session.h (struct systemtap_session): Added itrace_derived_probe

group.
elaborate.cxx (systemtap_session::systemtap_session): Added
initialization of itrace_derived_probes.
tapsets.cxx (struct itrace_derived_probe): Add derived_probe
struct for holding info needed by itrace probes.
(struct itrace_derived_probe_group): New derived_probe_group
to handle itrace probes.
(itrace_derived_probe::itrace_derived_probe): Needed for use with
task_finder.
(itrace_derived_probe_group::join_group): Ditto.
(itrace_derived_probe_group::enroll): Ditto.
(itrace_derived_probe_group::emit_probe_decl): Ditto.
(itrace_derived_probe_group::emit_module_decls): Ditto.
(itrace_derived_probe_group::emit_module_init): Ditto.
(itrace_derived_probe_group::emit_module_exit): Ditto.
stapprobes.5.in : Added documentation of itrace probe.

8 files changed, 854 insertions, 0 deletions

diff --git a/ChangeLog b/ChangeLog
index 11ef1787..53693fa1 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,23 @@
+2008-07-28  Dave Nomura  <dcnltc@us.ibm.com>
+
+	* session.h (struct systemtap_session): Added itrace_derived_probe
+	group.
+	* elaborate.cxx (systemtap_session::systemtap_session): Added
+	initialization of itrace_derived_probes.
+	* tapsets.cxx (struct itrace_derived_probe): Add derived_probe
+	struct for holding info needed by itrace probes.
+	(struct itrace_derived_probe_group): New derived_probe_group
+	to handle itrace probes.
+	(itrace_derived_probe::itrace_derived_probe): Needed for use with
+	task_finder.
+	(itrace_derived_probe_group::join_group): Ditto.
+	(itrace_derived_probe_group::enroll): Ditto.
+	(itrace_derived_probe_group::emit_probe_decl): Ditto.
+	(itrace_derived_probe_group::emit_module_decls): Ditto.
+	(itrace_derived_probe_group::emit_module_init): Ditto.
+	(itrace_derived_probe_group::emit_module_exit): Ditto.
+	* stapprobes.5.in : Added documentation of itrace probe.
+
 2008-07-24  Josh Stone  <joshua.i.stone@intel.com>
 
 	* buildrun.cxx (compile_pass): Remove STAPCONF_MODULE_NSECTIONS,
diff --git a/elaborate.cxx b/elaborate.cxx
index 73358a1d..0950b086 100644
--- a/elaborate.cxx
+++ b/elaborate.cxx
@@ -1187,6 +1187,7 @@ systemtap_session::systemtap_session ():
   dwarf_derived_probes(0), 
   uprobe_derived_probes(0), 
   utrace_derived_probes(0), 
+  itrace_derived_probes(0), 
   task_finder_derived_probes(0), 
   timer_derived_probes(0), 
   profile_derived_probes(0), 
diff --git a/runtime/itrace.c b/runtime/itrace.c
new file mode 100644
index 00000000..3ee48265
--- /dev/null
+++ b/runtime/itrace.c
@@ -0,0 +1,438 @@
+/*
+ * user space instruction tracing
+ * Copyright (C) 2005, 2006, 2007, 2008 IBM Corp.
