uprobes.c, uprobes.h: architecture-independent code

uprobes_i386.c, uprobes_i386.h: i386 code
uprobes_arch.c, uprobes_arch.h: map to architecture-specific files

6 files changed, 3073 insertions, 0 deletions

diff --git a/runtime/uprobes/uprobes.c b/runtime/uprobes/uprobes.c
new file mode 100644
index 00000000..41d0ef11
--- /dev/null
+++ b/runtime/uprobes/uprobes.c
@@ -0,0 +1,2297 @@
+/*
+ *  Userspace Probes (UProbes)
+ *  kernel/uprobes_core.c
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2006
+ */
+#include <linux/types.h>
+#include <linux/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/rcupdate.h>
+#include <linux/err.h>
+#include <linux/kref.h>
+#include <linux/utrace.h>
+#define UPROBES_IMPLEMENTATION 1
+#include "uprobes.h"
+#include <linux/tracehook.h>
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/uaccess.h>
+#include <asm/tracehook.h>
+#include <asm/errno.h>
+#include <asm/mman.h>
+
+#define SET_ENGINE_FLAGS	1
+#define CLEAR_ENGINE_FLAGS	0
+
+extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
+	void *buf, int len, int write);
+static int utask_fake_quiesce(struct uprobe_task *utask);
+static void uprobe_release_ssol_vma(struct uprobe_process *uproc);
+
+static void uretprobe_handle_entry(struct uprobe *u, struct pt_regs *regs,
+	struct uprobe_task *utask);
+static void uretprobe_handle_return(struct pt_regs *regs,
+	struct uprobe_task *utask);
+static void uretprobe_set_trampoline(struct uprobe_process *uproc);
+static void zap_uretprobe_instances(struct uprobe *u,
+	struct uprobe_process *uproc);
+
+typedef void (*uprobe_handler_t)(struct uprobe*, struct pt_regs*);
+#define URETPROBE_HANDLE_ENTRY ((uprobe_handler_t)-1L)
+#define is_uretprobe(u) (u->handler == URETPROBE_HANDLE_ENTRY)
+/* Point utask->active_probe at this while running uretprobe handler. */
+static struct uprobe_probept uretprobe_trampoline_dummy_probe;
+
+/*
+ * These data structures are shared by all SystemTap-generated modules
+ * that use uprobes.
+ */
+struct uprobe_globals {
+	struct hlist_head uproc_table[UPROBE_TABLE_SIZE];
+	struct mutex uproc_mutex;
+	struct hlist_head utask_table[UPROBE_TABLE_SIZE];
+	spinlock_t utask_table_lock;
+};
+static struct uprobe_globals *globals;
+static struct hlist_head *uproc_table;	/* = globals->uproc_table */
+static struct hlist_head *utask_table;	/* = globals->utask_table */
+
+#define lock_uproc_table() mutex_lock(&globals->uproc_mutex)
+#define unlock_uproc_table() mutex_unlock(&globals->uproc_mutex)
+
+#define lock_utask_table(flags) \
+	spin_lock_irqsave(&globals->utask_table_lock, (flags))
+#define unlock_utask_table(flags) \
+	spin_unlock_irqrestore(&globals->utask_table_lock, (flags))
+
+/*
+ * uprobes_data and uprobes_mutex are the only uprobes hooks in the kernel.
+ * A pointer to the uprobes_global area is stored in uprobe_data.
+ */
+extern void *uprobes_data;
+extern struct mutex uprobes_mutex;
+
+static int verify_uprobes(void);
+
+/* p_uprobe_utrace_ops = &uprobe_utrace_ops.  Fwd refs are a pain w/o this. */
+static const struct utrace_engine_ops *p_uprobe_utrace_ops;
+
+struct deferred_registration {
+	struct list_head list;
+	struct uprobe *uprobe;
+	int regflag;	/* 0 - unregister, 1 - register */
+	enum uprobe_type type;
+};
+
+static struct uprobe_task *uprobe_find_utask(struct task_struct *tsk)
+{
+	struct hlist_head *head;
+	struct hlist_node *node;
+	struct uprobe_task *utask;
+	unsigned long flags;
+
+	head = &utask_table[hash_ptr(tsk, UPROBE_HASH_BITS)];
+	lock_utask_table(flags);
+	hlist_for_each_entry(utask, node, head, hlist) {
+		if (utask->tsk == tsk) {
+			unlock_utask_table(flags);
+			return utask;
+		}
+	}
+	unlock_utask_table(flags);
+	return NULL;
+}
+
+static void uprobe_hash_utask(struct uprobe_task *utask)
+{
+	struct hlist_head *head;
+	unsigned long flags;
+
+	INIT_HLIST_NODE(&utask->hlist);
+	head = &utask_table[hash_ptr(utask->tsk, UPROBE_HASH_BITS)];
+	lock_utask_table(flags);
+	hlist_add_head(&utask->hlist, head);
+	unlock_utask_table(flags);
+}
+
+static void uprobe_unhash_utask(struct uprobe_task *utask)
+{
+	unsigned long flags;
+
+	lock_utask_table(flags);
+	hlist_del(&utask->hlist);
+	unlock_utask_table(flags);
+}
+
+static inline void uprobe_get_process(struct uprobe_process *uproc)
+{
+	atomic_inc(&uproc->refcount);
+}
+
+/*
+ * Decrement uproc's refcount in a situation where we "know" it can't
+ * reach zero.  It's OK to call this with uproc locked.  Compare with
+ * uprobe_put_process().
+ */
+static inline void uprobe_decref_process(struct uprobe_process *uproc)
+{
+	if (atomic_dec_and_test(&uproc->refcount))
+		BUG();
+}
+
+/*
+ * Runs with the uproc_mutex held.  Returns with uproc ref-counted and
+ * write-locked.
+ *
+ * Around exec time, briefly, it's possible to have one (finished) uproc
+ * for the old image and one for the new image.  We find the latter.
+ */
+static struct uprobe_process *uprobe_find_process(pid_t tgid)
+{
+	struct hlist_head *head;
+	struct hlist_node *node;
+	struct uprobe_process *uproc;
+
+	head = &uproc_table[hash_long(tgid, UPROBE_HASH_BITS)];
+	hlist_for_each_entry(uproc, node, head, hlist) {
+		if (uproc->tgid == tgid && !uproc->finished) {
+			uprobe_get_process(uproc);
+			down_write(&uproc->rwsem);
+			return uproc;
+		}
+	}
+	return NULL;
+}
+
+/*
+ * In the given uproc's hash table of probepoints, find the one with the
+ * specified virtual address.  Runs with uproc->rwsem locked.
+ */
+static struct uprobe_probept *uprobe_find_probept(struct uprobe_process *uproc,
+		unsigned long vaddr)
+{
+	struct uprobe_probept *ppt;
+	struct hlist_node *node;
+	struct hlist_head *head = &uproc->uprobe_table[hash_long(vaddr,
+		UPROBE_HASH_BITS)];
+
+	hlist_for_each_entry(ppt, node, head, ut_node) {
+		if (ppt->vaddr == vaddr && ppt->state != UPROBE_DISABLED)
+			return ppt;
+	}
+	return NULL;
+}
+
+/*
+ * set_bp: Store a breakpoint instruction at ppt->vaddr.
+ * Returns BP_INSN_SIZE on success.
+ *
+ * NOTE: BREAKPOINT_INSTRUCTION on all archs is the same size as
+ * uprobe_opcode_t.
+ */
+static int set_bp(struct uprobe_probept *ppt, struct task_struct *tsk)
+{
+	uprobe_opcode_t bp_insn = BREAKPOINT_INSTRUCTION;
+	return access_process_vm(tsk, ppt->vaddr, &bp_insn, BP_INSN_SIZE, 1);
+}
+
+/*
+ * set_orig_insn:  For probepoint ppt, replace the breakpoint instruction
+ * with the original opcode.  Returns BP_INSN_SIZE on success.
+ */
+static int set_orig_insn(struct uprobe_probept *ppt, struct task_struct *tsk)
+{
+	return access_process_vm(tsk, ppt->vaddr, &ppt->opcode, BP_INSN_SIZE,
+		1);
+}
+
+static void bkpt_insertion_failed(struct uprobe_probept *ppt, const char *why)
+{
+	printk(KERN_ERR "Can't place uprobe at pid %d vaddr %#lx: %s\n",
+			ppt->uproc->tgid, ppt->vaddr, why);
+}
+
+/*
+ * Save a copy of the original instruction (so it can be single-stepped
+ * out of line), insert the breakpoint instruction, and awake
+ * register_uprobe().
+ */
+static void insert_bkpt(struct uprobe_probept *ppt, struct task_struct *tsk)
+{
+	struct uprobe_kimg *uk;
+	long result = 0;
+	int len;
+
+	if (!tsk) {
+		/* No surviving tasks associated with ppt->uproc */
+		result = -ESRCH;
+		goto out;
+	}
+
+	/*
+	 * If access_process_vm() transfers fewer bytes than the maximum
+	 * instruction size, assume that the probed instruction is smaller
+	 * than the max and near the end of the last page of instructions.
+	 * But there must be room at least for a breakpoint-size instruction.
+	 */
+	len = access_process_vm(tsk, ppt->vaddr, ppt->insn, MAX_UINSN_BYTES, 0);
+	if (len < BP_INSN_SIZE) {
+		bkpt_insertion_failed(ppt,
+			"error reading original instruction");
+		result = -EIO;
+		goto out;
+	}
+	memcpy(&ppt->opcode, ppt->insn, BP_INSN_SIZE);
+	if (ppt->opcode == BREAKPOINT_INSTRUCTION) {
+		bkpt_insertion_failed(ppt, "bkpt already exists at that addr");
+		result = -EEXIST;
+		goto out;
+	}
+
+	if ((result = arch_validate_probed_insn(ppt)) < 0) {
+		bkpt_insertion_failed(ppt, "instruction type cannot be probed");
+		goto out;
+	}
+
+	len = set_bp(ppt, tsk);
+	if (len < BP_INSN_SIZE) {
+		bkpt_insertion_failed(ppt, "failed to insert bkpt instruction");
+		result = -EIO;
+		goto out;
+	}
+out:
+	ppt->state = (result ? UPROBE_DISABLED : UPROBE_BP_SET);
+	list_for_each_entry(uk, &ppt->uprobe_list, list)
+		uk->status = result;
+	wake_up_all(&ppt->waitq);
+}
+
+static void remove_bkpt(struct uprobe_probept *ppt, struct task_struct *tsk)
+{
+	int len;
+
+	if (tsk) {
+		len = set_orig_insn(ppt, tsk);
+		if (len < BP_INSN_SIZE) {
+			printk(KERN_ERR
+				"Error removing uprobe at pid %d vaddr %#lx:"
+				" can't restore original instruction\n",
+				tsk->tgid, ppt->vaddr);
+			/*
+			 * This shouldn't happen, since we were previously
+			 * able to write the breakpoint at that address.
+			 * There's not much we can do besides let the
+			 * process die with a SIGTRAP the next time the
+			 * breakpoint is hit.
+			 */
+		}
+	}
+	/* Wake up unregister_uprobe(). */
+	ppt->state = UPROBE_DISABLED;
+	wake_up_all(&ppt->waitq);
+}
+
+/*
+ * Runs with all of uproc's threads quiesced and uproc->rwsem write-locked.
+ * As specified, insert or remove the breakpoint instruction for each
+ * uprobe_probept on uproc's pending list.
+ * tsk = one of the tasks associated with uproc -- NULL if there are
+ * no surviving threads.
+ * It's OK for uproc->pending_uprobes to be empty here.  It can happen
+ * if a register and an unregister are requested (by different probers)
+ * simultaneously for the same pid/vaddr.
+ * Note that the current task may be a thread in uproc, or it may be
+ * a task running [un]register_uprobe() (or both).
+ */
+static void handle_pending_uprobes(struct uprobe_process *uproc,
+	struct task_struct *tsk)
+{
+	struct uprobe_probept *ppt, *tmp;
+
+	list_for_each_entry_safe(ppt, tmp, &uproc->pending_uprobes, pd_node) {
+		switch (ppt->state) {
+		case UPROBE_INSERTING:
+			insert_bkpt(ppt, tsk);
+			break;
+		case UPROBE_REMOVING:
+			remove_bkpt(ppt, tsk);
+			break;
+		default:
+			BUG();
+		}
+		list_del(&ppt->pd_node);
+	}
+}
+
+static void utask_adjust_flags(struct uprobe_task *utask, int set,
+	unsigned long flags)
+{
+	unsigned long newflags, oldflags;
+
+	newflags = oldflags = utask->engine->flags;
+
+	if (set)
+		newflags |= flags;
+	else
+		newflags &= ~flags;
+
+	if (newflags != oldflags)
+		utrace_set_flags(utask->tsk, utask->engine, newflags);
+}
+
+static inline void clear_utrace_quiesce(struct uprobe_task *utask)
+{
+	utask_adjust_flags(utask, CLEAR_ENGINE_FLAGS,
+			UTRACE_ACTION_QUIESCE | UTRACE_EVENT(QUIESCE));
+}
+
+/* Opposite of quiesce_all_threads().  Same locking applies. */
+static void rouse_all_threads(struct uprobe_process *uproc)
+{
+	struct uprobe_task *utask;
+
+	list_for_each_entry(utask, &uproc->thread_list, list) {
+		if (utask->quiescing) {
+			utask->quiescing = 0;
+			if (utask->state == UPTASK_QUIESCENT) {
+				utask->state = UPTASK_RUNNING;
+				uproc->n_quiescent_threads--;
+				clear_utrace_quiesce(utask);
+			}
+		}
+	}
+	/* Wake any threads that decided to sleep rather than quiesce. */
+	wake_up_all(&uproc->waitq);
+}
+
+/*
+ * If all of uproc's surviving threads have quiesced, do the necessary
+ * breakpoint insertions or removals and then un-quiesce everybody.
