
60
61uint32_t *vm_page_dump;
62int vm_page_dump_size;
63
64static struct kerneldumpheader kdh;
65static off_t dumplo;
66
67/* Handle chunked writes. */
68static size_t fragsz;
69static void *dump_va;
70static uint64_t counter, progress;
71
72CTASSERT(sizeof(*vm_page_dump) == 4);
73
74
75static int
76is_dumpable(vm_paddr_t pa)
77{
78	int i;
79
80	for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
81		if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
82			return (1);
83	}
84	return (0);
85}
86
87#define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
88
89static int
90blk_flush(struct dumperinfo *di)
91{
92	int error;
93
94	if (fragsz == 0)
95		return (0);
96
97	error = dump_write(di, dump_va, 0, dumplo, fragsz);
98	dumplo += fragsz;
99	fragsz = 0;
100	return (error);
101}
102
103static int
104blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
105{
106	size_t len;
107	int error, i, c;
108	u_int maxdumpsz;
109
110	maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE);
111	if (maxdumpsz == 0)	/* seatbelt */
112		maxdumpsz = PAGE_SIZE;
113	error = 0;
114	if ((sz % PAGE_SIZE) != 0) {
115		printf("size not page aligned\n");
116		return (EINVAL);
117	}
118	if (ptr != NULL && pa != 0) {
119		printf("cant have both va and pa!\n");
120		return (EINVAL);
121	}
122	if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) {
123		printf("address not page aligned\n");
124		return (EINVAL);
125	}
126	if (ptr != NULL) {
127		/* If we're doing a virtual dump, flush any pre-existing pa pages */
128		error = blk_flush(di);
129		if (error)
130			return (error);
131	}
132	while (sz) {
133		len = maxdumpsz - fragsz;
134		if (len > sz)
135			len = sz;
136		counter += len;
137		progress -= len;
138		if (counter >> 24) {
139			printf(" %lld", PG2MB(progress >> PAGE_SHIFT));
140			counter &= (1<<24) - 1;
141		}
142
143		wdog_kern_pat(WD_LASTVAL);
144
145		if (ptr) {
146			error = dump_write(di, ptr, 0, dumplo, len);
147			if (error)
148				return (error);
149			dumplo += len;
150			ptr += len;
151			sz -= len;
152		} else {
153			for (i = 0; i < len; i += PAGE_SIZE)
154				dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
155			fragsz += len;
156			pa += len;
157			sz -= len;
158			if (fragsz == maxdumpsz) {
159				error = blk_flush(di);
160				if (error)
161					return (error);
162			}
163		}
164
165		/* Check for user abort. */
166		c = cncheckc();
167		if (c == 0x03)
168			return (ECANCELED);
169		if (c != -1)
170			printf(" (CTRL-C to abort) ");
171	}
172
173	return (0);
174}
175
176/* A fake page table page, to avoid having to handle both 4K and 2M pages */
177static pt_entry_t fakept[NPTEPG];
178
179int
180minidumpsys(struct dumperinfo *di)
181{
182	uint64_t dumpsize;
183	uint32_t ptesize;
184	vm_offset_t va;
185	int error;
186	uint32_t bits;
187	uint64_t pa;
188	pd_entry_t *pd;
189	pt_entry_t *pt;
190	int i, j, k, bit;
191	struct minidumphdr mdhdr;
192
193	counter = 0;
194	/* Walk page table pages, set bits in vm_page_dump */
195	ptesize = 0;
196	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
197		/*
198		 * We always write a page, even if it is zero. Each
199		 * page written corresponds to 2MB of space
200		 */
201		ptesize += PAGE_SIZE;
202		pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE);	/* always mapped! */
203		j = va >> PDRSHIFT;
204		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
205			/* This is an entire 2M page. */
206			pa = pd[j] & PG_PS_FRAME;
207			for (k = 0; k < NPTEPG; k++) {
208				if (is_dumpable(pa))
209					dump_add_page(pa);
210				pa += PAGE_SIZE;
211			}
212			continue;
213		}
214		if ((pd[j] & PG_V) == PG_V) {
215			/* set bit for each valid page in this 2MB block */
216			pt = pmap_kenter_temporary(pd[j] & PG_FRAME, 0);
217			for (k = 0; k < NPTEPG; k++) {
218				if ((pt[k] & PG_V) == PG_V) {
219					pa = pt[k] & PG_FRAME;
220					if (is_dumpable(pa))
221						dump_add_page(pa);
222				}
223			}
224		} else {
225			/* nothing, we're going to dump a null page */
226		}
227	}
228
229	/* Calculate dump size. */
230	dumpsize = ptesize;
231	dumpsize += round_page(msgbufp->msg_size);
232	dumpsize += round_page(vm_page_dump_size);
233	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
234		bits = vm_page_dump[i];
235		while (bits) {
236			bit = bsfl(bits);
237			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