+ *
+ * This file is part of systemtap, and is free software.  You can
+ * redistribute it and/or modify it under the terms of the GNU General
+ * Public License (GPL); either version 2, or (at your option) any
+ * later version.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/err.h>
+#include <linux/sched.h>
+#include <linux/rcupdate.h>
+#include <linux/utrace.h>
+#include <asm/string.h>
+#include <asm/tracehook.h>
+#include <asm/ptrace.h>
+#include "uprobes/uprobes.h"
+
+#ifndef put_task_struct
+#define put_task_struct(t)	\
+	BUG_ON(atomic_dec_and_test(&tsk->usage))
+#endif
+
+#ifdef CONFIG_PPC
+struct bpt_info {
+        unsigned long addr;
+        unsigned int instr;
+};
+
+struct atomic_ss_info {
+        int step_over_atomic;
+        struct bpt_info end_bpt;
+        struct bpt_info br_bpt;
+};
+
+static int handle_ppc_atomic_seq(struct task_struct *tsk, struct pt_regs *regs,
+	struct atomic_ss_info *ss_info);
+static void remove_atomic_ss_breakpoint (struct task_struct *tsk, 
+	struct bpt_info *bpt);
+#endif
+
+struct itrace_info {
+	pid_t tid;
+	u32 step_flag;
+	struct stap_itrace_probe *itrace_probe;
+#ifdef CONFIG_PPC
+	struct atomic_ss_info ppc_atomic_ss;
+#endif
+	struct task_struct *tsk;
+	struct utrace_attached_engine *engine;
+	struct list_head link;
+};
+
+static u32 debug = 1;
+
+static LIST_HEAD(usr_itrace_info);
+static spinlock_t itrace_lock;
+static struct itrace_info *create_itrace_info(
+	struct task_struct *tsk, u32 step_flag,
+	struct stap_itrace_probe *itrace_probe);
+
+static u32 usr_itrace_report_signal(struct utrace_attached_engine *engine,
+			     struct task_struct *tsk,
+			     struct pt_regs *regs,
+			     u32 action, siginfo_t *info,
+			     const struct k_sigaction *orig_ka,
+			     struct k_sigaction *return_ka)
+{
+	struct itrace_info *ui;
+	u32 return_flags;
+	unsigned long data = 0;
+#ifdef CONFIG_PPC
+	data = mfspr(SPRN_SDAR);
+#endif
+
+	ui = rcu_dereference(engine->data);
+	WARN_ON(!ui);
+	
+	if (info->si_signo != SIGTRAP || !ui)
+		return UTRACE_ACTION_RESUME;
+
+	/* normal case: continue stepping, hide this trap from other engines */
+	return_flags =  ui->step_flag | UTRACE_ACTION_HIDE | UTRACE_SIGNAL_IGN |
+			   UTRACE_ACTION_NEWSTATE;
+
+#ifdef CONFIG_PPC
+	if (ui->ppc_atomic_ss.step_over_atomic) {
+		remove_atomic_ss_breakpoint(tsk, &ui->ppc_atomic_ss.end_bpt);
+		if (ui->ppc_atomic_ss.br_bpt.addr)
+			remove_atomic_ss_breakpoint(tsk,
+				&ui->ppc_atomic_ss.br_bpt);
+		ui->ppc_atomic_ss.step_over_atomic = 0;
+	}
+	
+	if (handle_ppc_atomic_seq(tsk, regs, &ui->ppc_atomic_ss))
+		return_flags = UTRACE_ACTION_RESUME | UTRACE_ACTION_NEWSTATE |
+			UTRACE_SIGNAL_IGN;
+#endif
+
+	enter_itrace_probe(ui->itrace_probe, regs, (void *)&data);
+
+	return return_flags;
+}
+
+static u32 usr_itrace_report_clone(struct utrace_attached_engine *engine,
+		struct task_struct *parent, unsigned long clone_flags,
+		struct task_struct *child)
+{
+	return UTRACE_ACTION_RESUME;
+}
+
+static u32 usr_itrace_report_death(struct utrace_attached_engine *e,
+	struct task_struct *tsk)
+{
+	struct itrace_info *ui = rcu_dereference(e->data);
+	WARN_ON(!ui);
+
+	return (UTRACE_ACTION_NEWSTATE | UTRACE_ACTION_DETACH);
+}
+
+static const struct utrace_engine_ops utrace_ops =
+{
+	.report_signal = usr_itrace_report_signal,
+	.report_clone = usr_itrace_report_clone,
+	.report_death = usr_itrace_report_death
+};
+
+
+static struct itrace_info *create_itrace_info(
+	struct task_struct *tsk, u32 step_flag,
+	struct stap_itrace_probe *itrace_probe)
