Add ia64 support. Various adjustments were made to existing targets to

cope with a few interface changes required by the ia64. In particular,
function pointers on ia64 need special treatment in rtld.

5 files changed, 981 insertions, 0 deletions

diff --git a/libexec/rtld-elf/ia64/Makefile.inc b/libexec/rtld-elf/ia64/Makefile.inc
new file mode 100644
index 0000000..3ea0ac3
--- /dev/null
+++ b/libexec/rtld-elf/ia64/Makefile.inc
@@ -0,0 +1,2 @@
+# $FreeBSD$
+LDFLAGS+=	-Wl,--export-dynamic
diff --git a/libexec/rtld-elf/ia64/lockdflt.c b/libexec/rtld-elf/ia64/lockdflt.c
new file mode 100644
index 0000000..5847abb
--- /dev/null
+++ b/libexec/rtld-elf/ia64/lockdflt.c
@@ -0,0 +1,181 @@
+/*-
+ * Copyright 1999, 2000 John D. Polstra.
+ * 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.
+ *
+ * $FreeBSD$
+ */
+
+/*
+ * Thread locking implementation for the dynamic linker.
+ *
+ * We use the "simple, non-scalable reader-preference lock" from:
+ *
+ *   J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer
+ *   Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on
+ *   Principles and Practice of Parallel Programming, April 1991.
+ *
+ * In this algorithm the lock is a single word.  Its low-order bit is
+ * set when a writer holds the lock.  The remaining high-order bits
+ * contain a count of readers desiring the lock.  The algorithm requires
+ * atomic "compare_and_store" and "add" operations, which we implement
+ * using assembly language sequences in "rtld_start.S".
+ *
+ * These are spinlocks.  When spinning we call nanosleep() for 1
+ * microsecond each time around the loop.  This will most likely yield
+ * the CPU to other threads (including, we hope, the lockholder) allowing
+ * them to make some progress.
+ */
+
+#include <signal.h>
+#include <stdlib.h>
+#include <time.h>
+
+#include "debug.h"
+#include "rtld.h"
+
+/*
+ * This value of CACHE_LINE_SIZE is conservative.  The actual size
+ * is 32 on the  21064, 21064A, 21066, 21066A, and 21164.  It is 64
+ * on the 21264.  Compaq recommends sequestering each lock in its own
+ * 128-byte block to allow for future implementations with larger
+ * cache lines.
+ */
+#define CACHE_LINE_SIZE		128
+
+#define WAFLAG		0x1	/* A writer holds the lock */
+#define RC_INCR		0x2	/* Adjusts count of readers desiring lock */
+
+typedef struct Struct_Lock {
+	volatile int lock;
+	void *base;
+} Lock;
+
+static const struct timespec usec = { 0, 1000 };	/* 1 usec. */
+static sigset_t fullsigmask, oldsigmask;
+
+static void *
+lock_create(void *context)
+{
+    void *base;
+    char *p;
+    uintptr_t r;
+    Lock *l;
+
+    /*
+     * Arrange for the lock to occupy its own cache line.  First, we
+     * optimistically allocate just a cache line, hoping that malloc
+     * will give us a well-aligned block of memory.  If that doesn't
+     * work, we allocate a larger block and take a well-aligned cache
+     * line from it.
