minidump_machdep.c revision 243132 
1/*-
2 * Copyright (c) 2006 Peter Wemm
3 * All rights reserved.
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 *
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 ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/amd64/amd64/minidump_machdep.c 243132 2012-11-16 05:55:56Z kib $");
29
30#include "opt_pmap.h"
31#include "opt_watchdog.h"
32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/conf.h>
36#include <sys/cons.h>
37#include <sys/kernel.h>
38#include <sys/kerneldump.h>
39#include <sys/msgbuf.h>
40#include <sys/watchdog.h>
41#include <vm/vm.h>
42#include <vm/vm_page.h>
43#include <vm/vm_phys.h>
44#include <vm/pmap.h>
45#include <machine/atomic.h>
46#include <machine/elf.h>
47#include <machine/md_var.h>
48#include <machine/vmparam.h>
49#include <machine/minidump.h>
50
51CTASSERT(sizeof(struct kerneldumpheader) == 512);
52
53/*
54 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
55 * is to protect us from metadata and to protect metadata from us.
56 */
57#define	SIZEOF_METADATA		(64*1024)
58
59#define	MD_ALIGN(x)	(((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
60#define	DEV_ALIGN(x)	(((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
61
62uint64_t *vm_page_dump;
63int vm_page_dump_size;
64
65static struct kerneldumpheader kdh;
66static off_t dumplo;
67
68/* Handle chunked writes. */
69static size_t fragsz;
70static void *dump_va;
71static size_t counter, progress, dumpsize;
72
73CTASSERT(sizeof(*vm_page_dump) == 8);
74
75static int
76is_dumpable(vm_paddr_t pa)
77{
78	vm_page_t m;
79	int i;
80
81	if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
82		return ((m->flags & PG_NODUMP) == 0);
83	for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
84		if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
85			return (1);
86	}
87	return (0);
88}
89
90#define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
91
92static int
93blk_flush(struct dumperinfo *di)
94{
95	int error;
96
97	if (fragsz == 0)
98		return (0);
99
100	error = dump_write(di, dump_va, 0, dumplo, fragsz);
101	dumplo += fragsz;
102	fragsz = 0;
103	return (error);
104}
105
106static struct {
107	int min_per;
108	int max_per;
109	int visited;
110} progress_track[10] = {
111	{  0,  10, 0},
112	{ 10,  20, 0},
113	{ 20,  30, 0},
114	{ 30,  40, 0},
115	{ 40,  50, 0},
116	{ 50,  60, 0},
117	{ 60,  70, 0},
118	{ 70,  80, 0},
119	{ 80,  90, 0},
120	{ 90, 100, 0}
121};
122
123static void
124report_progress(size_t progress, size_t dumpsize)
125{
126	int sofar, i;
127
128	sofar = 100 - ((progress * 100) / dumpsize);
129	for (i = 0; i < 10; i++) {
130		if (sofar < progress_track[i].min_per || sofar > progress_track[i].max_per)
131			continue;
132		if (progress_track[i].visited)
133			return;
134		progress_track[i].visited = 1;
135		printf("..%d%%", sofar);
136		return;
137	}
138}
139
140static int
141blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
142{
143	size_t len;
144	int error, i, c;
145	u_int maxdumpsz;
146
147	maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE);
148	if (maxdumpsz == 0)	/* seatbelt */
149		maxdumpsz = PAGE_SIZE;
150	error = 0;
151	if ((sz % PAGE_SIZE) != 0) {
152		printf("size not page aligned\n");
153		return (EINVAL);
154	}
155	if (ptr != NULL && pa != 0) {
156		printf("cant have both va and pa!\n");
157		return (EINVAL);
158	}
159	if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) {
160		printf("address not page aligned\n");
161		return (EINVAL);
162	}
163	if (ptr != NULL) {
164		/* If we're doing a virtual dump, flush any pre-existing pa pages */
165		error = blk_flush(di);
166		if (error)
167			return (error);
168	}
169	while (sz) {
170		len = maxdumpsz - fragsz;
171		if (len > sz)
172			len = sz;
173		counter += len;
174		progress -= len;
175		if (counter >> 24) {
176			report_progress(progress, dumpsize);
177			counter &= (1<<24) - 1;
178		}
179
180		wdog_kern_pat(WD_LASTVAL);
181
182		if (ptr) {