+{
+	struct itrace_info *ui;
+
+	if (debug)
+		printk(KERN_INFO "create_itrace_info: tid=%d\n", tsk->pid);
+	/* initialize ui */
+	ui = kzalloc(sizeof(struct itrace_info), GFP_USER);
+	ui->tsk = tsk;
+	ui->tid = tsk->pid;
+	ui->step_flag = step_flag;
+	ui->itrace_probe = itrace_probe;
+#ifdef CONFIG_PPC
+	ui->ppc_atomic_ss.step_over_atomic = 0;
+#endif
+	INIT_LIST_HEAD(&ui->link);
+
+	/* push ui onto usr_itrace_info */
+	spin_lock(&itrace_lock);
+	list_add(&ui->link, &usr_itrace_info);
+
+	/* attach a single stepping engine */
+	ui->engine = utrace_attach(ui->tsk, UTRACE_ATTACH_CREATE, &utrace_ops, ui);
+	if (IS_ERR(ui->engine)) {
+		printk(KERN_ERR "utrace_attach returns %ld\n",
+			PTR_ERR(ui->engine));
+		ui = NULL;
+	} else {
+		utrace_set_flags(tsk, ui->engine, ui->engine->flags |
+			ui->step_flag |
+			UTRACE_EVENT(CLONE) | UTRACE_EVENT_SIGNAL_ALL |
+			UTRACE_EVENT(DEATH));
+	}
+	spin_unlock(&itrace_lock);
+	return ui;
+}
+
+static struct itrace_info *find_itrace_info(pid_t tid)
+{
+	struct itrace_info *ui = NULL;
+
+	spin_lock(&itrace_lock);
+	list_for_each_entry(ui, &usr_itrace_info, link) {
+		if (ui->tid == tid)
+			goto done;
+	}
+	ui = NULL;
+done:
+	spin_unlock(&itrace_lock);
+	return ui;
+}
+
+
+int usr_itrace_init(int single_step, pid_t tid, struct stap_itrace_probe *p)
+{
+	struct itrace_info *ui;
+	struct task_struct *tsk;
+
+	rcu_read_lock();
+	tsk = find_task_by_pid(tid);
+	if (!tsk) {
+		printk(KERN_ERR "usr_itrace_init: Cannot find process %d\n", tid);
+		rcu_read_unlock();
+		return 1;
+	}
+
+	get_task_struct(tsk);
+	ui = create_itrace_info(tsk, 
+		(single_step ?
+			UTRACE_ACTION_SINGLESTEP : UTRACE_ACTION_BLOCKSTEP), p);
+	if (!ui)
+		return 1;
+
+	put_task_struct(tsk);
+	rcu_read_unlock();
+
+	spin_lock_init(&itrace_lock);
+
+	/* set initial state */
+	spin_lock(&itrace_lock);
+	spin_unlock(&itrace_lock);
+	printk(KERN_INFO "usr_itrace_init: completed for tid = %d\n", tid);
+
+	return 0;
+}
+
+void static remove_usr_itrace_info(struct itrace_info *ui)
+{
+	struct itrace_info *tmp;
+
+	if (!ui)
+		return;
+
+	if (debug)
+		printk(KERN_INFO "remove_usr_itrace_info: tid=%d\n", ui->tid);
+
+	spin_lock(&itrace_lock);
+	if (ui->tsk && ui->engine) {
+		(void) utrace_detach(ui->tsk, ui->engine);
+	}
+	list_del(&ui->link);
+	spin_unlock(&itrace_lock);
+	kfree(ui);
+}
+
+void static cleanup_usr_itrace(void)
+{
+	struct itrace_info *tmp;
+	struct itrace_info *ui;
+
+	if (debug)
+		printk(KERN_INFO "cleanup_usr_itrace called\n");
+
+	list_for_each_entry_safe(ui, tmp, &usr_itrace_info, link) {
+		remove_usr_itrace_info(ui);
+	}
+}
+
+
+#ifdef CONFIG_PPC
+#define PPC_INSTR_SIZE 4
+#define TEXT_SEGMENT_BASE 1
+
+/* Instruction masks used during single-stepping of atomic sequences.  */
+#define LWARX_MASK 0xfc0007fe
+#define LWARX_INSTR 0x7c000028
+#define LDARX_INSTR 0x7c0000A8
+#define STWCX_MASK 0xfc0007ff
+#define STWCX_INSTR 0x7c00012d
+#define STDCX_INSTR 0x7c0001ad
+#define BC_MASK 0xfc000000
+#define BC_INSTR 0x40000000
+#define ATOMIC_SEQ_LENGTH 16
+#define BPT_TRAP 0x7fe00008
+#define INSTR_SZ sizeof(int)
+
+static int get_instr(unsigned long addr, char *msg)
+{
+	unsigned int instr;
+
+	if (copy_from_user(&instr, (const void __user *) addr,
+			sizeof(instr))) {
+		printk(KERN_ERR "get_instr failed: %s\n", msg);
+		WARN_ON(1);
+	}
+	return instr;
+
+}
+
+static void insert_atomic_ss_breakpoint (struct task_struct *tsk,
+	struct bpt_info *bpt)
+{
+	unsigned int bp_instr = BPT_TRAP;
+	unsigned int cur_instr;
+