+     */
+    base = xmalloc(CACHE_LINE_SIZE);
+    p = (char *)base;
+    if ((uintptr_t)p % CACHE_LINE_SIZE != 0) {
+	free(base);
+	base = xmalloc(2 * CACHE_LINE_SIZE);
+	p = (char *)base;
+	if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0)
+	    p += CACHE_LINE_SIZE - r;
+    }
+    l = (Lock *)p;
+    l->base = base;
+    l->lock = 0;
+    return l;
+}
+
+static void
+lock_destroy(void *lock)
+{
+    Lock *l = (Lock *)lock;
+
+    free(l->base);
+}
+
+static void
+rlock_acquire(void *lock)
+{
+    Lock *l = (Lock *)lock;
+
+    atomic_add_int(&l->lock, RC_INCR);
+    while (l->lock & WAFLAG)
+	    nanosleep(&usec, NULL);
+}
+
+static void
+wlock_acquire(void *lock)
+{
+    Lock *l = (Lock *)lock;
+    sigset_t tmp_oldsigmask;
+
+    for ( ; ; ) {
+	sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
+	if (cmp0_and_store_int(&l->lock, WAFLAG) == 0)
+	    break;
+	sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
+	nanosleep(&usec, NULL);
+    }
+    oldsigmask = tmp_oldsigmask;
+}
+
+static void
+rlock_release(void *lock)
+{
+    Lock *l = (Lock *)lock;
+
+    atomic_add_int(&l->lock, -RC_INCR);
+}
+
+static void
+wlock_release(void *lock)
+{
+    Lock *l = (Lock *)lock;
+
+    atomic_add_int(&l->lock, -WAFLAG);
+    sigprocmask(SIG_SETMASK, &oldsigmask, NULL);
+}
+
+void
+lockdflt_init(LockInfo *li)
+{
+    li->context = NULL;
+    li->lock_create = lock_create;
+    li->rlock_acquire = rlock_acquire;
+    li->wlock_acquire = wlock_acquire;
+    li->rlock_release = rlock_release;
+    li->wlock_release = wlock_release;
+    li->lock_destroy = lock_destroy;
+    li->context_destroy = NULL;
+    /*
+     * Construct a mask to block all signals except traps which might
+     * conceivably be generated within the dynamic linker itself.
+     */
+    sigfillset(&fullsigmask);
+    sigdelset(&fullsigmask, SIGILL);
+    sigdelset(&fullsigmask, SIGTRAP);
+    sigdelset(&fullsigmask, SIGABRT);
+    sigdelset(&fullsigmask, SIGEMT);
+    sigdelset(&fullsigmask, SIGFPE);
+    sigdelset(&fullsigmask, SIGBUS);
+    sigdelset(&fullsigmask, SIGSEGV);
+    sigdelset(&fullsigmask, SIGSYS);
+}
diff --git a/libexec/rtld-elf/ia64/reloc.c b/libexec/rtld-elf/ia64/reloc.c
new file mode 100644
index 0000000..86b7ce6
--- /dev/null
+++ b/libexec/rtld-elf/ia64/reloc.c
@@ -0,0 +1,435 @@
+/*-
+ * Copyright 1996, 1997, 1998, 1999 John D. Polstra.
+ * 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.
+ *
+ * $FreeBSD$
+ */
+
+/*
+ * Dynamic linker for ELF.
+ *
+ * John Polstra <jdp@polstra.com>.
+ */
+
+#include <sys/param.h>
+#include <sys/mman.h>
+
+#include <dlfcn.h>
+#include <err.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "rtld.h"
+
+extern Elf_Dyn _DYNAMIC;
+
+/*
+ * Macros for loading/storing unaligned 64-bit values.  These are
+ * needed because relocations can point to unaligned data.  This
+ * occurs in the DWARF2 exception frame tables generated by the
+ * compiler, for instance.
+ *
+ * We don't use these when relocating jump slots and GOT entries,
+ * since they are guaranteed to be aligned.
+ *
+ * XXX dfr stub for now.
+ */
+#define load64(p)	(*(u_int64_t *) (p))
+#define store64(p, v)	(*(u_int64_t *) (p) = (v))
+
+/* Allocate an @fptr. */
+
+#define FPTR_CHUNK_SIZE		64
+
+struct fptr_chunk {
+	struct fptr fptrs[FPTR_CHUNK_SIZE];
+};
+
+static struct fptr_chunk first_chunk;
+static struct fptr_chunk *current_chunk = &first_chunk;
+static struct fptr *next_fptr = &first_chunk.fptrs[0];
+static struct fptr *last_fptr = &first_chunk.fptrs[FPTR_CHUNK_SIZE];
+
+/*
+ * We use static storage initially so that we don't have to call
+ * malloc during init_rtld().
