xref: /titanic_41/usr/src/uts/common/os/dumpsubr.c (revision b7d3956b92a285d8dac2c7f5f7e28d2ef5347ef8)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/vm.h>
32 #include <sys/proc.h>
33 #include <sys/file.h>
34 #include <sys/conf.h>
35 #include <sys/kmem.h>
36 #include <sys/mem.h>
37 #include <sys/mman.h>
38 #include <sys/vnode.h>
39 #include <sys/errno.h>
40 #include <sys/memlist.h>
41 #include <sys/dumphdr.h>
42 #include <sys/dumpadm.h>
43 #include <sys/ksyms.h>
44 #include <sys/compress.h>
45 #include <sys/stream.h>
46 #include <sys/strsun.h>
47 #include <sys/cmn_err.h>
48 #include <sys/bitmap.h>
49 #include <sys/modctl.h>
50 #include <sys/utsname.h>
51 #include <sys/systeminfo.h>
52 #include <sys/vmem.h>
53 #include <sys/log.h>
54 #include <sys/var.h>
55 #include <sys/debug.h>
56 #include <sys/sunddi.h>
57 #include <fs/fs_subr.h>
58 #include <sys/fs/snode.h>
59 #include <sys/ontrap.h>
60 #include <sys/panic.h>
61 #include <sys/dkio.h>
62 #include <sys/vtoc.h>
63 #include <sys/errorq.h>
64 #include <sys/fm/util.h>
65 #include <sys/fs/zfs.h>
66 
67 #include <vm/hat.h>
68 #include <vm/as.h>
69 #include <vm/page.h>
70 #include <vm/seg.h>
71 #include <vm/seg_kmem.h>
72 
73 kmutex_t	dump_lock;	/* lock for dump configuration */
74 dumphdr_t	*dumphdr;	/* dump header */
75 int		dump_conflags = DUMP_KERNEL; /* dump configuration flags */
76 vnode_t		*dumpvp;	/* dump device vnode pointer */
77 u_offset_t	dumpvp_size;	/* size of dump device, in bytes */
78 static u_offset_t dumpvp_limit;	/* maximum write offset */
79 char		*dumppath;	/* pathname of dump device */
80 int		dump_timeout = 120; /* timeout for dumping page during panic */
81 int		dump_timeleft;	/* portion of dump_timeout remaining */
82 int		dump_ioerr;	/* dump i/o error */
83 
84 #ifdef DEBUG
85 int		dumpfaildebug = 1;	/* enter debugger if dump fails */
86 #else
87 int		dumpfaildebug = 0;
88 #endif
89 
90 static ulong_t	*dump_bitmap;	/* bitmap for marking pages to dump */
91 static pgcnt_t	dump_bitmapsize; /* size of bitmap */
92 static pid_t	*dump_pids;	/* list of process IDs at dump time */
93 static offset_t	dumpvp_off;	/* current dump device offset */
94 static char	*dump_cmap;	/* VA for dump compression mapping */
95 static char	*dumpbuf_cur, *dumpbuf_start, *dumpbuf_end;
96 static char	*dump_cbuf;	/* compression buffer */
97 static char	*dump_uebuf;	/* memory error detection buffer */
98 static size_t	dumpbuf_size;	/* size of dumpbuf in bytes */
99 static size_t	dumpbuf_limit = 1UL << 23;	/* 8MB */
100 static size_t	dump_iosize;	/* device's best transfer size, if any */
101 static uint64_t	dumpbuf_thresh = 1ULL << 30;	/* 1GB */
102 static ulong_t	dumpbuf_mult = 8;
103 
104 /*
105  * The dump i/o buffer must be at least one page, at most xfer_size bytes, and
106  * should scale with physmem in between.  The transfer size passed in will
107  * either represent a global default (maxphys) or the best size for the device.
108  * Once the physical memory size exceeds dumpbuf_thresh (1GB by default), we
109  * increase the percentage of physical memory that dumpbuf can consume by a
110  * factor of dumpbuf_mult (8 by default) to improve large memory performance.
111  * The size of the dumpbuf i/o buffer is limited by dumpbuf_limit (8MB by
112  * default) because the dump performance saturates beyond a certain size.
