[PATCH] pi-futex: futex_lock_pi/futex_unlock_pi support

c87e2837 · Ingo Molnar <[email protected]> · 2006-06-27 02:54:58 -07:00

This adds the actual pi-futex implementation, based on rt-mutexes.

[[email protected]: fix an oops-causing race]
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Signed-off-by: Arjan van de Ven <[email protected]>
Signed-off-by: Dinakar Guniguntala <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Commit: c87e2837be82df479a6bae9f155c43516d2feebc
Parents: 0cdbee99

Changes

7 files changed · +828 −41

File	Status	Add	Del	Content
include/linux/futex.h	modified	+7	−0	View file
include/linux/sched.h	modified	+3	−0	View file
kernel/exit.c	modified	+8	−0	View file
kernel/fork.c	modified	+3	−0	View file
kernel/futex.c	modified	+791	−38	View file
kernel/futex_compat.c	modified	+8	−3	View file
kernel/rtmutex_common.h	modified	+8	−0	View file

Only the first 600 diff lines are shown. Download the complete patch.

include/linux/futex.h

View file

diff --git a/include/linux/futex.h b/include/linux/futex.h
index f05a3f4..34c3a21 100644
--- a/include/linux/futex.h
+++ b/include/linux/futex.h
@@ -12,6 +12,9 @@
 #define FUTEX_REQUEUE		3
 #define FUTEX_CMP_REQUEUE	4
 #define FUTEX_WAKE_OP		5
+#define FUTEX_LOCK_PI		6
+#define FUTEX_UNLOCK_PI		7
+#define FUTEX_TRYLOCK_PI	8
 
 /*
  * Support for robust futexes: the kernel cleans up held futexes at
@@ -97,10 +100,14 @@ extern int handle_futex_death(u32 __user *uaddr, struct task_struct *curr);
 
 #ifdef CONFIG_FUTEX
 extern void exit_robust_list(struct task_struct *curr);
+extern void exit_pi_state_list(struct task_struct *curr);
 #else
 static inline void exit_robust_list(struct task_struct *curr)
 {
 }
+static inline void exit_pi_state_list(struct task_struct *curr)
+{
+}
 #endif
 
 #define FUTEX_OP_SET		0	/* *(int *)UADDR2 = OPARG; */

include/linux/sched.h

View file

diff --git a/include/linux/sched.h b/include/linux/sched.h
index edadd13..b4e6be7 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -84,6 +84,7 @@ struct sched_param {
 #include <asm/processor.h>
 
 struct exec_domain;
+struct futex_pi_state;
 
 /*
  * List of flags we want to share for kernel threads,
@@ -915,6 +916,8 @@ struct task_struct {
 #ifdef CONFIG_COMPAT
 	struct compat_robust_list_head __user *compat_robust_list;
 #endif
+	struct list_head pi_state_list;
+	struct futex_pi_state *pi_state_cache;
 
 	atomic_t fs_excl;	/* holding fs exclusive resources */
 	struct rcu_head rcu;

kernel/exit.c

View file

diff --git a/kernel/exit.c b/kernel/exit.c
index 3e8a028..ab06b9f 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -926,6 +926,14 @@ fastcall NORET_TYPE void do_exit(long code)
 	tsk->mempolicy = NULL;
 #endif
 	/*
+	 * This must happen late, after the PID is not
+	 * hashed anymore:
+	 */
+	if (unlikely(!list_empty(&tsk->pi_state_list)))
+		exit_pi_state_list(tsk);
+	if (unlikely(current->pi_state_cache))
+		kfree(current->pi_state_cache);
+	/*
 	 * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 	 */
 	mutex_debug_check_no_locks_held(tsk);

kernel/fork.c

View file

diff --git a/kernel/fork.c b/kernel/fork.c
index b664a08..628198a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1092,6 +1092,9 @@ static task_t *copy_process(unsigned long clone_flags,
 #ifdef CONFIG_COMPAT
 	p->compat_robust_list = NULL;
 #endif
+	INIT_LIST_HEAD(&p->pi_state_list);
+	p->pi_state_cache = NULL;
+
 	/*
 	 * sigaltstack should be cleared when sharing the same VM
 	 */

