Merge minidumps from amd64 where they were originally developed.

Major differences:
 * since there is no direct map region, there is no custom uma memory
   allocator to modify to include its pages in the dumps.
 * Various data entries are reduced from 64 bit to 32 bit to match the
   native size.

dump_add_page() and dump_drop_page() are still present in case one wants to
arrange for arbitary pages to be dumped.  This is of marginal use though
because libkvm+kgdb cannot address physical memory that isn't mapped into
kvm.

4 files changed, 466 insertions, 0 deletions

diff --git a/sys/i386/i386/dump_machdep.c b/sys/i386/i386/dump_machdep.c
index 670fa464392f..eb7037137c17 100644
--- a/sys/i386/i386/dump_machdep.c
+++ b/sys/i386/i386/dump_machdep.c
@@ -31,6 +31,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/systm.h>
 #include <sys/conf.h>
 #include <sys/cons.h>
+#include <sys/sysctl.h>
 #include <sys/kernel.h>
 #include <sys/kerneldump.h>
 #include <vm/vm.h>
@@ -40,6 +41,11 @@ __FBSDID("$FreeBSD$");
 
 CTASSERT(sizeof(struct kerneldumpheader) == 512);
 
+int do_minidump = 1;
+TUNABLE_INT("debug.minidump", &do_minidump);
+SYSCTL_INT(_debug, OID_AUTO, minidump, CTLFLAG_RW, &do_minidump, 0,
+    "Enable mini crash dumps");
+
 /*
  * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
  * is to protect us from metadata and to protect metadata from us.
@@ -272,6 +278,10 @@ dumpsys(struct dumperinfo *di)
 	size_t hdrsz;
 	int error;
 