113  */
114 static size_t
115 dumpbuf_iosize(size_t xfer_size)
116 {
117 	pgcnt_t scale = physmem;
118 	size_t iosize;
119 
120 	if (scale >= dumpbuf_thresh / PAGESIZE) {
121 		scale *= dumpbuf_mult; /* increase scaling factor */
122 		iosize = MIN(xfer_size, scale) & PAGEMASK;
123 		if (dumpbuf_limit && iosize > dumpbuf_limit)
124 			iosize = MAX(PAGESIZE, dumpbuf_limit & PAGEMASK);
125 	} else
126 		iosize = MAX(PAGESIZE, MIN(xfer_size, scale) & PAGEMASK);
127 
128 	return (iosize);
129 }
130 
131 static void
132 dumpbuf_resize(void)
133 {
134 	char *old_buf = dumpbuf_start;
135 	size_t old_size = dumpbuf_size;
136 	char *new_buf;
137 	size_t new_size;
138 
139 	ASSERT(MUTEX_HELD(&dump_lock));
140 
141 	if ((new_size = dumpbuf_iosize(MAX(dump_iosize, maxphys))) <= old_size)
142 		return; /* no need to reallocate buffer */
143 
144 	new_buf = kmem_alloc(new_size, KM_SLEEP);
145 	dumpbuf_size = new_size;
146 	dumpbuf_start = new_buf;
147 	dumpbuf_end = new_buf + new_size;
148 	kmem_free(old_buf, old_size);
149 }
150 
151 static void
152 dumphdr_init(void)
153 {
154 	pgcnt_t npages = 0;
155 
156 	ASSERT(MUTEX_HELD(&dump_lock));
157 
158 	if (dumphdr == NULL) {
159 		dumphdr = kmem_zalloc(sizeof (dumphdr_t), KM_SLEEP);
160 		dumphdr->dump_magic = DUMP_MAGIC;
161 		dumphdr->dump_version = DUMP_VERSION;
162 		dumphdr->dump_wordsize = DUMP_WORDSIZE;
163 		dumphdr->dump_pageshift = PAGESHIFT;
164 		dumphdr->dump_pagesize = PAGESIZE;
165 		dumphdr->dump_utsname = utsname;
166 		(void) strcpy(dumphdr->dump_platform, platform);
167 		dump_cmap = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
168 		dumpbuf_size = dumpbuf_iosize(maxphys);
169 		dumpbuf_start = kmem_alloc(dumpbuf_size, KM_SLEEP);
170 		dumpbuf_end = dumpbuf_start + dumpbuf_size;
171 		dump_cbuf = kmem_alloc(PAGESIZE, KM_SLEEP); /* compress buf */
172 		dump_uebuf = kmem_alloc(PAGESIZE, KM_SLEEP); /* UE buf */
173 		dump_pids = kmem_alloc(v.v_proc * sizeof (pid_t), KM_SLEEP);
174 	}
175 
176 	npages = num_phys_pages();
177 
178 	if (dump_bitmapsize != npages) {
179 		void *map = kmem_alloc(BT_SIZEOFMAP(npages), KM_SLEEP);
180 		kmem_free(dump_bitmap, BT_SIZEOFMAP(dump_bitmapsize));
181 		dump_bitmap = map;
182 		dump_bitmapsize = npages;
183 	}
184 }
185 
186 /*
187  * Establish a new dump device.
188  */
189 int
190 dumpinit(vnode_t *vp, char *name, int justchecking)
191 {
192 	vnode_t *cvp;
193 	vattr_t vattr;
194 	vnode_t *cdev_vp;
195 	int error = 0;
196 
197 	ASSERT(MUTEX_HELD(&dump_lock));
198 
199 	dumphdr_init();
200 
201 	cvp = common_specvp(vp);
202 	if (cvp == dumpvp)
203 		return (0);
204 
205 	/*
206 	 * Determine whether this is a plausible dump device.  We want either:
207 	 * (1) a real device that's not mounted and has a cb_dump routine, or
208 	 * (2) a swapfile on some filesystem that has a vop_dump routine.