kernel/futex.c

View file

diff --git a/kernel/futex.c b/kernel/futex.c
index 50356fb..b305b7f 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -12,6 +12,10 @@
  *  (C) Copyright 2006 Red Hat Inc, All Rights Reserved
  *  Thanks to Thomas Gleixner for suggestions, analysis and fixes.
  *
+ *  PI-futex support started by Ingo Molnar and Thomas Gleixner
+ *  Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <[email protected]>
+ *  Copyright (C) 2006 Timesys Corp., Thomas Gleixner <[email protected]>
+ *
  *  Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
  *  enough at me, Linus for the original (flawed) idea, Matthew
  *  Kirkwood for proof-of-concept implementation.
@@ -46,6 +50,8 @@
 #include <linux/signal.h>
 #include <asm/futex.h>
 
+#include "rtmutex_common.h"
+
 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
 
 /*
@@ -75,6 +81,27 @@ union futex_key {
 };
 
 /*
+ * Priority Inheritance state:
+ */
+struct futex_pi_state {
+	/*
+	 * list of 'owned' pi_state instances - these have to be
+	 * cleaned up in do_exit() if the task exits prematurely:
+	 */
+	struct list_head list;
+
+	/*
+	 * The PI object:
+	 */
+	struct rt_mutex pi_mutex;
+
+	struct task_struct *owner;
+	atomic_t refcount;
+
+	union futex_key key;
+};
+
+/*
  * We use this hashed waitqueue instead of a normal wait_queue_t, so
  * we can wake only the relevant ones (hashed queues may be shared).
  *
@@ -96,6 +123,10 @@ struct futex_q {
 	/* For fd, sigio sent using these: */
 	int fd;
 	struct file *filp;
+
+	/* Optional priority inheritance state: */
+	struct futex_pi_state *pi_state;
+	struct task_struct *task;
 };
 
 /*
@@ -259,6 +290,232 @@ static inline int get_futex_value_locked(u32 *dest, u32 __user *from)
 }
 