+	cur_instr = get_instr(bpt->addr, "insert_atomic_ss_breakpoint");
+	if (cur_instr != BPT_TRAP) {
+		bpt->instr = cur_instr;
+		WARN_ON(access_process_vm(tsk, bpt->addr, &bp_instr, INSTR_SZ, 1) !=
+			INSTR_SZ);
+	}
+}
+
+static void remove_atomic_ss_breakpoint (struct task_struct *tsk,
+	struct bpt_info *bpt)
+{
+	WARN_ON(access_process_vm(tsk, bpt->addr, &bpt->instr, INSTR_SZ, 1) !=
+		INSTR_SZ);
+}
+
+/* locate the branch destination.  Return -1 if not a branch.  */
+static unsigned long
+branch_dest (int opcode, int instr, struct pt_regs *regs, unsigned long pc)
+{
+	unsigned long dest;
+	int immediate;
+	int absolute;
+	int ext_op;
+
+	absolute = (int) ((instr >> 1) & 1);
+
+	switch (opcode) {
+	case 18:
+		immediate = ((instr & ~3) << 6) >> 6;	/* br unconditional */
+		if (absolute)
+			dest = immediate;
+		else
+			dest = pc + immediate;
+		break;
+
+	case 16:
+		immediate = ((instr & ~3) << 16) >> 16;	/* br conditional */
+		if (absolute)
+			dest = immediate;
+		else
+			dest = pc + immediate;
+		break;
+
+	case 19:
+		ext_op = (instr >> 1) & 0x3ff;
+
+		if (ext_op == 16) {
+			/* br conditional register */
+			dest = regs->link & ~3;
+			/* FIX: we might be in a signal handler */
+			WARN_ON(dest > 0);
+		} else if (ext_op == 528) {
+			/* br cond to ctr reg */
+			dest = regs->ctr & ~3;
+
+			/* for system call dest < TEXT_SEGMENT_BASE */
+			if (dest < TEXT_SEGMENT_BASE)
+				dest = regs->link & ~3;
+		} else
+			return -1;
+		break;
+
+	default:
+		return -1;
+	}
+	return dest;
+}
+
+/* Checks for an atomic sequence of instructions beginning with a LWARX/LDARX
+   instruction and ending with a STWCX/STDCX instruction.  If such a sequence
+   is found, attempt to step through it.  A breakpoint is placed at the end of 
+   the sequence.  */
+
+static int handle_ppc_atomic_seq(struct task_struct *tsk, struct pt_regs *regs,
+	struct atomic_ss_info *ss_info)
+{
+	unsigned long ip = regs->nip;
+	unsigned long start_addr;
+	unsigned int instr;
+	int got_stx = 0;
+	int i;
+	int ret;
+
+	unsigned long br_dest; /* bpt at branch instr's destination */
+	int bc_instr_count = 0; /* conditional branch instr count  */
+
+	instr = get_instr(regs->nip, "handle_ppc_atomic_seq:1");
+	/* Beginning of atomic sequence starts with lwarx/ldarx instr */
+	if ((instr & LWARX_MASK) != LWARX_INSTR
+		&& (instr & LWARX_MASK) != LDARX_INSTR)
+		return 0;
+
+	start_addr = regs->nip;
+	for (i = 0; i < ATOMIC_SEQ_LENGTH; ++i) {
+		ip += INSTR_SZ;
+		instr = get_instr(ip, "handle_ppc_atomic_seq:2");
+
+		/* look for at most one conditional branch in the sequence
+		 * and put a bpt at it's destination address
+		 */
+		if ((instr & BC_MASK) == BC_INSTR) {
+			if (bc_instr_count >= 1)
+				return 0; /* only handle a single branch */
+
+			br_dest = branch_dest (BC_INSTR >> 26, instr, regs, ip);
+
+			if (br_dest != -1 &&
+				br_dest >= TEXT_SEGMENT_BASE) {
+				ss_info->br_bpt.addr = br_dest;
+				bc_instr_count++;
+			}
+		}
+
+		if ((instr & STWCX_MASK) == STWCX_INSTR
+			|| (instr & STWCX_MASK) == STDCX_INSTR) {
+			got_stx = 1;
+			break;
+		}
+	}
+
+	/* Atomic sequence ends with a stwcx/stdcx instr */
+	if (!got_stx)
+		return 0;
+
+	ip += INSTR_SZ;
+	instr = get_instr(ip, "handle_ppc_atomic_seq:3");
+	if ((instr & BC_MASK) == BC_INSTR) {
+		ip += INSTR_SZ;
+		instr = get_instr(ip, "handle_ppc_atomic_seq:4");
+	}
+
+	/* Insert a breakpoint right after the end of the atomic sequence.  */
+	ss_info->end_bpt.addr = ip;
+
+	/* Check for duplicate bpts */
