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)
110 if ((sz & PAGE_MASK)) {
111 kprintf("size not page aligned\n");
114 if (ptr != NULL && pa != 0) {
115 kprintf("can't have both va and pa!\n");
118 if (pa != 0 && (((uintptr_t)pa) & PAGE_MASK) != 0) {
119 kprintf("address not page aligned\n");
124 * If we're doing a virtual dump, flush any
125 * pre-existing pa pages
127 error = blk_flush(di);
131 max_iosize = min(MAXPHYS, di->maxiosize);
133 len = max_iosize - fragsz;
139 kprintf(" %ld", PG2MB(progress >> PAGE_SHIFT));
140 counter &= (1<<24) - 1;
144 error = dev_ddump(di->priv, ptr, 0, dumplo, len);
152 for (i = 0; i < len; i += PAGE_SIZE) {
153 dump_va = pmap_kenter_temporary(pa + i,
154 (i + fragsz) >> PAGE_SHIFT);
160 if (fragsz == max_iosize) {
161 error = blk_flush(di);
168 /* Check for user abort. */
173 kprintf(" (CTRL-C to abort) ");
178 /* A fake page table page, to avoid having to handle both 4K and 2M pages */
179 static pt_entry_t fakept[NPTEPG];
182 minidumpsys(struct dumperinfo *di)
187 vm_offset_t kern_end;
190 uint64_t *pdp, *pd, *pt, pa;
192 struct minidumphdr mdhdr;
193 struct mdglobaldata *md;
197 * Walk page table pages, set bits in vm_page_dump.
199 * NOTE: kernel_vm_end can actually be below KERNBASE.
200 * Just use KvaEnd. Also note that loops which go
201 * all the way to the end of the address space might
202 * overflow the loop variable.
206 md = (struct mdglobaldata *)globaldata_find(0);
209 if (kern_end < (vm_offset_t)&(md[ncpus]))
210 kern_end = (vm_offset_t)&(md[ncpus]);
212 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
213 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
215 * The loop probably overflows a 64-bit int due to NBPDR.
217 if (va < VM_MIN_KERNEL_ADDRESS)
221 * We always write a page, even if it is zero. Each
222 * page written corresponds to 2MB of space
224 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
225 ptesize += PAGE_SIZE;
226 if ((pdp[i] & PG_V) == 0)
228 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
229 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
230 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
231 /* This is an entire 2M page. */
232 pa = pd[j] & PG_PS_FRAME;
233 for (k = 0; k < NPTEPG; k++) {
240 if ((pd[j] & PG_V) == PG_V) {
241 /* set bit for each valid page in this 2MB block */
242 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
243 for (k = 0; k < NPTEPG; k++) {
244 if ((pt[k] & PG_V) == PG_V) {
245 pa = pt[k] & PG_FRAME;
251 /* nothing, we're going to dump a null page */
255 /* Calculate dump size. */
257 dumpsize += round_page(msgbufp->msg_size);
258 dumpsize += round_page(vm_page_dump_size);
259 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
260 bits = vm_page_dump[i];
263 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
264 /* Clear out undumpable pages now if needed */
265 if (is_dumpable(pa)) {
266 dumpsize += PAGE_SIZE;
270 bits &= ~(1ul << bit);
273 dumpsize += PAGE_SIZE;
275 /* Determine dump offset on device. */
276 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
280 dumplo = di->mediaoffset + di->mediasize - dumpsize;
281 dumplo -= sizeof(kdh) * 2;
284 /* Initialize mdhdr */
285 bzero(&mdhdr, sizeof(mdhdr));
286 strcpy(mdhdr.magic, MINIDUMP_MAGIC);
287 mdhdr.version = MINIDUMP_VERSION;
288 mdhdr.msgbufsize = msgbufp->msg_size;
289 mdhdr.bitmapsize = vm_page_dump_size;
290 mdhdr.ptesize = ptesize;
291 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
292 mdhdr.dmapbase = DMAP_MIN_ADDRESS;
293 mdhdr.dmapend = DMAP_MAX_ADDRESS;
295 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION,
296 dumpsize, di->blocksize);
298 kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576);
299 kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20);
302 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
305 dumplo += sizeof(kdh);
308 bzero(&fakept, sizeof(fakept));
309 bcopy(&mdhdr, &fakept, sizeof(mdhdr));
310 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
314 /* Dump msgbuf up front */
315 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
320 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
324 /* Dump kernel page table pages */
325 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
326 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
328 * The loop probably overflows a 64-bit int due to NBPDR.
330 if (va < VM_MIN_KERNEL_ADDRESS)
334 * We always write a page, even if it is zero
336 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
337 if ((pdp[i] & PG_V) == 0) {
338 bzero(fakept, sizeof(fakept));
339 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
342 /* flush, in case we reuse fakept in the same block */
343 error = blk_flush(di);
348 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
349 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
350 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) {
351 /* This is a single 2M block. Generate a fake PTP */
352 pa = pd[j] & PG_PS_FRAME;
353 for (k = 0; k < NPTEPG; k++) {
354 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
356 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
359 /* flush, in case we reuse fakept in the same block */
360 error = blk_flush(di);
365 if ((pd[j] & PG_V) == PG_V) {
366 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
367 error = blk_write(di, (char *)pt, 0, PAGE_SIZE);
371 bzero(fakept, sizeof(fakept));
372 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
375 /* flush, in case we reuse fakept in the same block */
376 error = blk_flush(di);
382 /* Dump memory chunks */
383 /* XXX cluster it up and use blk_dump() */
384 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
385 bits = vm_page_dump[i];
388 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
389 error = blk_write(di, 0, pa, PAGE_SIZE);
392 bits &= ~(1ul << bit);
396 error = blk_flush(di);
401 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
404 dumplo += sizeof(kdh);
406 /* Signal completion, signoff and exit stage left. */
407 dev_ddump(di->priv, NULL, 0, 0, 0);
408 kprintf("\nDump complete\n");
415 if (error == ECANCELED)
416 kprintf("\nDump aborted\n");
417 else if (error == ENOSPC)
418 kprintf("\nDump failed. Partition too small.\n");
420 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error);
424 dump_add_page(vm_paddr_t pa)
429 idx = pa >> 6; /* 2^6 = 64 */
431 atomic_set_long(&vm_page_dump[idx], 1ul << bit);
435 dump_drop_page(vm_paddr_t pa)
440 idx = pa >> 6; /* 2^6 = 64 */
442 atomic_clear_long(&vm_page_dump[idx], 1ul << bit);