1 /*-
2  * Copyright (c) 2002 Marcel Moolenaar
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  * $FreeBSD: src/sys/amd64/amd64/dump_machdep.c,v 1.18 2009/05/29 21:27:12 jamie Exp $
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/conf.h>
32 #include <sys/cons.h>
33 #include <sys/sysctl.h>
34 #include <sys/device.h>
35 #include <sys/kernel.h>
36 #include <sys/kerneldump.h>
37 #include <vm/vm.h>
38 #include <vm/pmap.h>
39 #include <machine/elf.h>
40 #include <machine/md_var.h>
41 #include <machine/thread.h>
42 #include <sys/thread2.h>
43 
44 CTASSERT(sizeof(struct kerneldumpheader) == 512);
45 
46 int do_minidump = 1;
47 TUNABLE_INT("debug.minidump", &do_minidump);
48 SYSCTL_INT(_debug, OID_AUTO, minidump, CTLFLAG_RW, &do_minidump, 0,
49     "Enable mini crash dumps");
50 
51 /*
52  * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
53  * is to protect us from metadata and to protect metadata from us.
54  */
55 #define	SIZEOF_METADATA		(64*1024)
56 
57 #define	MD_ALIGN(x)	(((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
58 #define	DEV_ALIGN(x)	(((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
59 
60 struct md_pa {
61 	vm_paddr_t md_start;
62 	vm_paddr_t md_size;
63 };
64 
65 typedef int callback_t(struct md_pa *, int, void *);
66 
67 static struct kerneldumpheader kdh;
68 static off_t dumplo, fileofs;
69 
70 /* Handle buffered writes. */
71 static char buffer[DEV_BSIZE];
72 static size_t fragsz;
73 
74 /* 20 phys_avail entry pairs correspond to 10 md_pa's */
75 static struct md_pa dump_map[10];
76 
77 static void
78 md_pa_init(void)
79 {
80 	int n, idx;
81 
82 	bzero(dump_map, sizeof(dump_map));
83 	for (n = 0; n < NELEM(dump_map); n++) {
84 		idx = n * 2;
85 		if (dump_avail[idx] == 0 && dump_avail[idx + 1] == 0)
86 			break;
87 		dump_map[n].md_start = dump_avail[idx];
88 		dump_map[n].md_size = dump_avail[idx + 1] - dump_avail[idx];
89 	}
90 }
91 
92 static struct md_pa *
93 md_pa_first(void)
94 {
95 
96 	return (&dump_map[0]);
97 }
98 
99 static struct md_pa *
100 md_pa_next(struct md_pa *mdp)
101 {
102 
103 	mdp++;
104 	if (mdp->md_size == 0)
105 		mdp = NULL;
106 	return (mdp);
107 }
108 
109 static int
110 buf_write(struct dumperinfo *di, char *ptr, size_t sz)
111 {
112 	size_t len;
113 	int error;
114 
115 	while (sz) {
116 		len = DEV_BSIZE - fragsz;
117 		if (len > sz)
118 			len = sz;
119 		bcopy(ptr, buffer + fragsz, len);
120 		fragsz += len;
121 		ptr += len;
122 		sz -= len;
123 		if (fragsz == DEV_BSIZE) {
124 			error = dev_ddump(di->priv, buffer, 0, dumplo,
125 			    DEV_BSIZE);
126 			if (error)
127 				return error;
128 			dumplo += DEV_BSIZE;
129 			fragsz = 0;
130 		}
131 	}
132 
133 	return (0);
134 }
135 
136 static int
137 buf_flush(struct dumperinfo *di)
138 {
139 	int error;
140 
141 	if (fragsz == 0)
142 		return (0);
143 
144 	error = dev_ddump(di->priv, buffer, 0, dumplo, DEV_BSIZE);
145 	dumplo += DEV_BSIZE;