 /*
+ * Fault handling. Called with current->mm->mmap_sem held.
+ */
+static int futex_handle_fault(unsigned long address, int attempt)
+{
+	struct vm_area_struct * vma;
+	struct mm_struct *mm = current->mm;
+
+	if (attempt >= 2 || !(vma = find_vma(mm, address)) ||
+	    vma->vm_start > address || !(vma->vm_flags & VM_WRITE))
+		return -EFAULT;
+
+	switch (handle_mm_fault(mm, vma, address, 1)) {
+	case VM_FAULT_MINOR:
+		current->min_flt++;
+		break;
+	case VM_FAULT_MAJOR:
+		current->maj_flt++;
+		break;
+	default:
+		return -EFAULT;
+	}
+	return 0;
+}
+
+/*
+ * PI code:
+ */
+static int refill_pi_state_cache(void)
+{
+	struct futex_pi_state *pi_state;
+
+	if (likely(current->pi_state_cache))
+		return 0;
+
+	pi_state = kmalloc(sizeof(*pi_state), GFP_KERNEL);
+
+	if (!pi_state)
+		return -ENOMEM;
+
+	memset(pi_state, 0, sizeof(*pi_state));
+	INIT_LIST_HEAD(&pi_state->list);
+	/* pi_mutex gets initialized later */
+	pi_state->owner = NULL;
+	atomic_set(&pi_state->refcount, 1);
+
+	current->pi_state_cache = pi_state;
+
+	return 0;
+}
+
+static struct futex_pi_state * alloc_pi_state(void)
+{
+	struct futex_pi_state *pi_state = current->pi_state_cache;
+
+	WARN_ON(!pi_state);
+	current->pi_state_cache = NULL;
+
+	return pi_state;
+}
+
+static void free_pi_state(struct futex_pi_state *pi_state)
+{
+	if (!atomic_dec_and_test(&pi_state->refcount))
+		return;
+
+	/*
+	 * If pi_state->owner is NULL, the owner is most probably dying
+	 * and has cleaned up the pi_state already
+	 */
+	if (pi_state->owner) {
+		spin_lock_irq(&pi_state->owner->pi_lock);
+		list_del_init(&pi_state->list);
+		spin_unlock_irq(&pi_state->owner->pi_lock);
+
+		rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
+	}
+
+	if (current->pi_state_cache)
+		kfree(pi_state);
+	else {
+		/*
+		 * pi_state->list is already empty.
+		 * clear pi_state->owner.
+		 * refcount is at 0 - put it back to 1.
+		 */
+		pi_state->owner = NULL;
+		atomic_set(&pi_state->refcount, 1);
+		current->pi_state_cache = pi_state;
+	}
+}
+
+/*
+ * Look up the task based on what TID userspace gave us.
+ * We dont trust it.
+ */
+static struct task_struct * futex_find_get_task(pid_t pid)
+{
+	struct task_struct *p;
+
+	read_lock(&tasklist_lock);
+	p = find_task_by_pid(pid);
+	if (!p)
+		goto out_unlock;
+	if ((current->euid != p->euid) && (current->euid != p->uid)) {
+		p = NULL;
+		goto out_unlock;
+	}
+	if (p->state == EXIT_ZOMBIE || p->exit_state == EXIT_ZOMBIE) {
+		p = NULL;
+		goto out_unlock;
+	}
+	get_task_struct(p);
+out_unlock:
+	read_unlock(&tasklist_lock);
+
+	return p;
+}
+
+/*
+ * This task is holding PI mutexes at exit time => bad.
+ * Kernel cleans up PI-state, but userspace is likely hosed.
+ * (Robust-futex cleanup is separate and might save the day for userspace.)
+ */
+void exit_pi_state_list(struct task_struct *curr)
+{
+	struct futex_hash_bucket *hb;
+	struct list_head *next, *head = &curr->pi_state_list;
+	struct futex_pi_state *pi_state;
+	union futex_key key;
+
+	/*
+	 * We are a ZOMBIE and nobody can enqueue itself on
+	 * pi_state_list anymore, but we have to be careful
+	 * versus waiters unqueueing themselfs
+	 */
+	spin_lock_irq(&curr->pi_lock);
+	while (!list_empty(head)) {
+
+		next = head->next;
+		pi_state = list_entry(next, struct futex_pi_state, list);
+		key = pi_state->key;
+		spin_unlock_irq(&curr->pi_lock);
+
+		hb = hash_futex(&key);
+		spin_lock(&hb->lock);
+
+		spin_lock_irq(&curr->pi_lock);
+		if (head->next != next) {
+			spin_unlock(&hb->lock);
+			continue;
+		}
+
+		list_del_init(&pi_state->list);
+
+		WARN_ON(pi_state->owner != curr);
+
+		pi_state->owner = NULL;
+		spin_unlock_irq(&curr->pi_lock);
+
+		rt_mutex_unlock(&pi_state->pi_mutex);
+
+		spin_unlock(&hb->lock);
+
+		spin_lock_irq(&curr->pi_lock);
+	}
+	spin_unlock_irq(&curr->pi_lock);
+}
+
+static int
+lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, struct futex_q *me)
+{
+	struct futex_pi_state *pi_state = NULL;
+	struct futex_q *this, *next;
+	struct list_head *head;
+	struct task_struct *p;
+	pid_t pid;
+
+	head = &hb->chain;
+
+	list_for_each_entry_safe(this, next, head, list) {
+		if (match_futex (&this->key, &me->key)) {
+			/*
+			 * Another waiter already exists - bump up
+			 * the refcount and return its pi_state:
+			 */
+			pi_state = this->pi_state;
+			atomic_inc(&pi_state->refcount);
+			me->pi_state = pi_state;
+
+			return 0;
+		}
+	}
+
+	/*
+	 * We are the first waiter - try to look up the real owner and
+	 * attach the new pi_state to it:
+	 */
+	pid = uval & FUTEX_TID_MASK;
+	p = futex_find_get_task(pid);
+	if (!p)
+		return -ESRCH;
+
+	pi_state = alloc_pi_state();
+
+	/*
+	 * Initialize the pi_mutex in locked state and make 'p'
+	 * the owner of it:
+	 */
+	rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
+
+	/* Store the key for possible exit cleanups: */
+	pi_state->key = me->key;
+
+	spin_lock_irq(&p->pi_lock);
+	list_add(&pi_state->list, &p->pi_state_list);
+	pi_state->owner = p;
+	spin_unlock_irq(&p->pi_lock);
+
+	put_task_struct(p);
+
+	me->pi_state = pi_state;
+
+	return 0;
+}
+
+/*
  * The hash bucket lock must be held when this is called.
  * Afterwards, the futex_q must not be accessed.
  */
@@ -285,6 +542,70 @@ static void wake_futex(struct futex_q *q)
 	q->lock_ptr = NULL;
 }
 