209 	 */
210 	if ((error = VOP_OPEN(&cvp, FREAD | FWRITE, kcred, NULL)) != 0)
211 		return (error);
212 
213 	vattr.va_mask = AT_SIZE | AT_TYPE | AT_RDEV;
214 	if ((error = VOP_GETATTR(cvp, &vattr, 0, kcred, NULL)) == 0) {
215 		if (vattr.va_type == VBLK || vattr.va_type == VCHR) {
216 			if (devopsp[getmajor(vattr.va_rdev)]->
217 			    devo_cb_ops->cb_dump == nodev)
218 				error = ENOTSUP;
219 			else if (vfs_devismounted(vattr.va_rdev))
220 				error = EBUSY;
221 		} else {
222 			if (vn_matchopval(cvp, VOPNAME_DUMP, fs_nosys) ||
223 			    !IS_SWAPVP(cvp))
224 				error = ENOTSUP;
225 		}
226 	}
227 
228 	if (error == 0 && vattr.va_size < 2 * DUMP_LOGSIZE + DUMP_ERPTSIZE)
229 		error = ENOSPC;
230 
231 	if (error || justchecking) {
232 		(void) VOP_CLOSE(cvp, FREAD | FWRITE, 1, (offset_t)0,
233 		    kcred, NULL);
234 		return (error);
235 	}
236 
237 	VN_HOLD(cvp);
238 
239 	if (dumpvp != NULL)
240 		dumpfini();	/* unconfigure the old dump device */
241 
242 	dumpvp = cvp;
243 	dumpvp_size = vattr.va_size & -DUMP_OFFSET;
244 	dumppath = kmem_alloc(strlen(name) + 1, KM_SLEEP);
245 	(void) strcpy(dumppath, name);
246 	dump_iosize = 0;
247 
248 	/*
249 	 * If the dump device is a block device, attempt to open up the
250 	 * corresponding character device and determine its maximum transfer
251 	 * size.  We use this information to potentially resize dumpbuf to a
252 	 * larger and more optimal size for performing i/o to the dump device.
253 	 */
254 	if (cvp->v_type == VBLK &&
255 	    (cdev_vp = makespecvp(VTOS(cvp)->s_dev, VCHR)) != NULL) {
256 		if (VOP_OPEN(&cdev_vp, FREAD | FWRITE, kcred, NULL) == 0) {
257 			size_t blk_size;
258 			struct dk_cinfo dki;
259 			struct vtoc vtoc;
260 
261 			if (VOP_IOCTL(cdev_vp, DKIOCGVTOC, (intptr_t)&vtoc,
262 			    FKIOCTL, kcred, NULL, NULL) == 0 &&
263 			    vtoc.v_sectorsz != 0)
264 				blk_size = vtoc.v_sectorsz;
265 			else
266 				blk_size = DEV_BSIZE;
267 
268 			if (VOP_IOCTL(cdev_vp, DKIOCINFO, (intptr_t)&dki,
269 			    FKIOCTL, kcred, NULL, NULL) == 0) {
270 				dump_iosize = dki.dki_maxtransfer * blk_size;
271 				dumpbuf_resize();
272 			}
273 			/*
274 			 * If we are working with a zvol then call into
275 			 * it to dumpify itself.
276 			 */
277 			if (strcmp(dki.dki_dname, ZVOL_DRIVER) == 0) {
278 				if ((error = VOP_IOCTL(cdev_vp,
279 				    DKIOCDUMPINIT, NULL, FKIOCTL, kcred,
280 				    NULL, NULL)) != 0) {
281 					dumpfini();
282 				}
283 			}
284 
285 			(void) VOP_CLOSE(cdev_vp, FREAD | FWRITE, 1, 0,
286 			    kcred, NULL);
287 		}
288 
289 		VN_RELE(cdev_vp);
290 	}
291 
292 	cmn_err(CE_CONT, "?dump on %s size %llu MB\n", name, dumpvp_size >> 20);
293 
294 	return (error);
295 }
296 
297 void
298 dumpfini(void)
299 {
300 	vattr_t vattr;
301 	boolean_t is_zfs = B_FALSE;
302 	vnode_t *cdev_vp;
303 	ASSERT(MUTEX_HELD(&dump_lock));
304 
305 	kmem_free(dumppath, strlen(dumppath) + 1);
306 
307 	/*
308 	 * Determine if we are using zvols for our dump device
309 	 */
310 	vattr.va_mask = AT_RDEV;
311 	if (VOP_GETATTR(dumpvp, &vattr, 0, kcred, NULL) == 0) {
312 		is_zfs = (getmajor(vattr.va_rdev) ==
313 		    ddi_name_to_major(ZFS_DRIVER)) ? B_TRUE : B_FALSE;
314 	}
315 
316 	/*
317 	 * If we have a zvol dump device then we call into zfs so
318 	 * that it may have a chance to cleanup.
