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: releng/6.2/sys/i386/i386/minidump_machdep.c 160633 2006-07-24 23:28:09Z peter $");
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
45 CTASSERT(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
56 uint32_t *vm_page_dump;
57 int vm_page_dump_size;
58
59 static struct kerneldumpheader kdh;
60 static off_t dumplo;
61
62 /* Handle chunked writes. */
63 static size_t fragsz;
64 static void *dump_va;
65 static uint64_t counter, progress;
66
67 CTASSERT(sizeof(*vm_page_dump) == 4);
68
69 static int
70 is_dumpable(vm_paddr_t pa)
71 {
72 int i;
73
74 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
75 if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
76 return (1);
77 }
78 return (0);
79 }
80
81 /* XXX should be MI */
82 static void
83 mkdumpheader(struct kerneldumpheader *kdh, uint32_t archver, uint64_t dumplen,
84 uint32_t blksz)
85 {
86
87 bzero(kdh, sizeof(*kdh));
88 strncpy(kdh->magic, KERNELDUMPMAGIC, sizeof(kdh->magic));
89 strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
90 kdh->version = htod32(KERNELDUMPVERSION);
91 kdh->architectureversion = htod32(archver);
92 kdh->dumplength = htod64(dumplen);
93 kdh->dumptime = htod64(time_second);
94 kdh->blocksize = htod32(blksz);
95 strncpy(kdh->hostname, hostname, sizeof(kdh->hostname));
96 strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
97 if (panicstr != NULL)
98 strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
99 kdh->parity = kerneldump_parity(kdh);
100 }
101
102 #define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
103
104 static int
105 blk_flush(struct dumperinfo *di)
106 {
107 int error;
108
109 if (fragsz == 0)
110 return (0);
111
112 error = di->dumper(di->priv, dump_va, 0, dumplo, fragsz);
113 dumplo += fragsz;
114 fragsz = 0;
115 return (error);
116 }
117
118 static int
119 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
120 {
121 size_t len;
122 int error, i, c;
123
124 error = 0;
125 if ((sz % PAGE_SIZE) != 0) {
126 printf("size not page aligned\n");
127 return (EINVAL);
128 }
129 if (ptr != NULL && pa != 0) {
130 printf("cant have both va and pa!\n");
131 return (EINVAL);
132 }
133 if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) {
134 printf("address not page aligned\n");
135 return (EINVAL);
136 }
137 if (ptr != NULL) {
138 /* If we're doing a virtual dump, flush any pre-existing pa pages */
139 error = blk_flush(di);
140 if (error)
141 return (error);
142 }
143 while (sz) {
144 len = (MAXDUMPPGS * PAGE_SIZE) - fragsz;
145 if (len > sz)
146 len = sz;
147 counter += len;
148 progress -= len;
149 if (counter >> 24) {
150 printf(" %lld", PG2MB(progress >> PAGE_SHIFT));
151 counter &= (1<<24) - 1;
152 }
153 if (ptr) {
154 error = di->dumper(di->priv, ptr, 0, dumplo, len);
155 if (error)
156 return (error);
157 dumplo += len;
158 ptr += len;
159 sz -= len;
160 } else {
161 for (i = 0; i < len; i += PAGE_SIZE)
162 dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
163 fragsz += len;
164 pa += len;
165 sz -= len;
166 if (fragsz == (MAXDUMPPGS * PAGE_SIZE)) {
167 error = blk_flush(di);
168 if (error)
169 return (error);
170 }
171 }
172
173 /* Check for user abort. */
174 c = cncheckc();
175 if (c == 0x03)
176 return (ECANCELED);
177 if (c != -1)
178 printf(" (CTRL-C to abort) ");
179 }
180
181 return (0);
182 }
183
184 /* A fake page table page, to avoid having to handle both 4K and 2M pages */
185 static pt_entry_t fakept[NPTEPG];
186
187 void
188 minidumpsys(struct dumperinfo *di)
189 {
190 uint64_t dumpsize;
191 uint32_t ptesize;
192 vm_offset_t va;
193 int error;
194 uint32_t bits;
195 uint64_t pa;
196 pd_entry_t *pd;
197 pt_entry_t *pt;
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 for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
205 /*
206 * We always write a page, even if it is zero. Each
207 * page written corresponds to 2MB of space
208 */
209 ptesize += PAGE_SIZE;
210 pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE); /* always mapped! */
211 j = va >> PDRSHIFT;
212 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
213 /* This is an entire 2M page. */
214 pa = pd[j] & PG_FRAME & ~PDRMASK;
215 for (k = 0; k < NPTEPG; k++) {
216 if (is_dumpable(pa))
217 dump_add_page(pa);
218 pa += PAGE_SIZE;
219 }
220 continue;
221 }
222 if ((pd[j] & PG_V) == PG_V) {
223 /* set bit for each valid page in this 2MB block */
224 pt = pmap_kenter_temporary(pd[j] & PG_FRAME, 0);
225 for (k = 0; k < NPTEPG; k++) {
226 if ((pt[k] & PG_V) == PG_V) {
227 pa = pt[k] & PG_FRAME;