+ * tsk is a surviving thread, or NULL if there is none.  Runs with
+ * uproc->rwsem write-locked.
+ */
+static void check_uproc_quiesced(struct uprobe_process *uproc,
+		struct task_struct *tsk)
+{
+	if (uproc->n_quiescent_threads >= uproc->nthreads) {
+		handle_pending_uprobes(uproc, tsk);
+		rouse_all_threads(uproc);
+	}
+}
+
+/*
+ * Quiesce all threads in the specified process -- e.g., prior to
+ * breakpoint insertion.  Runs with uproc->rwsem write-locked.
+ * Returns the number of threads that haven't died yet.
+ */
+static int quiesce_all_threads(struct uprobe_process *uproc,
+		struct uprobe_task **cur_utask_quiescing)
+{
+	struct uprobe_task *utask;
+	struct task_struct *survivor = NULL;	// any survivor
+	int survivors = 0;
+
+	*cur_utask_quiescing = NULL;
+	list_for_each_entry(utask, &uproc->thread_list, list) {
+		survivor = utask->tsk;
+		survivors++;
+		if (!utask->quiescing) {
+			/*
+			 * If utask is currently handling a probepoint, it'll
+			 * check utask->quiescing and quiesce when it's done.
+			 */
+			utask->quiescing = 1;
+			if (utask->tsk == current)
+				*cur_utask_quiescing = utask;
+			else if (utask->state == UPTASK_RUNNING) {
+				utask->quiesce_master = current;
+				utask_adjust_flags(utask, SET_ENGINE_FLAGS,
+					UTRACE_ACTION_QUIESCE
+					| UTRACE_EVENT(QUIESCE));
+				utask->quiesce_master = NULL;
+			}
+		}
+	}
+	/*
+	 * If any task was already quiesced (in utrace's opinion) when we
+	 * called utask_adjust_flags() on it, uprobe_report_quiesce() was
+	 * called, but wasn't in a position to call check_uproc_quiesced().
+	 */
+	check_uproc_quiesced(uproc, survivor);
+	return survivors;
+}
+
+/* Called with utask->uproc write-locked. */
+static void uprobe_free_task(struct uprobe_task *utask)
+{
+	struct deferred_registration *dr, *d;
+	struct uretprobe_instance *ri;
+	struct hlist_node *r1, *r2;
+
+	uprobe_unhash_utask(utask);
+	list_del(&utask->list);
+	list_for_each_entry_safe(dr, d, &utask->deferred_registrations, list) {
+		list_del(&dr->list);
+		kfree(dr);
+	}
+	hlist_for_each_entry_safe(ri, r1, r2, &utask->uretprobe_instances,
+			hlist) {
+		hlist_del(&ri->hlist);
+		kfree(ri);
+		uprobe_decref_process(utask->uproc);
+	}
+	kfree(utask);
+}
+
+/* Runs with uproc_mutex held and uproc->rwsem write-locked. */
+static void uprobe_free_process(struct uprobe_process *uproc)
+{
+	struct uprobe_task *utask, *tmp;
+	struct uprobe_ssol_area *area = &uproc->ssol_area;
+
+	if (!uproc->finished)
+		uprobe_release_ssol_vma(uproc);
+	if (area->slots)
+		kfree(area->slots);
+	if (!hlist_unhashed(&uproc->hlist))
+		hlist_del(&uproc->hlist);
+	list_for_each_entry_safe(utask, tmp, &uproc->thread_list, list) {
+		/*
+		 * utrace_detach() is OK here (required, it seems) even if
+		 * utask->tsk == current and we're in a utrace callback.
+		 */
+		if (utask->engine)
+			utrace_detach(utask->tsk, utask->engine);
+		uprobe_free_task(utask);
+	}
+	up_write(&uproc->rwsem);	// So kfree doesn't complain
+	kfree(uproc);
+}
+
+/*
+ * Decrement uproc's ref count.  If it's zero, free uproc and return 1.
+ * Else return 0.  If uproc is locked, don't call this; use
+ * uprobe_decref_process().
+ */
+static int uprobe_put_process(struct uprobe_process *uproc)
+{
+	int ret = 0;
+	if (atomic_dec_and_test(&uproc->refcount)) {
+		lock_uproc_table();
+		down_write(&uproc->rwsem);
+		if (unlikely(atomic_read(&uproc->refcount) != 0)) {
+			/*
+			 * The works because uproc_mutex is held any
+			 * time the ref count can go from 0 to 1 -- e.g.,
+			 * register_uprobe() sneaks in with a new probe.
+			 */
+			up_write(&uproc->rwsem);
+		} else {
+			uprobe_free_process(uproc);
+			ret = 1;
+		}
+		unlock_uproc_table();
+	}
+	return ret;
+}
+
+static struct uprobe_kimg *uprobe_mk_kimg(struct uprobe *u)
+{
+	struct uprobe_kimg *uk = (struct uprobe_kimg*)kzalloc(sizeof *uk,
+		GFP_USER);
+	if (unlikely(!uk))
+		return ERR_PTR(-ENOMEM);
+	u->kdata = uk;
+	uk->uprobe = u;
+	uk->ppt = NULL;
+	INIT_LIST_HEAD(&uk->list);
+	uk->status = -EBUSY;
+	return uk;
+}
+
+/*
+ * Allocate a uprobe_task object for t and add it to uproc's list.
+ * Called with t "got" and uproc->rwsem write-locked.  Called in one of
+ * the following cases:
+ * - before setting the first uprobe in t's process
+ * - we're in uprobe_report_clone() and t is the newly added thread
+ * Returns:
+ * - pointer to new uprobe_task on success
+ * - NULL if t dies before we can utrace_attach it
+ * - negative errno otherwise
+ */
+static struct uprobe_task *uprobe_add_task(struct task_struct *t,
+		struct uprobe_process *uproc)
+{
+	struct uprobe_task *utask;
+	struct utrace_attached_engine *engine;
+
+	utask = (struct uprobe_task *)kzalloc(sizeof *utask, GFP_USER);
+	if (unlikely(utask == NULL))
+		return ERR_PTR(-ENOMEM);
+
+	utask->tsk = t;
+	utask->state = UPTASK_RUNNING;
+	utask->quiescing = 0;
+	utask->uproc = uproc;
+	utask->active_probe = NULL;
+	utask->doomed = 0;
+	INIT_HLIST_HEAD(&utask->uretprobe_instances);
+	INIT_LIST_HEAD(&utask->deferred_registrations);
+	INIT_LIST_HEAD(&utask->list);
+	list_add_tail(&utask->list, &uproc->thread_list);
+	uprobe_hash_utask(utask);
+
+	engine = utrace_attach(t, UTRACE_ATTACH_CREATE, p_uprobe_utrace_ops,
+		utask);
+	if (IS_ERR(engine)) {
+		long err = PTR_ERR(engine);
+		printk("uprobes: utrace_attach failed, returned %ld\n", err);
+		uprobe_free_task(utask);
+		if (err == -ESRCH)
+			 return NULL;
+		return ERR_PTR(err);
+	}
+	utask->engine = engine;
+	/*
+	 * Always watch for traps, clones, execs and exits. Caller must
+	 * set any other engine flags.
+	 */
+	utask_adjust_flags(utask, SET_ENGINE_FLAGS,
+			UTRACE_EVENT(SIGNAL) | UTRACE_EVENT(SIGNAL_IGN) |
+			UTRACE_EVENT(SIGNAL_CORE) | UTRACE_EVENT(EXEC) |
+			UTRACE_EVENT(CLONE) | UTRACE_EVENT(EXIT));
+	/*
+	 * Note that it's OK if t dies just after utrace_attach, because
+	 * with the engine in place, the appropriate report_* callback
+	 * should handle it after we release uproc->rwsem.
+	 */
+	return utask;
+}
+
+/* See comment in uprobe_mk_process(). */
+static struct task_struct *find_next_thread_to_add(struct uprobe_process *uproc,		struct task_struct *start)
+{
+	struct task_struct *t;
+	struct uprobe_task *utask;
+
+	read_lock(&tasklist_lock);
+	t = start;
+	do {
+		if (unlikely(t->flags & PF_EXITING))
+			goto dont_add;
+		list_for_each_entry(utask, &uproc->thread_list, list) {
+			if (utask->tsk == t)
+				/* Already added */
+				goto dont_add;
+		}
+		/* Found thread/task to add. */
+		get_task_struct(t);
+		read_unlock(&tasklist_lock);
+		return t;
+dont_add:
+		t = next_thread(t);
+	} while (t != start);
+
+	read_unlock(&tasklist_lock);
+	return NULL;
+}
+
+/* Runs with uproc_mutex held; returns with uproc->rwsem write-locked. */
+static struct uprobe_process *uprobe_mk_process(struct task_struct *p)
+{
+	struct uprobe_process *uproc;
+	struct uprobe_task *utask;
+	struct task_struct *add_me;
+	int i;
+	long err;
+
+	uproc = (struct uprobe_process *)kzalloc(sizeof *uproc, GFP_USER);
+	if (unlikely(uproc == NULL))
+		return ERR_PTR(-ENOMEM);
+
+	/* Initialize fields */
+	atomic_set(&uproc->refcount, 1);
+	init_rwsem(&uproc->rwsem);
+	down_write(&uproc->rwsem);
+	init_waitqueue_head(&uproc->waitq);
+	for (i = 0; i < UPROBE_TABLE_SIZE; i++)
+		INIT_HLIST_HEAD(&uproc->uprobe_table[i]);
+	uproc->nppt = 0;
+	INIT_LIST_HEAD(&uproc->pending_uprobes);
+	INIT_LIST_HEAD(&uproc->thread_list);
+	uproc->nthreads = 0;
+	uproc->n_quiescent_threads = 0;
+	INIT_HLIST_NODE(&uproc->hlist);
+	uproc->tgid = p->tgid;
+	uproc->finished = 0;
+	uproc->uretprobe_trampoline_addr = NULL;
+
+	uproc->ssol_area.insn_area = NULL;
+	uproc->ssol_area.initialized = 0;
+	mutex_init(&uproc->ssol_area.setup_mutex);
+#ifdef CONFIG_UPROBES_SSOL
+	uproc->sstep_out_of_line = 1;
+#else
+	uproc->sstep_out_of_line = 0;
+#endif
+
+	/*
+	 * Create and populate one utask per thread in this process.  We
+	 * can't call uprobe_add_task() while holding tasklist_lock, so we:
+	 *	1. Lock task list.
+	 *	2. Find the next task, add_me, in this process that's not
+	 *	already on uproc's thread_list.  (Start search at previous
+	 *	one found.)
+	 *	3. Unlock task list.
+	 *	4. uprobe_add_task(add_me, uproc)
+	 *	Repeat 1-4 'til we have utasks for all tasks.
+	 */
+	add_me = p;
+	while ((add_me = find_next_thread_to_add(uproc, add_me)) != NULL) {
+		utask = uprobe_add_task(add_me, uproc);
+		put_task_struct(add_me);
+		if (IS_ERR(utask)) {
+			err = PTR_ERR(utask);
+			goto fail;
+		}
+		if (utask)
+			uproc->nthreads++;
+	}
+
+	if (uproc->nthreads == 0) {
+		/* All threads -- even p -- are dead. */
+		err = -ESRCH;
+		goto fail;
+	}
+	return uproc;
+
+fail:
+	uprobe_free_process(uproc);
+	return ERR_PTR(err);
+}
+
+/*
+ * Creates a uprobe_probept and connects it to uk and uproc.  Runs with
+ * uproc->rwsem write-locked.
+ */
+static struct uprobe_probept *uprobe_add_probept(struct uprobe_kimg *uk,
+	struct uprobe_process *uproc)
+{
+	struct uprobe_probept *ppt;
+
+	ppt = (struct uprobe_probept *)kzalloc(sizeof *ppt, GFP_USER);
+	if (unlikely(ppt == NULL))
+		return ERR_PTR(-ENOMEM);
+	init_waitqueue_head(&ppt->waitq);
+	mutex_init(&ppt->ssil_mutex);
+	mutex_init(&ppt->slot_mutex);
+	ppt->slot = NULL;
+
+	/* Connect to uk. */
+	INIT_LIST_HEAD(&ppt->uprobe_list);
+	list_add_tail(&uk->list, &ppt->uprobe_list);
+	uk->ppt = ppt;
+	uk->status = -EBUSY;
+	ppt->vaddr = uk->uprobe->vaddr;
+
+	/* Connect to uproc. */
+	ppt->state = UPROBE_INSERTING;
+	ppt->uproc = uproc;
+	INIT_LIST_HEAD(&ppt->pd_node);
+	list_add_tail(&ppt->pd_node, &uproc->pending_uprobes);
+	INIT_HLIST_NODE(&ppt->ut_node);
+	hlist_add_head(&ppt->ut_node,
+		&uproc->uprobe_table[hash_long(ppt->vaddr, UPROBE_HASH_BITS)]);
+	uproc->nppt++;
+	uprobe_get_process(uproc);
+	return ppt;
+}
+
+/* ppt is going away.  Free its slot (if it owns one) in the SSOL area. */
+static void uprobe_free_slot(struct uprobe_probept *ppt)
+{
+	struct uprobe_ssol_slot *slot = ppt->slot;
+	if (slot) {
+		down_write(&slot->rwsem);
+		if (slot->owner == ppt) {
+			unsigned long flags;
+			struct uprobe_ssol_area *area = &ppt->uproc->ssol_area;
+			spin_lock_irqsave(&area->lock, flags);
+			slot->state = SSOL_FREE;
+			slot->owner = NULL;
+			area->nfree++;
+			spin_unlock_irqrestore(&area->lock, flags);
+		}
+		up_write(&slot->rwsem);
+	}
+}
+
+/*
+ * Runs with ppt->uproc write-locked.  Frees ppt and decrements the ref count
+ * on ppt->uproc (but ref count shouldn't hit 0).
