MFC 259910, 260023, 260028, 260600 & 260701:

o   Fix "kptdir is itself virtual" error, caused by having the kptdir in PBVM.
o   Allow building a cross libkvm for ia64.
o   Add support for virtual cores (aka minidumps).
o   We don't have to worry about page sizes when working on virtual cores.
o   Handle truncation of the size returned by _kvm_kvatop().

1 files changed, 158 insertions, 75 deletions

diff --git a/lib/libkvm/kvm_ia64.c b/lib/libkvm/kvm_ia64.c
index 74e2b80..5db7e1e 100644
--- a/lib/libkvm/kvm_ia64.c
+++ b/lib/libkvm/kvm_ia64.c
@@ -32,12 +32,21 @@
 #include <sys/elf64.h>
 #include <sys/mman.h>
 
+#ifndef CROSS_LIBKVM
 #include <machine/atomic.h>
 #include <machine/bootinfo.h>
+#include <machine/elf.h>
 #include <machine/pte.h>
+#else
+#include "../../sys/ia64/include/atomic.h"
+#include "../../sys/ia64/include/bootinfo.h"
+#include "../../sys/ia64/include/elf.h"
+#include "../../sys/ia64/include/pte.h"
+#endif
 
 #include <kvm.h>
 #include <limits.h>
+#include <stdint.h>
 #include <stdlib.h>
 #include <unistd.h>
 
@@ -55,6 +64,8 @@
 #define	PBVM_BASE		0x9ffc000000000000UL
 #define	PBVM_PGSZ		(64 * 1024)
 
+typedef size_t (a2p_f)(kvm_t *, uint64_t, off_t *);
+
 struct vmstate {
 	void	*mmapbase;
 	size_t	mmapsize;
@@ -62,6 +73,7 @@ struct vmstate {
 	u_long	kptdir;
 	u_long	*pbvm_pgtbl;
 	u_int	pbvm_pgtblsz;
+	a2p_f	*kvatop;
 };
 
 /*
@@ -70,7 +82,7 @@ struct vmstate {
  * set of headers.
  */
 static int
-_kvm_maphdrs(kvm_t *kd, size_t sz)
+ia64_maphdrs(kvm_t *kd, size_t sz)
 {
 	struct vmstate *vm = kd->vmst;
 
@@ -91,38 +103,103 @@ _kvm_maphdrs(kvm_t *kd, size_t sz)
 }
 
 /*
- * Translate a physical memory address to a file-offset in the crash-dump.
+ * Physical core support.
  */
+
 static size_t
-_kvm_pa2off(kvm_t *kd, uint64_t pa, off_t *ofs, size_t pgsz)
+phys_addr2off(kvm_t *kd, uint64_t pa, off_t *ofs, size_t pgsz)
 {
-	Elf64_Ehdr *e = kd->vmst->mmapbase;
-	Elf64_Phdr *p = (Elf64_Phdr*)((char*)e + e->e_phoff);
-	int n = e->e_phnum;
+	Elf64_Ehdr *e;
+	Elf64_Phdr *p;
+	int n;
 
-	if (pa != REGION_ADDR(pa)) {
-		_kvm_err(kd, kd->program, "internal error");
-		return (0);
-	}
+	if (pa != REGION_ADDR(pa))
+		goto fail;
 
+	e = (Elf64_Ehdr *)(kd->vmst->mmapbase);
+	n = e->e_phnum;
+	p = (Elf64_Phdr *)(void *)((uintptr_t)(void *)e + e->e_phoff);
 	while (n && (pa < p->p_paddr || pa >= p->p_paddr + p->p_memsz))
 		p++, n--;
 	if (n == 0)
-		return (0);
+		goto fail;
 