319 	 */
320 	if (is_zfs &&
321 	    (cdev_vp = makespecvp(VTOS(dumpvp)->s_dev, VCHR)) != NULL) {
322 		if (VOP_OPEN(&cdev_vp, FREAD | FWRITE, kcred, NULL) == 0) {
323 			(void) VOP_IOCTL(cdev_vp, DKIOCDUMPFINI, NULL, FKIOCTL,
324 			    kcred, NULL, NULL);
325 			(void) VOP_CLOSE(cdev_vp, FREAD | FWRITE, 1, 0,
326 			    kcred, NULL);
327 		}
328 		VN_RELE(cdev_vp);
329 	}
330 
331 	(void) VOP_CLOSE(dumpvp, FREAD | FWRITE, 1, (offset_t)0, kcred, NULL);
332 
333 	VN_RELE(dumpvp);
334 
335 	dumpvp = NULL;
336 	dumpvp_size = 0;
337 	dumppath = NULL;
338 }
339 
340 static pfn_t
341 dump_bitnum_to_pfn(pgcnt_t bitnum)
342 {
343 	struct memlist *mp;
344 
345 	for (mp = phys_install; mp != NULL; mp = mp->next) {
346 		if (bitnum < (mp->size >> PAGESHIFT))
347 			return ((mp->address >> PAGESHIFT) + bitnum);
348 		bitnum -= mp->size >> PAGESHIFT;
349 	}
350 	return (PFN_INVALID);
351 }
352 
353 static pgcnt_t
354 dump_pfn_to_bitnum(pfn_t pfn)
355 {
356 	struct memlist *mp;
357 	pgcnt_t bitnum = 0;
358 
359 	for (mp = phys_install; mp != NULL; mp = mp->next) {
360 		if (pfn >= (mp->address >> PAGESHIFT) &&
361 		    pfn < ((mp->address + mp->size) >> PAGESHIFT))
362 			return (bitnum + pfn - (mp->address >> PAGESHIFT));
363 		bitnum += mp->size >> PAGESHIFT;
364 	}
365 	return ((pgcnt_t)-1);
366 }
367 
368 static offset_t
369 dumpvp_flush(void)
370 {
371 	size_t size = P2ROUNDUP(dumpbuf_cur - dumpbuf_start, PAGESIZE);
372 	int err;
373 
374 	if (dumpvp_off + size > dumpvp_limit) {
375 		dump_ioerr = ENOSPC;
376 	} else if (size != 0) {
377 		if (panicstr)
378 			err = VOP_DUMP(dumpvp, dumpbuf_start,
379 			    lbtodb(dumpvp_off), btod(size), NULL);
380 		else
381 			err = vn_rdwr(UIO_WRITE, dumpvp, dumpbuf_start, size,
382 			    dumpvp_off, UIO_SYSSPACE, 0, dumpvp_limit,
383 			    kcred, 0);
384 		if (err && dump_ioerr == 0)
385 			dump_ioerr = err;
386 	}
387 	dumpvp_off += size;
388 	dumpbuf_cur = dumpbuf_start;
389 	dump_timeleft = dump_timeout;
390 	return (dumpvp_off);
391 }
392 
393 void
394 dumpvp_write(const void *va, size_t size)
395 {
396 	while (size != 0) {
397 		size_t len = MIN(size, dumpbuf_end - dumpbuf_cur);
398 		if (len == 0) {
399 			(void) dumpvp_flush();
400 		} else {
401 			bcopy(va, dumpbuf_cur, len);
402 			va = (char *)va + len;
403 			dumpbuf_cur += len;
404 			size -= len;
405 		}
406 	}
407 }
408 
409 /*ARGSUSED*/
410 static void
411 dumpvp_ksyms_write(const void *src, void *dst, size_t size)
412 {
413 	dumpvp_write(src, size);
414 }
415 
416 /*
417  * Mark 'pfn' in the bitmap and dump its translation table entry.