+ */
+static void uprobe_free_probept(struct uprobe_probept *ppt)
+{
+	struct uprobe_process *uproc = ppt->uproc;
+	uprobe_free_slot(ppt);
+	hlist_del(&ppt->ut_node);
+	uproc->nppt--;
+	kfree(ppt);
+	uprobe_decref_process(uproc);
+}
+
+static void uprobe_free_kimg(struct uprobe_kimg *uk)
+{
+	uk->uprobe->kdata = NULL;
+	kfree(uk);
+}
+
+/*
+ * Runs with uprobe_process write-locked.
+ * Note that we never free u, because the user owns that.
+ */
+static void purge_uprobe(struct uprobe_kimg *uk)
+{
+	struct uprobe_probept *ppt = uk->ppt;
+	list_del(&uk->list);
+	uprobe_free_kimg(uk);
+	if (list_empty(&ppt->uprobe_list))
+		uprobe_free_probept(ppt);
+}
+
+/* Probed address must be in an executable VM area, outside the SSOL area. */
+static int uprobe_validate_vaddr(struct task_struct *p, unsigned long vaddr,
+	struct uprobe_process *uproc)
+{
+	struct vm_area_struct *vma;
+	struct mm_struct *mm = p->mm;
+	if (!mm)
+		return -EINVAL;
+	down_read(&mm->mmap_sem);
+	vma = find_vma(mm, vaddr);
+	if (!vma || vaddr < vma->vm_start || !(vma->vm_flags & VM_EXEC)
+	    || vma->vm_start == (unsigned long) uproc->ssol_area.insn_area) {
+		up_read(&mm->mmap_sem);
+		return -EINVAL;
+	}
+	up_read(&mm->mmap_sem);
+	return 0;
+}
+
+static struct task_struct *uprobe_get_task(pid_t pid)
+{
+	struct task_struct *p;
+	rcu_read_lock();
+	p = find_task_by_pid(pid);
+	if (p)
+		get_task_struct(p);
+	rcu_read_unlock();
+	return p;
+}
+
+/* Runs with utask->uproc read-locked.  Returns -EINPROGRESS on success. */
+static int defer_registration(struct uprobe *u, int regflag,
+		struct uprobe_task *utask)
+{
+	struct deferred_registration *dr =
+		kmalloc(sizeof(struct deferred_registration), GFP_USER);
+	if (!dr)
+		return -ENOMEM;
+	dr->type = (is_uretprobe(u) ? UPTY_URETPROBE : UPTY_UPROBE);
+	dr->uprobe = u;
+	dr->regflag = regflag;
+	INIT_LIST_HEAD(&dr->list);
+	list_add_tail(&dr->list, &utask->deferred_registrations);
+	return -EINPROGRESS;
+}
+
+/* See Documentation/uprobes.txt. */
+static
+int register_uprobe(struct uprobe *u)
+{
+	struct task_struct *p;
+	struct uprobe_process *uproc;
+	struct uprobe_kimg *uk;
+	struct uprobe_probept *ppt;
+	struct uprobe_task *cur_utask, *cur_utask_quiescing = NULL;
+	int survivors, ret = 0, uproc_is_new = 0;
+	if ((ret = verify_uprobes()) < 0)
+		return ret;
+
+	if (!u || !u->handler)
+		return -EINVAL;
+
+	p = uprobe_get_task(u->pid);
+	if (!p)
+		return -ESRCH;
+
+	cur_utask = uprobe_find_utask(current);
+	if (cur_utask && cur_utask->active_probe) {
+		/*
+		 * Called from handler; cur_utask->uproc is read-locked.
+		 * Do this registration later.
+		 */
+		put_task_struct(p);
+		return defer_registration(u, 1, cur_utask);
+	}
+
+	/* Get the uprobe_process for this pid, or make a new one. */
+	lock_uproc_table();
+	uproc = uprobe_find_process(p->tgid);
+
+	if (uproc)
+		unlock_uproc_table();
+	else {
+		uproc = uprobe_mk_process(p);
+		if (IS_ERR(uproc)) {
+			ret = (int) PTR_ERR(uproc);
+			unlock_uproc_table();
+			goto fail_tsk;
+		}
+		/* Hold uproc_mutex until we've added uproc to uproc_table. */
+		uproc_is_new = 1;
+	}
+
+	if (is_uretprobe(u) && IS_ERR(uproc->uretprobe_trampoline_addr)) {
+		/* Previously failed to set up trampoline. */
+		ret = -ENOMEM;
+		goto fail_uproc;
+	}
+
+	if ((ret = uprobe_validate_vaddr(p, u->vaddr, uproc)) < 0)
+		goto fail_uproc;
+
+	if (u->kdata) {
+		/*
+		 * Probe is already/still registered.  This is the only
+		 * place we return -EBUSY to the user.
+		 */
+		ret = -EBUSY;
+		goto fail_uproc;
+	}
+
+	uk = uprobe_mk_kimg(u);
+	if (IS_ERR(uk)) {
+		ret = (int) PTR_ERR(uk);
+		goto fail_uproc;
+	}
+
+	/* See if we already have a probepoint at the vaddr. */
+	ppt = (uproc_is_new ? NULL : uprobe_find_probept(uproc, u->vaddr));
+	if (ppt) {
+		/* Breakpoint is already in place, or soon will be. */
+		uk->ppt = ppt;
+		list_add_tail(&uk->list, &ppt->uprobe_list);
+		switch (ppt->state) {
+		case UPROBE_INSERTING:
+			uk->status = -EBUSY;	// in progress
+			if (uproc->tgid == current->tgid) {
+				cur_utask_quiescing = cur_utask;
+				BUG_ON(!cur_utask_quiescing);
+			}
+			break;
+		case UPROBE_REMOVING:
+			/* Wait!  Don't remove that bkpt after all! */
+			ppt->state = UPROBE_BP_SET;
+			list_del(&ppt->pd_node);  // Remove from pending list.
+			wake_up_all(&ppt->waitq); // Wake unregister_uprobe().
+			/*FALLTHROUGH*/
+		case UPROBE_BP_SET:
+			uk->status = 0;
+			break;
+		default:
+			BUG();
+		}
+		up_write(&uproc->rwsem);
+		put_task_struct(p);
+		if (uk->status == 0) {
+			uprobe_put_process(uproc);
+			return 0;
+		}
+		goto await_bkpt_insertion;
+	} else {
+		ppt = uprobe_add_probept(uk, uproc);
+		if (IS_ERR(ppt)) {
+			ret = (int) PTR_ERR(ppt);
+			goto fail_uk;
+		}
+	}
+
+	if (uproc_is_new) {
+		hlist_add_head(&uproc->hlist,
+			&uproc_table[hash_long(uproc->tgid, UPROBE_HASH_BITS)]);
+		unlock_uproc_table();
+	}
+	put_task_struct(p);
+	survivors = quiesce_all_threads(uproc, &cur_utask_quiescing);
+
+	if (survivors == 0) {
+		purge_uprobe(uk);
+		up_write(&uproc->rwsem);
+		uprobe_put_process(uproc);
+		return -ESRCH;
+	}
+	up_write(&uproc->rwsem);
+
+await_bkpt_insertion:
+	if (cur_utask_quiescing)
+		/* Current task is probing its own process. */
+		(void) utask_fake_quiesce(cur_utask_quiescing);
+	else
+		wait_event(ppt->waitq, ppt->state != UPROBE_INSERTING);
+	ret = uk->status;
+	if (ret != 0) {
+		down_write(&uproc->rwsem);
+		purge_uprobe(uk);
+		up_write(&uproc->rwsem);
+	}
+	uprobe_put_process(uproc);
+	return ret;
+
+fail_uk:
+	uprobe_free_kimg(uk);
+
+fail_uproc:
+	if (uproc_is_new) {
+		uprobe_free_process(uproc);
+		unlock_uproc_table();
+	} else {
+		up_write(&uproc->rwsem);
+		uprobe_put_process(uproc);
+	}
+
+fail_tsk:
+	put_task_struct(p);
+	return ret;
+}
+
+/* See Documentation/uprobes.txt. */
+static
+void unregister_uprobe(struct uprobe *u)
+{
+	struct task_struct *p;
+	struct uprobe_process *uproc;
+	struct uprobe_kimg *uk;
+	struct uprobe_probept *ppt;
+	struct uprobe_task *cur_utask, *cur_utask_quiescing = NULL;
+
+	if (verify_uprobes() < 0)
+		return;
+	if (!u)
+		return;
+	p = uprobe_get_task(u->pid);
+	if (!p)
+		return;
+
+	cur_utask = uprobe_find_utask(current);
+	if (cur_utask && cur_utask->active_probe) {
+		/* Called from handler; uproc is read-locked; do this later */
+		put_task_struct(p);
+		(void) defer_registration(u, 0, cur_utask);
+		return;
+	}
+
+	/*
+	 * Lock uproc before walking the graph, in case the process we're
+	 * probing is exiting.
+	 */
+	lock_uproc_table();
+	uproc = uprobe_find_process(p->tgid);
+	unlock_uproc_table();
+	put_task_struct(p);
+	if (!uproc)
+		return;
+
+	uk = (struct uprobe_kimg *)u->kdata;
+	if (!uk)
+		/*
+		 * This probe was never successfully registered, or
+		 * has already been unregistered.
+		 */
+		goto done;
+	if (uk->status == -EBUSY)
+		/* Looks like register or unregister is already in progress. */
+		goto done;
+	ppt = uk->ppt;
+
+	list_del(&uk->list);
+	uprobe_free_kimg(uk);
+
+	if (is_uretprobe(u))
+		zap_uretprobe_instances(u, uproc);
+
+	if (!list_empty(&ppt->uprobe_list))
+		goto done;
+
+	/*
+	 * The last uprobe at ppt's probepoint is being unregistered.
+	 * Queue the breakpoint for removal.
+	 */
+	ppt->state = UPROBE_REMOVING;
+	list_add_tail(&ppt->pd_node, &uproc->pending_uprobes);
+
+	(void) quiesce_all_threads(uproc, &cur_utask_quiescing);
+	up_write(&uproc->rwsem);
+	if (cur_utask_quiescing)
+		/* Current task is probing its own process. */
+		(void) utask_fake_quiesce(cur_utask_quiescing);
+	else
+		wait_event(ppt->waitq, ppt->state != UPROBE_REMOVING);
+
+	if (likely(ppt->state == UPROBE_DISABLED)) {
+		down_write(&uproc->rwsem);
+		uprobe_free_probept(ppt);
+		/* else somebody else's register_uprobe() resurrected ppt. */
+		up_write(&uproc->rwsem);
+	}
+	uprobe_put_process(uproc);
+	return;
+
+done:
+	up_write(&uproc->rwsem);
+	uprobe_put_process(uproc);
+}
+
+/* Find a surviving thread in uproc.  Runs with uproc->rwsem locked. */
+static struct task_struct *find_surviving_thread(struct uprobe_process *uproc)
+{
+	struct uprobe_task *utask;
+
+	list_for_each_entry(utask, &uproc->thread_list, list)
+		return utask->tsk;
+	return NULL;
+}
+
+/*
+ * Run all the deferred_registrations previously queued by the current utask.
+ * Runs with no locks or mutexes held.  The current utask could disappear
+ * as the result of unregister_u*probe() called here.
+ */
+static void uprobe_run_def_regs(struct list_head *drlist)
+{
+	struct deferred_registration *dr, *d;
+
+	list_for_each_entry_safe(dr, d, drlist, list) {
+		int result = 0;
+		struct uprobe *u = dr->uprobe;
+
+		if (dr->type == UPTY_URETPROBE) {
+			struct uretprobe *rp =
+				container_of(u, struct uretprobe, u);
+			if (dr->regflag)
+				result = register_uretprobe(rp);
+			else
+				unregister_uretprobe(rp);
+		} else {
+			if (dr->regflag)
+				result = register_uprobe(u);
+			else
+				unregister_uprobe(u);
+		}
+		if (u && u->registration_callback)
+			u->registration_callback(u, dr->regflag, dr->type,
+					result);
+		list_del(&dr->list);
+		kfree(dr);
+	}
+}
+
+/*
+ * Functions for allocation of the SSOL area, and the instruction slots
+ * therein
+ */
+
+/*
+ * We leave the SSOL vma in place even after all the probes are gone.
+ * We used to remember its address in current->mm->context.uprobes_ssol_area,
+ * but adding that field to mm_context broke KAPI compatibility.
+ * Instead, when we shut down the uproc for lack of probes, we "tag" the vma
+ * for later identification.  This is not particularly robust, but it's
+ * no more vulnerable to ptrace or mprotect mischief than any other part
+ * of the address space.
+ */
+#define UPROBES_SSOL_VMA_TAG \
+	"This is the SSOL area for uprobes.  Mess with it at your own risk."
+#define UPROBES_SSOL_TAGSZ ((int)sizeof(UPROBES_SSOL_VMA_TAG))
+
+/*
+ * Searching downward from ceiling address (0 signifies top of memory),
+ * find the next vma whose flags indicate it could be an SSOL area.
+ * Return its address, or 0 for no match.