+ */
+static struct fptr *
+alloc_fptr(Elf_Addr target, Elf_Addr gp)
+{
+	struct fptr* fptr;
+
+	if (next_fptr == last_fptr) {
+		current_chunk = malloc(sizeof(struct fptr_chunk));
+		next_fptr = &current_chunk->fptrs[0];
+		last_fptr = &current_chunk->fptrs[FPTR_CHUNK_SIZE];
+	}
+	fptr = next_fptr;
+	next_fptr++;
+	fptr->target = target;
+	fptr->gp = gp;
+	return fptr;
+}
+
+/* Relocate a non-PLT object with addend. */
+static int
+reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
+		  SymCache *cache, struct fptr **fptrs)
+{
+	Elf_Addr *where = (Elf_Addr *) (obj->relocbase + rela->r_offset);
+
+	switch (ELF_R_TYPE(rela->r_info)) {
+	case R_IA64_REL64LSB:
+		/*
+		 * We handle rtld's relocations in rtld_start.S
+		 */
+		if (obj != obj_rtld)
+			store64(where,
+				load64(where) + (Elf_Addr) obj->relocbase);
+		break;
+
+	case R_IA64_DIR64LSB: {
+		const Elf_Sym *def;
+		const Obj_Entry *defobj;
+		Elf_Addr target;
+
+		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
+				  false, cache);
+		if (def == NULL)
+			return -1;
+		target = (Elf_Addr) (defobj->relocbase + def->st_value);
+		store64(where, target + rela->r_addend);
+		break;
+	}
+
+	case R_IA64_FPTR64LSB: {
+		/*
+		 * We have to make sure that all @fptr references to
+		 * the same function are identical so that code can
+		 * compare function pointers. We actually only bother
+		 * to ensure this within a single object. If the
+		 * caller's alloca failed, we don't even ensure that.
+		 */
+		const Elf_Sym *def;
+		const Obj_Entry *defobj;
+		struct fptr *fptr = 0;
+		Elf_Addr target, gp;
+
+		/*
+		 * Not sure why the call to find_symdef() doesn't work 
+		 * properly (it fails if the symbol is local). Perhaps 
+		 * this is a toolchain issue - revisit after we
+		 * upgrade the ia64 toolchain.
+		 */
+		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
+				  false, cache);
+		if (def == NULL) {
+			def = &obj->symtab[ELF_R_SYM(rela->r_info)];
+			defobj = obj;
+		}
+		target = (Elf_Addr) (defobj->relocbase + def->st_value);
+		gp = (Elf_Addr) defobj->pltgot;
+
+		/*
+		 * Find the @fptr, using fptrs as a helper.
+		 */
+		if (fptrs)
+			fptr = fptrs[ELF_R_SYM(rela->r_info)];
+		if (!fptr) {
+			fptr = alloc_fptr(target, gp);
+			if (fptrs)
+				fptrs[ELF_R_SYM(rela->r_info)] = fptr;
+		}
+		store64(where, (Elf_Addr) fptr);
+		break;
+	}
+
+	default:
+		_rtld_error("%s: Unsupported relocation type %d"
+			    " in non-PLT relocations\n", obj->path,
+			    ELF_R_TYPE(rela->r_info));
+		return -1;
+	}
+
+	return(0);
+}
+
+/* Process the non-PLT relocations. */
+int
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld)
+{
+	const Elf_Rel *rellim;
+	const Elf_Rel *rel;
+	const Elf_Rela *relalim;
+	const Elf_Rela *rela;
+	SymCache *cache;
+	struct fptr **fptrs;
+
+	cache = (SymCache *)alloca(obj->nchains * sizeof(SymCache));
+	if (cache != NULL)
+		memset(cache, 0, obj->nchains * sizeof(SymCache));
+
+	/*
+	 * When relocating rtld itself, we need to avoid using malloc.