238			/* Clear out undumpable pages now if needed */
239			if (is_dumpable(pa)) {
240				dumpsize += PAGE_SIZE;
241			} else {
242				dump_drop_page(pa);
243			}
244			bits &= ~(1ul << bit);
245		}
246	}
247	dumpsize += PAGE_SIZE;
248
249	/* Determine dump offset on device. */
250	if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
251		error = ENOSPC;
252		goto fail;
253	}
254	dumplo = di->mediaoffset + di->mediasize - dumpsize;
255	dumplo -= sizeof(kdh) * 2;
256	progress = dumpsize;
257
258	/* Initialize mdhdr */
259	bzero(&mdhdr, sizeof(mdhdr));
260	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
261	mdhdr.version = MINIDUMP_VERSION;
262	mdhdr.msgbufsize = msgbufp->msg_size;
263	mdhdr.bitmapsize = vm_page_dump_size;
264	mdhdr.ptesize = ptesize;
265	mdhdr.kernbase = KERNBASE;
266#if defined(PAE) || defined(PAE_TABLES)
267	mdhdr.paemode = 1;
268#endif
269
270	mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_I386_VERSION, dumpsize, di->blocksize);
271
272	printf("Physical memory: %ju MB\n", ptoa((uintmax_t)physmem) / 1048576);
273	printf("Dumping %llu MB:", (long long)dumpsize >> 20);
274
275	/* Dump leader */
276	error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
277	if (error)
278		goto fail;
279	dumplo += sizeof(kdh);
280
281	/* Dump my header */
282	bzero(&fakept, sizeof(fakept));
283	bcopy(&mdhdr, &fakept, sizeof(mdhdr));
284	error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
285	if (error)
286		goto fail;
287
288	/* Dump msgbuf up front */
289	error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
290	if (error)
291		goto fail;
292
293	/* Dump bitmap */
294	error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
295	if (error)
296		goto fail;
297
298	/* Dump kernel page table pages */
299	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
300		/* We always write a page, even if it is zero */
301		pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE);	/* always mapped! */
302		j = va >> PDRSHIFT;
303		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
304			/* This is a single 2M block. Generate a fake PTP */
305			pa = pd[j] & PG_PS_FRAME;
306			for (k = 0; k < NPTEPG; k++) {
307				fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
308			}
309			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
310			if (error)
311				goto fail;
312			/* flush, in case we reuse fakept in the same block */
313			error = blk_flush(di);
314			if (error)
315				goto fail;
316			continue;
317		}
318		if ((pd[j] & PG_V) == PG_V) {
319			pa = pd[j] & PG_FRAME;
320			error = blk_write(di, 0, pa, PAGE_SIZE);
321			if (error)
322				goto fail;
323		} else {
324			bzero(fakept, sizeof(fakept));
325			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
326			if (error)
327				goto fail;
328			/* flush, in case we reuse fakept in the same block */
329			error = blk_flush(di);
330			if (error)
331				goto fail;
332		}
333	}
334
335	/* Dump memory chunks */
336	/* XXX cluster it up and use blk_dump() */
337	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
338		bits = vm_page_dump[i];
339		while (bits) {
340			bit = bsfl(bits);
341			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
342			error = blk_write(di, 0, pa, PAGE_SIZE);
343			if (error)
344				goto fail;
345			bits &= ~(1ul << bit);
346		}
347	}
348
349	error = blk_flush(di);
350	if (error)
351		goto fail;
352
353	/* Dump trailer */
354	error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
355	if (error)
356		goto fail;
357	dumplo += sizeof(kdh);
358
359	/* Signal completion, signoff and exit stage left. */
360	dump_write(di, NULL, 0, 0, 0);
361	printf("\nDump complete\n");
362	return (0);
363
364 fail:
365	if (error < 0)
366		error = -error;
367
368	if (error == ECANCELED)
369		printf("\nDump aborted\n");
370	else if (error == ENOSPC)
371		printf("\nDump failed. Partition too small.\n");
372	else
373		printf("\n** DUMP FAILED (ERROR %d) **\n", error);
374	return (error);
375}
376
377void
378dump_add_page(vm_paddr_t pa)
379{
380	int idx, bit;
381
382	pa >>= PAGE_SHIFT;
383	idx = pa >> 5;		/* 2^5 = 32 */
384	bit = pa & 31;
385	atomic_set_int(&vm_page_dump[idx], 1ul << bit);
386}
387
388void
389dump_drop_page(vm_paddr_t pa)
390{
391	int idx, bit;
392
393	pa >>= PAGE_SHIFT;
394	idx = pa >> 5;		/* 2^5 = 32 */
395	bit = pa & 31;
396	atomic_clear_int(&vm_page_dump[idx], 1ul << bit);
397}
398
399