+ */
+static unsigned long find_next_possible_ssol_vma(unsigned long ceiling)
+{
+	struct mm_struct *mm = current->mm;
+	struct rb_node *rb_node;
+	struct vm_area_struct *vma;
+	unsigned long good_flags = VM_EXEC | VM_DONTCOPY | VM_DONTEXPAND;
+	unsigned long bad_flags = VM_WRITE | VM_GROWSDOWN | VM_GROWSUP;
+	unsigned long addr = 0;
+
+	down_read(&mm->mmap_sem);
+	for (rb_node=rb_last(&mm->mm_rb); rb_node; rb_node=rb_prev(rb_node)) {
+		vma = rb_entry(rb_node, struct vm_area_struct, vm_rb);
+		if (ceiling && vma->vm_start >= ceiling)
+			continue;
+		if ((vma->vm_flags & good_flags) != good_flags)
+			continue;
+		if ((vma->vm_flags & bad_flags) != 0)
+			continue;
+		addr = vma->vm_start;
+		break;
+	}
+	up_read(&mm->mmap_sem);
+	return addr;
+}
+
+static noinline unsigned long find_old_ssol_vma(void)
+{
+	unsigned long addr;
+	unsigned long ceiling = 0;	// top of memory
+	char buf[UPROBES_SSOL_TAGSZ];
+	while ((addr = find_next_possible_ssol_vma(ceiling)) != 0) {
+		ceiling = addr;
+		if (copy_from_user(buf, (const void __user*)addr,
+						UPROBES_SSOL_TAGSZ))
+			continue;
+		if (!strcmp(buf, UPROBES_SSOL_VMA_TAG))
+			return addr;
+	}
+	return 0;
+}
+
+/*
+ * Mmap nbytes bytes for the uprobes SSOL area for the current process.
+ * Returns the address of the page, or a negative errno.
+ * This approach was suggested by Roland McGrath.
+ */
+static noinline unsigned long uprobe_setup_ssol_vma(unsigned long nbytes)
+{
+	unsigned long addr;
+	struct mm_struct *mm = current->mm;
+	struct vm_area_struct *vma;
+
+	BUG_ON(nbytes & ~PAGE_MASK);
+	if ((addr = find_old_ssol_vma()) != 0)
+		return addr;
+
+	down_write(&mm->mmap_sem);
+	/*
+	 * Find the end of the top mapping and skip a page.
+	 * If there is no space for PAGE_SIZE above
+	 * that, mmap will ignore our address hint.
+	 */
+	vma = rb_entry(rb_last(&mm->mm_rb), struct vm_area_struct, vm_rb);
+	addr = vma->vm_end + PAGE_SIZE;
+	addr = do_mmap_pgoff(NULL, addr, nbytes, PROT_EXEC,
+					MAP_PRIVATE|MAP_ANONYMOUS, 0);
+	if (addr & ~PAGE_MASK) {
+		up_write(&mm->mmap_sem);
+		printk(KERN_ERR "Uprobes failed to allocate a vma for"
+			" pid/tgid %d/%d for single-stepping out of line.\n",
+			current->pid, current->tgid);
+		return addr;
+	}
+
+	vma = find_vma(mm, addr);
+	BUG_ON(!vma);
+	/* avoid vma copy on fork() and don't expand when mremap() */
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+
+	up_write(&mm->mmap_sem);
+	return addr;
+}
+
+/*
+ * Initialize per-process area for single stepping out-of-line.
+ * Must be run by a thread in the probed process.  Returns with
+ * area->insn_area pointing to the initialized area, or set to a
+ * negative errno.
+ */
+static noinline void uprobe_init_ssol(struct uprobe_process *uproc)
+{
+	struct uprobe_ssol_area *area = &uproc->ssol_area;
+	struct uprobe_ssol_slot *slot;
+	int i;
+	char *slot_addr;	// Simplify pointer arithmetic
+
+	/* Trampoline setup will either fail or succeed here. */
+	uproc->uretprobe_trampoline_addr = ERR_PTR(-ENOMEM);
+
+	area->insn_area = (uprobe_opcode_t *) uprobe_setup_ssol_vma(PAGE_SIZE);
+	if (IS_ERR(area->insn_area))
+		return;
+
+	area->nfree = area->nslots = PAGE_SIZE / MAX_UINSN_BYTES;
+	area->slots = (struct uprobe_ssol_slot *)
+		kzalloc(sizeof(struct uprobe_ssol_slot) * area->nslots,
+								GFP_USER);
+	if (!area->slots) {
+		area->insn_area = ERR_PTR(-ENOMEM);
+		return;
+	}
+	spin_lock_init(&area->lock);
+	area->next_slot = 0;
+	slot_addr = (char*) area->insn_area;
+	for (i = 0; i < area->nslots; i++) {
+		slot = &area->slots[i];
+		init_rwsem(&slot->rwsem);
+		slot->state = SSOL_FREE;
+		slot->owner = NULL;
+		slot->last_used = 0;
+		slot->insn = (__user uprobe_opcode_t *) slot_addr;
+		slot_addr += MAX_UINSN_BYTES;
+	}
+	uretprobe_set_trampoline(uproc);
+}
+
+/*
+ * Verify that the SSOL area has been set up for uproc.  Returns a
+ * pointer to the SSOL area, or a negative erro if we couldn't set it up.
+ */
+static __user uprobe_opcode_t
+			*uprobe_verify_ssol(struct uprobe_process *uproc)
+{
+	struct uprobe_ssol_area *area = &uproc->ssol_area;
+
+	if (unlikely(!area->initialized)) {
+		/* First time through for this probed process */
+		mutex_lock(&uproc->ssol_area.setup_mutex);
+		if (likely(!area->initialized)) {
+			/* Nobody snuck in and set things up ahead of us. */
+			uprobe_init_ssol(uproc);
+			area->initialized = 1;
+		}
+		mutex_unlock(&uproc->ssol_area.setup_mutex);
+	}
+	return area->insn_area;
+}
+
+/*
+ * uproc is going away, but the process lives on.  Tag the SSOL vma so a new
+ * uproc can reuse it if more probes are requested.
+ */
+static void uprobe_release_ssol_vma(struct uprobe_process *uproc)
+{
+	unsigned long addr;
+	struct task_struct *tsk;
+	static const char *buf = UPROBES_SSOL_VMA_TAG;
+	int nb;
+
+	/* No need to muck with dying image's mm_struct. */
+	BUG_ON(uproc->finished);
+	addr = (unsigned long) uproc->ssol_area.insn_area;
+	if (!addr || IS_ERR_VALUE(addr))
+		return;
+	tsk = find_surviving_thread(uproc);
+	if (!tsk)
+		return;
+	nb = access_process_vm(tsk, addr, (void*)buf, UPROBES_SSOL_TAGSZ, 1);
+	if (nb != UPROBES_SSOL_TAGSZ)
+		printk(KERN_ERR "Failed to tag uprobes SSOL vma: "
+			"pid/tgid=%d/%d, vaddr=%#lx\n", tsk->pid, tsk->tgid,
+			addr);
+}
+
+static inline int advance_slot(int slot, struct uprobe_ssol_area *area)
+{
+	/* Slot 0 is reserved for uretprobe trampoline. */
+	slot++;
+	if (unlikely(slot >= area->nslots))
+		slot = 1;
+	return slot;
+}
+
+/*
+ * Return the slot number of the least-recently-used slot in the
+ * neighborhood of area->next_slot.  Limit the number of slots we test
+ * to keep it fast.  Nobody dies if this isn't the best choice.
+ */
+static int uprobe_lru_insn_slot(struct uprobe_ssol_area *area)
+{
+#define MAX_LRU_TESTS 10
+	struct uprobe_ssol_slot *s;
+	int lru_slot = -1;
+	unsigned long lru_time = ULONG_MAX;
+	int nr_lru_tests = 0;
+	int slot = area->next_slot;
+	do {
+		s = &area->slots[slot];
+		if (likely(s->state == SSOL_ASSIGNED)) {
+			if( lru_time > s->last_used) {
+				lru_time = s->last_used;
+				lru_slot = slot;
+			}
+			if (++nr_lru_tests >= MAX_LRU_TESTS)
+				break;
+		}
+		slot = advance_slot(slot, area);
+	} while (slot != area->next_slot);
+
+	if (unlikely(lru_slot < 0))
+		/* All slots are in the act of being stolen.  Join the melee. */
+		return area->next_slot;
+	else
+		return lru_slot;
+}
+
+/*
+ * Choose an instruction slot and take it.  Choose a free slot if there is one.
+ * Otherwise choose the least-recently-used slot.  Returns with slot
+ * read-locked and containing the desired instruction.  Runs with
+ * ppt->slot_mutex locked.
+ */
+static struct uprobe_ssol_slot
+		*uprobe_take_insn_slot(struct uprobe_probept *ppt)
+{
+	struct uprobe_process *uproc = ppt->uproc;
+	struct uprobe_ssol_area *area = &uproc->ssol_area;
+	struct uprobe_ssol_slot *s;
+	int len, slot;
+	unsigned long flags;
+
+	spin_lock_irqsave(&area->lock, flags);
+
+	if (area->nfree) {
+		for (slot = 0; slot < area->nslots; slot++) {
+			if (area->slots[slot].state == SSOL_FREE) {
+				area->nfree--;
+				goto found_slot;
+			}
+		}
+		/* Shouldn't get here.  Fix nfree and get on with life. */
+		area->nfree = 0;
+	}
+	slot = uprobe_lru_insn_slot(area);
+
+found_slot:
+	area->next_slot = advance_slot(slot, area);
+	s = &area->slots[slot];
+	s->state = SSOL_BEING_STOLEN;
+
+	spin_unlock_irqrestore(&area->lock, flags);
+
+	/* Wait for current users of slot to finish. */
+	down_write(&s->rwsem);
+	ppt->slot = s;
+	s->owner = ppt;
+	s->last_used = jiffies;
+	s->state = SSOL_ASSIGNED;
+	/* Copy the original instruction to the chosen slot. */
+	len = access_process_vm(current, (unsigned long)s->insn,
+					 ppt->insn, MAX_UINSN_BYTES, 1);
+        if (unlikely(len < MAX_UINSN_BYTES)) {
+		up_write(&s->rwsem);
+		printk(KERN_ERR "Failed to copy instruction at %#lx"
+			" to SSOL area (%#lx)\n", ppt->vaddr,
+			(unsigned long) area->slots);
+		return NULL;
+	}
+	/* Let other threads single-step in this slot. */
+	downgrade_write(&s->rwsem);
+	return s;
+}
+
+/* ppt doesn't own a slot.  Get one for ppt, and return it read-locked. */
+static struct uprobe_ssol_slot
+		*uprobe_find_insn_slot(struct uprobe_probept *ppt)
+{
+	struct uprobe_ssol_slot *slot;
+
+	mutex_lock(&ppt->slot_mutex);
+	slot = ppt->slot;
+	if (unlikely(slot && slot->owner == ppt)) {
+		/* Looks like another thread snuck in and got a slot for us. */
+		down_read(&slot->rwsem);
+		if (likely(slot->owner == ppt)) {
+			slot->last_used = jiffies;
+			mutex_unlock(&ppt->slot_mutex);
+			return slot;
+		}
+		/* ... but then somebody stole it. */
+		up_read(&slot->rwsem);
+	}
+	slot = uprobe_take_insn_slot(ppt);
+	mutex_unlock(&ppt->slot_mutex);
+	return slot;
+}
+
+/*
+ * Ensure that ppt owns an instruction slot for single-stepping.
+ * Returns with the slot read-locked and ppt->slot pointing at it.
+ */
+static
+struct uprobe_ssol_slot *uprobe_get_insn_slot(struct uprobe_probept *ppt)
+{
+	struct uprobe_ssol_slot *slot = ppt->slot;
+
+	if (unlikely(!slot))
+		return uprobe_find_insn_slot(ppt);
+
+	down_read(&slot->rwsem);
+	if (unlikely(slot->owner != ppt)) {
+		up_read(&slot->rwsem);
+		return uprobe_find_insn_slot(ppt);
+	}
+	slot->last_used = jiffies;
+	return slot;
+}
+
+/*
+ * utrace engine report callbacks
+ */
+
+/*
+ * We've been asked to quiesce, but aren't in a position to do so.
+ * This could happen in either of the following cases:
+ *
+ * 1) Our own thread is doing a register or unregister operation --
+ * e.g., as called from a u[ret]probe handler or a non-uprobes utrace
+ * callback.  We can't wait_event() for ourselves in [un]register_uprobe().
+ *
+ * 2) We've been asked to quiesce, but we hit a probepoint first.  Now
+ * we're in the report_signal callback, having handled the probepoint.
+ * We'd like to just set the UTRACE_ACTION_QUIESCE and
+ * UTRACE_EVENT(QUIESCE) flags and coast into quiescence.  Unfortunately,
+ * it's possible to hit a probepoint again before we quiesce.  When
+ * processing the SIGTRAP, utrace would call uprobe_report_quiesce(),
+ * which must decline to take any action so as to avoid removing the
+ * uprobe just hit.  As a result, we could keep hitting breakpoints
+ * and never quiescing.
+ *
+ * So here we do essentially what we'd prefer to do in uprobe_report_quiesce().
+ * If we're the last thread to quiesce, handle_pending_uprobes() and
+ * rouse_all_threads().  Otherwise, pretend we're quiescent and sleep until
+ * the last quiescent thread handles that stuff and then wakes us.
+ *
+ * Called and returns with no mutexes held.  Returns 1 if we free utask->uproc,
+ * else 0.
+ */
+static int utask_fake_quiesce(struct uprobe_task *utask)
+{
+	struct uprobe_process *uproc = utask->uproc;
+	enum uprobe_task_state prev_state = utask->state;
+
+	down_write(&uproc->rwsem);
+
+	/* In case we're somehow set to quiesce for real... */
+	clear_utrace_quiesce(utask);
+
+	if (uproc->n_quiescent_threads == uproc->nthreads-1) {
+		/* We're the last thread to "quiesce." */
+		handle_pending_uprobes(uproc, utask->tsk);
+		rouse_all_threads(uproc);
+		up_write(&uproc->rwsem);
+		return 0;
+	} else {
+		utask->state = UPTASK_SLEEPING;
+		uproc->n_quiescent_threads++;
+		up_write(&uproc->rwsem);
+		/* We ref-count sleepers. */
+		uprobe_get_process(uproc);
+
+		wait_event(uproc->waitq, !utask->quiescing);
+
+		down_write(&uproc->rwsem);
+		utask->state = prev_state;
+		uproc->n_quiescent_threads--;
+		up_write(&uproc->rwsem);
+
+		/*
+		 * If uproc's last uprobe has been unregistered, and
+		 * unregister_uprobe() woke up before we did, it's up
+		 * to us to free uproc.