+	if (bc_instr_count && (ss_info->br_bpt.addr >= start_addr &&
+		ss_info->br_bpt.addr <= ss_info->end_bpt.addr))
+		ss_info->br_bpt.addr = 0;
+
+	insert_atomic_ss_breakpoint (tsk, &ss_info->end_bpt);
+	if (ss_info->br_bpt.addr)
+		insert_atomic_ss_breakpoint (tsk, &ss_info->br_bpt);
+
+	ss_info->step_over_atomic = 1;
+	return 1;
+}
+#endif
diff --git a/session.h b/session.h
index 734c8d7d..9f38372f 100644
--- a/session.h
+++ b/session.h
@@ -32,6 +32,7 @@ struct be_derived_probe_group;
 struct dwarf_derived_probe_group;
 struct uprobe_derived_probe_group;
 struct utrace_derived_probe_group;
+struct itrace_derived_probe_group;
 struct task_finder_derived_probe_group;
 struct timer_derived_probe_group;
 struct profile_derived_probe_group;
@@ -146,6 +147,7 @@ struct systemtap_session
   dwarf_derived_probe_group* dwarf_derived_probes;
   uprobe_derived_probe_group* uprobe_derived_probes;
   utrace_derived_probe_group* utrace_derived_probes;
+  itrace_derived_probe_group* itrace_derived_probes;
   task_finder_derived_probe_group* task_finder_derived_probes;
   timer_derived_probe_group* timer_derived_probes;
   profile_derived_probe_group* profile_derived_probes;
diff --git a/stapprobes.5.in b/stapprobes.5.in
index 700452a7..fcc44df1 100644
--- a/stapprobes.5.in
+++ b/stapprobes.5.in
@@ -367,6 +367,8 @@ process(PID).syscall
 process("PATH").syscall
 process(PID).syscall.return
 process("PATH").syscall.return
+process(PID).itrace
+process("PATH").itrace
 .ESAMPLE
 .PP
 A
@@ -391,6 +393,9 @@ A
 probe gets called when a thread described by PID or PATH returns from a
 system call.  The system call number is available in the "$syscall"
 context variable.
+A
+.B .itrace
+probe gets called for every single step of the process described by PID or PATH.
 .PP
 Note that
 .I PATH
diff --git a/systemtap.base/itrace.exp b/systemtap.base/itrace.exp
new file mode 100644
index 00000000..4b73ac1c
--- /dev/null
+++ b/systemtap.base/itrace.exp
@@ -0,0 +1,102 @@
+# itrace test
+
+# Initialize variables
+set utrace_support_found 0
+set exepath "[pwd]/ls_[pid]"
+
+set itrace1_script {
+    global instrs = 0
+    probe begin { printf("systemtap starting probe\n") }
+    probe process("%s").itrace
+    {
+        instrs += 1
+	if (instrs == 5)
+		exit()
+    }
+
+
+    probe end { printf("systemtap ending probe\n")
+          printf("itraced = %%d\n", instrs)
+    }
+}
+set itrace1_script_output "itraced = 5\r\n"
+
+set itrace2_script {
+    global instrs = 0, itrace_on = 0, start_timer = 0
+    probe begin { start_timer = 1; printf("systemtap starting probe\n") }
+    probe process("%s").itrace if (itrace_on)
+    {
+        instrs += 1
+	if (instrs == 5)
+		exit()
+    }
+
+
+    probe timer.ms(1) if (start_timer)
+    {   
+        itrace_on = 1
+    }
+
+    probe timer.ms(10) if (start_timer)
+    {   
+        itrace_on = 0
+    }
+    probe end { printf("systemtap ending probe\n")
+          printf("itraced = %%d\n", instrs)
+    }
+}
+set itrace2_script_output "itraced = 5\r\n"
+
+
+# Set up our own copy of /bin/ls, to make testing for a particular
+# executable easy.  We can't use 'ln' here, since we might be creating
+# a cross-device link.  We can't use 'ln -s' here, since the kernel
+# resolves the symbolic link and reports that /bin/ls is being
+# exec'ed (instead of our local copy).
+if {[catch {exec cp /bin/ls $exepath} res]} {
+    fail "unable to copy /bin/ls: $res"
+    return
+}
+
+# "load" generation function for stap_run.  It spawns our own copy of
+# /bin/ls, waits 5 seconds, then kills it.