+static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
+{
+	struct task_struct *new_owner;
+	struct futex_pi_state *pi_state = this->pi_state;
+	u32 curval, newval;
+
+	if (!pi_state)
+		return -EINVAL;
+
+	new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
+
+	/*
+	 * This happens when we have stolen the lock and the original
+	 * pending owner did not enqueue itself back on the rt_mutex.
+	 * Thats not a tragedy. We know that way, that a lock waiter
+	 * is on the fly. We make the futex_q waiter the pending owner.
+	 */
+	if (!new_owner)
+		new_owner = this->task;
+
+	/*
+	 * We pass it to the next owner. (The WAITERS bit is always
+	 * kept enabled while there is PI state around. We must also
+	 * preserve the owner died bit.)
+	 */
+	newval = (uval & FUTEX_OWNER_DIED) | FUTEX_WAITERS | new_owner->pid;
+
+	inc_preempt_count();
+	curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
+	dec_preempt_count();
+
+	if (curval == -EFAULT)
+		return -EFAULT;
+	if (curval != uval)
+		return -EINVAL;
+
+	list_del_init(&pi_state->owner->pi_state_list);
+	list_add(&pi_state->list, &new_owner->pi_state_list);
+	pi_state->owner = new_owner;
+	rt_mutex_unlock(&pi_state->pi_mutex);
+
+	return 0;
+}
+
+static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
+{
+	u32 oldval;
+
+	/*
+	 * There is no waiter, so we unlock the futex. The owner died
+	 * bit has not to be preserved here. We are the owner:
+	 */
+	inc_preempt_count();
+	oldval = futex_atomic_cmpxchg_inatomic(uaddr, uval, 0);
+	dec_preempt_count();
+
+	if (oldval == -EFAULT)
+		return oldval;
+	if (oldval != uval)
+		return -EAGAIN;
+
+	return 0;
+}
+
 /*
  * Wake up all waiters hashed on the physical page that is mapped
  * to this virtual address:
@@ -309,6 +630,8 @@ static int futex_wake(u32 __user *uaddr, int nr_wake)
 
 	list_for_each_entry_safe(this, next, head, list) {
 		if (match_futex (&this->key, &key)) {
+			if (this->pi_state)
+				return -EINVAL;
 			wake_futex(this);
 			if (++ret >= nr_wake)
 				break;
@@ -385,27 +708,9 @@ retry:
 		 * still holding the mmap_sem.
 		 */
 		if (attempt++) {
-			struct vm_area_struct * vma;
-			struct mm_struct *mm = current->mm;
-			unsigned long address = (unsigned long)uaddr2;
-
-			ret = -EFAULT;
-			if (attempt >= 2 ||
-			    !(vma = find_vma(mm, address)) ||
-			    vma->vm_start > address ||
-			    !(vma->vm_flags & VM_WRITE))
+			if (futex_handle_fault((unsigned long)uaddr2,
+					       attempt))
 				goto out;
-
-			switch (handle_mm_fault(mm, vma, address, 1)) {
-			case VM_FAULT_MINOR:
-				current->min_flt++;
-				break;
-			case VM_FAULT_MAJOR:
-				current->maj_flt++;
-				break;
-			default:
-				goto out;
-			}
 			goto retry;
 		}
 