+		 */
+		return uprobe_put_process(uproc);
+	}
+}
+
+/* Prepare to single-step ppt's probed instruction inline. */
+static inline void uprobe_pre_ssin(struct uprobe_task *utask,
+	struct uprobe_probept *ppt, struct pt_regs *regs)
+{
+	int len;
+	arch_reset_ip_for_sstep(regs);
+	mutex_lock(&ppt->ssil_mutex);
+	len = set_orig_insn(ppt, utask->tsk);
+	if (unlikely(len != BP_INSN_SIZE)) {
+		printk("Failed to temporarily restore original "
+			"instruction for single-stepping: "
+			"pid/tgid=%d/%d, vaddr=%#lx\n",
+			utask->tsk->pid, utask->tsk->tgid, ppt->vaddr);
+		utask->doomed = 1;
+	}
+}
+
+/* Prepare to continue execution after single-stepping inline. */
+static inline void uprobe_post_ssin(struct uprobe_task *utask,
+	struct uprobe_probept *ppt)
+{
+
+	int len = set_bp(ppt, utask->tsk);
+	if (unlikely(len != BP_INSN_SIZE)) {
+		printk("Couldn't restore bp: pid/tgid=%d/%d, addr=%#lx\n",
+			utask->tsk->pid, utask->tsk->tgid, ppt->vaddr);
+		ppt->state = UPROBE_DISABLED;
+	}
+	mutex_unlock(&ppt->ssil_mutex);
+}
+
+/* uprobe_pre_ssout() and uprobe_post_ssout() are architecture-specific. */
+
+/*
+ * Signal callback:
+ *
+ * We get called here with:
+ *	state = UPTASK_RUNNING => we are here due to a breakpoint hit
+ *		- Read-lock the process
+ *		- Figure out which probepoint, based on regs->IP
+ *		- Set state = UPTASK_BP_HIT
+ *		- Reset regs->IP to beginning of the insn, if necessary
+ *		- Invoke handler for each uprobe at this probepoint
+ *		- Set singlestep in motion (UTRACE_ACTION_SINGLESTEP),
+ *			with state = UPTASK_SSTEP
+ *
+ *	state = UPTASK_SSTEP => here after single-stepping
+ *		- Validate we are here per the state machine
+ *		- Clean up after single-stepping
+ *		- Set state = UPTASK_RUNNING
+ *		- Read-unlock the process
+ *		- If it's time to quiesce, take appropriate action.
+ *		- If the handler(s) we ran called [un]register_uprobe(),
+ *			complete those via uprobe_run_def_regs().
+ *
+ *	state = ANY OTHER STATE
+ *		- Not our signal, pass it on (UTRACE_ACTION_RESUME)
+ * Note: Intermediate states such as UPTASK_POST_SSTEP help
+ * uprobe_report_exit() decide what to unlock if we die.
+ */
+static u32 uprobe_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 uprobe_task *utask;
+	struct uprobe_probept *ppt;
+	struct uprobe_process *uproc;
+	struct uprobe_kimg *uk;
+	u32 ret;
+	unsigned long probept;
+	int hit_uretprobe_trampoline = 0;
+	LIST_HEAD(def_reg_list);
+
+	utask = rcu_dereference((struct uprobe_task *)engine->data);
+	BUG_ON(!utask);
+
+	if (info->si_signo != BREAKPOINT_SIGNAL &&
+					info->si_signo != SSTEP_SIGNAL)
+		goto no_interest;
+
+	/*
+	 * Set up the SSOL area if it's not already there.  We do this
+	 * here because we have to do it before handling the first
+	 * probepoint hit, the probed process has to do it, and this may
+	 * be the first time our probed process runs uprobes code.
+	 *
+	 * We need the SSOL area for the uretprobe trampoline even if
+	 * this architectures doesn't single-step out of line.
+	 */
+	uproc = utask->uproc;
+#ifdef CONFIG_UPROBES_SSOL
+	if (uproc->sstep_out_of_line &&
+			unlikely(IS_ERR(uprobe_verify_ssol(uproc))))
+		uproc->sstep_out_of_line = 0;
+#elif defined(CONFIG_URETPROBES)
+	(void) uprobe_verify_ssol(uproc);
+#endif
+
+	switch (utask->state) {
+	case UPTASK_RUNNING:
+		if (info->si_signo != BREAKPOINT_SIGNAL)
+			goto no_interest;
+		down_read(&uproc->rwsem);
+		clear_utrace_quiesce(utask);
+		probept = arch_get_probept(regs);
+
+		hit_uretprobe_trampoline = (probept == (unsigned long)
+			uproc->uretprobe_trampoline_addr);
+		if (hit_uretprobe_trampoline) {
+			uretprobe_handle_return(regs, utask);
+			goto bkpt_done;
+		}
+
+		ppt = uprobe_find_probept(uproc, probept);
+		if (!ppt) {
+			up_read(&uproc->rwsem);
+			goto no_interest;
+		}
+		utask->active_probe = ppt;
+		utask->state = UPTASK_BP_HIT;
+
+		if (likely(ppt->state == UPROBE_BP_SET)) {
+			list_for_each_entry(uk, &ppt->uprobe_list, list) {
+				struct uprobe *u = uk->uprobe;
+				if (is_uretprobe(u))
+					uretprobe_handle_entry(u, regs, utask);
+				else if (u->handler)
+					u->handler(u, regs);
+			}
+		}
+
+		utask->state = UPTASK_PRE_SSTEP;
+#ifdef CONFIG_UPROBES_SSOL
+		if (uproc->sstep_out_of_line)
+			uprobe_pre_ssout(utask, ppt, regs);
+		else
+#endif
+			uprobe_pre_ssin(utask, ppt, regs);
+		if (unlikely(utask->doomed))
+			do_exit(SIGSEGV);
+		utask->state = UPTASK_SSTEP;
+		/*
+		 * No other engines must see this signal, and the
+		 * signal shouldn't be passed on either.
+		 */
+		ret = UTRACE_ACTION_HIDE | UTRACE_SIGNAL_IGN |
+			UTRACE_ACTION_SINGLESTEP | UTRACE_ACTION_NEWSTATE;
+		break;
+	case UPTASK_SSTEP:
+		if (info->si_signo != SSTEP_SIGNAL)
+			goto no_interest;
+		ppt = utask->active_probe;
+		BUG_ON(!ppt);
+		utask->state = UPTASK_POST_SSTEP;
+#ifdef CONFIG_UPROBES_SSOL
+		if (uproc->sstep_out_of_line)
+			uprobe_post_ssout(utask, ppt, regs);
+		else
+#endif
+			uprobe_post_ssin(utask, ppt);
+bkpt_done:
+		/* Note: Can come here after running uretprobe handlers */
+		if (unlikely(utask->doomed))
+			do_exit(SIGSEGV);
+
+		utask->active_probe = NULL;
+
+		/*
+		 * The deferred_registrations list accumulates in utask,
+		 * but utask could go away when we uprobe_run_def_regs.
+		 * So switch the list head to a local variable.
+		 */
+		list_splice_init(&utask->deferred_registrations, &def_reg_list);
+
+		ret = UTRACE_ACTION_HIDE | UTRACE_SIGNAL_IGN
+			| UTRACE_ACTION_NEWSTATE;
+		utask->state = UPTASK_RUNNING;
+		if (utask->quiescing) {
+			up_read(&uproc->rwsem);
+			if (utask_fake_quiesce(utask) == 1)
+				ret |= UTRACE_ACTION_DETACH;
+		} else
+			up_read(&uproc->rwsem);
+
+		if (hit_uretprobe_trampoline)
+			/*
+			 * It's possible that the uretprobe_instance
+			 * we just recycled was the last reason for
+			 * keeping uproc around.
+			 */
+			uprobe_put_process(uproc);
+
+		uprobe_run_def_regs(&def_reg_list);
+		break;
+	default:
+		goto no_interest;
+	}
+	return ret;
+
+no_interest:
+	return UTRACE_ACTION_RESUME;
+}
+
+/*
+ * utask_quiesce_pending_sigtrap: The utask entered the quiesce callback
+ * through the signal delivery path, apparently. Check if the associated
+ * signal happened due to a uprobe hit.
+ *
+ * Called with utask->uproc write-locked.  Returns 1 if quiesce was
+ * entered with SIGTRAP pending due to a uprobe hit.
+ */
+static int utask_quiesce_pending_sigtrap(struct uprobe_task *utask)
+{
+	const struct utrace_regset_view *view;
+	const struct utrace_regset *regset;
+	struct uprobe_probept *ppt;
+	unsigned long insn_ptr;
+
+	view = utrace_native_view(utask->tsk);
+	regset = utrace_regset(utask->tsk, utask->engine, view, 0);
+	if (unlikely(regset == NULL))
+		return -EIO;
+
+	if ((*regset->get)(utask->tsk, regset, SLOT_IP * regset->size,
+			regset->size, &insn_ptr, NULL) != 0)
+		return -EIO;
+
+	if (regset->size != sizeof(insn_ptr)) {
+		/* Assume 32-bit app and 64-bit kernel. */
+		u32 *insn_ptr32 = (u32*) &insn_ptr;
+		BUG_ON(regset->size != sizeof(u32));
+		insn_ptr = *insn_ptr32;
+	}
+
+	ppt = uprobe_find_probept(utask->uproc, ARCH_BP_INST_PTR(insn_ptr));
+	return (ppt != NULL);
+}
+
+/*
+ * Quiesce callback: The associated process has one or more breakpoint
+ * insertions or removals pending.  If we're the last thread in this
+ * process to quiesce, do the insertion(s) and/or removal(s).
+ */
+static u32 uprobe_report_quiesce(struct utrace_attached_engine *engine,
+		struct task_struct *tsk)
+{
+	struct uprobe_task *utask;
+	struct uprobe_process *uproc;
+
+	utask = rcu_dereference((struct uprobe_task *)engine->data);
+	BUG_ON(!utask);
+	uproc = utask->uproc;
+	if (current == utask->quiesce_master) {
+		/*
+		 * tsk was already quiescent when quiesce_all_threads()
+		 * called utrace_set_flags(), which in turned called
+		 * here.  uproc is already locked.  Do as little as possible
+		 * and get out.
+		 */
+		utask->state = UPTASK_QUIESCENT;
+		uproc->n_quiescent_threads++;
+		return UTRACE_ACTION_RESUME;
+	}
+
+	BUG_ON(utask->active_probe);
+	down_write(&uproc->rwsem);
+
+	/*
+	 * When a thread hits a breakpoint or single-steps, utrace calls
+	 * this quiesce callback before our signal callback.  We must
+	 * let uprobe_report_signal() handle the uprobe hit and THEN
+	 * quiesce, because (a) there's a chance that we're quiescing
+	 * in order to remove that very uprobe, and (b) there's a tiny
+	 * chance that even though that uprobe isn't marked for removal
+	 * now, it may be before all threads manage to quiesce.
+	 */
+	if (!utask->quiescing || utask_quiesce_pending_sigtrap(utask) == 1) {
+		clear_utrace_quiesce(utask);
+		goto done;
+	}
+
+	utask->state = UPTASK_QUIESCENT;
+	uproc->n_quiescent_threads++;
+	check_uproc_quiesced(uproc, tsk);
+done:
+	up_write(&uproc->rwsem);
+	return UTRACE_ACTION_RESUME;
+}
+
+/*
+ * uproc's process is exiting or exec-ing, so zap all the (now irrelevant)
+ * probepoints.  Runs with uproc->rwsem write-locked.  Caller must ref-count
+ * uproc before calling this function, to ensure that uproc doesn't get
+ * freed in the middle of this.
+ */
+static void uprobe_cleanup_process(struct uprobe_process *uproc)
+{
+	int i;
+	struct uprobe_probept *ppt;
+	struct hlist_node *pnode1, *pnode2;
+	struct hlist_head *head;
+	struct uprobe_kimg *uk, *unode;
+
+	uproc->finished = 1;
+
+	for (i = 0; i < UPROBE_TABLE_SIZE; i++) {
+		head = &uproc->uprobe_table[i];
+		hlist_for_each_entry_safe(ppt, pnode1, pnode2, head, ut_node) {
+			if (ppt->state == UPROBE_INSERTING ||
+					ppt->state == UPROBE_REMOVING) {
+				/*
+				 * This task is (exec/exit)ing with
+				 * a [un]register_uprobe pending.
+				 * [un]register_uprobe will free ppt.
+				 */
+				ppt->state = UPROBE_DISABLED;
+				list_del(&ppt->pd_node);
+				list_for_each_entry_safe(uk, unode,
+					       &ppt->uprobe_list, list)
+					uk->status = -ESRCH;
+				wake_up_all(&ppt->waitq);
+			} else if (ppt->state == UPROBE_BP_SET) {
+				list_for_each_entry_safe(uk, unode,
+					       &ppt->uprobe_list, list) {
+					list_del(&uk->list);
+					uprobe_free_kimg(uk);
+				}
+				uprobe_free_probept(ppt);
+			/* else */
+				/*
+				 * If ppt is UPROBE_DISABLED, assume that
+				 * [un]register_uprobe() has been notified
+				 * and will free it soon.
+				 */
+			}
+		}
+	}
+}
+
+/*
+ * Exit callback: The associated task/thread is exiting.
