minidump_machdep.c revision 164912 
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 164912 2006-12-05 11:31:33Z ru $");
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/conf.h>
33#include <sys/cons.h>
34#include <sys/kernel.h>
35#include <sys/kerneldump.h>
36#include <sys/msgbuf.h>
37#include <vm/vm.h>
38#include <vm/pmap.h>
39#include <machine/atomic.h>
40#include <machine/elf.h>
41#include <machine/md_var.h>
42#include <machine/vmparam.h>
43#include <machine/minidump.h>
44
45CTASSERT(sizeof(struct kerneldumpheader) == 512);
46
47/*
48 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
49 * is to protect us from metadata and to protect metadata from us.
50 */
51#define	SIZEOF_METADATA		(64*1024)
52
53#define	MD_ALIGN(x)	(((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
54#define	DEV_ALIGN(x)	(((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
55
56extern uint64_t KPDPphys;
57
58uint64_t *vm_page_dump;
59int vm_page_dump_size;
60
61static struct kerneldumpheader kdh;
62static off_t dumplo;
63
64/* Handle chunked writes. */
65static size_t fragsz;
66static void *dump_va;
67static size_t counter, progress;
68
69CTASSERT(sizeof(*vm_page_dump) == 8);
70
71static int
72is_dumpable(vm_paddr_t pa)
73{
74	int i;
75
76	for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
77		if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
78			return (1);
79	}
80	return (0);
81}
82
83/* XXX should be MI */
84static void
85mkdumpheader(struct kerneldumpheader *kdh, uint32_t archver, uint64_t dumplen,
86    uint32_t blksz)
87{
88
89	bzero(kdh, sizeof(*kdh));
90	strncpy(kdh->magic, KERNELDUMPMAGIC, sizeof(kdh->magic));
91	strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
92	kdh->version = htod32(KERNELDUMPVERSION);
93	kdh->architectureversion = htod32(archver);
94	kdh->dumplength = htod64(dumplen);
95	kdh->dumptime = htod64(time_second);
96	kdh->blocksize = htod32(blksz);
97	strncpy(kdh->hostname, hostname, sizeof(kdh->hostname));
98	strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
99	if (panicstr != NULL)
100		strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
101	kdh->parity = kerneldump_parity(kdh);
102}
103
104#define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
105
106static int
107blk_flush(struct dumperinfo *di)
108{
109	int error;
110
111	if (fragsz == 0)
112		return (0);
113
114	error = di->dumper(di->priv, dump_va, 0, dumplo, fragsz);
115	dumplo += fragsz;
116	fragsz = 0;
117	return (error);
118}
119
120static int
121blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
122{
123	size_t len;
124	int error, i, c;
125
126	error = 0;
127	if ((sz % PAGE_SIZE) != 0) {
128		printf("size not page aligned\n");
129		return (EINVAL);
130	}
131	if (ptr != NULL && pa != 0) {
132		printf("cant have both va and pa!\n");
133		return (EINVAL);
134	}
135	if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) {
136		printf("address not page aligned\n");
137		return (EINVAL);
138	}
139	if (ptr != NULL) {
140		/* If we're doing a virtual dump, flush any pre-existing pa pages */
141		error = blk_flush(di);
142		if (error)
143			return (error);
144	}
145	while (sz) {
146		len = (MAXDUMPPGS * PAGE_SIZE) - fragsz;
147		if (len > sz)
148			len = sz;
149		counter += len;
150		progress -= len;
151		if (counter >> 24) {
152			printf(" %ld", PG2MB(progress >> PAGE_SHIFT));
153			counter &= (1<<24) - 1;
154		}
155		if (ptr) {
156			error = di->dumper(di->priv, ptr, 0, dumplo, len);
157			if (error)
158				return (error);
159			dumplo += len;
160			ptr += len;
161			sz -= len;
162		} else {
163			for (i = 0; i < len; i += PAGE_SIZE)
164				dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
165			fragsz += len;
166			pa += len;
167			sz -= len;
168			if (fragsz == (MAXDUMPPGS * PAGE_SIZE)) {
169				error = blk_flush(di);
170				if (error)
171					return (error);
172			}
173		}
174
175		/* Check for user abort. */
176		c = cncheckc();
177		if (c == 0x03)
178			return (ECANCELED);
179		if (c != -1)
180			printf(" (CTRL-C to abort) ");
181	}
182
183	return (0);
184}
185
186/* A fake page table page, to avoid having to handle both 4K and 2M pages */
187static pt_entry_t fakept[NPTEPG];
188
189void
190minidumpsys(struct dumperinfo *di)
191{
192	uint64_t dumpsize;
193	uint32_t ptesize;
194	vm_offset_t va;
195	int error;
196	uint64_t bits;
197	uint64_t *pdp, *pd, *pt, pa;
198	int i, j, k, bit;
199	struct minidumphdr mdhdr;
200
201	counter = 0;
202	/* Walk page table pages, set bits in vm_page_dump */
203	ptesize = 0;
204	pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
205	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
206		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
207		/*
208		 * We always write a page, even if it is zero. Each
209		 * page written corresponds to 2MB of space
210		 */
211		ptesize += PAGE_SIZE;
212		if ((pdp[i] & PG_V) == 0)
213			continue;
214		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
215		j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
216		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
217			/* This is an entire 2M page. */
218			pa = pd[j] & PG_PS_FRAME;
219			for (k = 0; k < NPTEPG; k++) {
220				if (is_dumpable(pa))
221					dump_add_page(pa);
222				pa += PAGE_SIZE;
223			}
224			continue;
225		}
226		if ((pd[j] & PG_V) == PG_V) {
227			/* set bit for each valid page in this 2MB block */
228			pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