+proc run_ls_5_sec {} {
+    global exepath
+
+    spawn $exepath
+    set exe_id $spawn_id
+    after 5000;
+    exec kill -INT -[exp_pid -i $exe_id]
+    return 0;
+}
+
+
+# Try to find utrace_attach symbol in /proc/kallsyms
+set path "/proc/kallsyms"
+if {! [catch {exec grep -q utrace_attach $path} dummy]} {
+    set utrace_support_found 1
+}
+
+set TEST_NAME "itrace1"
+if {$utrace_support_found == 0} {
+    untested "$TEST_NAME : no kernel utrace support found"
+} elseif {![installtest_p]} {
+    untested "$TEST_NAME : not installtest_p"
+} else {
+    set script [format $itrace1_script $exepath]
+    stap_run $TEST_NAME run_ls_5_sec $itrace1_script_output -e $script
+}
+
+
+set TEST_NAME "itrace2"
+if {$utrace_support_found == 0} {
+    untested "$TEST_NAME : no kernel utrace support found"
+} elseif {![installtest_p]} {
+    untested "$TEST_NAME : not installtest_p"
+} else {
+    set script [format $itrace2_script $exepath]
+    stap_run $TEST_NAME run_ls_5_sec $itrace2_script_output -e $script
+}
+
+# Cleanup
+exec rm -f $exepath
diff --git a/tapsets.cxx b/tapsets.cxx
index 5a3fbdc8..83463f85 100644
--- a/tapsets.cxx
+++ b/tapsets.cxx
@@ -5342,6 +5342,280 @@ task_finder_derived_probe_group::emit_module_exit (systemtap_session& s)
   s.op->newline() << "stap_stop_task_finder();";
 }
 
+// ------------------------------------------------------------------------
+// itrace user-space probes
+// ------------------------------------------------------------------------
+
+
+struct itrace_derived_probe: public derived_probe
+{
+  bool has_path;
+  string path;
+  int64_t pid;
+  int single_step;
+
+  itrace_derived_probe (systemtap_session &s, probe* p, probe_point* l,
+                        bool hp, string &pn, int64_t pd, int ss
+			);
+  void join_group (systemtap_session& s);
+};
+
+
+struct itrace_derived_probe_group: public generic_dpg<itrace_derived_probe>
+{
+private:
+  map<string, vector<itrace_derived_probe*> > probes_by_path;
+  typedef map<string, vector<itrace_derived_probe*> >::iterator p_b_path_iterator;
+  map<int64_t, vector<itrace_derived_probe*> > probes_by_pid;
+  typedef map<int64_t, vector<itrace_derived_probe*> >::iterator p_b_pid_iterator;
+  unsigned num_probes;
+
+  void emit_probe_decl (systemtap_session& s, itrace_derived_probe *p);
+
+public:
+  itrace_derived_probe_group(): num_probes(0) { }
+
+  void enroll (itrace_derived_probe* probe);
+  void emit_module_decls (systemtap_session& s);
+  void emit_module_init (systemtap_session& s);
+  void emit_module_exit (systemtap_session& s);
+};
+
+
+itrace_derived_probe::itrace_derived_probe (systemtap_session &s,
+                                            probe* p, probe_point* l,
+                                            bool hp, string &pn, int64_t pd,
+											int ss
+					    ):
+  derived_probe(p, l), has_path(hp), path(pn), pid(pd), single_step(ss)
+{
+}
+
+
+void
+itrace_derived_probe::join_group (systemtap_session& s)
+{
+  if (! s.itrace_derived_probes)
+      s.itrace_derived_probes = new itrace_derived_probe_group ();
+
+  s.itrace_derived_probes->enroll (this);
+
+  task_finder_derived_probe_group::create_session_group (s);
+}
+
+struct itrace_builder: public derived_probe_builder
+{
+  itrace_builder() {}
+  virtual void build(systemtap_session & sess,
+		     probe * base,
+		     probe_point * location,
+		     std::map<std::string, literal *> const & parameters,
+		     vector<derived_probe *> & finished_results)
+  {
+    string path;
+    int64_t pid;
+    int single_step;
+
+    bool has_path = get_param (parameters, TOK_PROCESS, path);
+    bool has_pid = get_param (parameters, TOK_PROCESS, pid);
+    assert (has_path || has_pid);
+
+    single_step = 1;
+
+    // If we have a path, we need to validate it.