183			error = dump_write(di, ptr, 0, dumplo, len);
184			if (error)
185				return (error);
186			dumplo += len;
187			ptr += len;
188			sz -= len;
189		} else {
190			for (i = 0; i < len; i += PAGE_SIZE)
191				dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
192			fragsz += len;
193			pa += len;
194			sz -= len;
195			if (fragsz == maxdumpsz) {
196				error = blk_flush(di);
197				if (error)
198					return (error);
199			}
200		}
201
202		/* Check for user abort. */
203		c = cncheckc();
204		if (c == 0x03)
205			return (ECANCELED);
206		if (c != -1)
207			printf(" (CTRL-C to abort) ");
208	}
209
210	return (0);
211}
212
213/* A fake page table page, to avoid having to handle both 4K and 2M pages */
214static pd_entry_t fakepd[NPDEPG];
215
216void
217minidumpsys(struct dumperinfo *di)
218{
219	uint32_t pmapsize;
220	vm_offset_t va;
221	int error;
222	uint64_t bits;
223	uint64_t *pdp, *pd, *pt, pa;
224	int i, j, k, n, bit;
225	int retry_count;
226	struct minidumphdr mdhdr;
227
228	retry_count = 0;
229 retry:
230	retry_count++;
231	counter = 0;
232	/* Walk page table pages, set bits in vm_page_dump */
233	pmapsize = 0;
234	pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
235	for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + NKPT * NBPDR,
236	    kernel_vm_end); ) {
237		/*
238		 * We always write a page, even if it is zero. Each
239		 * page written corresponds to 1GB of space
240		 */
241		pmapsize += PAGE_SIZE;
242		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
243		if ((pdp[i] & PG_V) == 0) {
244			va += NBPDP;
245			continue;
246		}
247
248		/*
249		 * 1GB page is represented as 512 2MB pages in a dump.
250		 */
251		if ((pdp[i] & PG_PS) != 0) {
252			va += NBPDP;
253			pa = pdp[i] & PG_PS_FRAME;
254			for (n = 0; n < NPDEPG * NPTEPG; n++) {
255				if (is_dumpable(pa))
256					dump_add_page(pa);
257				pa += PAGE_SIZE;
258			}
259			continue;
260		}
261
262		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
263		for (n = 0; n < NPDEPG; n++, va += NBPDR) {
264			j = (va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1);
265
266			if ((pd[j] & PG_V) == 0)
267				continue;
268
269			if ((pd[j] & PG_PS) != 0) {
270				/* This is an entire 2M page. */
271				pa = pd[j] & PG_PS_FRAME;
272				for (k = 0; k < NPTEPG; k++) {
273					if (is_dumpable(pa))
274						dump_add_page(pa);
275					pa += PAGE_SIZE;
276				}
277				continue;
278			}
279
280			pa = pd[j] & PG_FRAME;
281			/* set bit for this PTE page */
282			if (is_dumpable(pa))
283				dump_add_page(pa);
284			/* and for each valid page in this 2MB block */
285			pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
286			for (k = 0; k < NPTEPG; k++) {
287				if ((pt[k] & PG_V) == 0)
288					continue;
289				pa = pt[k] & PG_FRAME;
290				if (is_dumpable(pa))
291					dump_add_page(pa);
292			}
293		}
294	}
295
296	/* Calculate dump size. */
297	dumpsize = pmapsize;
298	dumpsize += round_page(msgbufp->msg_size);
299	dumpsize += round_page(vm_page_dump_size);
300	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
301		bits = vm_page_dump[i];
302		while (bits) {
303			bit = bsfq(bits);
304			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
305			/* Clear out undumpable pages now if needed */
306			if (is_dumpable(pa)) {
307				dumpsize += PAGE_SIZE;
308			} else {
309				dump_drop_page(pa);
310			}
311			bits &= ~(1ul << bit);
312		}
313	}
314	dumpsize += PAGE_SIZE;
315
316	/* Determine dump offset on device. */
317	if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
318		error = E2BIG;
319		goto fail;
320	}
321	dumplo = di->mediaoffset + di->mediasize - dumpsize;
322	dumplo -= sizeof(kdh) * 2;
323	progress = dumpsize;
324
325	/* Initialize mdhdr */
326	bzero(&mdhdr, sizeof(mdhdr));
327	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
328	mdhdr.version = MINIDUMP_VERSION;
329	mdhdr.msgbufsize = msgbufp->msg_size;
330	mdhdr.bitmapsize = vm_page_dump_size;
331	mdhdr.pmapsize = pmapsize;
332	mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
