2 * Copyright (c) 2006 Peter Wemm
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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.
26 * $FreeBSD: src/sys/amd64/amd64/minidump_machdep.c,v 1.10 2009/05/29 21:27:12 jamie Exp $
29 #include <sys/param.h>
30 #include <sys/systm.h>
33 #include <sys/device.h>
34 #include <sys/globaldata.h>
35 #include <sys/kernel.h>
36 #include <sys/kerneldump.h>
37 #include <sys/msgbuf.h>
39 #include <vm/vm_kern.h>
41 #include <machine/atomic.h>
42 #include <machine/elf.h>
43 #include <machine/globaldata.h>
44 #include <machine/md_var.h>
45 #include <machine/vmparam.h>
46 #include <machine/minidump.h>
48 CTASSERT(sizeof(struct kerneldumpheader) == 512);
51 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
52 * is to protect us from metadata and to protect metadata from us.
54 #define SIZEOF_METADATA (64*1024)
56 #define MD_ALIGN(x) (((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
57 #define DEV_ALIGN(x) (((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
59 extern uint64_t KPDPphys;
61 uint64_t *vm_page_dump;
62 int vm_page_dump_size;
64 static struct kerneldumpheader kdh;
67 /* Handle chunked writes. */
70 static size_t counter, progress;
72 CTASSERT(sizeof(*vm_page_dump) == 8);
75 is_dumpable(vm_paddr_t pa)
79 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
80 if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
86 #define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
89 blk_flush(struct dumperinfo *di)
96 error = dev_ddump(di->priv, dump_va, 0, dumplo, fragsz);
103 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
109 if ((sz & PAGE_MASK)) {
110 kprintf("size not page aligned\n");
113 if (ptr != NULL && pa != 0) {
114 kprintf("can't have both va and pa!\n");
117 if (pa != 0 && (((uintptr_t)pa) & PAGE_MASK) != 0) {
118 kprintf("address not page aligned\n");
123 * If we're doing a virtual dump, flush any
124 * pre-existing pa pages
126 error = blk_flush(di);
131 len = (MAXDUMPPGS * PAGE_SIZE) - fragsz;
137 kprintf(" %ld", PG2MB(progress >> PAGE_SHIFT));
138 counter &= (1<<24) - 1;
142 error = dev_ddump(di->priv, ptr, 0, dumplo, len);
150 for (i = 0; i < len; i += PAGE_SIZE) {
151 dump_va = pmap_kenter_temporary(pa + i,
152 (i + fragsz) >> PAGE_SHIFT);
158 if (fragsz == (MAXDUMPPGS * PAGE_SIZE)) {
159 error = blk_flush(di);
166 /* Check for user abort. */
171 kprintf(" (CTRL-C to abort) ");
176 /* A fake page table page, to avoid having to handle both 4K and 2M pages */
177 static pt_entry_t fakept[NPTEPG];
180 minidumpsys(struct dumperinfo *di)
185 vm_offset_t kern_end;
188 uint64_t *pdp, *pd, *pt, pa;
190 struct minidumphdr mdhdr;
191 struct mdglobaldata *md;
195 * Walk page table pages, set bits in vm_page_dump.
197 * NOTE: kernel_vm_end can actually be below KERNBASE.
198 * Just use KvaEnd. Also note that loops which go
199 * all the way to the end of the address space might
200 * overflow the loop variable.
204 md = (struct mdglobaldata *)globaldata_find(0);
207 if (kern_end < (vm_offset_t)&(md[ncpus]))
208 kern_end = (vm_offset_t)&(md[ncpus]);
210 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
211 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
213 * The loop probably overflows a 64-bit int due to NBPDR.