146 	fragsz = 0;
147 	return (error);
148 }
149 
150 #define PG2MB(pgs) ((pgs + (1 << 8) - 1) >> 8)
151 
152 static int
153 cb_dumpdata(struct md_pa *mdp, int seqnr, void *arg)
154 {
155 	struct dumperinfo *di = (struct dumperinfo*)arg;
156 	vm_paddr_t a, pa;
157 	void *va;
158 	uint64_t pgs;
159 	size_t counter, sz, chunk;
160 	int i, c, error;
161 
162 	error = 0;	/* catch case in which chunk size is 0 */
163 	counter = 0;	/* Update twiddle every 16MB */
164 	va = 0;
165 	pgs = mdp->md_size / PAGE_SIZE;
166 	pa = mdp->md_start;
167 
168 	kprintf("  chunk %d: %ldMB (%ld pages)", seqnr, PG2MB(pgs), pgs);
169 
170 	while (pgs) {
171 		chunk = pgs;
172 		if (chunk > MAXDUMPPGS)
173 			chunk = MAXDUMPPGS;
174 		sz = chunk << PAGE_SHIFT;
175 		counter += sz;
176 		if (counter >> 24) {
177 			kprintf(" %ld", PG2MB(pgs));
178 			counter &= (1<<24) - 1;
179 		}
180 		for (i = 0; i < chunk; i++) {
181 			a = pa + i * PAGE_SIZE;
182 			va = pmap_kenter_temporary(trunc_page(a), i);
183 		}
184 		smp_invltlb();
185 		error = dev_ddump(di->priv, va, 0, dumplo, sz);
186 		if (error)
187 			break;
188 		dumplo += sz;
189 		pgs -= chunk;
190 		pa += sz;
191 
192 		/* Check for user abort. */
193 		c = cncheckc();
194 		if (c == 0x03)
195 			return (ECANCELED);
196 		if (c != -1)
197 			kprintf(" (CTRL-C to abort) ");
198 	}
199 	kprintf(" ... %s\n", (error) ? "fail" : "ok");
200 	return (error);
201 }
202 
203 static int
204 cb_dumphdr(struct md_pa *mdp, int seqnr, void *arg)
205 {
206 	struct dumperinfo *di = (struct dumperinfo*)arg;
207 	Elf_Phdr phdr;
208 	uint64_t size;
209 	int error;
210 
211 	size = mdp->md_size;
212 	bzero(&phdr, sizeof(phdr));
213 	phdr.p_type = PT_LOAD;
214 	phdr.p_flags = PF_R;			/* XXX */
215 	phdr.p_offset = fileofs;
216 	phdr.p_vaddr = mdp->md_start;
217 	phdr.p_paddr = mdp->md_start;
218 	phdr.p_filesz = size;
219 	phdr.p_memsz = size;
220 	phdr.p_align = PAGE_SIZE;
221 
222 	error = buf_write(di, (char*)&phdr, sizeof(phdr));
223 	fileofs += phdr.p_filesz;
224 	return (error);
225 }
226 
227 static int
228 cb_size(struct md_pa *mdp, int seqnr, void *arg)
229 {
230 	uint64_t *sz = (uint64_t*)arg;
231 
232 	*sz += (uint64_t)mdp->md_size;
233 	return (0);
234 }
235 
236 static int
237 foreach_chunk(callback_t cb, void *arg)
238 {
239 	struct md_pa *mdp;
240 	int error, seqnr;
241 
242 	seqnr = 0;
243 	mdp = md_pa_first();
244 	while (mdp != NULL) {
245 		error = (*cb)(mdp, seqnr++, arg);
246 		if (error)
247 			return (-error);
248 		mdp = md_pa_next(mdp);
249 	}
250 	return (seqnr);
251 }
252 
253 void
254 md_dumpsys(struct dumperinfo *di)
255 {
256 	Elf_Ehdr ehdr;
257 	uint64_t dumpsize;
258 	off_t hdrgap;
259 	size_t hdrsz;
260 	int error;
261 
262 	savectx(&dumppcb);
263 	dumpthread = curthread;
264 
265 	if (do_minidump) {
266 		minidumpsys(di);
267 		return;
268 	}
269 	bzero(&ehdr, sizeof(ehdr));
270 	ehdr.e_ident[EI_MAG0] = ELFMAG0;