418  */
419 void
420 dump_addpage(struct as *as, void *va, pfn_t pfn)
421 {
422 	mem_vtop_t mem_vtop;
423 	pgcnt_t bitnum;
424 
425 	if ((bitnum = dump_pfn_to_bitnum(pfn)) != (pgcnt_t)-1) {
426 		if (!BT_TEST(dump_bitmap, bitnum)) {
427 			dumphdr->dump_npages++;
428 			BT_SET(dump_bitmap, bitnum);
429 		}
430 		dumphdr->dump_nvtop++;
431 		mem_vtop.m_as = as;
432 		mem_vtop.m_va = va;
433 		mem_vtop.m_pfn = pfn;
434 		dumpvp_write(&mem_vtop, sizeof (mem_vtop_t));
435 	}
436 	dump_timeleft = dump_timeout;
437 }
438 
439 /*
440  * Mark 'pfn' in the bitmap
441  */
442 void
443 dump_page(pfn_t pfn)
444 {
445 	pgcnt_t bitnum;
446 
447 	if ((bitnum = dump_pfn_to_bitnum(pfn)) != (pgcnt_t)-1) {
448 		if (!BT_TEST(dump_bitmap, bitnum)) {
449 			dumphdr->dump_npages++;
450 			BT_SET(dump_bitmap, bitnum);
451 		}
452 	}
453 	dump_timeleft = dump_timeout;
454 }
455 
456 /*
457  * Dump the <as, va, pfn> information for a given address space.
458  * SEGOP_DUMP() will call dump_addpage() for each page in the segment.
459  */
460 static void
461 dump_as(struct as *as)
462 {
463 	struct seg *seg;
464 
465 	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
466 	for (seg = AS_SEGFIRST(as); seg; seg = AS_SEGNEXT(as, seg)) {
467 		if (seg->s_as != as)
468 			break;
469 		if (seg->s_ops == NULL)
470 			continue;
471 		SEGOP_DUMP(seg);
472 	}
473 	AS_LOCK_EXIT(as, &as->a_lock);
474 
475 	if (seg != NULL)
476 		cmn_err(CE_WARN, "invalid segment %p in address space %p",
477 		    (void *)seg, (void *)as);
478 }
479 
480 static int
481 dump_process(pid_t pid)
482 {
483 	proc_t *p = sprlock(pid);
484 
485 	if (p == NULL)
486 		return (-1);
487 	if (p->p_as != &kas) {
488 		mutex_exit(&p->p_lock);
489 		dump_as(p->p_as);
490 		mutex_enter(&p->p_lock);
491 	}
492 
493 	sprunlock(p);
494 
495 	return (0);
496 }
497 
498 void
499 dump_ereports(void)
500 {
501 	u_offset_t dumpvp_start;
502 	erpt_dump_t ed;
503 
504 	if (dumpvp == NULL || dumphdr == NULL)
505 		return;
506 
507 	dumpbuf_cur = dumpbuf_start;
508 	dumpvp_limit = dumpvp_size - (DUMP_OFFSET + DUMP_LOGSIZE);
509 	dumpvp_start = dumpvp_limit - DUMP_ERPTSIZE;
510 	dumpvp_off = dumpvp_start;
511 
512 	fm_ereport_dump();
513 	if (panicstr)
514 		errorq_dump();
515 
516 	bzero(&ed, sizeof (ed)); /* indicate end of ereports */
517 	dumpvp_write(&ed, sizeof (ed));
518 	(void) dumpvp_flush();
519 
520 	if (!panicstr) {
521 		(void) VOP_PUTPAGE(dumpvp, dumpvp_start,
522 		    (size_t)(dumpvp_off - dumpvp_start),
523 		    B_INVAL | B_FORCE, kcred, NULL);
524 	}
525 }
526 
527 void
528 dump_messages(void)
529 {
530 	log_dump_t ld;
531 	mblk_t *mctl, *mdata;
532 	queue_t *q, *qlast;
533 	u_offset_t dumpvp_start;
534 
535 	if (dumpvp == NULL || dumphdr == NULL || log_consq == NULL)
536 		return;
537 
538 	dumpbuf_cur = dumpbuf_start;
539 	dumpvp_limit = dumpvp_size - DUMP_OFFSET;
540 	dumpvp_start = dumpvp_limit - DUMP_LOGSIZE;
541 	dumpvp_off = dumpvp_start;
542 
543 	qlast = NULL;
544 	do {
545 		for (q = log_consq; q->q_next != qlast; q = q->q_next)
546 			continue;
547 		for (mctl = q->q_first; mctl != NULL; mctl = mctl->b_next) {
548 			dump_timeleft = dump_timeout;
549 			mdata = mctl->b_cont;
550 			ld.ld_magic = LOG_MAGIC;