333	mdhdr.dmapbase = DMAP_MIN_ADDRESS;
334	mdhdr.dmapend = DMAP_MAX_ADDRESS;
335
336	mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, dumpsize, di->blocksize);
337
338	printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20,
339	    ptoa((uintmax_t)physmem) / 1048576);
340
341	/* Dump leader */
342	error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
343	if (error)
344		goto fail;
345	dumplo += sizeof(kdh);
346
347	/* Dump my header */
348	bzero(&fakepd, sizeof(fakepd));
349	bcopy(&mdhdr, &fakepd, sizeof(mdhdr));
350	error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
351	if (error)
352		goto fail;
353
354	/* Dump msgbuf up front */
355	error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
356	if (error)
357		goto fail;
358
359	/* Dump bitmap */
360	error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
361	if (error)
362		goto fail;
363
364	/* Dump kernel page directory pages */
365	bzero(fakepd, sizeof(fakepd));
366	pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
367	for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + NKPT * NBPDR,
368	    kernel_vm_end); va += NBPDP) {
369		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
370
371		/* We always write a page, even if it is zero */
372		if ((pdp[i] & PG_V) == 0) {
373			error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
374			if (error)
375				goto fail;
376			/* flush, in case we reuse fakepd in the same block */
377			error = blk_flush(di);
378			if (error)
379				goto fail;
380			continue;
381		}
382
383		/* 1GB page is represented as 512 2MB pages in a dump */
384		if ((pdp[i] & PG_PS) != 0) {
385			/* PDPE and PDP have identical layout in this case */
386			fakepd[0] = pdp[i];
387			for (j = 1; j < NPDEPG; j++)
388				fakepd[j] = fakepd[j - 1] + NBPDR;
389			error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
390			if (error)
391				goto fail;
392			/* flush, in case we reuse fakepd in the same block */
393			error = blk_flush(di);
394			if (error)
395				goto fail;
396			bzero(fakepd, sizeof(fakepd));
397			continue;
398		}
399
400		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
401		error = blk_write(di, (char *)pd, 0, PAGE_SIZE);
402		if (error)
403			goto fail;
404		error = blk_flush(di);
405		if (error)
406			goto fail;
407	}
408
409	/* Dump memory chunks */
410	/* XXX cluster it up and use blk_dump() */
411	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
412		bits = vm_page_dump[i];
413		while (bits) {
414			bit = bsfq(bits);
415			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
416			error = blk_write(di, 0, pa, PAGE_SIZE);
417			if (error)
418				goto fail;
419			bits &= ~(1ul << bit);
420		}
421	}
422
423	error = blk_flush(di);
424	if (error)
425		goto fail;
426
427	/* Dump trailer */
428	error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
429	if (error)
430		goto fail;
431	dumplo += sizeof(kdh);
432
433	/* Signal completion, signoff and exit stage left. */
434	dump_write(di, NULL, 0, 0, 0);
435	printf("\nDump complete\n");
436	return;
437
438 fail:
439	if (error < 0)
440		error = -error;
441
442	printf("\n");
443	if (error == ENOSPC) {
444		printf("Dump map grown while dumping. ");
445		if (retry_count < 5) {
446			printf("Retrying...\n");
447			goto retry;
448		}
449		printf("Dump failed.\n");
450	}
451	else if (error == ECANCELED)
452		printf("Dump aborted\n");
453	else if (error == E2BIG)
454		printf("Dump failed. Partition too small.\n");
455	else
456		printf("** DUMP FAILED (ERROR %d) **\n", error);
457}
458
459void
460dump_add_page(vm_paddr_t pa)
461{
462	int idx, bit;
463
464	pa >>= PAGE_SHIFT;
465	idx = pa >> 6;		/* 2^6 = 64 */
466	bit = pa & 63;
467	atomic_set_long(&vm_page_dump[idx], 1ul << bit);
468}
469
470void
471dump_drop_page(vm_paddr_t pa)
472{
473	int idx, bit;
474
475	pa >>= PAGE_SHIFT;
476	idx = pa >> 6;		/* 2^6 = 64 */
477	bit = pa & 63;
478	atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
479}
480