+ */
+static u32 uprobe_report_exit(struct utrace_attached_engine *engine,
+		struct task_struct *tsk, long orig_code, long *code)
+{
+	struct uprobe_task *utask;
+	struct uprobe_process *uproc;
+	struct uprobe_probept *ppt;
+	int utask_quiescing;
+
+	utask = rcu_dereference((struct uprobe_task *)engine->data);
+	uproc = utask->uproc;
+	uprobe_get_process(uproc);
+
+	ppt = utask->active_probe;
+	if (ppt) {
+		if (utask->state == UPTASK_TRAMPOLINE_HIT)
+			printk(KERN_WARNING "Task died during uretprobe return:"
+				"  pid/tgid = %d/%d\n", tsk->pid, tsk->tgid);
+		else
+			printk(KERN_WARNING "Task died at uprobe probepoint:"
+				"  pid/tgid = %d/%d, probepoint = %#lx\n",
+				tsk->pid, tsk->tgid, ppt->vaddr);
+		/* Mutex cleanup depends on where we died and SSOL vs. SSIL. */
+		if (uproc->sstep_out_of_line) {
+			if (utask->state == UPTASK_SSTEP
+					&& ppt->slot && ppt->slot->owner == ppt)
+				up_read(&ppt->slot->rwsem);
+		} else {
+			switch (utask->state) {
+			case UPTASK_PRE_SSTEP:
+			case UPTASK_SSTEP:
+			case UPTASK_POST_SSTEP:
+				mutex_unlock(&ppt->ssil_mutex);
+				break;
+			default:
+				break;
+			}
+		}
+		up_read(&uproc->rwsem);
+		if (utask->state == UPTASK_TRAMPOLINE_HIT)
+			uprobe_decref_process(uproc);
+	}
+
+	down_write(&uproc->rwsem);
+	utask_quiescing = utask->quiescing;
+	uprobe_free_task(utask);
+
+	uproc->nthreads--;
+	if (uproc->nthreads) {
+		if (utask_quiescing)
+			/*
+			 * In case other threads are waiting for
+			 * us to quiesce...
+			 */
+			check_uproc_quiesced(uproc,
+				       find_surviving_thread(uproc));
+	} else {
+		/*
+		 * We were the last remaining thread - clean up the uprobe
+		 * remnants a la unregister_uprobe(). We don't have to
+		 * remove the breakpoints, though.
+		 */
+		uprobe_cleanup_process(uproc);
+	}
+	up_write(&uproc->rwsem);
+	uprobe_put_process(uproc);
+
+	return UTRACE_ACTION_DETACH;
+}
+
+/*
+ * Clone callback: The current task has spawned a thread/process.
+ *
+ * NOTE: For now, we don't pass on uprobes from the parent to the
+ * child. We now do the necessary clearing of breakpoints in the
+ * child's address space.
+ *
+ * TODO:
+ *	- Provide option for child to inherit uprobes.
+ */
+static u32 uprobe_report_clone(struct utrace_attached_engine *engine,
+		struct task_struct *parent, unsigned long clone_flags,
+		struct task_struct *child)
+{
+	int len;
+	struct uprobe_process *uproc;
+	struct uprobe_task *ptask, *ctask;
+
+	ptask = rcu_dereference((struct uprobe_task *)engine->data);
+	uproc = ptask->uproc;
+
+	/*
+	 * Lock uproc so no new uprobes can be installed 'til all
+	 * report_clone activities are completed
+	 */
+	down_write(&uproc->rwsem);
+	get_task_struct(child);
+
+	if (clone_flags & CLONE_THREAD) {
+		/* New thread in the same process */
+		ctask = uprobe_add_task(child, uproc);
+		BUG_ON(!ctask);
+		if (IS_ERR(ctask)) {
+			put_task_struct(child);
+			up_write(&uproc->rwsem);
+			goto fail;
+		}
+		if (ctask)
+			uproc->nthreads++;
+		/*
+		 * FIXME: Handle the case where uproc is quiescing
+		 * (assuming it's possible to clone while quiescing).
+		 */
+	} else {
+		/*
+		 * New process spawned by parent.  Remove the probepoints
+		 * in the child's text.
+		 *
+		 * Its not necessary to quiesce the child as we are assured
+		 * by utrace that this callback happens *before* the child
+		 * gets to run userspace.
+		 *
+		 * We also hold the uproc->rwsem for the parent - so no
+		 * new uprobes will be registered 'til we return.
+		 */
+		int i;
+		struct uprobe_probept *ppt;
+		struct hlist_node *node;
+		struct hlist_head *head;
+
+		for (i = 0; i < UPROBE_TABLE_SIZE; i++) {
+			head = &uproc->uprobe_table[i];
+			hlist_for_each_entry(ppt, node, head, ut_node) {
+				len = set_orig_insn(ppt, child);
+				if (len != BP_INSN_SIZE) {
+					/* Ratelimit this? */
+					printk(KERN_ERR "Pid %d forked %d;"
+						" failed to remove probepoint"
+						" at %#lx in child\n",
+						parent->pid, child->pid,
+						ppt->vaddr);
+				}
+			}
+		}
+	}
+
+	put_task_struct(child);
+	up_write(&uproc->rwsem);
+
+fail:
+	return UTRACE_ACTION_RESUME;
+}
+
+/*
+ * Exec callback: The associated process called execve() or friends
+ *
+ * The new program is about to start running and so there is no
+ * possibility of a uprobe from the previous user address space
+ * to be hit.
+ *
+ * NOTE:
+ *	Typically, this process would have passed through the clone
+ *	callback, where the necessary action *should* have been
+ *	taken. However, if we still end up at this callback:
+ *		- We don't have to clear the uprobes - memory image
+ *		  will be overlaid.
+ *		- We have to free up uprobe resources associated with
+ *		  this process.
+ */
+static u32 uprobe_report_exec(struct utrace_attached_engine *engine,
+		struct task_struct *tsk, const struct linux_binprm *bprm,
+		struct pt_regs *regs)
+{
+	struct uprobe_process *uproc;
+	struct uprobe_task *utask;
+	int uproc_freed;
+
+	utask = rcu_dereference((struct uprobe_task *)engine->data);
+	uproc = utask->uproc;
+	uprobe_get_process(uproc);
+
+	down_write(&uproc->rwsem);
+	uprobe_cleanup_process(uproc);
+	/*
+	 * If [un]register_uprobe() is in progress, cancel the quiesce.
+	 * Otherwise, utrace_report_exec() might call uprobe_report_exec()
+	 * while the [un]register_uprobe thread is freeing the uproc.
+	 */
+	clear_utrace_quiesce(utask);
+	up_write(&uproc->rwsem);
+
+	/* If any [un]register_uprobe is pending, it'll clean up. */
+	uproc_freed = uprobe_put_process(uproc);
+	return (uproc_freed ? UTRACE_ACTION_DETACH : UTRACE_ACTION_RESUME);
+}
+
+static const struct utrace_engine_ops uprobe_utrace_ops =
+{
+	.report_quiesce = uprobe_report_quiesce,
+	.report_signal = uprobe_report_signal,
+	.report_exit = uprobe_report_exit,
+	.report_clone = uprobe_report_clone,
+	.report_exec = uprobe_report_exec
+};
+
+/*
+ * Initialize the uprobes global data area, and return a pointer
+ * to it.  Caller will initialize uprobes_data pointer afterward, to
+ * ensure that no other module sees a non-null uprobes_data until it's
+ * completely initialized.
+ */
+static struct uprobe_globals *init_uprobes(void)
+{
+	int i;
+	struct uprobe_globals *g = kmalloc(sizeof(*g), GFP_USER);
+	if (!g)
+		return ERR_PTR(-ENOMEM);
+	for (i = 0; i < UPROBE_TABLE_SIZE; i++) {
+		INIT_HLIST_HEAD(&g->uproc_table[i]);
+		INIT_HLIST_HEAD(&g->utask_table[i]);
+	}
+	mutex_init(&g->uproc_mutex);
+	spin_lock_init(&g->utask_table_lock);
+	return g;
+}
+
+/*
+ * Verify that the uprobes globals area has been set up, and that the
+ * current module's globals variable points at it.  Returns 0 if the
+ * area is successfully set up, or a negative erro otherwise.
+ */
+static int verify_uprobes(void)
+{
+	if (unlikely(!globals)) {
+		/*
+		 * First time through for this instrumentation module.
+		 * uprobes_mutex protects both the global uprobes
+		 * initialization and this module's local initialization.
+		 */
+		struct uprobe_globals *g;
+
+		mutex_lock(&uprobes_mutex);
+		if (!uprobes_data) {
+			/* First time through since boot time */
+			g = init_uprobes();
+			uprobes_data = g;
+		} else
+			g = uprobes_data;
+		if (!IS_ERR(g)) {
+			p_uprobe_utrace_ops = &uprobe_utrace_ops;
+			uproc_table = g->uproc_table;
+			utask_table = g->utask_table;
+		}
+
+		/* Set globals pointer to signify all is initialized. */
+		globals = g;
+		mutex_unlock(&uprobes_mutex);
+	}
+	if (unlikely(IS_ERR(globals)))
+		return (int) PTR_ERR(globals);
+	return 0;
+}
+
+#ifdef CONFIG_URETPROBES
+
+/* Called when the entry-point probe u is hit. */
+static void uretprobe_handle_entry(struct uprobe *u, struct pt_regs *regs,
+	struct uprobe_task *utask)
+{
+	struct uretprobe_instance *ri;
+	unsigned long trampoline_addr;
+
+	if (IS_ERR(utask->uproc->uretprobe_trampoline_addr))
+		return;
+	trampoline_addr = (unsigned long)
+		utask->uproc->uretprobe_trampoline_addr;
+	ri = (struct uretprobe_instance *)
+		kmalloc(sizeof(struct uretprobe_instance), GFP_USER);
+	if (!ri)
+		return;
+	ri->ret_addr = arch_hijack_uret_addr(trampoline_addr, regs, utask);
+	if (likely(ri->ret_addr)) {
+		ri->rp = container_of(u, struct uretprobe, u);
+		INIT_HLIST_NODE(&ri->hlist);
+		hlist_add_head(&ri->hlist, &utask->uretprobe_instances);
+		/* We ref-count outstanding uretprobe_instances. */
+		uprobe_get_process(utask->uproc);
+	} else
+		kfree(ri);
+}
+
+/*
+ * For each uretprobe_instance pushed onto the LIFO for the function
+ * instance that's now returning, call the handler, free the ri, and
+ * decrement the uproc's ref count.  Caller ref-counts uproc, so we
+ * should never hit zero in this function.
+ *
+ * Returns the original return address.
+ *
+ * TODO: Handle longjmp out of uretprobed function.
+ */
+static unsigned long uretprobe_run_handlers(struct uprobe_task *utask,
+		struct pt_regs *regs, unsigned long trampoline_addr)
+{
+	unsigned long ret_addr;
+	struct hlist_head *head = &utask->uretprobe_instances;
+	struct uretprobe_instance *ri;
+	struct hlist_node *r1, *r2;
+
+	hlist_for_each_entry_safe(ri, r1, r2, head, hlist) {
+		if (ri->rp && ri->rp->handler)
+			ri->rp->handler(ri, regs);
+		ret_addr = ri->ret_addr;
+		hlist_del(&ri->hlist);
+		kfree(ri);
+		uprobe_decref_process(utask->uproc);
+		if (ret_addr != trampoline_addr)
+			/*
+			 * This is the first ri (chronologically) pushed for
+			 * this particular instance of the probed function.
+			 */
+			return ret_addr;
+	}
+	printk(KERN_ERR "No uretprobe instance with original return address!"
+		" pid/tgid=%d/%d", utask->tsk->pid, utask->tsk->tgid);
+	utask->doomed = 1;
+	return 0;
+}
+
+/* Called when the uretprobe trampoline is hit. */
+static void uretprobe_handle_return(struct pt_regs *regs,
+	struct uprobe_task *utask)
+{
+	unsigned long orig_ret_addr;
+	/* Delay recycling of uproc until end of uprobe_report_signal() */
+	uprobe_get_process(utask->uproc);
+	utask->state = UPTASK_TRAMPOLINE_HIT;
+	utask->active_probe = &uretprobe_trampoline_dummy_probe;
+	orig_ret_addr = uretprobe_run_handlers(utask, regs,
+		(unsigned long) utask->uproc->uretprobe_trampoline_addr);
+	arch_restore_uret_addr(orig_ret_addr, regs);
+}
+
+static
+int register_uretprobe(struct uretprobe *rp)
+{
+	if (!rp || !rp->handler)
+		return -EINVAL;
+	rp->u.handler = URETPROBE_HANDLE_ENTRY;
+	return register_uprobe(&rp->u);
+}
+
+/*
+ * The uretprobe containing u is being unregistered.  Its uretprobe_instances
+ * have to hang around 'til their associated instances return (but we can't
+ * run rp's handler).  Zap ri->rp for each one to indicate unregistration.
+ *
+ * Runs with uproc write-locked.
+ */
+static void zap_uretprobe_instances(struct uprobe *u,
+		struct uprobe_process *uproc)
+{
+	struct uprobe_task *utask;
+	struct uretprobe *rp = container_of(u, struct uretprobe, u);
+
+	if (!uproc)
+		return;
+
+	list_for_each_entry(utask, &uproc->thread_list, list) {
+		struct hlist_node *r;
+		struct uretprobe_instance *ri;
+
+		hlist_for_each_entry(ri, r, &utask->uretprobe_instances, hlist)
+			if (ri->rp == rp)
+				ri->rp = NULL;
+	}
+}
+
+static
+void unregister_uretprobe(struct uretprobe *rp)
+{
+	if (!rp)
+		return;
+	unregister_uprobe(&rp->u);
+}
+
+/*
+ * uproc->ssol_area has been successfully set up.  Establish the
+ * uretprobe trampoline in slot 0.