+	 */
+        if (obj == obj_rtld)
+		fptrs = (struct fptr **)
+			alloca(obj->nchains * sizeof(struct fptr *));
+	else
+		fptrs = (struct fptr **)
+			malloc(obj->nchains * sizeof(struct fptr *));
+
+	if (fptrs == NULL)
+		return -1;
+	memset(fptrs, 0, obj->nchains * sizeof(struct fptr *));
+
+	/* Perform relocations without addend if there are any: */
+	rellim = (const Elf_Rel *) ((caddr_t) obj->rel + obj->relsize);
+	for (rel = obj->rel;  obj->rel != NULL && rel < rellim;  rel++) {
+		Elf_Rela locrela;
+
+		locrela.r_info = rel->r_info;
+		locrela.r_offset = rel->r_offset;
+		locrela.r_addend = 0;
+		if (reloc_non_plt_obj(obj_rtld, obj, &locrela, cache, fptrs))
+			return -1;
+	}
+
+	/* Perform relocations with addend if there are any: */
+	relalim = (const Elf_Rela *) ((caddr_t) obj->rela + obj->relasize);
+	for (rela = obj->rela;  obj->rela != NULL && rela < relalim;  rela++) {
+		if (reloc_non_plt_obj(obj_rtld, obj, rela, cache, fptrs))
+			return -1;
+	}
+
+	/*
+	 * Remember the fptrs in case of later calls to dlsym(). Don't 
+	 * bother for rtld - we will lazily create a table in
+	 * make_function_pointer(). At this point we still can't risk
+	 * calling malloc().
+	 */
+	if (obj != obj_rtld)
+		obj->priv = fptrs;
+	else
+		obj->priv = NULL;
+
+	return 0;
+}
+
+/* Process the PLT relocations. */
+int
+reloc_plt(Obj_Entry *obj)
+{
+	/* All PLT relocations are the same kind: Elf_Rel or Elf_Rela. */
+	if (obj->pltrelsize != 0) {
+		const Elf_Rel *rellim;
+		const Elf_Rel *rel;
+
+		rellim = (const Elf_Rel *)
+			((char *)obj->pltrel + obj->pltrelsize);
+		for (rel = obj->pltrel;  rel < rellim;  rel++) {
+			Elf_Addr *where;
+
+			assert(ELF_R_TYPE(rel->r_info) == R_IA64_IPLTLSB);
+
+			/* Relocate the @fptr pointing into the PLT. */
+			where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
+			*where += (Elf_Addr)obj->relocbase;
+		}
+	} else {
+		const Elf_Rela *relalim;
+		const Elf_Rela *rela;
+
+		relalim = (const Elf_Rela *)
+			((char *)obj->pltrela + obj->pltrelasize);
+		for (rela = obj->pltrela;  rela < relalim;  rela++) {
+			Elf_Addr *where;
+
+			assert(ELF_R_TYPE(rela->r_info) == R_IA64_IPLTLSB);
+
+			/* Relocate the @fptr pointing into the PLT. */
+			where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
+			*where += (Elf_Addr)obj->relocbase;
+		}
+	}
+	return 0;
+}
+
+/* Relocate the jump slots in an object. */
+int
+reloc_jmpslots(Obj_Entry *obj)
+{
+	if (obj->jmpslots_done)
+		return 0;
+	/* All PLT relocations are the same kind: Elf_Rel or Elf_Rela. */
+	if (obj->pltrelsize != 0) {
+		const Elf_Rel *rellim;
+		const Elf_Rel *rel;
+
+		rellim = (const Elf_Rel *)
+			((char *)obj->pltrel + obj->pltrelsize);
+		for (rel = obj->pltrel;  rel < rellim;  rel++) {
+			Elf_Addr *where;
+			const Elf_Sym *def;
+			const Obj_Entry *defobj;
+
+			assert(ELF_R_TYPE(rel->r_info) == R_IA64_IPLTLSB);
+			where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
+			def = find_symdef(ELF_R_SYM(rel->r_info), obj,
+					  &defobj, true, NULL);
+			if (def == NULL)