551 			ld.ld_msgsize = MBLKL(mctl->b_cont);
552 			ld.ld_csum = checksum32(mctl->b_rptr, MBLKL(mctl));
553 			ld.ld_msum = checksum32(mdata->b_rptr, MBLKL(mdata));
554 			dumpvp_write(&ld, sizeof (ld));
555 			dumpvp_write(mctl->b_rptr, MBLKL(mctl));
556 			dumpvp_write(mdata->b_rptr, MBLKL(mdata));
557 		}
558 	} while ((qlast = q) != log_consq);
559 
560 	ld.ld_magic = 0;		/* indicate end of messages */
561 	dumpvp_write(&ld, sizeof (ld));
562 	(void) dumpvp_flush();
563 	if (!panicstr) {
564 		(void) VOP_PUTPAGE(dumpvp, dumpvp_start,
565 		    (size_t)(dumpvp_off - dumpvp_start),
566 		    B_INVAL | B_FORCE, kcred, NULL);
567 	}
568 }
569 
570 static void
571 dump_pagecopy(void *src, void *dst)
572 {
573 	long *wsrc = (long *)src;
574 	long *wdst = (long *)dst;
575 	const ulong_t ncopies = PAGESIZE / sizeof (long);
576 	volatile int w = 0;
577 	volatile int ueoff = -1;
578 	on_trap_data_t otd;
579 
580 	if (on_trap(&otd, OT_DATA_EC)) {
581 		if (ueoff == -1) {
582 			uint64_t pa;
583 
584 			ueoff = w * sizeof (long);
585 			pa = ptob((uint64_t)hat_getpfnum(kas.a_hat, src))
586 			    + ueoff;
587 			cmn_err(CE_WARN, "memory error at PA 0x%08x.%08x",
588 			    (uint32_t)(pa >> 32), (uint32_t)pa);
589 		}
590 #ifdef _LP64
591 		wdst[w++] = 0xbadecc00badecc;
592 #else
593 		wdst[w++] = 0xbadecc;
594 #endif
595 	}
596 	while (w < ncopies) {
597 		wdst[w] = wsrc[w];
598 		w++;
599 	}
600 	no_trap();
601 }
602 
603 /*
604  * Dump the system.
605  */
606 void
607 dumpsys(void)
608 {
609 	pfn_t pfn;
610 	pgcnt_t bitnum;
611 	int npages = 0;
612 	int percent_done = 0;
613 	uint32_t csize;
614 	u_offset_t total_csize = 0;
615 	int compress_ratio;
616 	proc_t *p;
617 	pid_t npids, pidx;
618 	char *content;
619 
620 	if (dumpvp == NULL || dumphdr == NULL) {
621 		uprintf("skipping system dump - no dump device configured\n");
622 		return;
623 	}
624 	dumpbuf_cur = dumpbuf_start;
625 
626 	/*
627 	 * Calculate the starting block for dump.  If we're dumping on a
628 	 * swap device, start 1/5 of the way in; otherwise, start at the
629 	 * beginning.  And never use the first page -- it may be a disk label.
630 	 */
631 	if (dumpvp->v_flag & VISSWAP)
632 		dumphdr->dump_start = P2ROUNDUP(dumpvp_size / 5, DUMP_OFFSET);
633 	else
634 		dumphdr->dump_start = DUMP_OFFSET;
635 
636 	dumphdr->dump_flags = DF_VALID | DF_COMPLETE | DF_LIVE;
637 	dumphdr->dump_crashtime = gethrestime_sec();
638 	dumphdr->dump_npages = 0;
639 	dumphdr->dump_nvtop = 0;
640 	bzero(dump_bitmap, BT_SIZEOFMAP(dump_bitmapsize));
641 	dump_timeleft = dump_timeout;
642 
643 	if (panicstr) {
644 		dumphdr->dump_flags &= ~DF_LIVE;
645 		(void) VOP_DUMPCTL(dumpvp, DUMP_FREE, NULL, NULL);
646 		(void) VOP_DUMPCTL(dumpvp, DUMP_ALLOC, NULL, NULL);
647 		(void) vsnprintf(dumphdr->dump_panicstring, DUMP_PANICSIZE,
648 		    panicstr, panicargs);
649 	}
650 
651 	if (dump_conflags & DUMP_ALL)
652 		content = "all";
653 	else if (dump_conflags & DUMP_CURPROC)
654 		content = "kernel + curproc";
655 	else
656 		content = "kernel";
657 	uprintf("dumping to %s, offset %lld, content: %s\n", dumppath,
658 	    dumphdr->dump_start, content);
659 
660 	/*
661 	 * Leave room for the message and ereport save areas and terminal dump
662 	 * header.