+ */
+static void uretprobe_set_trampoline(struct uprobe_process *uproc)
+{
+	uprobe_opcode_t bp_insn = BREAKPOINT_INSTRUCTION;
+	struct uprobe_ssol_area *area = &uproc->ssol_area;
+	struct uprobe_ssol_slot *slot = &area->slots[0];
+
+	if (access_process_vm(current, (unsigned long) slot->insn,
+			&bp_insn, BP_INSN_SIZE, 1) == BP_INSN_SIZE) {
+		uproc->uretprobe_trampoline_addr = slot->insn;
+		slot->state = SSOL_RESERVED;
+		area->next_slot = 1;
+		area->nfree--;
+	} else {
+		printk(KERN_ERR "uretprobes disabled for pid %d:"
+			" cannot set uretprobe trampoline at %p\n",
+			uproc->tgid, slot->insn);
+	}
+}
+
+#else	/* ! CONFIG_URETPROBES */
+
+static void uretprobe_handle_entry(struct uprobe *u, struct pt_regs *regs,
+	struct uprobe_task *utask)
+{
+}
+static void uretprobe_handle_return(struct pt_regs *regs,
+	struct uprobe_task *utask)
+{
+}
+static void uretprobe_set_trampoline(struct uprobe_process *uproc)
+{
+}
+static void zap_uretprobe_instances(struct uprobe *u,
+	struct uprobe_process *uproc)
+{
+}
+#endif /* CONFIG_URETPROBES */
+
+#include "uprobes_arch.c"
diff --git a/runtime/uprobes/uprobes.h b/runtime/uprobes/uprobes.h
new file mode 100644
index 00000000..574bee62
--- /dev/null
+++ b/runtime/uprobes/uprobes.h
@@ -0,0 +1,385 @@
+#ifndef _LINUX_UPROBES_H
+#define _LINUX_UPROBES_H
+/*
+ * Userspace Probes (UProbes)
+ * include/linux/uprobes.h
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2006
+ */
+#include <linux/types.h>
+#include <linux/list.h>
+
+struct pt_regs;
+
+enum uprobe_type {
+	UPTY_UPROBE,
+	UPTY_URETPROBE
+};
+
+/* This is what the user supplies us. */
+struct uprobe {
+	/*
+	 * The pid of the probed process.  Currently, this can be the
+	 * thread ID (task->pid) of any active thread in the process.
+	 */
+	pid_t pid;
+
+	/* Location of the probepoint */
+	unsigned long vaddr;
+
+	/* Handler to run when the probepoint is hit */
+	void (*handler)(struct uprobe*, struct pt_regs*);
+
+	/*
+	 * This function, if non-NULL, will be called upon completion of
+	 * an ASYNCHRONOUS registration (i.e., one initiated by a uprobe
+	 * handler).  reg = 1 for register, 0 for unregister.  type
+	 * specifies the type of [un]register call (uprobe or uretprobe).
+	 */
+	void (*registration_callback)(struct uprobe *u, int reg,
+			enum uprobe_type type, int result);
+
+	/* Reserved for use by uprobes */
+	void *kdata;
+};
+
+struct uretprobe_instance;
+
+struct uretprobe {
+	struct uprobe u;
+	void (*handler)(struct uretprobe_instance*, struct pt_regs*);
+};
+
+struct uretprobe_instance {
+	struct uretprobe *rp;
+	unsigned long ret_addr;
+	struct hlist_node hlist;
+	unsigned long reserved1;
+	unsigned long reserved2;
+};
+
+static int register_uprobe(struct uprobe *u);
+static void unregister_uprobe(struct uprobe *u);
+/* For runtime, assume uprobes support includes uretprobes. */
+static int register_uretprobe(struct uretprobe *rp);
+static void unregister_uretprobe(struct uretprobe *rp);
+
+#ifdef UPROBES_IMPLEMENTATION
+
+#include <linux/mutex.h>
+#include <linux/rwsem.h>
+#include <linux/wait.h>
+#include <asm/atomic.h>
+#include "uprobes_arch.h"
+
+struct task_struct;
+struct utrace_attached_engine;
+
+enum uprobe_probept_state {
+	UPROBE_INSERTING,	// process quiescing prior to insertion
+	UPROBE_BP_SET,		// breakpoint in place
+	UPROBE_REMOVING,	// process quiescing prior to removal
+	UPROBE_DISABLED		// removal completed
+};
+
+enum uprobe_task_state {
+	UPTASK_QUIESCENT,
+	UPTASK_SLEEPING,	// used when task may not be able to quiesce
+	UPTASK_RUNNING,
+	UPTASK_BP_HIT,
+	UPTASK_TRAMPOLINE_HIT,
+	UPTASK_PRE_SSTEP,
+	UPTASK_SSTEP,
+	UPTASK_POST_SSTEP
+};
+
+#define UPROBE_HASH_BITS 5
+#define UPROBE_TABLE_SIZE (1 << UPROBE_HASH_BITS)
+
+/* Used when deciding which instruction slot to steal. */
+enum uprobe_slot_state {
+	SSOL_FREE,
+	SSOL_ASSIGNED,
+	SSOL_BEING_STOLEN,
+	SSOL_RESERVED		// e.g., for uretprobe trampoline
+};
+
+/*
+ * For a uprobe_process that uses an SSOL area, there's an array of these
+ * objects matching the array of instruction slots in the SSOL area.
+ */
+struct uprobe_ssol_slot {
+	/* The slot in the SSOL area that holds the instruction-copy */
+	__user uprobe_opcode_t	*insn;
+
+	enum uprobe_slot_state state;
+
+	/* The probepoint that currently owns this slot */
+	struct uprobe_probept *owner;
+
+	/*
+	 * Read-locked when slot is in use during single-stepping.
+	 * Write-locked by stealing task.
+	 */
+	struct rw_semaphore rwsem;
+
+	/* Used for LRU heuristics.  If this overflows, it's OK. */
+	unsigned long last_used;
+};
+
+/*
+ * The per-process single-stepping out-of-line (SSOL) area
+ */
+struct uprobe_ssol_area {
+	/* Array of instruction slots in the vma we allocate */
+	__user uprobe_opcode_t *insn_area;
+
+	int nslots;
+	int nfree;
+
+	/* Array of slot objects, one per instruction slot */
+	struct uprobe_ssol_slot *slots;
+
+	/* lock held while finding a free slot */
+	spinlock_t lock;
+
+	/* Next slot to steal */
+	int next_slot;
+
+	/* Ensures 2 threads don't try to set up the vma simultaneously. */
+	struct mutex setup_mutex;
+
+	/* 1 = we've at least tried.  IS_ERR(insn_area) if we failed. */
+	int initialized;
+};
+
+/*
+ * uprobe_process -- not a user-visible struct.
+ * A uprobe_process represents a probed process.  A process can have
+ * multiple probepoints (each represented by a uprobe_probept) and
+ * one or more threads (each represented by a uprobe_task).
+ */
+struct uprobe_process {
+	/*
+	 * rwsem is write-locked for any change to the uprobe_process's
+	 * graph (including uprobe_tasks, uprobe_probepts, and uprobe_kimgs) --
+	 * e.g., due to probe [un]registration or special events like exit.
+	 * It's read-locked during the whole time we process a probepoint hit.
+	 */
+	struct rw_semaphore rwsem;
+
+	/* Table of uprobe_probepts registered for this process */
+	/* TODO: Switch to list_head[] per Ingo. */
+	struct hlist_head uprobe_table[UPROBE_TABLE_SIZE];
+	int nppt;	/* number of probepoints */
+
+	/* List of uprobe_probepts awaiting insertion or removal */
+	struct list_head pending_uprobes;
+
+	/* List of uprobe_tasks in this task group */
+	struct list_head thread_list;
+	int nthreads;
+	int n_quiescent_threads;
+
+	/* this goes on the uproc_table */
+	struct hlist_node hlist;
+
+	/*
+	 * All threads (tasks) in a process share the same uprobe_process.
+	 */
+	pid_t tgid;
+
+	/* Threads in SLEEPING state wait here to be roused. */
+	wait_queue_head_t waitq;
+
+	/*
+	 * We won't free the uprobe_process while...
+	 * - any register/unregister operations on it are in progress; or
+	 * - uprobe_table[] is not empty; or
+	 * - any tasks are SLEEPING in the waitq; or
+	 * - any uretprobe_instances are outstanding.
+	 * refcount reflects this.  We do NOT ref-count tasks (threads),
+	 * since once the last thread has exited, the rest is academic.
+	 */
+	atomic_t refcount;
+
+	/* Return-probed functions return via this trampoline. */
+	__user uprobe_opcode_t *uretprobe_trampoline_addr;
+
+	/*
+	 * finished = 1 means the process is execing or the last thread
+	 * is exiting, and we're cleaning up the uproc.  If the execed
+	 * process is probed, a new uproc will be created.
+	 */
+	int finished;
+
+	/*
+	 * Manages slots for instruction-copies to be single-stepped
+	 * out of line.
+	 */
+	struct uprobe_ssol_area ssol_area;
+
+	/*
+	 * 1 to single-step out of line; 0 for inline.  This can drop to
+	 * 0 if we can't set up the SSOL area, but never goes from 0 to 1.
+	 */
+	int sstep_out_of_line;
+};
+
+/*
+ * uprobe_kimg -- not a user-visible struct.
+ * Holds implementation-only per-uprobe data.
+ * uprobe->kdata points to this.
+ */
+struct uprobe_kimg {
+	struct uprobe *uprobe;
+	struct uprobe_probept *ppt;
+
+	/*
+	 * -EBUSY while we're waiting for all threads to quiesce so the
+	 * associated breakpoint can be inserted or removed.
+	 * 0 if the the insert/remove operation has succeeded, or -errno
+	 * otherwise.
+	 */
+	int status;
+
+	/* on ppt's list */
+	struct list_head list;
+};
+
+/*
+ * uprobe_probept -- not a user-visible struct.
+ * A probepoint, at which several uprobes can be registered.
+ * Guarded by uproc->rwsem.
+ */
+struct uprobe_probept {
+	/* vaddr copied from (first) uprobe */
+	unsigned long vaddr;
+
+	/* The uprobe_kimg(s) associated with this uprobe_probept */
+	struct list_head uprobe_list;
+
+	enum uprobe_probept_state state;
+
+	/* Saved opcode (which has been replaced with breakpoint) */
+	uprobe_opcode_t opcode;
+
+	/* Saved original instruction */
+	uprobe_opcode_t insn[MAX_UINSN_BYTES / sizeof(uprobe_opcode_t)];
+
+	/* The parent uprobe_process */
+	struct uprobe_process *uproc;
+
+	/*
+	 * ppt goes in the uprobe_process->uprobe_table when registered --
+	 * even before the breakpoint has been inserted.
+	 */
+	struct hlist_node ut_node;
+
+	/*
+	 * ppt sits in the uprobe_process->pending_uprobes queue while
+	 * awaiting insertion or removal of the breakpoint.
+	 */
+	struct list_head pd_node;
+
+	/* [un]register_uprobe() waits 'til bkpt inserted/removed. */
+	wait_queue_head_t waitq;
+
+	/*
+	 * Serialize single-stepping inline, so threads don't clobber
+	 * each other swapping the breakpoint instruction in and out.
+	 * This helps prevent crashing the probed app, but it does NOT
+	 * prevent probe misses while the breakpoint is swapped out.
+	 */
+	struct mutex ssil_mutex;
+
+	/*
+	 * We put the instruction-copy here to single-step it.
+	 * We don't own it unless slot->owner points back to us.
+	 */
+	struct uprobe_ssol_slot *slot;
+
+	/*
+	 * Hold this while stealing an insn slot to ensure that no
+	 * other thread, having also hit this probepoint, simultaneously
+	 * steals a slot for it.
+	 */
+	struct mutex slot_mutex;
+};
+
+/*
+ * uprobe_utask -- not a user-visible struct.
+ * Corresponds to a thread in a probed process.
+ * Guarded by uproc->rwsem.
+ */
+struct uprobe_task {
+	/* Lives in the global utask_table */
+	struct hlist_node hlist;
+
+	/* Lives on the thread_list for the uprobe_process */
+	struct list_head list;
+
+	/* This is a back pointer to the task_struct for this task */
+	struct task_struct *tsk;
+
+	/* The utrace engine for this task */
+	struct utrace_attached_engine *engine;
+
+	/* Back pointer to the associated uprobe_process */
+	struct uprobe_process *uproc;
+
+	enum uprobe_task_state state;
+
+	/*
+	 * quiescing = 1 means this task has been asked to quiesce.
+	 * It may not be able to comply immediately if it's hit a bkpt.
+	 */
+	int quiescing;
+
+	/* Task currently running quiesce_all_threads() */
+	struct task_struct *quiesce_master;
+
+	/* Set before running handlers; cleared after single-stepping. */
+	struct uprobe_probept *active_probe;
+
+	/* Saved address of copied original instruction */
+	long singlestep_addr;
+
+	/*
+	 * Unexpected error in probepoint handling has left task's
+	 * text or stack corrupted.  Kill task ASAP.