229			for (k = 0; k < NPTEPG; k++) {
230				if ((pt[k] & PG_V) == PG_V) {
231					pa = pt[k] & PG_FRAME;
232					if (is_dumpable(pa))
233						dump_add_page(pa);
234				}
235			}
236		} else {
237			/* nothing, we're going to dump a null page */
238		}
239	}
240
241	/* Calculate dump size. */
242	dumpsize = ptesize;
243	dumpsize += round_page(msgbufp->msg_size);
244	dumpsize += round_page(vm_page_dump_size);
245	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
246		bits = vm_page_dump[i];
247		while (bits) {
248			bit = bsfq(bits);
249			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
250			/* Clear out undumpable pages now if needed */
251			if (is_dumpable(pa)) {
252				dumpsize += PAGE_SIZE;
253			} else {
254				dump_drop_page(pa);
255			}
256			bits &= ~(1ul << bit);
257		}
258	}
259	dumpsize += PAGE_SIZE;
260
261	/* Determine dump offset on device. */
262	if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
263		error = ENOSPC;
264		goto fail;
265	}
266	dumplo = di->mediaoffset + di->mediasize - dumpsize;
267	dumplo -= sizeof(kdh) * 2;
268	progress = dumpsize;
269
270	/* Initialize mdhdr */
271	bzero(&mdhdr, sizeof(mdhdr));
272	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
273	mdhdr.version = MINIDUMP_VERSION;
274	mdhdr.msgbufsize = msgbufp->msg_size;
275	mdhdr.bitmapsize = vm_page_dump_size;
276	mdhdr.ptesize = ptesize;
277	mdhdr.kernbase = KERNBASE;
278	mdhdr.dmapbase = DMAP_MIN_ADDRESS;
279	mdhdr.dmapend = DMAP_MAX_ADDRESS;
280
281	mkdumpheader(&kdh, KERNELDUMP_AMD64_VERSION, dumpsize, di->blocksize);
282
283	printf("Physical memory: %ju MB\n", ptoa((uintmax_t)physmem) / 1048576);
284	printf("Dumping %llu MB:", (long long)dumpsize >> 20);
285
286	/* Dump leader */
287	error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
288	if (error)
289		goto fail;
290	dumplo += sizeof(kdh);
291
292	/* Dump my header */
293	bzero(&fakept, sizeof(fakept));
294	bcopy(&mdhdr, &fakept, sizeof(mdhdr));
295	error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
296	if (error)
297		goto fail;
298
299	/* Dump msgbuf up front */
300	error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
301	if (error)
302		goto fail;
303
304	/* Dump bitmap */
305	error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
306	if (error)
307		goto fail;
308
309	/* Dump kernel page table pages */
310	pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
311	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
312		i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
313		/* We always write a page, even if it is zero */
314		if ((pdp[i] & PG_V) == 0) {
315			bzero(fakept, sizeof(fakept));
316			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
317			if (error)
318				goto fail;
319			/* flush, in case we reuse fakept in the same block */
320			error = blk_flush(di);
321			if (error)
322				goto fail;
323			continue;
324		}
325		pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
326		j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
327		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
328			/* This is a single 2M block. Generate a fake PTP */
329			pa = pd[j] & PG_PS_FRAME;
330			for (k = 0; k < NPTEPG; k++) {
331				fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
332			}
333			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
334			if (error)
335				goto fail;
336			/* flush, in case we reuse fakept in the same block */
337			error = blk_flush(di);
338			if (error)
339				goto fail;
340			continue;
341		}
342		if ((pd[j] & PG_V) == PG_V) {
343			pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
344			error = blk_write(di, (char *)pt, 0, PAGE_SIZE);
345			if (error)
346				goto fail;
347		} else {
348			bzero(fakept, sizeof(fakept));
349			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
350			if (error)
351				goto fail;
352			/* flush, in case we reuse fakept in the same block */
353			error = blk_flush(di);
354			if (error)
355				goto fail;
356		}
357	}
358
359	/* Dump memory chunks */
360	/* XXX cluster it up and use blk_dump() */
361	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
362		bits = vm_page_dump[i];
363		while (bits) {
364			bit = bsfq(bits);
365			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
366			error = blk_write(di, 0, pa, PAGE_SIZE);
367			if (error)
368				goto fail;
369			bits &= ~(1ul << bit);
370		}
371	}
372
373	error = blk_flush(di);
374	if (error)
375		goto fail;
376
377	/* Dump trailer */
378	error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
379	if (error)
380		goto fail;
381	dumplo += sizeof(kdh);
382
383	/* Signal completion, signoff and exit stage left. */
384	di->dumper(di->priv, NULL, 0, 0, 0);
385	printf("\nDump complete\n");
386	return;
387
388 fail:
389	if (error < 0)
390		error = -error;
391
392	if (error == ECANCELED)
393		printf("\nDump aborted\n");
394	else if (error == ENOSPC)
395		printf("\nDump failed. Partition too small.\n");
396	else
397		printf("\n** DUMP FAILED (ERROR %d) **\n", error);
398}
399
400void
401dump_add_page(vm_paddr_t pa)
402{
403	int idx, bit;
404
405	pa >>= PAGE_SHIFT;
406	idx = pa >> 6;		/* 2^6 = 64 */
407	bit = pa & 63;
408	atomic_set_long(&vm_page_dump[idx], 1ul << bit);
409}
410
411void
412dump_drop_page(vm_paddr_t pa)
413{
414	int idx, bit;
415
416	pa >>= PAGE_SHIFT;
417	idx = pa >> 6;		/* 2^6 = 64 */
418	bit = pa & 63;
419	atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
420}
421