663 	 */
664 	dumpvp_limit = dumpvp_size - DUMP_LOGSIZE - DUMP_OFFSET - DUMP_ERPTSIZE;
665 
666 	/*
667 	 * Write out the symbol table.  It's no longer compressed,
668 	 * so its 'size' and 'csize' are equal.
669 	 */
670 	dumpvp_off = dumphdr->dump_ksyms = dumphdr->dump_start + PAGESIZE;
671 	dumphdr->dump_ksyms_size = dumphdr->dump_ksyms_csize =
672 	    ksyms_snapshot(dumpvp_ksyms_write, NULL, LONG_MAX);
673 
674 	/*
675 	 * Write out the translation map.
676 	 */
677 	dumphdr->dump_map = dumpvp_flush();
678 	dump_as(&kas);
679 	dumphdr->dump_nvtop += dump_plat_addr();
680 
681 	/*
682 	 * call into hat, which may have unmapped pages that also need to
683 	 * be in the dump
684 	 */
685 	hat_dump();
686 
687 	if (dump_conflags & DUMP_ALL) {
688 		mutex_enter(&pidlock);
689 
690 		for (npids = 0, p = practive; p != NULL; p = p->p_next)
691 			dump_pids[npids++] = p->p_pid;
692 
693 		mutex_exit(&pidlock);
694 
695 		for (pidx = 0; pidx < npids; pidx++)
696 			(void) dump_process(dump_pids[pidx]);
697 
698 		for (bitnum = 0; bitnum < dump_bitmapsize; bitnum++) {
699 			dump_timeleft = dump_timeout;
700 			BT_SET(dump_bitmap, bitnum);
701 		}
702 		dumphdr->dump_npages = dump_bitmapsize;
703 		dumphdr->dump_flags |= DF_ALL;
704 
705 	} else if (dump_conflags & DUMP_CURPROC) {
706 		/*
707 		 * Determine which pid is to be dumped.  If we're panicking, we
708 		 * dump the process associated with panic_thread (if any).  If
709 		 * this is a live dump, we dump the process associated with
710 		 * curthread.
711 		 */
712 		npids = 0;
713 		if (panicstr) {
714 			if (panic_thread != NULL &&
715 			    panic_thread->t_procp != NULL &&
716 			    panic_thread->t_procp != &p0) {
717 				dump_pids[npids++] =
718 				    panic_thread->t_procp->p_pid;
719 			}
720 		} else {
721 			dump_pids[npids++] = curthread->t_procp->p_pid;
722 		}
723 
724 		if (npids && dump_process(dump_pids[0]) == 0)
725 			dumphdr->dump_flags |= DF_CURPROC;
726 		else
727 			dumphdr->dump_flags |= DF_KERNEL;
728 
729 	} else {
730 		dumphdr->dump_flags |= DF_KERNEL;
731 	}
732 
733 	dumphdr->dump_hashmask = (1 << highbit(dumphdr->dump_nvtop - 1)) - 1;
734 
735 	/*
736 	 * Write out the pfn table.
737 	 */
738 	dumphdr->dump_pfn = dumpvp_flush();
739 	for (bitnum = 0; bitnum < dump_bitmapsize; bitnum++) {
740 		dump_timeleft = dump_timeout;
741 		if (!BT_TEST(dump_bitmap, bitnum))
742 			continue;
743 		pfn = dump_bitnum_to_pfn(bitnum);
744 		ASSERT(pfn != PFN_INVALID);
745 		dumpvp_write(&pfn, sizeof (pfn_t));
746 	}
747 	dump_plat_pfn();
748 
749 	/*
750 	 * Write out all the pages.
