minidump_machdep.c revision 302408 
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: stable/11/sys/amd64/amd64/minidump_machdep.c 298076 2016-04-15 17:45:12Z cem $");
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_param.h>
43#include <vm/vm_page.h>
44#include <vm/vm_phys.h>
45#include <vm/pmap.h>
46#include <machine/atomic.h>
47#include <machine/elf.h>
48#include <machine/md_var.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
59uint64_t *vm_page_dump;
60int vm_page_dump_size;
61
62static struct kerneldumpheader kdh;
63static off_t dumplo;
64
65/* Handle chunked writes. */
66static size_t fragsz;
67static void *dump_va;
68static size_t counter, progress, dumpsize;
69
70CTASSERT(sizeof(*vm_page_dump) == 8);
71
72static int
73is_dumpable(vm_paddr_t pa)
74{
75	vm_page_t m;
76	int i;
77
78	if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
79		return ((m->flags & PG_NODUMP) == 0);
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 struct {
104	int min_per;
105	int max_per;
106	int visited;
107} progress_track[10] = {
108	{  0,  10, 0},
109	{ 10,  20, 0},
110	{ 20,  30, 0},
111	{ 30,  40, 0},
112	{ 40,  50, 0},
113	{ 50,  60, 0},
114	{ 60,  70, 0},
115	{ 70,  80, 0},
116	{ 80,  90, 0},
117	{ 90, 100, 0}
118};
119
120static void
121report_progress(size_t progress, size_t dumpsize)
122{
123	int sofar, i;
124
125	sofar = 100 - ((progress * 100) / dumpsize);
126	for (i = 0; i < nitems(progress_track); i++) {
127		if (sofar < progress_track[i].min_per ||
128		    sofar > progress_track[i].max_per)
129			continue;
130		if (progress_track[i].visited)
131			return;
132		progress_track[i].visited = 1;
133		printf("..%d%%", sofar);
134		return;
135	}
136}
137
138static int
139blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
140{
141	size_t len;
142	int error, i, c;
143	u_int maxdumpsz;
144
145	maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE);
146	if (maxdumpsz == 0)	/* seatbelt */
147		maxdumpsz = PAGE_SIZE;
148	error = 0;
149	if ((sz % PAGE_SIZE) != 0) {
150		printf("size not page aligned\n");
151		return (EINVAL);
152	}
153	if (ptr != NULL && pa != 0) {
154		printf("cant have both va and pa!\n");
155		return (EINVAL);
156	}
157	if ((((uintptr_t)pa) % PAGE_SIZE) != 0) {
158		printf("address not page aligned %p\n", ptr);
159		return (EINVAL);
160	}
161	if (ptr != NULL) {
162		/* If we're doing a virtual dump, flush any pre-existing pa pages */
163		error = blk_flush(di);
164		if (error)
165			return (error);
166	}
167	while (sz) {
168		len = maxdumpsz - fragsz;
169		if (len > sz)
170			len = sz;
171		counter += len;
172		progress -= len;
173		if (counter >> 24) {
174			report_progress(progress, dumpsize);
175			counter &= (1<<24) - 1;
176		}
177
178		wdog_kern_pat(WD_LASTVAL);
179
180		if (ptr) {
181			error = dump_write(di, ptr, 0, dumplo, len);
182			if (error)
183				return (error);
184			dumplo += len;
185			ptr += len;
186			sz -= len;
187		} else {
188			for (i = 0; i < len; i += PAGE_SIZE)
189				dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
190			fragsz += len;
191			pa += len;
192			sz -= len;
193			if (fragsz == maxdumpsz) {
194				error = blk_flush(di);
195				if (error)
196					return (error);
197			}
198		}
199
200		/* Check for user abort. */
201		c = cncheckc();
202		if (c == 0x03)
203			return (ECANCELED);
204		if (c != -1)
205			printf(" (CTRL-C to abort) ");
206	}
207
208	return (0);
209}
210
211/* A fake page table page, to avoid having to handle both 4K and 2M pages */
212static pd_entry_t fakepd[NPDEPG];
213
214int
215minidumpsys(struct dumperinfo *di)
216{
217	uint32_t pmapsize;
218	vm_offset_t va;
219	int error;
220	uint64_t bits;
221	uint64_t *pml4, *pdp, *pd, *pt, pa;
222	size_t size;
223	int i, ii, j, k, n, bit;
224	int retry_count;
225	struct minidumphdr mdhdr;
226
227	retry_count = 0;
228 retry:
229	retry_count++;
230	counter = 0;
231	for (i = 0; i < nitems(progress_track); i++)
232		progress_track[i].visited = 0;
233	/* Walk page table pages, set bits in vm_page_dump */
234	pmapsize = 0;
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		ii = (va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1);
243		pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii;
244		pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME);
245		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
246		if ((pdp[i] & PG_V) == 0) {
247			va += NBPDP;
248			continue;
249		}
250
251		/*
252		 * 1GB page is represented as 512 2MB pages in a dump.