215 if (va < VM_MIN_KERNEL_ADDRESS)
219 * We always write a page, even if it is zero. Each
220 * page written corresponds to 2MB of space
222 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
223 ptesize += PAGE_SIZE;
224 if ((pdp[i] & PG_V) == 0)
226 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
227 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
228 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
229 /* This is an entire 2M page. */
230 pa = pd[j] & PG_PS_FRAME;
231 for (k = 0; k < NPTEPG; k++) {
238 if ((pd[j] & PG_V) == PG_V) {
239 /* set bit for each valid page in this 2MB block */
240 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
241 for (k = 0; k < NPTEPG; k++) {
242 if ((pt[k] & PG_V) == PG_V) {
243 pa = pt[k] & PG_FRAME;
249 /* nothing, we're going to dump a null page */
253 /* Calculate dump size. */
255 dumpsize += round_page(msgbufp->msg_size);
256 dumpsize += round_page(vm_page_dump_size);
257 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
258 bits = vm_page_dump[i];
261 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
262 /* Clear out undumpable pages now if needed */
263 if (is_dumpable(pa)) {
264 dumpsize += PAGE_SIZE;
268 bits &= ~(1ul << bit);
271 dumpsize += PAGE_SIZE;
273 /* Determine dump offset on device. */
274 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
278 dumplo = di->mediaoffset + di->mediasize - dumpsize;
279 dumplo -= sizeof(kdh) * 2;
282 /* Initialize mdhdr */
283 bzero(&mdhdr, sizeof(mdhdr));
284 strcpy(mdhdr.magic, MINIDUMP_MAGIC);
285 mdhdr.version = MINIDUMP_VERSION;
286 mdhdr.msgbufsize = msgbufp->msg_size;
287 mdhdr.bitmapsize = vm_page_dump_size;
288 mdhdr.ptesize = ptesize;
289 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
290 mdhdr.dmapbase = DMAP_MIN_ADDRESS;
291 mdhdr.dmapend = DMAP_MAX_ADDRESS;
293 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION,
294 dumpsize, di->blocksize);
296 kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576);
297 kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20);
300 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
303 dumplo += sizeof(kdh);
306 bzero(&fakept, sizeof(fakept));
307 bcopy(&mdhdr, &fakept, sizeof(mdhdr));
308 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
312 /* Dump msgbuf up front */
313 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
318 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
322 /* Dump kernel page table pages */
323 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
324 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
326 * The loop probably overflows a 64-bit int due to NBPDR.
328 if (va < VM_MIN_KERNEL_ADDRESS)
332 * We always write a page, even if it is zero
334 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
335 if ((pdp[i] & PG_V) == 0) {
336 bzero(fakept, sizeof(fakept));
337 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
340 /* flush, in case we reuse fakept in the same block */
341 error = blk_flush(di);
346 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
347 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
348 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
349 /* This is a single 2M block. Generate a fake PTP */
350 pa = pd[j] & PG_PS_FRAME;
351 for (k = 0; k < NPTEPG; k++) {
352 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
354 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
357 /* flush, in case we reuse fakept in the same block */
358 error = blk_flush(di);
363 if ((pd[j] & PG_V) == PG_V) {
364 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
365 error = blk_write(di, (char *)pt, 0, PAGE_SIZE);
369 bzero(fakept, sizeof(fakept));
370 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
373 /* flush, in case we reuse fakept in the same block */
374 error = blk_flush(di);
380 /* Dump memory chunks */
381 /* XXX cluster it up and use blk_dump() */
382 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
383 bits = vm_page_dump[i];
386 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
387 error = blk_write(di, 0, pa, PAGE_SIZE);
390 bits &= ~(1ul << bit);
394 error = blk_flush(di);
399 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
402 dumplo += sizeof(kdh);
404 /* Signal completion, signoff and exit stage left. */
405 dev_ddump(di->priv, NULL, 0, 0, 0);
406 kprintf("\nDump complete\n");
413 if (error == ECANCELED)
414 kprintf("\nDump aborted\n");
415 else if (error == ENOSPC)
416 kprintf("\nDump failed. Partition too small.\n");
418 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error);
422 dump_add_page(vm_paddr_t pa)
427 idx = pa >> 6; /* 2^6 = 64 */
429 atomic_set_long(&vm_page_dump[idx], 1ul << bit);
433 dump_drop_page(vm_paddr_t pa)
438 idx = pa >> 6; /* 2^6 = 64 */
440 atomic_clear_long(&vm_page_dump[idx], 1ul << bit);