+				return -1;
+			reloc_jmpslot(where,
+				      (Elf_Addr)(defobj->relocbase
+						 + def->st_value),
+				      defobj);
+		}
+	} else {
+		const Elf_Rela *relalim;
+		const Elf_Rela *rela;
+
+		relalim = (const Elf_Rela *)
+			((char *)obj->pltrela + obj->pltrelasize);
+		for (rela = obj->pltrela;  rela < relalim;  rela++) {
+			Elf_Addr *where;
+			const Elf_Sym *def;
+			const Obj_Entry *defobj;
+
+			/* assert(ELF_R_TYPE(rela->r_info) == R_ALPHA_JMP_SLOT); */
+			where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
+			def = find_symdef(ELF_R_SYM(rela->r_info), obj,
+					  &defobj, true, NULL);
+			if (def == NULL)
+				return -1;
+			reloc_jmpslot(where,
+				      (Elf_Addr)(defobj->relocbase
+						 + def->st_value),
+				      defobj);
+		}
+	}
+	obj->jmpslots_done = true;
+	return 0;
+}
+
+/* Fixup the jump slot at "where" to transfer control to "target". */
+Elf_Addr
+reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *obj)
+{
+	Elf_Addr stubaddr;
+
+	dbg(" reloc_jmpslot: where=%p, target=%p, gp=%p",
+	    (void *)where, (void *)target, (void *)obj->pltgot);
+	stubaddr = *where;
+	if (stubaddr != target) {
+
+		/*
+		 * Point this @fptr directly at the target. Update the
+		 * gp value first so that we don't break another cpu
+		 * which is currently executing the PLT entry.
+		 */
+		where[1] = (Elf_Addr) obj->pltgot;
+		ia64_mf();
+		where[0] = target;
+		ia64_mf();
+	}
+
+	/*
+	 * The caller needs an @fptr for the adjusted entry. The PLT
+	 * entry serves this purpose nicely.
+	 */
+	return (Elf_Addr) where;
+}
+
+/*
+ * XXX ia64 doesn't seem to have copy relocations.
+ *
+ * Returns 0 on success, -1 on failure.
+ */
+int
+do_copy_relocations(Obj_Entry *dstobj)
+{
+
+	return 0;
+}
+
+/*
+ * Return the @fptr representing a given function symbol.
+ */
+void *
+make_function_pointer(const Elf_Sym *sym, const Obj_Entry *obj)
+{
+	struct fptr **fptrs = obj->priv;
+	int index = sym - obj->symtab;
+
+	if (!fptrs) {
+		/*
+		 * This should only happen for something like
+		 * dlsym("dlopen"). Actually, I'm not sure it can ever 
+		 * happen.
+		 */
+		fptrs = (struct fptr **)
+			malloc(obj->nchains * sizeof(struct fptr *));
+		memset(fptrs, 0, obj->nchains * sizeof(struct fptr *));
+		((Obj_Entry*) obj)->priv = fptrs;
+	}
+	if (!fptrs[index]) {
+		Elf_Addr target, gp;
+		target = (Elf_Addr) (obj->relocbase + sym->st_value);
+		gp = (Elf_Addr) obj->pltgot;
+		fptrs[index] = alloc_fptr(target, gp);
+	}
+	return fptrs[index];
+}
+
+/* Initialize the special PLT entries. */
+void
+init_pltgot(Obj_Entry *obj)
+{
+	const Elf_Dyn *dynp;
+	Elf_Addr *pltres = 0;
+
+	/*
+	 * Find the PLT RESERVE section.
+	 */
+	for (dynp = obj->dynamic;  dynp->d_tag != DT_NULL;  dynp++) {
+		if (dynp->d_tag == DT_IA64_PLT_RESERVE)
+			pltres = (u_int64_t *)
+				(obj->relocbase + dynp->d_un.d_ptr);
+	}
+	if (!pltres)
+		errx(1, "Can't find DT_IA64_PLT_RESERVE entry");
+
+	/*
+	 * The PLT RESERVE section is used to get values to pass to
+	 * _rtld_bind when lazy binding.