+	if (do_minidump) {
+		minidumpsys(di);
+		return;
+	}
 	bzero(&ehdr, sizeof(ehdr));
 	ehdr.e_ident[EI_MAG0] = ELFMAG0;
 	ehdr.e_ident[EI_MAG1] = ELFMAG1;
diff --git a/sys/i386/i386/minidump_machdep.c b/sys/i386/i386/minidump_machdep.c
new file mode 100644
index 000000000000..66db12383169
--- /dev/null
+++ b/sys/i386/i386/minidump_machdep.c
@@ -0,0 +1,405 @@
+/*-
+ * Copyright (c) 2006 Peter Wemm
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/conf.h>
+#include <sys/cons.h>
+#include <sys/kernel.h>
+#include <sys/kerneldump.h>
+#include <sys/msgbuf.h>
+#include <vm/vm.h>
+#include <vm/pmap.h>
+#include <machine/atomic.h>
+#include <machine/elf.h>
+#include <machine/md_var.h>
+#include <machine/vmparam.h>
+#include <machine/minidump.h>
+
+CTASSERT(sizeof(struct kerneldumpheader) == 512);
+
+/*
+ * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
+ * is to protect us from metadata and to protect metadata from us.
+ */
+#define	SIZEOF_METADATA		(64*1024)
+
+#define	MD_ALIGN(x)	(((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
+#define	DEV_ALIGN(x)	(((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
+
+uint32_t *vm_page_dump;
+int vm_page_dump_size;
+
+static struct kerneldumpheader kdh;
+static off_t dumplo;
+
+/* Handle chunked writes. */
+static size_t fragsz;
+static void *dump_va;
+static uint64_t counter, progress;
+
+CTASSERT(sizeof(*vm_page_dump) == 4);
+
+static int
+is_dumpable(vm_paddr_t pa)
+{
+	int i;
+
+	for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
+		if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
+			return (1);
+	}
+	return (0);
+}
+
+/* XXX should be MI */
+static void
+mkdumpheader(struct kerneldumpheader *kdh, uint32_t archver, uint64_t dumplen,
+    uint32_t blksz)
+{
+
+	bzero(kdh, sizeof(*kdh));
+	strncpy(kdh->magic, KERNELDUMPMAGIC, sizeof(kdh->magic));
+	strncpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
+	kdh->version = htod32(KERNELDUMPVERSION);
+	kdh->architectureversion = htod32(archver);
+	kdh->dumplength = htod64(dumplen);
+	kdh->dumptime = htod64(time_second);
+	kdh->blocksize = htod32(blksz);
+	strncpy(kdh->hostname, hostname, sizeof(kdh->hostname));
+	strncpy(kdh->versionstring, version, sizeof(kdh->versionstring));
+	if (panicstr != NULL)
+		strncpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
+	kdh->parity = kerneldump_parity(kdh);
+}
+
+#define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
+
+static int
+blk_flush(struct dumperinfo *di)
+{
+	int error;
+
+	if (fragsz == 0)
+		return (0);
+
+	error = di->dumper(di->priv, dump_va, 0, dumplo, fragsz);
+	dumplo += fragsz;
+	fragsz = 0;
+	return (error);
+}
+
+static int
+blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
+{
+	size_t len;
+	int error, i, c;
+
+	error = 0;
+	if ((sz % PAGE_SIZE) != 0) {
+		printf("size not page aligned\n");
+		return (EINVAL);
+	}
+	if (ptr != NULL && pa != 0) {
+		printf("cant have both va and pa!\n");
+		return (EINVAL);
+	}
+	if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) {
+		printf("address not page aligned\n");
+		return (EINVAL);
+	}
+	if (ptr != NULL) {
+		/* If we're doing a virtual dump, flush any pre-existing pa pages */
+		error = blk_flush(di);
+		if (error)
+			return (error);
+	}
+	while (sz) {
+		len = (MAXDUMPPGS * PAGE_SIZE) - fragsz;
+		if (len > sz)
+			len = sz;
+		counter += len;
+		progress -= len;
+		if (counter >> 24) {
+			printf(" %lld", PG2MB(progress >> PAGE_SHIFT));
+			counter &= (1<<24) - 1;
+		}
+		if (ptr) {
+			error = di->dumper(di->priv, ptr, 0, dumplo, len);
+			if (error)
+				return (error);
+			dumplo += len;
+			ptr += len;
+			sz -= len;
+		} else {
+			for (i = 0; i < len; i += PAGE_SIZE)
+				dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
+			fragsz += len;
+			pa += len;
+			sz -= len;
+			if (fragsz == (MAXDUMPPGS * PAGE_SIZE)) {
+				error = blk_flush(di);
+				if (error)
+					return (error);
+			}
+		}
+
+		/* Check for user abort. */
+		c = cncheckc();
+		if (c == 0x03)
+			return (ECANCELED);
+		if (c != -1)
+			printf(" (CTRL-C to abort) ");
+	}
+
+	return (0);
+}
+
+/* A fake page table page, to avoid having to handle both 4K and 2M pages */
+static pt_entry_t fakept[NPTEPG];
+
+void
+minidumpsys(struct dumperinfo *di)
+{
+	uint64_t dumpsize;
+	uint32_t ptesize;
+	vm_offset_t va;
+	int error;
+	uint32_t bits;
+	uint64_t pa;
+	pd_entry_t *pd;
+	pt_entry_t *pt;
+	int i, j, k, bit;
+	struct minidumphdr mdhdr;
+
+	counter = 0;
+	/* Walk page table pages, set bits in vm_page_dump */
+	ptesize = 0;
+	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
+		/*
+		 * We always write a page, even if it is zero. Each
+		 * page written corresponds to 2MB of space
+		 */
+		ptesize += PAGE_SIZE;
+		pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE);	/* always mapped! */
+		j = va >> PDRSHIFT;
+		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
+			/* This is an entire 2M page. */
+			pa = pd[j] & PG_FRAME & ~PDRMASK;
+			for (k = 0; k < NPTEPG; k++) {
+				if (is_dumpable(pa))
+					dump_add_page(pa);
+				pa += PAGE_SIZE;