 	*ofs = (pa - p->p_paddr) + p->p_offset;
 	if (pgsz == 0)
 		return (p->p_memsz - (pa - p->p_paddr));
 	return (pgsz - ((size_t)pa & (pgsz - 1)));
+
+ fail:
+	_kvm_err(kd, kd->program, "invalid physical address %#jx",
+	    (uintmax_t)pa);
+	return (0);
+}
+
+static size_t
+phys_kvatop(kvm_t *kd, uint64_t va, off_t *ofs)
+{
+	struct ia64_lpte pte;
+	uint64_t pa, pgaddr, pt0addr, pt1addr;
+	size_t pgno, pgsz, pt0no, pt1no;
+
+	if (va >= REGION_BASE(6)) {
+		/* Regions 6 and 7: direct mapped. */
+		pa = REGION_ADDR(va);
+		return (phys_addr2off(kd, pa, ofs, 0));
+	} else if (va >= REGION_BASE(5)) {
+		/* Region 5: Kernel Virtual Memory. */
+		va = REGION_ADDR(va);
+		pgsz = kd->vmst->pagesize;
+		pt0no = KPTE_DIR0_INDEX(va, pgsz);
+		pt1no = KPTE_DIR1_INDEX(va, pgsz);
+		pgno = KPTE_PTE_INDEX(va, pgsz);
+		if (pt0no >= NKPTEDIR(pgsz))
+			goto fail;
+		pt0addr = kd->vmst->kptdir + (pt0no << 3);
+		if (kvm_read(kd, pt0addr, &pt1addr, 8) != 8)
+			goto fail;
+		if (pt1addr == 0)
+			goto fail;
+		pt1addr += pt1no << 3;
+		if (kvm_read(kd, pt1addr, &pgaddr, 8) != 8)
+			goto fail;
+		if (pgaddr == 0)
+			goto fail;
+		pgaddr += pgno * sizeof(pte);
+		if (kvm_read(kd, pgaddr, &pte, sizeof(pte)) != sizeof(pte))
+			goto fail;
+		if (!(pte.pte & PTE_PRESENT))
+			goto fail;
+		pa = (pte.pte & PTE_PPN_MASK) + (va & (pgsz - 1));
+		return (phys_addr2off(kd, pa, ofs, pgsz));
+	} else if (va >= PBVM_BASE) {
+		/* Region 4: Pre-Boot Virtual Memory (PBVM). */
+		va -= PBVM_BASE;
+		pgsz = PBVM_PGSZ;
+		pt0no = va / pgsz;
+		if (pt0no >= (kd->vmst->pbvm_pgtblsz >> 3))
+			goto fail;
+		pt0addr = kd->vmst->pbvm_pgtbl[pt0no];
+		if (!(pt0addr & PTE_PRESENT))
+			goto fail;
+		pa = (pt0addr & PTE_PPN_MASK) + va % pgsz;
+		return (phys_addr2off(kd, pa, ofs, pgsz));
+	}
+
+ fail:
+	_kvm_err(kd, kd->program, "invalid kernel virtual address %#jx",
+	    (uintmax_t)va);
+	*ofs = -1;
+	return (0);
 }
 
 static ssize_t
-_kvm_read_phys(kvm_t *kd, uint64_t pa, void *buf, size_t bufsz)
+phys_read(kvm_t *kd, uint64_t pa, void *buf, size_t bufsz)
 {
 	off_t ofs;
 	size_t sz;
 
-	sz = _kvm_pa2off(kd, pa, &ofs, 0);
+	sz = phys_addr2off(kd, pa, &ofs, 0);
 	if (sz < bufsz)
 		return ((ssize_t)sz);
 
@@ -131,6 +208,50 @@ _kvm_read_phys(kvm_t *kd, uint64_t pa, void *buf, size_t bufsz)
 	return (read(kd->pmfd, buf, bufsz));
 }
 