+	 */
+	pltres[0] = (Elf_Addr) obj;
+	pltres[1] = FPTR_TARGET(_rtld_bind_start);
+	pltres[2] = FPTR_GP(_rtld_bind_start);
+}
diff --git a/libexec/rtld-elf/ia64/rtld_machdep.h b/libexec/rtld-elf/ia64/rtld_machdep.h
new file mode 100644
index 0000000..74a3f62
--- /dev/null
+++ b/libexec/rtld-elf/ia64/rtld_machdep.h
@@ -0,0 +1,57 @@
+/*-
+ * Copyright (c) 1999, 2000 John D. Polstra.
+ * 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 AND CONTRIBUTORS ``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 OR CONTRIBUTORS 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 RTLD_MACHDEP_H
+#define RTLD_MACHDEP_H	1
+
+/*
+ * Macros for cracking ia64 function pointers.
+ */
+struct fptr {
+	Elf_Addr	target;
+	Elf_Addr	gp;
+};
+
+#define FPTR_TARGET(f)	(((struct fptr *) (f))->target)
+#define FPTR_GP(f)	(((struct fptr *) (f))->gp)
+
+/* Return the address of the .dynamic section in the dynamic linker. */
+#define rtld_dynamic(obj)	(&_DYNAMIC)
+
+struct Struct_Obj_Entry;
+
+Elf_Addr reloc_jmpslot(Elf_Addr *, Elf_Addr, const struct Struct_Obj_Entry *);
+void *make_function_pointer(const Elf_Sym *, const struct Struct_Obj_Entry *);
+
+/* Atomic operations. */
+int cmp0_and_store_int(volatile int *, int);
+void atomic_add_int(volatile int *, int);
+void atomic_incr_int(volatile int *);
+void atomic_decr_int(volatile int *);
+
+#endif
diff --git a/libexec/rtld-elf/ia64/rtld_start.S b/libexec/rtld-elf/ia64/rtld_start.S
new file mode 100644
index 0000000..6dbb1db
--- /dev/null
+++ b/libexec/rtld-elf/ia64/rtld_start.S
@@ -0,0 +1,306 @@
+/* $FreeBSD$	*/
+/*	From: NetBSD: rtld_start.S,v 1.1 1996/12/16 20:38:09 cgd Exp 	*/
+
+/*
+ * Copyright 1996 Matt Thomas <matt@3am-software.com>
+ * Copyright 2000 John D. Polstra
+ * 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.
+ * 3. The name of the author may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * 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 <machine/asm.h>
+#include <sys/syscall.h>
+
+ENTRY(_rtld_start, 0)
+	alloc	r2=ar.pfs,3,0,3,0
+	;;
+1:	mov	r14=ip			// calculate gp
+	addl	r3=@gprel(1b),r0
+	;;
+	sub	gp=r14,r3
+	;;
+	.section .sdata
+2:	data4	@ltv(1b)		// unrelocated address of 1b
+	.align	8
+	.previous
+	add	r15=@gprel(2b),gp
+	;;
+	ld8	r15=[r15]
+	;;
+	sub	out0=r14,r15		// out0 is image base address
+	br.call.sptk.many rp=_rtld_reloc // fixup image
+	
+	add	sp=-16,sp		// 16 bytes for us, 16 for _rtld
+	;;
+	mov	out0=in0
+	add	out1=16,sp		// address for exit proc
+	add	out2=24,sp		// address for obj_main
+
+	br.call.sptk.many rp=_rtld	// r8=_rtld(sp, &exit_proc, &obj_main)
+
+	add	r16=16,sp		// address for exit proc
+	;; 
+	ld8	r15=[r16]		// read exit proc
+	add	sp=16,sp		// readjust stack
+	mov	b7=r8			// address of real _start
+	;;
+	alloc	r2=ar.pfs,0,0,3,0	// dump register frame
+	mov	out2=r15
+
+	br.call.sptk.many rp=b7		// transfer to main program
+	br.call.sptk.many rp=exit	// die
+END(_rtld_start)
+
+/*
+ * _rtld_bind_start:	lookup a lazy binding and transfer to real target
+ *
+ * Arguments:
+ *	r1	gp value for rtld
+ *	r15	Index in plt
+ *	r16	Obj_Entry of caller
+ *	in0-in7	Arguments for target procedure