751 	 */
752 	dumphdr->dump_data = dumpvp_flush();
753 	for (bitnum = 0; bitnum < dump_bitmapsize; bitnum++) {
754 		dump_timeleft = dump_timeout;
755 		if (!BT_TEST(dump_bitmap, bitnum))
756 			continue;
757 		pfn = dump_bitnum_to_pfn(bitnum);
758 		ASSERT(pfn != PFN_INVALID);
759 
760 		/*
761 		 * Map in page frame 'pfn', scan it for UE's while copying
762 		 * the data to dump_uebuf, unmap it, compress dump_uebuf into
763 		 * dump_cbuf, and write out dump_cbuf.  The UE check ensures
764 		 * that we don't lose the whole dump because of a latent UE.
765 		 */
766 		hat_devload(kas.a_hat, dump_cmap, PAGESIZE, pfn, PROT_READ,
767 		    HAT_LOAD_NOCONSIST);
768 		dump_pagecopy(dump_cmap, dump_uebuf);
769 		hat_unload(kas.a_hat, dump_cmap, PAGESIZE, HAT_UNLOAD);
770 		csize = (uint32_t)compress(dump_uebuf, dump_cbuf, PAGESIZE);
771 		dumpvp_write(&csize, sizeof (uint32_t));
772 		dumpvp_write(dump_cbuf, csize);
773 		if (dump_ioerr) {
774 			dumphdr->dump_flags &= ~DF_COMPLETE;
775 			dumphdr->dump_npages = npages;
776 			break;
777 		}
778 		total_csize += csize;
779 		if (++npages * 100LL / dumphdr->dump_npages > percent_done) {
780 			uprintf("^\r%3d%% done", ++percent_done);
781 			if (!panicstr)
782 				delay(1);	/* let the output be sent */
783 		}
784 	}
785 	dumphdr->dump_npages += dump_plat_data(dump_cbuf);
786 
787 	(void) dumpvp_flush();
788 
789 	/*
790 	 * Write out the initial and terminal dump headers.
791 	 */
792 	dumpvp_off = dumphdr->dump_start;
793 	dumpvp_write(dumphdr, sizeof (dumphdr_t));
794 	(void) dumpvp_flush();
795 
796 	dumpvp_limit = dumpvp_size;
797 	dumpvp_off = dumpvp_limit - DUMP_OFFSET;
798 	dumpvp_write(dumphdr, sizeof (dumphdr_t));
799 	(void) dumpvp_flush();
800 
801 	compress_ratio = (int)(100LL * npages / (btopr(total_csize + 1)));
802 
803 	uprintf("\r%3d%% done: %d pages dumped, compression ratio %d.%02d, ",
804 	    percent_done, npages, compress_ratio / 100, compress_ratio % 100);
805 
806 	if (dump_ioerr == 0) {
807 		uprintf("dump succeeded\n");
808 	} else {
809 		uprintf("dump failed: error %d\n", dump_ioerr);
810 		if (panicstr && dumpfaildebug)
811 			debug_enter("dump failed");
812 	}
813 
814 	/*
815 	 * Write out all undelivered messages.  This has to be the *last*
816 	 * thing we do because the dump process itself emits messages.
817 	 */
818 	if (panicstr) {
819 		dump_ereports();
820 		dump_messages();
821 	}
822 
823 	delay(2 * hz);	/* let people see the 'done' message */
824 	dump_timeleft = 0;
825 	dump_ioerr = 0;
826 }
827 
828 /*
829  * This function is called whenever the memory size, as represented
830  * by the phys_install list, changes.
831  */
832 void
833 dump_resize()
834 {
835 	mutex_enter(&dump_lock);
836 	dumphdr_init();
837 	dumpbuf_resize();
838 	mutex_exit(&dump_lock);
839 }
840 
841 /*
842  * This function allows for dynamic resizing of a dump area. It assumes that
843  * the underlying device has update its appropriate size(9P).
844  */
845 int
846 dumpvp_resize()
847 {
848 	int error;
849 	vattr_t vattr;
850 
851 	mutex_enter(&dump_lock);
852 	vattr.va_mask = AT_SIZE;
853 	if ((error = VOP_GETATTR(dumpvp, &vattr, 0, kcred, NULL)) != 0) {
854 		mutex_exit(&dump_lock);
855 		return (error);
856 	}
857 
858 	if (error == 0 && vattr.va_size < 2 * DUMP_LOGSIZE + DUMP_ERPTSIZE) {
859 		mutex_exit(&dump_lock);
860 		return (ENOSPC);
861 	}
862 
863 	dumpvp_size = vattr.va_size & -DUMP_OFFSET;
864 	mutex_exit(&dump_lock);
865 	return (0);
866 }
867