+			}
+			continue;
+		}
+		if ((pd[j] & PG_V) == PG_V) {
+			/* set bit for each valid page in this 2MB block */
+			pt = pmap_kenter_temporary(pd[j] & PG_FRAME, 0);
+			for (k = 0; k < NPTEPG; k++) {
+				if ((pt[k] & PG_V) == PG_V) {
+					pa = pt[k] & PG_FRAME;
+					if (is_dumpable(pa))
+						dump_add_page(pa);
+				}
+			}
+		} else {
+			/* nothing, we're going to dump a null page */
+		}
+	}
+
+	/* Calculate dump size. */
+	dumpsize = ptesize;
+	dumpsize += round_page(msgbufp->msg_size);
+	dumpsize += round_page(vm_page_dump_size);
+	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
+		bits = vm_page_dump[i];
+		while (bits) {
+			bit = bsfl(bits);
+			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
+			/* Clear out undumpable pages now if needed */
+			if (is_dumpable(pa)) {
+				dumpsize += PAGE_SIZE;
+			} else {
+				dump_drop_page(pa);
+			}
+			bits &= ~(1ul << bit);
+		}
+	}
+	dumpsize += PAGE_SIZE;
+
+	/* Determine dump offset on device. */
+	if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
+		error = ENOSPC;
+		goto fail;
+	}
+	dumplo = di->mediaoffset + di->mediasize - dumpsize;
+	dumplo -= sizeof(kdh) * 2;
+	progress = dumpsize;
+
+	/* Initialize mdhdr */
+	bzero(&mdhdr, sizeof(mdhdr));
+	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
+	mdhdr.version = MINIDUMP_VERSION;
+	mdhdr.msgbufsize = msgbufp->msg_size;
+	mdhdr.bitmapsize = vm_page_dump_size;
+	mdhdr.ptesize = ptesize;
+	mdhdr.kernbase = KERNBASE;
+#ifdef PAE
+	mdhdr.paemode = 1;
+#endif
+
+	mkdumpheader(&kdh, KERNELDUMP_AMD64_VERSION, dumpsize, di->blocksize);
+
+	printf("Physical memory: %ju MB\n", ptoa((uintmax_t)physmem) / 1048576);
+	printf("Dumping %llu MB:", (long long)dumpsize >> 20);
+
+	/* Dump leader */
+	error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
+	if (error)
+		goto fail;
+	dumplo += sizeof(kdh);
+
+	/* Dump my header */
+	bzero(&fakept, sizeof(fakept));
+	bcopy(&mdhdr, &fakept, sizeof(mdhdr));
+	error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
+	if (error)
+		goto fail;
+
+	/* Dump msgbuf up front */
+	error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
+	if (error)
+		goto fail;
+
+	/* Dump bitmap */
+	error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
+	if (error)
+		goto fail;
+
+	/* Dump kernel page table pages */
+	for (va = KERNBASE; va < kernel_vm_end; va += NBPDR) {
+		/* We always write a page, even if it is zero */
+		pd = (pd_entry_t *)((uintptr_t)IdlePTD + KERNBASE);	/* always mapped! */
+		j = va >> PDRSHIFT;
+		if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
+			/* This is a single 2M block. Generate a fake PTP */
+			pa = pd[j] & PG_FRAME & ~PDRMASK;
+			for (k = 0; k < NPTEPG; k++) {
+				fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
+			}
+			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
+			if (error)
+				goto fail;
+			/* flush, in case we reuse fakept in the same block */
+			error = blk_flush(di);
+			if (error)
+				goto fail;
+			continue;
+		}
+		if ((pd[j] & PG_V) == PG_V) {
+			pa = pd[j] & PG_FRAME;
+			error = blk_write(di, 0, pa, PAGE_SIZE);
+			if (error)
+				goto fail;
+		} else {
+			bzero(fakept, sizeof(fakept));
+			error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
+			if (error)
+				goto fail;
+			/* flush, in case we reuse fakept in the same block */
+			error = blk_flush(di);
+			if (error)
+				goto fail;
+		}
+	}
+
+	/* Dump memory chunks */
+	/* XXX cluster it up and use blk_dump() */
+	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
+		bits = vm_page_dump[i];
+		while (bits) {
+			bit = bsfl(bits);
+			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
+			error = blk_write(di, 0, pa, PAGE_SIZE);
+			if (error)
+				goto fail;
+			bits &= ~(1ul << bit);
+		}
+	}
+
+	error = blk_flush(di);
+	if (error)
+		goto fail;
+
+	/* Dump trailer */
+	error = di->dumper(di->priv, &kdh, 0, dumplo, sizeof(kdh));
+	if (error)
+		goto fail;
+	dumplo += sizeof(kdh);
+
+	/* Signal completion, signoff and exit stage left. */
+	di->dumper(di->priv, NULL, 0, 0, 0);
+	printf("\nDump complete\n");
+	return;
+
+ fail:
+	if (error < 0)
+		error = -error;
+
+	if (error == ECANCELED)
+		printf("\nDump aborted\n");
+	else if (error == ENOSPC)
+		printf("\nDump failed. Partition too small.\n");
+	else
+		printf("\n** DUMP FAILED (ERROR %d) **\n", error);
+}
+
+void
+dump_add_page(vm_paddr_t pa)
+{
+	int idx, bit;
+
+	pa >>= PAGE_SHIFT;
+	idx = pa >> 5;		/* 2^5 = 32 */
+	bit = pa & 31;
+	atomic_set_int(&vm_page_dump[idx], 1ul << bit);
+}
+
+void
+dump_drop_page(vm_paddr_t pa)
+{
+	int idx, bit;
+
+	pa >>= PAGE_SHIFT;
+	idx = pa >> 5;		/* 2^5 = 32 */
+	bit = pa & 31;
+	atomic_clear_int(&vm_page_dump[idx], 1ul << bit);
+}
+
diff --git a/sys/i386/include/md_var.h b/sys/i386/include/md_var.h
index afccf1d0c1a9..92b7152fa6d1 100644
--- a/sys/i386/include/md_var.h
+++ b/sys/i386/include/md_var.h
@@ -65,12 +65,15 @@ extern	int	szfreebsd4_sigcode;
 #ifdef COMPAT_43
 extern	int	szosigcode;
 #endif
+extern	uint32_t *vm_page_dump;
+extern	int	vm_page_dump_size;
 