253		 */
254		if ((pdp[i] & PG_PS) != 0) {
255			va += NBPDP;
256			pa = pdp[i] & PG_PS_FRAME;
257			for (n = 0; n < NPDEPG * NPTEPG; n++) {
258				if (is_dumpable(pa))
259					dump_add_page(pa);
260				pa += PAGE_SIZE;
261			}
262			continue;
263		}
264
265		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
266		for (n = 0; n < NPDEPG; n++, va += NBPDR) {
267			j = (va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1);
268
269			if ((pd[j] & PG_V) == 0)
270				continue;
271
272			if ((pd[j] & PG_PS) != 0) {
273				/* This is an entire 2M page. */
274				pa = pd[j] & PG_PS_FRAME;
275				for (k = 0; k < NPTEPG; k++) {
276					if (is_dumpable(pa))
277						dump_add_page(pa);
278					pa += PAGE_SIZE;
279				}
280				continue;
281			}
282
283			pa = pd[j] & PG_FRAME;
284			/* set bit for this PTE page */
285			if (is_dumpable(pa))
286				dump_add_page(pa);
287			/* and for each valid page in this 2MB block */
288			pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
289			for (k = 0; k < NPTEPG; k++) {
290				if ((pt[k] & PG_V) == 0)
291					continue;
292				pa = pt[k] & PG_FRAME;
293				if (is_dumpable(pa))
294					dump_add_page(pa);
295			}
296		}
297	}
298
299	/* Calculate dump size. */
300	dumpsize = pmapsize;
301	dumpsize += round_page(msgbufp->msg_size);
302	dumpsize += round_page(vm_page_dump_size);
303	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
304		bits = vm_page_dump[i];
305		while (bits) {
306			bit = bsfq(bits);
307			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
308			/* Clear out undumpable pages now if needed */
309			if (is_dumpable(pa)) {
310				dumpsize += PAGE_SIZE;
311			} else {
312				dump_drop_page(pa);
313			}
314			bits &= ~(1ul << bit);
315		}
316	}
317	dumpsize += PAGE_SIZE;
318
319	/* Determine dump offset on device. */
320	if (di->mediasize < SIZEOF_METADATA + dumpsize + di->blocksize * 2) {
321		error = E2BIG;
322		goto fail;
323	}
324	dumplo = di->mediaoffset + di->mediasize - dumpsize;
325	dumplo -= di->blocksize * 2;
326	progress = dumpsize;
327
328	/* Initialize mdhdr */
329	bzero(&mdhdr, sizeof(mdhdr));
330	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
331	mdhdr.version = MINIDUMP_VERSION;
332	mdhdr.msgbufsize = msgbufp->msg_size;
333	mdhdr.bitmapsize = vm_page_dump_size;
334	mdhdr.pmapsize = pmapsize;
335	mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
336	mdhdr.dmapbase = DMAP_MIN_ADDRESS;
337	mdhdr.dmapend = DMAP_MAX_ADDRESS;
338
339	mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, dumpsize, di->blocksize);
340
341	printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20,
342	    ptoa((uintmax_t)physmem) / 1048576);
343
344	/* Dump leader */
345	error = dump_write_pad(di, &kdh, 0, dumplo, sizeof(kdh), &size);
346	if (error)
347		goto fail;
348	dumplo += size;
349
350	/* Dump my header */
351	bzero(&fakepd, sizeof(fakepd));
352	bcopy(&mdhdr, &fakepd, sizeof(mdhdr));
353	error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
354	if (error)
355		goto fail;
356
357	/* Dump msgbuf up front */
358	error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
359	if (error)
360		goto fail;
361
362	/* Dump bitmap */
363	error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
364	if (error)