+ *	rp	Return address back to caller
+ */
+ENTRY(_rtld_bind_start, 0)
+{ .mii
+	alloc	loc0=ar.pfs,8,6,3,0	// space to save r8-r11
+	add	r17=16-8*16,sp		// leave 16 bytes for _rtld_bind
+	add	r18=32-8*16,sp
+	;;
+} { .mii
+	mov	loc2=r8			// structure return address
+	add	sp=-8*16,sp		// space to save f8-f15
+	mov	loc1=rp
+	;;
+} { .mii
+	stf.spill [r17]=f8,32		// save float arguments
+	mov	loc3=r9			// language specific
+	mov	loc4=r10		// language specific
+} { .mii
+	stf.spill [r18]=f9,32
+	mov	loc5=r11		// language specific
+	shl	out1=r15,4		// 16 * index
+	;;
+} { .mmi
+	stf.spill [r17]=f10,32
+	stf.spill [r18]=f11,32
+	mov	out0=r16		// Obj_Entry for caller
+	;; 
+} { .mmi
+	stf.spill [r17]=f12,32
+	stf.spill [r18]=f13,32
+	shladd	out1=r15,3,out1		// rela offset = 24 * index
+	;; 
+} { .mmb
+	stf.spill [r17]=f14,32
+	stf.spill [r18]=f15,32
+	br.call.sptk.many rp=_rtld_bind
+} { .mii
+	ld8	r14=[r8],8		// target address
+	add	r17=16,sp
+	add	r18=32,sp
+	;;
+} { .mii
+	ld8	r1=[r8]			// target gp
+	mov	ar.pfs=loc0		// clean up
+	mov	rp=loc1
+} { .mmi	
+	ldf.fill f8=[r17],32		// restore float arguments
+	ldf.fill f9=[r18],32
+	mov	r8=loc2			// restore structure pointer
+	;; 
+} { .mmi
+	ldf.fill f10=[r17],32
+	ldf.fill f11=[r18],32
+	mov	r9=loc3
+	;; 
+} { .mmi
+	ldf.fill f12=[r17],32
+	ldf.fill f13=[r18],32
+	mov	r10=loc4
+	;; 
+} { .mmi
+	ldf.fill f14=[r17],32
+	ldf.fill f15=[r18],32
+	mov	r11=loc5
+	;;
+} { .mii
+	nop.m	0
+	mov	b7=r14
+	add	sp=8*16,sp
+	;;
+} { .mib
+	alloc	r14=ar.pfs,0,0,8,0	// drop our register frame
+	nop.i	0
+	br.sptk.many b7			// jump to target
+}
+END(_rtld_bind_start)
+
+/*
+ * int cmp0_and_store_int(volatile int *p, int newval);
+ *
+ * If an int holds 0, store newval into it; else do nothing.  Returns
+ * the previous value.
+ */
+ENTRY(cmp0_and_store_int, 2)
+	mov	ar.ccv=0
+	;;
+	cmpxchg4.acq r8=[in0],in1,ar.ccv
+	br.ret.sptk.many rp
+END(cmp0_and_store_int)
+
+ENTRY(atomic_add_int, 2)
+1:	ld4	r14=[in0]
+	;;
+	mov	ar.ccv=r14
+	add	r15=in1,r14
+	;;
+	cmpxchg4.acq r16=[in0],r15,ar.ccv
+	;;
+	cmp.ne	p6,p0=r14,r16
+(p6)	br.cond.spnt.few 1b
+	br.ret.sptk.many rp
+END(atomic_add_int)
+ 
+/* Atomically increment an int. */
+ENTRY(atomic_incr_int, 1)
+1:	ld4	r14=[in0]
+	;;
+	mov	ar.ccv=r14
+	add	r15=1,r14
+	;;
+	cmpxchg4.acq r16=[in0],r15,ar.ccv
+	;;
+	cmp.ne	p6,p0=r14,r16
+(p6)	br.cond.spnt.few 1b
+	br.ret.sptk.many rp
+END(atomic_incr_int)
+
+/* Atomically decrement an int. */
+ENTRY(atomic_decr_int, 1)
+1:	ld4	r14=[in0]
+	;;
+	mov	ar.ccv=r14
+	add	r15=-1,r14
+	;;
+	cmpxchg4.acq r16=[in0],r15,ar.ccv
+	;;
+	cmp.ne	p6,p0=r14,r16
+(p6)	br.cond.spnt.few 1b
+	br.ret.sptk.many rp
+END(atomic_decr_int)
+
+#define DT_NULL		0	/* Terminating entry. */
+#define DT_RELA		7	/* Address of ElfNN_Rela relocations. */
+#define DT_RELASZ	8	/* Total size of ElfNN_Rela relocations. */
+#define DT_RELAENT	9	/* Size of each ElfNN_Rela relocation entry. */
+	
+#define R_IA64_NONE		0	/* None */
+#define R_IA64_DIR64LSB		0x27	/* word64 LSB	S + A */
+#define R_IA64_REL64LSB		0x6f	/* word64 LSB	BD + A */
+	
+/*
+ * _rtld_reloc:	relocate the rtld image, apart from @fptrs.