+    if (has_path)
+    {
+	string::size_type start_pos, end_pos;
+	string component;
+
+	// Make sure it starts with '/'.
+	if (path[0] != '/')
+	    throw semantic_error ("process path must start with a '/'",
+				  location->tok);
+
+	start_pos = 1;			// get past the initial '/'
+	while ((end_pos = path.find('/', start_pos)) != string::npos)
+        {
+	    component = path.substr(start_pos, end_pos - start_pos);
+	    // Make sure it isn't empty.
+	    if (component.size() == 0)
+		throw semantic_error ("process path component cannot be empty",
+				      location->tok);
+	    // Make sure it isn't relative.
+	    else if (component == "." || component == "..")
+		throw semantic_error ("process path cannot be relative (and contain '.' or '..')", location->tok);
+
+	    start_pos = end_pos + 1;
+	}
+	component = path.substr(start_pos);
+	// Make sure it doesn't end with '/'.
+	if (component.size() == 0)
+	    throw semantic_error ("process path cannot end with a '/'", location->tok);
+	// Make sure it isn't relative.
+	else if (component == "." || component == "..")
+	    throw semantic_error ("process path cannot be relative (and contain '.' or '..')", location->tok);
+    }
+
+    finished_results.push_back(new itrace_derived_probe(sess, base, location,
+                                                        has_path, path, pid,
+																	single_step
+							));
+  }
+};
+
+
+void
+itrace_derived_probe_group::enroll (itrace_derived_probe* p)
+{
+  if (p->has_path)
+    probes_by_path[p->path].push_back(p);
+  else
+    probes_by_pid[p->pid].push_back(p);
+  num_probes++;
+
+  // XXX: multiple exec probes (for instance) for the same path (or
+  // pid) should all share a itrace report function, and have their
+  // handlers executed sequentially.
+}
+
+
+void
+itrace_derived_probe_group::emit_probe_decl (systemtap_session& s,
+					     itrace_derived_probe *p)
+{
+  s.op->newline() << "{";
+  s.op->line() << " .tgt={";
+
+  if (p->has_path)
+    {
+      s.op->line() << " .pathname=\"" << p->path << "\",";
+      s.op->line() << " .pid=0,";
+    }
+  else
+    {
+      s.op->line() << " .pathname=NULL,";
+      s.op->line() << " .pid=" << p->pid << ",";
+    }
+
+  s.op->line() << " .callback=&_stp_itrace_probe_cb,";
+  s.op->line() << " },";
+  s.op->line() << " .pp=" << lex_cast_qstring (*p->sole_location()) << ",";
+  s.op->line() << " .single_step=" << p->single_step << ",";
+  s.op->line() << " .ph=&" << p->name << ",";
+
+  s.op->line() << " },";
+}
+
+
+void
+itrace_derived_probe_group::emit_module_decls (systemtap_session& s)
+{
+  if (probes_by_path.empty() && probes_by_pid.empty())
+    return;
+
+  s.op->newline();
+  s.op->newline() << "/* ---- itrace probes ---- */";
+  s.op->newline() << "struct stap_itrace_probe {";
+  s.op->indent(1);
+  s.op->newline() << "struct stap_task_finder_target tgt;";
+  s.op->newline() << "const char *pp;";
+  s.op->newline() << "void (*ph) (struct context*);";
+  s.op->newline() << "int single_step;";
+  s.op->newline(-1) << "};";
+  s.op->newline() << "static void enter_itrace_probe(struct stap_itrace_probe *p, struct pt_regs *regs, void *data);";
+  s.op->newline() << "#include \"itrace.c\"";
+
+  // output routine to call itrace probe
+  s.op->newline() << "static void enter_itrace_probe(struct stap_itrace_probe *p, struct pt_regs *regs, void *data) {";
+  s.op->indent(1);
+
+  common_probe_entryfn_prologue (s.op, "STAP_SESSION_RUNNING");
+  s.op->newline() << "c->probe_point = p->pp;";
+  s.op->newline() << "c->regs = regs;";
+  s.op->newline() << "c->data = data;";
+
+  // call probe function
+  s.op->newline() << "(*p->ph) (c);";
+  common_probe_entryfn_epilogue (s.op);
+
+  s.op->newline() << "return;";
+  s.op->newline(-1) << "}";
+
+  // Output task finder callback routine that gets called for all
+  // itrace probe types.