365		goto fail;
366
367	/* Dump kernel page directory pages */
368	bzero(fakepd, sizeof(fakepd));
369	for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + nkpt * NBPDR,
370	    kernel_vm_end); va += NBPDP) {
371		ii = (va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1);
372		pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii;
373		pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME);
374		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
375
376		/* We always write a page, even if it is zero */
377		if ((pdp[i] & PG_V) == 0) {
378			error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
379			if (error)
380				goto fail;
381			/* flush, in case we reuse fakepd in the same block */
382			error = blk_flush(di);
383			if (error)
384				goto fail;
385			continue;
386		}
387
388		/* 1GB page is represented as 512 2MB pages in a dump */
389		if ((pdp[i] & PG_PS) != 0) {
390			/* PDPE and PDP have identical layout in this case */
391			fakepd[0] = pdp[i];
392			for (j = 1; j < NPDEPG; j++)
393				fakepd[j] = fakepd[j - 1] + NBPDR;
394			error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
395			if (error)
396				goto fail;
397			/* flush, in case we reuse fakepd in the same block */
398			error = blk_flush(di);
399			if (error)
400				goto fail;
401			bzero(fakepd, sizeof(fakepd));
402			continue;
403		}
404
405		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
406		error = blk_write(di, (char *)pd, 0, PAGE_SIZE);
407		if (error)
408			goto fail;
409		error = blk_flush(di);
410		if (error)
411			goto fail;
412	}
413
414	/* Dump memory chunks */
415	/* XXX cluster it up and use blk_dump() */
416	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
417		bits = vm_page_dump[i];
418		while (bits) {
419			bit = bsfq(bits);
420			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
421			error = blk_write(di, 0, pa, PAGE_SIZE);
422			if (error)
423				goto fail;
424			bits &= ~(1ul << bit);
425		}
426	}
427
428	error = blk_flush(di);
429	if (error)
430		goto fail;
431
432	/* Dump trailer */
433	error = dump_write_pad(di, &kdh, 0, dumplo, sizeof(kdh), &size);
434	if (error)
435		goto fail;
436	dumplo += size;
437
438	/* Signal completion, signoff and exit stage left. */
439	dump_write(di, NULL, 0, 0, 0);
440	printf("\nDump complete\n");
441	return (0);
442
443 fail:
444	if (error < 0)
445		error = -error;
446
447	printf("\n");
448	if (error == ENOSPC) {
449		printf("Dump map grown while dumping. ");
450		if (retry_count < 5) {
451			printf("Retrying...\n");
452			goto retry;
453		}
454		printf("Dump failed.\n");
455	}
456	else if (error == ECANCELED)
457		printf("Dump aborted\n");
458	else if (error == E2BIG)
459		printf("Dump failed. Partition too small.\n");
460	else
461		printf("** DUMP FAILED (ERROR %d) **\n", error);
462	return (error);
463}
464
465void
466dump_add_page(vm_paddr_t pa)
467{
468	int idx, bit;
469
470	pa >>= PAGE_SHIFT;
471	idx = pa >> 6;		/* 2^6 = 64 */
472	bit = pa & 63;
473	atomic_set_long(&vm_page_dump[idx], 1ul << bit);
474}
475
476void
477dump_drop_page(vm_paddr_t pa)
478{
479	int idx, bit;
480
481	pa >>= PAGE_SHIFT;
482	idx = pa >> 6;		/* 2^6 = 64 */
483	bit = pa & 63;
484	atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
485}
486