@@ -572,6 +877,7 @@ queue_lock(struct futex_q *q, int fd, struct file *filp)
 static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
 {
 	list_add_tail(&q->list, &hb->chain);
+	q->task = current;
 	spin_unlock(&hb->lock);
 }
 
@@ -626,6 +932,9 @@ static int unqueue_me(struct futex_q *q)
 		}
 		WARN_ON(list_empty(&q->list));
 		list_del(&q->list);
+
+		BUG_ON(q->pi_state);
+
 		spin_unlock(lock_ptr);
 		ret = 1;
 	}
@@ -634,16 +943,36 @@ static int unqueue_me(struct futex_q *q)
 	return ret;
 }
 
+/*
+ * PI futexes can not be requeued and must remove themself from the
+ * hash bucket. The hash bucket lock is held on entry and dropped here.
+ */
+static void unqueue_me_pi(struct futex_q *q, struct futex_hash_bucket *hb)
+{
+	WARN_ON(list_empty(&q->list));
+	list_del(&q->list);
+
+	BUG_ON(!q->pi_state);
+	free_pi_state(q->pi_state);
+	q->pi_state = NULL;
+
+	spin_unlock(&hb->lock);
+
+	drop_key_refs(&q->key);
+}
+
 static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time)
 {
-	DECLARE_WAITQUEUE(wait, current);
+	struct task_struct *curr = current;
+	DECLARE_WAITQUEUE(wait, curr);
 	struct futex_hash_bucket *hb;
 	struct futex_q q;
 	u32 uval;
 	int ret;
 
+	q.pi_state = NULL;
  retry:
-	down_read(&current->mm->mmap_sem);
+	down_read(&curr->mm->mmap_sem);
 
 	ret = get_futex_key(uaddr, &q.key);
 	if (unlikely(ret != 0))
@@ -680,7 +1009,7 @@ static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time)
 		 * If we would have faulted, release mmap_sem, fault it in and
 		 * start all over again.
 		 */
-		up_read(&current->mm->mmap_sem);
+		up_read(&curr->mm->mmap_sem);
 
 		ret = get_user(uval, uaddr);
 
@@ -688,11 +1017,9 @@ static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time)
 			goto retry;
 		return ret;
 	}
-	if (uval != val) {
-		ret = -EWOULDBLOCK;
-		queue_unlock(&q, hb);
-		goto out_release_sem;
-	}
+	ret = -EWOULDBLOCK;
+	if (uval != val)
+		goto out_unlock_release_sem;
 
 	/* Only actually queue if *uaddr contained val.  */
 	__queue_me(&q, hb);
@@ -700,8 +1027,8 @@ static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time)
 	/*
 	 * Now the futex is queued and we have checked the data, we
 	 * don't want to hold mmap_sem while we sleep.
-	 */	
-	up_read(&current->mm->mmap_sem);
+	 */
+	up_read(&curr->mm->mmap_sem);
 
 	/*
 	 * There might have been scheduling since the queue_me(), as we
@@ -739,8 +1066,415 @@ static int futex_wait(u32 __user *uaddr, u32 val, unsigned long time)
 	 */
 	return -EINTR;
 
+ out_unlock_release_sem:
+	queue_unlock(&q, hb);
+
  out_release_sem:
+	up_read(&curr->mm->mmap_sem);
+	return ret;
+}
+
+/*
+ * Userspace tried a 0 -> TID atomic transition of the futex value
+ * and failed. The kernel side here does the whole locking operation:
+ * if there are waiters then it will block, it does PI, etc. (Due to
+ * races the kernel might see a 0 value of the futex too.)
+ */
+static int do_futex_lock_pi(u32 __user *uaddr, int detect, int trylock,

This diff is incomplete. Download the complete patch.

kernel/futex_compat.c

View file

This diff is incomplete. Download the complete patch.

kernel/rtmutex_common.h

View file

This diff is incomplete. Download the complete patch.