+/*
+ * Virtual core support (aka minidump).
+ */
+
+static size_t
+virt_addr2off(kvm_t *kd, uint64_t va, off_t *ofs, size_t pgsz)
+{
+	Elf64_Ehdr *e;
+	Elf64_Phdr *p;
+	int n;
+
+	if (va < REGION_BASE(4))
+		goto fail;
+
+	e = (Elf64_Ehdr *)(kd->vmst->mmapbase);
+	n = e->e_phnum;
+	p = (Elf64_Phdr *)(void *)((uintptr_t)(void *)e + e->e_phoff); 
+	while (n && (va < p->p_vaddr || va >= p->p_vaddr + p->p_memsz))
+		p++, n--;
+	if (n == 0)
+		goto fail;
+
+	*ofs = (va - p->p_vaddr) + p->p_offset;
+	if (pgsz == 0)
+		return (p->p_memsz - (va - p->p_vaddr));
+	return (pgsz - ((size_t)va & (pgsz - 1)));
+
+ fail:
+	_kvm_err(kd, kd->program, "invalid virtual address %#jx",
+	    (uintmax_t)va);
+	return (0);
+}
+
+static size_t
+virt_kvatop(kvm_t *kd, uint64_t va, off_t *ofs)
+{
+
+	return (virt_addr2off(kd, va, ofs, 0));
+}
+
+/*
+ * KVM architecture support functions.
+ */
+
 void
 _kvm_freevtop(kvm_t *kd)
 {
@@ -160,27 +281,37 @@ _kvm_initvtop(kvm_t *kd)
 		return (-1);
 	}
 
+#ifndef CROSS_LIBKVM
 	kd->vmst->pagesize = getpagesize();
+#else
+	kd->vmst->pagesize = 8192;
+#endif
 
-	if (_kvm_maphdrs(kd, sizeof(Elf64_Ehdr)) == -1)
+	if (ia64_maphdrs(kd, sizeof(Elf64_Ehdr)) == -1)
 		return (-1);
 
 	ehdr = kd->vmst->mmapbase;
 	hdrsz = ehdr->e_phoff + ehdr->e_phentsize * ehdr->e_phnum;
-	if (_kvm_maphdrs(kd, hdrsz) == -1)
+	if (ia64_maphdrs(kd, hdrsz) == -1)
 		return (-1);
 
+	kd->vmst->kvatop = (ehdr->e_flags & EF_IA_64_ABSOLUTE) ?
+	    phys_kvatop : virt_kvatop;
+
 	/*
 	 * Load the PBVM page table. We need this to resolve PBVM addresses.
 	 * The PBVM page table is obtained from the bootinfo structure, of
-	 * which the physical address is given to us in e_entry. If e_entry
-	 * is 0, then this is assumed to be a pre-PBVM kernel.
+	 * which the address is given to us in e_entry. If e_entry is 0, then
+	 * this is assumed to be a pre-PBVM kernel.
+	 * Note that the address of the bootinfo structure is either physical
+	 * or virtual, depending on whether the core is physical or virtual.
 	 */
-	if (ehdr->e_entry != 0) {
-		sz = _kvm_read_phys(kd, ehdr->e_entry, &bi, sizeof(bi));
+	if (ehdr->e_entry != 0 && (ehdr->e_flags & EF_IA_64_ABSOLUTE) != 0) {
+		sz = phys_read(kd, ehdr->e_entry, &bi, sizeof(bi));
 		if (sz != sizeof(bi)) {
 			_kvm_err(kd, kd->program,
-			    "cannot read bootinfo from PA %#lx", ehdr->e_entry);
+			    "cannot read bootinfo at physical address %#jx",
+			    (uintmax_t)ehdr->e_entry);
 			return (-1);
 		}
 		if (bi.bi_magic != BOOTINFO_MAGIC) {
@@ -193,12 +324,12 @@ _kvm_initvtop(kvm_t *kd)
 			return (-1);
 		}
 		kd->vmst->pbvm_pgtblsz = bi.bi_pbvm_pgtblsz;
-		sz = _kvm_read_phys(kd, bi.bi_pbvm_pgtbl, kd->vmst->pbvm_pgtbl,
+		sz = phys_read(kd, bi.bi_pbvm_pgtbl, kd->vmst->pbvm_pgtbl,
 		    bi.bi_pbvm_pgtblsz);
 		if (sz != bi.bi_pbvm_pgtblsz) {
 			_kvm_err(kd, kd->program,
-			    "cannot read page table from PA %#lx",
-			    bi.bi_pbvm_pgtbl);
+			    "cannot read page table at physical address %#jx",
+			    (uintmax_t)bi.bi_pbvm_pgtbl);
 			return (-1);
 		}
 	} else {
@@ -225,7 +356,7 @@ _kvm_initvtop(kvm_t *kd)
 		return (-1);
 	}
 