+ *
+ * Assumes that rtld was linked at zero and that we only need to
+ * handle REL64LSB and DIR64LSB relocations.
+ *
+ * Arguments:
+ *	r1	gp value for rtld
+ *	in0	rtld base address
+ */
+STATIC_ENTRY(_rtld_reloc, 1)
+	alloc	loc0=ar.pfs,1,2,0,0
+	mov	loc1=rp
+	;; 
+	movl	r15=@gprel(_DYNAMIC)	// find _DYNAMIC etc.
+	;;
+	add	r15=r15,gp		// relocate _DYNAMIC etc.
+	;;
+1:	ld8	r16=[r15],8		// read r15->d_tag
+	;;
+	ld8	r17=[r15],8		// and r15->d_val
+	;;
+	cmp.eq	p6,p0=DT_NULL,r16	// done?
+(p6)	br.cond.dpnt.few 2f
+	;; 
+	cmp.eq	p6,p0=DT_RELA,r16
+	;; 
+(p6)	add	r18=r17,in0		// found rela section
+	;; 
+	cmp.eq	p6,p0=DT_RELASZ,r16
+	;; 
+(p6)	mov	r19=r17			// found rela size
+	;; 
+	cmp.eq	p6,p0=DT_RELAENT,r16
+	;; 
+(p6)	mov	r22=r17			// found rela entry size
+	;;
+	br.sptk.few 1b
+	
+2:	
+	ld8	r15=[r18],8		// read r_offset
+	;; 
+	ld8	r16=[r18],8		// read r_info
+	add	r15=r15,in0		// relocate r_offset
+	;;
+	ld8	r17=[r18],8		// read r_addend
+	sub	r19=r19,r22		// update relasz
+
+	extr.u	r23=r16,0,32		// ELF64_R_TYPE(r16)
+	;;
+	cmp.eq	p6,p0=R_IA64_NONE,r23
+(p6)	br.cond.dpnt.few 3f
+	;;
+	cmp.eq	p6,p0=R_IA64_DIR64LSB,r23
+	;;
+(p6)	br.cond.dptk.few 4f
+	;;
+	cmp.eq	p6,p0=R_IA64_REL64LSB,r23
+	;;
+(p6)	br.cond.dptk.few 4f
+	;;
+
+3:	cmp.ltu	p6,p0=0,r19		// more?
+(p6)	br.cond.dptk.few 2b		// loop
+	
+	mov	r8=0			// success return value
+	;;
+	br.cond.sptk.few 9f		// done
+
+4:
+	ld8	r16=[r15]		// read value
+	;;
+	add	r16=r16,in0		// relocate it
+	;;
+	st8	[r15]=r16		// and store it back
+	br.cond.sptk.few 3b
+
+9:
+	mov	ar.pfs=loc0
+	mov	rp=loc1
+	;;
+	br.ret.sptk.few rp
+
+END(_rtld_reloc)