+  s.op->newline() << "static int _stp_itrace_probe_cb(struct task_struct *tsk, int register_p, int process_p, struct stap_task_finder_target *tgt) {";
+  s.op->indent(1);
+  s.op->newline() << "int rc = 0;";
+  s.op->newline() << "struct stap_itrace_probe *p = container_of(tgt, struct stap_itrace_probe, tgt);";
+
+  s.op->newline() << "if (register_p) ";
+  s.op->indent(1);
+
+  s.op->newline() << "rc = usr_itrace_init(p->single_step, tsk->pid, p);";
+  s.op->newline(-1) << "else";
+  s.op->newline(1) << "remove_usr_itrace_info(find_itrace_info(p->tgt.pid));";
+  s.op->newline(-1) << "return rc;";
+  s.op->newline(-1) << "}";
+
+  s.op->newline() << "struct stap_itrace_probe stap_itrace_probes[] = {";
+  s.op->indent(1);
+
+  // Set up 'process(PATH)' probes
+  if (! probes_by_path.empty())
+    {
+      for (p_b_path_iterator it = probes_by_path.begin();
+	   it != probes_by_path.end(); it++)
+        {
+	  for (unsigned i = 0; i < it->second.size(); i++)
+	    {
+	      itrace_derived_probe *p = it->second[i];
+	      emit_probe_decl(s, p);
+	    }
+	}
+    }
+
+  // Set up 'process(PID)' probes
+  if (! probes_by_pid.empty())
+    {
+      for (p_b_pid_iterator it = probes_by_pid.begin();
+	   it != probes_by_pid.end(); it++)
+        {
+	  for (unsigned i = 0; i < it->second.size(); i++)
+	    {
+	      itrace_derived_probe *p = it->second[i];
+	      emit_probe_decl(s, p);
+	    }
+	}
+    }
+  s.op->newline(-1) << "};";
+}
+
+
+void
+itrace_derived_probe_group::emit_module_init (systemtap_session& s)
+{
+  if (probes_by_path.empty() && probes_by_pid.empty())
+    return;
+
+  s.op->newline();
+  s.op->newline() << "/* ---- itrace probes ---- */";
+
+  s.op->newline() << "for (i=0; i<" << num_probes << "; i++) {";
+  s.op->indent(1);
+  s.op->newline() << "struct stap_itrace_probe *p = &stap_itrace_probes[i];";
+  s.op->newline() << "rc = stap_register_task_finder_target(&p->tgt);";
+  s.op->newline(-1) << "}";
+}
+
+
+void
+itrace_derived_probe_group::emit_module_exit (systemtap_session& s)
+{
+  if (probes_by_path.empty() && probes_by_pid.empty()) return;
+  s.op->newline();
+  s.op->newline() << "/* ---- itrace probes ---- */";
+  s.op->newline() << "cleanup_usr_itrace();";
+}
 
 // ------------------------------------------------------------------------
 // utrace user-space probes
@@ -8483,6 +8757,13 @@ register_standard_tapsets(systemtap_session & s)
     ->bind(new utrace_builder ());
   s.pattern_root->bind_num(TOK_PROCESS)->bind(TOK_THREAD)->bind(TOK_END)
     ->bind(new utrace_builder ());
+
+  // itrace user-space probes
+  s.pattern_root->bind_str(TOK_PROCESS)->bind("itrace")
+    ->bind(new itrace_builder ());
+  s.pattern_root->bind_num(TOK_PROCESS)->bind("itrace")
+    ->bind(new itrace_builder ());
+
   s.pattern_root->bind_str(TOK_PROCESS)->bind(TOK_SYSCALL)
     ->bind(new utrace_builder ());
   s.pattern_root->bind_num(TOK_PROCESS)->bind(TOK_SYSCALL)
@@ -8534,6 +8815,7 @@ all_session_groups(systemtap_session& s)
   // "register" the dummy task_finder probe group after all probe
   // groups that use the task_finder.
   DOONE(utrace);
+  DOONE(itrace);
   DOONE(task_finder);
 #undef DOONE
   return g;
diff --git a/testsuite/ChangeLog b/testsuite/ChangeLog
index 14243b80..b69a8dfa 100644
--- a/testsuite/ChangeLog
+++ b/testsuite/ChangeLog
@@ -1,3 +1,7 @@
+2008-07-24  Dave Nomura  <dcnltc@us.ibm.com>
+
+	* systemtap.base/itrace.stp: Added simple tests of itrace probe.
+
 2008-07-14  Dave Brolley  <brolley@redhat.com>
 
 	* Makefile.am (TOOL_OPTS): New variable.