-	if (va < REGION_BASE(6)) {
+	if (va == REGION_BASE(5)) {
 		_kvm_err(kd, kd->program, "kptdir is itself virtual");
 		return (-1);
 	}
@@ -237,56 +368,8 @@ _kvm_initvtop(kvm_t *kd)
 int
 _kvm_kvatop(kvm_t *kd, u_long va, off_t *ofs)
 {
-	struct ia64_lpte pte;
-	uint64_t pa, pgaddr, pt0addr, pt1addr;
-	size_t pgno, pgsz, pt0no, pt1no;
-
-	if (va >= REGION_BASE(6)) {
-		/* Regions 6 and 7: direct mapped. */
-		pa = REGION_ADDR(va);
-		return (_kvm_pa2off(kd, pa, ofs, 0));
-	} else if (va >= REGION_BASE(5)) {
-		/* Region 5: Kernel Virtual Memory. */
-		va = REGION_ADDR(va);
-		pgsz = kd->vmst->pagesize;
-		pt0no = KPTE_DIR0_INDEX(va, pgsz);
-		pt1no = KPTE_DIR1_INDEX(va, pgsz);
-		pgno = KPTE_PTE_INDEX(va, pgsz);
-		if (pt0no >= NKPTEDIR(pgsz))
-			goto fail;
-		pt0addr = kd->vmst->kptdir + (pt0no << 3);
-		if (kvm_read(kd, pt0addr, &pt1addr, 8) != 8)
-			goto fail;
-		if (pt1addr == 0)
-			goto fail;
-		pt1addr += pt1no << 3;
-		if (kvm_read(kd, pt1addr, &pgaddr, 8) != 8)
-			goto fail;
-		if (pgaddr == 0)
-			goto fail;
-		pgaddr += pgno * sizeof(pte);
-		if (kvm_read(kd, pgaddr, &pte, sizeof(pte)) != sizeof(pte))
-			goto fail;
-		if (!(pte.pte & PTE_PRESENT))
-			goto fail;
-		pa = (pte.pte & PTE_PPN_MASK) + (va & (pgsz - 1));
-		return (_kvm_pa2off(kd, pa, ofs, pgsz));
-	} else if (va >= PBVM_BASE) {
-		/* Region 4: Pre-Boot Virtual Memory (PBVM). */
-		va -= PBVM_BASE;
-		pgsz = PBVM_PGSZ;
-		pt0no = va / pgsz;
-		if (pt0no >= (kd->vmst->pbvm_pgtblsz >> 3))
-			goto fail;
-		pt0addr = kd->vmst->pbvm_pgtbl[pt0no];
-		if (!(pt0addr & PTE_PRESENT))
-			goto fail;
-		pa = (pt0addr & PTE_PPN_MASK) + va % pgsz;
-		return (_kvm_pa2off(kd, pa, ofs, pgsz));
-	}
+	size_t sz;
 
- fail:
-	_kvm_err(kd, kd->program, "invalid kernel virtual address");
-	*ofs = ~0UL;
-	return (0);
+	sz = kd->vmst->kvatop(kd, va, ofs);
+	return ((sz > INT_MAX) ? INT_MAX : sz);
 }