271 	ehdr.e_ident[EI_MAG1] = ELFMAG1;
272 	ehdr.e_ident[EI_MAG2] = ELFMAG2;
273 	ehdr.e_ident[EI_MAG3] = ELFMAG3;
274 	ehdr.e_ident[EI_CLASS] = ELF_CLASS;
275 #if BYTE_ORDER == LITTLE_ENDIAN
276 	ehdr.e_ident[EI_DATA] = ELFDATA2LSB;
277 #else
278 	ehdr.e_ident[EI_DATA] = ELFDATA2MSB;
279 #endif
280 	ehdr.e_ident[EI_VERSION] = EV_CURRENT;
281 	ehdr.e_ident[EI_OSABI] = ELFOSABI_STANDALONE;	/* XXX big picture? */
282 	ehdr.e_type = ET_CORE;
283 	ehdr.e_machine = EM_X86_64;
284 	ehdr.e_phoff = sizeof(ehdr);
285 	ehdr.e_flags = 0;
286 	ehdr.e_ehsize = sizeof(ehdr);
287 	ehdr.e_phentsize = sizeof(Elf_Phdr);
288 	ehdr.e_shentsize = sizeof(Elf_Shdr);
289 
290 	md_pa_init();
291 
292 	/* Calculate dump size. */
293 	dumpsize = 0L;
294 	ehdr.e_phnum = foreach_chunk(cb_size, &dumpsize);
295 	hdrsz = ehdr.e_phoff + ehdr.e_phnum * ehdr.e_phentsize;
296 	fileofs = MD_ALIGN(hdrsz);
297 	dumpsize += fileofs;
298 	hdrgap = fileofs - DEV_ALIGN(hdrsz);
299 
300 	/* Determine dump offset on device. */
301 	if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
302 		error = ENOSPC;
303 		goto fail;
304 	}
305 	dumplo = di->mediaoffset + di->mediasize - dumpsize;
306 	dumplo -= sizeof(kdh) * 2;
307 
308 	mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION,
309 	    dumpsize, di->blocksize);
310 
311 	kprintf("Dumping %llu MB (%d chunks)\n", (long long)dumpsize >> 20,
312 	    ehdr.e_phnum);
313 
314 	/* Dump leader */
315 	error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
316 	if (error)
317 		goto fail;
318 	dumplo += sizeof(kdh);
319 
320 	/* Dump ELF header */
321 	error = buf_write(di, (char*)&ehdr, sizeof(ehdr));
322 	if (error)
323 		goto fail;
324 
325 	/* Dump program headers */
326 	error = foreach_chunk(cb_dumphdr, di);
327 	if (error < 0)
328 		goto fail;
329 	buf_flush(di);
330 
331 	/*
332 	 * All headers are written using blocked I/O, so we know the
333 	 * current offset is (still) block aligned. Skip the alignement
334 	 * in the file to have the segment contents aligned at page
335 	 * boundary. We cannot use MD_ALIGN on dumplo, because we don't
336 	 * care and may very well be unaligned within the dump device.
337 	 */
338 	dumplo += hdrgap;
339 
340 	/* Dump memory chunks (updates dumplo) */
341 	error = foreach_chunk(cb_dumpdata, di);
342 	if (error < 0)
343 		goto fail;
344 
345 	/* Dump trailer */
346 	error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
347 	if (error)
348 		goto fail;
349 
350 	/* Signal completion, signoff and exit stage left. */
351 	dev_ddump(di->priv, NULL, 0, 0, 0);
352 	kprintf("\nDump complete\n");
353 	return;
354 
355  fail:
356 	if (error < 0)
357 		error = -error;
358 
359 	if (error == ECANCELED)
360 		kprintf("\nDump aborted\n");
361 	else if (error == ENOSPC)
362 		kprintf("\nDump failed. Partition too small.\n");
363 	else
364 		kprintf("\n** DUMP FAILED (ERROR %d) **\n", error);
365 }
366