 typedef void alias_for_inthand_t(u_int cs, u_int ef, u_int esp, u_int ss);
 struct	thread;
 struct	reg;
 struct	fpreg;
 struct  dbreg;
+struct	dumperinfo;
 
 void	bcopyb(const void *from, void *to, size_t len);
 void	busdma_swi(void);
@@ -84,6 +87,8 @@ void	doreti_popl_es(void) __asm(__STRING(doreti_popl_es));
 void	doreti_popl_es_fault(void) __asm(__STRING(doreti_popl_es_fault));
 void	doreti_popl_fs(void) __asm(__STRING(doreti_popl_fs));
 void	doreti_popl_fs_fault(void) __asm(__STRING(doreti_popl_fs_fault));
+void	dump_add_page(vm_paddr_t);
+void	dump_drop_page(vm_paddr_t);
 void	enable_sse(void);
 void	fillw(int /*u_short*/ pat, void *base, size_t cnt);
 void	i486_bzero(void *buf, size_t len);
@@ -99,5 +104,6 @@ int	isa_nmi(int cd);
 vm_paddr_t kvtop(void *addr);
 void	setidt(int idx, alias_for_inthand_t *func, int typ, int dpl, int selec);
 int     user_dbreg_trap(void);
+void	minidumpsys(struct dumperinfo *);
 
 #endif /* !_MACHINE_MD_VAR_H_ */
diff --git a/sys/i386/include/minidump.h b/sys/i386/include/minidump.h
new file mode 100644
index 000000000000..55508a370e70
--- /dev/null
+++ b/sys/i386/include/minidump.h
@@ -0,0 +1,45 @@
+/*-
+ * Copyright (c) 2006 Peter Wemm
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef	_MACHINE_MINIDUMP_H_
+#define	_MACHINE_MINIDUMP_H_ 1
+
+#define	MINIDUMP_MAGIC		"minidump FreeBSD/i386"
+#define	MINIDUMP_VERSION	1
+
+struct minidumphdr {
+	char magic[24];
+	uint32_t version;
+	uint32_t msgbufsize;
+	uint32_t bitmapsize;
+	uint32_t ptesize;
+	uint32_t kernbase;
+	uint32_t paemode;
+};
+
+#endif /* _MACHINE_MINIDUMP_H_ */