+	 */
+	int doomed;
+
+	/* LIFO -- active instances */
+	struct hlist_head uretprobe_instances;
+
+	/* [un]registrations initiated by handlers must be asynchronous. */
+	struct list_head deferred_registrations;
+};
+
+#ifdef CONFIG_UPROBES_SSOL
+static struct uprobe_ssol_slot *uprobe_get_insn_slot(struct uprobe_probept*);
+static void uprobe_pre_ssout(struct uprobe_task*, struct uprobe_probept*,
+			struct pt_regs*);
+static void uprobe_post_ssout(struct uprobe_task*, struct uprobe_probept*,
+			struct pt_regs*);
+#endif
+
+#endif	/* UPROBES_IMPLEMENTATION */
+
+#endif	/* _LINUX_UPROBES_H */
diff --git a/runtime/uprobes/uprobes_arch.c b/runtime/uprobes/uprobes_arch.c
new file mode 100644
index 00000000..99ef54c8
--- /dev/null
+++ b/runtime/uprobes/uprobes_arch.c
@@ -0,0 +1,11 @@
+#if defined (__x86_64__)
+#include "uprobes_x86_64.c"
+#elif defined (__i386__)
+#include "uprobes_i386.c"
+#elif defined (__powerpc64__)
+#include "uprobes_ppc64.c"
+#elif defined (__s390__) || defined (__s390x__)
+#include "uprobes_s390.c"
+#else
+#error "Unsupported architecture"
+#endif
diff --git a/runtime/uprobes/uprobes_arch.h b/runtime/uprobes/uprobes_arch.h
new file mode 100644
index 00000000..0223e280
--- /dev/null
+++ b/runtime/uprobes/uprobes_arch.h
@@ -0,0 +1,11 @@
+#if defined (__x86_64__)
+#include "uprobes_x86_64.h"
+#elif defined (__i386__)
+#include "uprobes_i386.h"
+#elif defined (__powerpc64__)
+#include "uprobes_ppc64.h"
+#elif defined (__s390__) || defined (__s390x__)
+#include "uprobes_s390.h"
+#else
+#error "Unsupported architecture"
+#endif
diff --git a/runtime/uprobes/uprobes_i386.c b/runtime/uprobes/uprobes_i386.c
new file mode 100644
index 00000000..90d50ba0
--- /dev/null
+++ b/runtime/uprobes/uprobes_i386.c
@@ -0,0 +1,302 @@
+/*
+ *  Userspace Probes (UProbes)
+ *  arch/i386/kernel/uprobes_i386.c
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2006
+ */
+#define UPROBES_IMPLEMENTATION 1
+#include "uprobes.h"
+#include <linux/uaccess.h>
+
+/* Adapted from arch/x86_64/kprobes.c */
+#undef W
+#define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf)		      \
+	(((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) |   \
+	  (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) |   \
+	  (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) |   \
+	  (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf))    \
+	 << (row % 32))
+	static const unsigned long good_insns[256 / 32] = {
+		/*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
+		/*      -------------------------------         */
+		W(0x00, 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,0)| /* 00 */
+		W(0x10, 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,0), /* 10 */
+		W(0x20, 1,1,1,1,1,1,0,1,1,1,1,1,1,1,0,1)| /* 20 */
+		W(0x30, 1,1,1,1,1,1,0,1,1,1,1,1,1,1,0,1), /* 30 */
+		W(0x40, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 40 */
+		W(0x50, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1), /* 50 */
+		W(0x60, 1,1,1,0,0,0,0,0,1,1,1,1,0,0,0,0)| /* 60 */
+		W(0x70, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1), /* 70 */
+		W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
+		W(0x90, 1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1), /* 90 */
+		W(0xa0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* a0 */
+		W(0xb0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1), /* b0 */
+		W(0xc0, 1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0)| /* c0 */
+		W(0xd0, 1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1), /* d0 */
+		W(0xe0, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* e0 */
+		W(0xf0, 0,0,0,0,0,1,1,1,1,1,0,0,1,1,1,1)  /* f0 */
+		/*      -------------------------------         */
+		/*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
+	};
+
+	static const unsigned long good_2byte_insns[256 / 32] = {
+		/*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
+		/*      -------------------------------         */
+		W(0x00, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 00 */
+		W(0x10, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 10 */
+		W(0x20, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 20 */
+		W(0x30, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 30 */
+		W(0x40, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 40 */
+		W(0x50, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 50 */
+		W(0x60, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 60 */
+		W(0x70, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 70 */
+		W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
+		W(0x90, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1), /* 90 */
+		W(0xa0, 1,1,1,1,1,1,0,0,1,1,1,1,1,1,0,1)| /* a0 */
+		W(0xb0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1), /* b0 */
+		W(0xc0, 1,1,0,0,0,0,1,1,1,1,1,1,1,1,1,1)| /* c0 */
+		W(0xd0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* d0 */
+		W(0xe0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* e0 */
+		W(0xf0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)  /* f0 */
+		/*      -------------------------------         */
+		/*      0 1 2 3 4 5 6 7 8 9 a b c d e f         */
+	};
+
+/*
+ * TODO:
+ * - Allow valid 2-byte opcodes (first byte = 0x0f).
+ * - Where necessary, examine the modrm byte and allow valid instructions
+ * in the different Groups and fpu instructions.
+ * - Allow at least some instruction prefixes.
+ * - Note: If we go past the first byte, do we need to verify that
+ * subsequent bytes were actually there, rather than off the last page?
+ * Probably overkill.  We don't verify that they specified the first byte
+ * of the instruction, either.
+ * - Be clearer about which instructions we'll never probe.
+ */
+
+/*
+ * opcodes we'll probably never support:
+ * 63 - arpl
+ * 6c-6d, e4-e5, ec-ed - in
+ * 6e-6f, e6-e7, ee-ef - out
+ * cc, cd - int3, int
+ * cf - iret
+ * d6 - illegal instruction
+ * f1 - int1/icebp
+ * f4 - hlt
+ * fa, fb - cli, sti
+ *
+ * opcodes we may need to refine support for:
+ * 0f - valid 2-byte opcodes
+ * 66 - data16 prefix
+ * 8f - Group 1 - only reg = 0 is OK
+ * c6-c7 - Group 11 - only reg = 0 is OK
+ * d9-df - fpu insns with some illegal encodings
+ * fe - Group 4 - only reg = 1 or 2 is OK
+ * ff - Group 5 - only reg = 0-6 is OK
+ *
+ * others -- Do we need to support these?
+ * 07, 17, 1f - pop es, pop ss, pop ds
+ * 26, 2e, 36, 3e, 64, 65 - es:, cs:, ss:, ds:, fs:, gs: segment prefixes
+ * 67 - addr16 prefix
+ * 9b - wait/fwait
+ * ce - into
+ * f0 - lock prefix
+ * f2, f3 - repnz, repz prefixes
+ */
+
+static
+int arch_validate_probed_insn(struct uprobe_probept *ppt)
+{
+	uprobe_opcode_t *insn = ppt->insn;
+
+	if (insn[0] == 0x66)
+		/* Skip operand-size prefix */
+		insn++;
+	if (test_bit(insn[0], good_insns))
+		return 0;
+	if (insn[0] == 0x0f) {
+		if (test_bit(insn[1], good_2byte_insns))
+			return 0;
+		printk(KERN_ERR "uprobes does not currently support probing "
+			"instructions with the 2-byte opcode 0x0f 0x%2.2x\n",
+			insn[1]);
+	}
+	printk(KERN_ERR "uprobes does not currently support probing "
+		"instructions whose first byte is 0x%2.2x\n", insn[0]);
+	return -EPERM;
+}
+
+/*
+ * Get an instruction slot from the process's SSOL area, containing the
+ * instruction at ppt's probepoint.  Point the eip at that slot, in
+ * preparation for single-stepping out of line.
+ */
+static
+void uprobe_pre_ssout(struct uprobe_task *utask, struct uprobe_probept *ppt,
+		struct pt_regs *regs)
+{
+	struct uprobe_ssol_slot *slot;
+
+	slot = uprobe_get_insn_slot(ppt);
+	if (!slot) {
+		utask->doomed = 1;
+		return;
+	}
+	regs->eip = (long)slot->insn;
+	utask->singlestep_addr = regs->eip;
+}
+
+/*
+ * Called by uprobe_post_ssout() to adjust the return address
+ * pushed by a call instruction executed out-of-line.
+ */
+static void adjust_ret_addr(long esp, long correction,
+		struct uprobe_task *utask)
+{
+	int nleft;
+	long ra;
+
+	nleft = copy_from_user(&ra, (const void __user *) esp, 4);
+	if (unlikely(nleft != 0))
+		goto fail;
+	ra +=  correction;
+	nleft = copy_to_user((void __user *) esp, &ra, 4);
+	if (unlikely(nleft != 0))
+		goto fail;
+	return;
+
+fail:
+	printk(KERN_ERR
+		"uprobes: Failed to adjust return address after"
+		" single-stepping call instruction;"
+		" pid=%d, esp=%#lx\n", current->pid, esp);
+	utask->doomed = 1;
+}
+
+/*
+ * Called after single-stepping.  ppt->vaddr is the address of the
+ * instruction whose first byte has been replaced by the "int3"
+ * instruction.  To avoid the SMP problems that can occur when we
+ * temporarily put back the original opcode to single-step, we
+ * single-stepped a copy of the instruction.  The address of this
+ * copy is utask->singlestep_addr.
+ *
+ * This function prepares to return from the post-single-step
+ * interrupt.  We have to fix up the stack as follows:
+ *
+ * 0) Typically, the new eip is relative to the copied instruction.  We
+ * need to make it relative to the original instruction.  Exceptions are
+ * return instructions and absolute or indirect jump or call instructions.
+ *
+ * 1) If the single-stepped instruction was a call, the return address
+ * that is atop the stack is the address following the copied instruction.
+ * We need to make it the address following the original instruction.
+ */
+static
+void uprobe_post_ssout(struct uprobe_task *utask, struct uprobe_probept *ppt,
+		struct pt_regs *regs)
+{
+	long next_eip = 0;
+	long copy_eip = utask->singlestep_addr;
+	long orig_eip = ppt->vaddr;
+	uprobe_opcode_t *insn = ppt->insn;
+
+	up_read(&ppt->slot->rwsem);
+
+	if (insn[0] == 0x66)
+		/* Skip operand-size prefix */
+		insn++;
+
+	switch (insn[0]) {
+	case 0xc3:		/* ret/lret */
+	case 0xcb:
+	case 0xc2:
+	case 0xca:
+		next_eip = regs->eip;
+		/* eip is already adjusted, no more changes required*/
+		break;
+	case 0xe8:		/* call relative - Fix return addr */
+		adjust_ret_addr(regs->esp, (orig_eip - copy_eip), utask);
+		break;
+	case 0xff:
+		if ((insn[1] & 0x30) == 0x10) {
+			/* call absolute, indirect */
+			/* Fix return addr; eip is correct. */
+			next_eip = regs->eip;
+			adjust_ret_addr(regs->esp, (orig_eip - copy_eip),
+				utask);
+		} else if ((insn[1] & 0x31) == 0x20 ||
+			   (insn[1] & 0x31) == 0x21) {
+			/* jmp near or jmp far  absolute indirect */
+			/* eip is correct. */
+			next_eip = regs->eip;
+		}
+		break;
+	case 0xea:		/* jmp absolute -- eip is correct */
+		next_eip = regs->eip;
+		break;
+	default:
+		break;
+	}
+
+	if (next_eip)
+		regs->eip = next_eip;
+	else
+		regs->eip = orig_eip + (regs->eip - copy_eip);
+}
+
+/*
+ * Replace the return address with the trampoline address.  Returns
+ * the original return address.
+ */
+static
+unsigned long arch_hijack_uret_addr(unsigned long trampoline_address,
+		struct pt_regs *regs, struct uprobe_task *utask)
+{
+	int nleft;
+	unsigned long orig_ret_addr;
+#define RASIZE (sizeof(unsigned long))
+
+	nleft = copy_from_user(&orig_ret_addr,
+		       (const void __user *)regs->esp, RASIZE);
+	if (unlikely(nleft != 0))
+		return 0;
+
+	if (orig_ret_addr == trampoline_address)
+		/*
+		 * There's another uretprobe on this function, and it was
+		 * processed first, so the return address has already
+		 * been hijacked.
+		 */
+		return orig_ret_addr;
+
+	nleft = copy_to_user((void __user *)regs->esp,
+		       &trampoline_address, RASIZE);
+	if (unlikely(nleft != 0)) {
+		if (nleft != RASIZE) {
+			printk(KERN_ERR "uretprobe_entry_handler: "
+					"return address partially clobbered -- "
+					"pid=%d, %%esp=%#lx, %%eip=%#lx\n",
+					current->pid, regs->esp, regs->eip);
+			utask->doomed = 1;
+		} /* else nothing written, so no harm */
+		return 0;
+	}
+	return orig_ret_addr;
+}
diff --git a/runtime/uprobes/uprobes_i386.h b/runtime/uprobes/uprobes_i386.h
new file mode 100644
index 00000000..6e2aae70
--- /dev/null
+++ b/runtime/uprobes/uprobes_i386.h
@@ -0,0 +1,67 @@
+#ifndef _ASM_UPROBES_H
+#define _ASM_UPROBES_H
+/*
+ *  Userspace Probes (UProbes)
+ *  include/asm-i386/uprobes.h
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2006
+ */
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/signal.h>
+
+/* Normally defined in Kconfig */
+#define CONFIG_URETPROBES 1
+#define CONFIG_UPROBES_SSOL 1
+
+typedef u8 uprobe_opcode_t;
+#define BREAKPOINT_INSTRUCTION	0xcc
+#define BP_INSN_SIZE 1
+#define MAX_UINSN_BYTES 16
+#define SLOT_IP 12	/* instruction pointer slot from include/asm/elf.h */
+
+#define BREAKPOINT_SIGNAL SIGTRAP
+#define SSTEP_SIGNAL SIGTRAP
+
+/* Architecture specific switch for where the IP points after a bp hit */
+#define ARCH_BP_INST_PTR(inst_ptr)	(inst_ptr - BP_INSN_SIZE)
+
+struct uprobe_probept;
+struct uprobe_task;
+static int arch_validate_probed_insn(struct uprobe_probept *ppt);
+
+/* On i386, the int3 traps leaves eip pointing past the int3 instruction. */
+static inline unsigned long arch_get_probept(struct pt_regs *regs)
+{
+	return (unsigned long) (regs->eip - BP_INSN_SIZE);
+}
+
+static inline void arch_reset_ip_for_sstep(struct pt_regs *regs)
+{
+	regs->eip -= BP_INSN_SIZE;
+}
+
+static inline void arch_restore_uret_addr(unsigned long ret_addr,
+		struct pt_regs *regs)
+{
+	regs->eip = ret_addr;
+}
+
+static unsigned long arch_hijack_uret_addr(unsigned long trampoline_addr,
+		struct pt_regs *regs, struct uprobe_task *utask);
+
+#endif				/* _ASM_UPROBES_H */