228 if (is_dumpable(pa))
229 dump_add_page(pa);
230 }
231 }
232 } else {
233 /* nothing, we're going to dump a null page */
234 }
235 }
236
237 /* Calculate dump size. */
238 dumpsize = ptesize;
239 dumpsize += round_page(msgbufp->msg_size);
240 dumpsize += round_page(vm_page_dump_size);
241 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
242 bits = vm_page_dump[i];
243 while (bits) {
244 bit = bsfl(bits);
245 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
246 /* Clear out undumpable pages now if needed */
247 if (is_dumpable(pa)) {
248 dumpsize += PAGE_SIZE;
249 } else {
250 dump_drop_page(pa);
251 }
252 bits &= ~(1ul << bit);
253 }
254 }
255 dumpsize += PAGE_SIZE;
256
257 /* Determine dump offset on device. */
258 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
259 error = ENOSPC;
260 goto fail;
261 }
262 dumplo = di->mediaoffset + di->mediasize - dumpsize;
263 dumplo -= sizeof(kdh) * 2;
264 progress = dumpsize;
265
266 /* Initialize mdhdr */
267 bzero(&mdhdr, sizeof(mdhdr));
268 strcpy(mdhdr.magic, MINIDUMP_MAGIC);
269 mdhdr.version = MINIDUMP_VERSION;
270 mdhdr.msgbufsize = msgbufp->msg_size;
271 mdhdr.bitmapsize = vm_page_dump_size;
272 mdhdr.ptesize = ptesize;
273 mdhdr.kernbase = KERNBASE;
274 #ifdef PAE
275 mdhdr.paemode = 1;
276 #endif
277
278 mkdumpheader(&kdh, KERNELDUMP_I386_VERSION, dumpsize, di->blocksize);
279
280 printf("Physical memory: %ju MB\n", ptoa((uintmax_t)physmem) / 1048576);
281 printf("Dumping %llu MB:", (long long)dumpsize >> 20);
282
283 /* Dump leader */
284 error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
285 if (error)
286 goto fail;
287 dumplo += sizeof(kdh);
288
289 /* Dump my header */
290 bzero(&fakept, sizeof(fakept));
291 bcopy(&mdhdr, &fakept, sizeof(mdhdr));
292 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
293 if (error)
294 goto fail;
295
296 /* Dump msgbuf up front */
297 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
298 if (error)
299 goto fail;
300
301 /* Dump bitmap */
302 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
303 if (error)
304 goto fail;
305
306 /* Dump kernel page table pages */
307 for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
308 /* We always write a page, even if it is zero */
309 pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE); /* always mapped! */
310 j = va >> PDRSHIFT;
311 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
312 /* This is a single 2M block. Generate a fake PTP */
313 pa = pd[j] & PG_FRAME & ~PDRMASK;
314 for (k = 0; k < NPTEPG; k++) {
315 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
316 }
317 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
318 if (error)
319 goto fail;
320 /* flush, in case we reuse fakept in the same block */
321 error = blk_flush(di);
322 if (error)
323 goto fail;
324 continue;
325 }
326 if ((pd[j] & PG_V) == PG_V) {
327 pa = pd[j] & PG_FRAME;
328 error = blk_write(di, 0, pa, PAGE_SIZE);
329 if (error)
330 goto fail;
331 } else {
332 bzero(fakept, sizeof(fakept));
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 }
341 }
342
343 /* Dump memory chunks */
344 /* XXX cluster it up and use blk_dump() */
345 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
346 bits = vm_page_dump[i];
347 while (bits) {
348 bit = bsfl(bits);
349 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
350 error = blk_write(di, 0, pa, PAGE_SIZE);
351 if (error)
352 goto fail;
353 bits &= ~(1ul << bit);
354 }
355 }
356
357 error = blk_flush(di);
358 if (error)
359 goto fail;
360
361 /* Dump trailer */
362 error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
363 if (error)
364 goto fail;
365 dumplo += sizeof(kdh);
366
367 /* Signal completion, signoff and exit stage left. */
368 di->dumper(di->priv, NULL, 0, 0, 0);
369 printf("\nDump complete\n");
370 return;
371
372 fail:
373 if (error < 0)
374 error = -error;
375
376 if (error == ECANCELED)
377 printf("\nDump aborted\n");
378 else if (error == ENOSPC)
379 printf("\nDump failed. Partition too small.\n");
380 else
381 printf("\n** DUMP FAILED (ERROR %d) **\n", error);
382 }
383
384 void
385 dump_add_page(vm_paddr_t pa)
386 {
387 int idx, bit;
388
389 pa >>= PAGE_SHIFT;
390 idx = pa >> 5; /* 2^5 = 32 */
391 bit = pa & 31;
392 atomic_set_int(&vm_page_dump[idx], 1ul << bit);
393 }
394
395 void
396 dump_drop_page(vm_paddr_t pa)
397 {
398 int idx, bit;
399
400 pa >>= PAGE_SHIFT;
401 idx = pa >> 5; /* 2^5 = 32 */
402 bit = pa & 31;
403 atomic_clear_int(&vm_page_dump[idx], 1ul << bit);
404 }
405
Cache object: d13d7c55e4122843ea84c9399410f4f7
|