futex-pi updates
cover futex-pi updates
Andrew,
please apply the following updates to the futex-pi code in 2.6.17-rc1-mm3:
- Remove buggy BUG_ON in the PI boosting code
- Enforce waiter bit in owner died situations
- Printk output fixlet
- Add restart handling for interrupted futex_pi lock operations
tglx
--
PATCH 1/4 rtmutex: Remove buggy BUG_ON in PI boosting code
From: Steven Rostedt <[email protected]>
The condition in that particular BUG_ON can legitimately be the case, if you have processes A, B, C, D, and E holding the following locks in this scenario:
L1 <=blocks= A
<=owns= L2 <=blocks= B <=owns= L4 <=blocks= D
<=owns= L3 <=blocks= C <=owns= L5 <=blocks= EWhere the priorities of these tasks are
B,C < A < D = E
B and C are less than A and A is less than D and E where D and E are equal (actually it probably works when D and E are not equal too).
As D and E climb the chain, there's a very slight race condition that could allow for the condition in the offending BUG_ON to be true.
kernel/rtmutex.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
kernel/rtmutex.c +1 -3
@@ -209,10 +209,8 @@ static int rt_mutex_adjust_prio_chain(ta* When deadlock detection is off then we check, if further * priority adjustment is necessary. */ -if (!detect_deadlock && waiter->list_entry.prio == task->prio) { -BUG_ON (waiter->pi_list_entry.prio != waiter->list_entry.prio); +if (!detect_deadlock && waiter->list_entry.prio == task->prio) goto out_unlock_pi; - } lock = waiter->lock;if (!spin_trylock (&lock->wait_lock)) {
--
PATCH 2/4 futex-pi: Enforce waiter bit when owner died is detected
In reply to Thomas GleixnerEnforce the waiter bit to be set, when the previous owner has died. This simplifies the glibc handling of the possible race from userspace tasks which try to get hold of the lock and cleanup the mess which was leftover by the unexpectedly died previous owner.
kernel/futex.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
kernel/futex.c +7 -2
@@ -1161,11 +1161,16 @@ static int futex_lock_pi(u32 __user *uad* failed. When the OWNER_DIED bit is set, then we * know that this is a robust futex and we actually * take the lock. This is safe as we are protected by -* the hash bucket lock. +* the hash bucket lock. We also set the waiters bit +* unconditionally here, to simplify glibc handling of +* multiple tasks racing to acquire the lock and +* cleanup the problems which were left by the dead +* owner. */ if (curval &FUTEX_OWNER_DIED ) { uval = newval; - newval = current->pid |FUTEX_OWNER_DIED ; + newval = current->pid | +FUTEX_OWNER_DIED |FUTEX_WAITERS ;inc_preempt_count (); curval = futex_atomic_cmpxchg_inatomic(uaddr,
--
PATCH 3/4 rtmutex debug: printk correct task information
In reply to Thomas GleixnerPrint the information of the current task rather than some random picked task information.
kernel/rtmutex.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/rtmutex.c +1 -1
@@ -180,7 +180,7 @@ static int rt_mutex_adjust_prio_chain(ta prev_max = max_lock_depth;printk (KERN_WARNING "Maximum lock depth %d reached " "task: %s (%d)\n" , max_lock_depth, - task->comm, task->pid); + current->comm, current->pid); }put_task_struct (task);
--
PATCH 4/4 futex-pi: Make use of restart_block when interrupted
In reply to Thomas GleixnerMake use of restart_block when the lock operation has been interrupted.
kernel/futex.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 72 insertions(+), 10 deletions(-)
kernel/futex.c +72 -10
@@ -1077,26 +1077,18 @@ static int futex_wait(u32 __user *uaddr,* 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 futex_lock_pi (u32 __user *uaddr,int detect, -unsigned long sec,long nsec,int trylock) +static int do_futex_lock_pi (u32 __user *uaddr,int detect,int trylock, +struct hrtimer_sleeper *to) {struct task_struct *curr = current;struct futex_hash_bucket *hb;u32 uval, newval, curval;struct futex_q q; -struct hrtimer_sleeper timeout, *to =NULL ;int ret, attempt =0 ;if (refill_pi_state_cache ())return -ENOMEM ; -if (sec !=MAX_SCHEDULE_TIMEOUT ) { - to = &timeout; -hrtimer_init (&to->timer,CLOCK_REALTIME ,HRTIMER_ABS ); -hrtimer_init_sleeper (to, current); - to->timer.expires =ktime_set (sec, nsec); - } - q.pi_state =NULL ; retry:down_read (&curr->mm->mmap_sem);
@@ -1304,6 +1296,76 @@ static int futex_lock_pi(u32 __user *uad }/* +* Restart handler +*/ +static long futex_lock_pi_restart (struct restart_block *restart) +{ +struct hrtimer_sleeper timeout, *to =NULL ; +int ret; + + restart->fn = do_no_restart_syscall; + +if (restart->arg2 || restart->arg3) { + to = &timeout; +hrtimer_init (&to->timer,CLOCK_REALTIME ,HRTIMER_ABS ); +hrtimer_init_sleeper (to, current); + to->timer.expires.tv64 = ((u64 )restart->arg1 <<32 ) | + (u64 ) restart->arg0; + } + +pr_debug ("lock_pi restart: %p, %d (%d)\n" , + (u32 __user *)restart->arg0, current->pid); + + ret =do_futex_lock_pi ((u32 __user *)restart->arg0, restart->arg1, +0 , to); + +if (ret != -EINTR ) +return ret; + + restart->fn = futex_lock_pi_restart; + +/* The other values are filled in */ +return -ERESTART_RESTARTBLOCK ; +} + +/* +* Called from the syscall entry below. +*/ +static int futex_lock_pi (u32 __user *uaddr,int detect,unsigned long sec, +long nsec,int trylock) +{ +struct hrtimer_sleeper timeout, *to =NULL ; +struct restart_block *restart; +int ret; + +if (sec !=MAX_SCHEDULE_TIMEOUT ) { + to = &timeout; +hrtimer_init (&to->timer,CLOCK_REALTIME ,HRTIMER_ABS ); +hrtimer_init_sleeper (to, current); + to->timer.expires =ktime_set (sec, nsec); + } + + ret =do_futex_lock_pi (uaddr, detect, trylock, to); + +if (ret != -EINTR ) +return ret; + +pr_debug ("lock_pi interrupted: %p, %d (%d)\n" , uaddr, current->pid); + + restart = ¤t_thread_info ()->restart_block; + restart->fn = futex_lock_pi_restart; + restart->arg0 = (unsigned long ) uaddr; + restart->arg1 = detect; +if (to) { + restart->arg2 = to->timer.expires.tv64 &0xFFFFFFFF ; + restart->arg3 = to->timer.expires.tv64 >>32 ; + }else + restart->arg2 = restart->arg3 =0 ; + +return -ERESTART_RESTARTBLOCK ; +} + +/* * Userspace attempted a TID -> 0 atomic transition, and failed. * This is the in-kernel slowpath: we look up the PI state (if any), * and do the rt-mutex unlock.
--