Re: [PATCH v3 07/13] futex: Throughput-optimized (TP) futexes

From: Waiman Long
Date: Sat Oct 01 2016 - 21:12:54 EST


On 10/01/2016 02:47 AM, Thomas Gleixner wrote:
On Fri, 30 Sep 2016, Waiman Long wrote:
+ WRITE_ONCE(state->owner, current);
+ preempt_disable();
+ for (;;) {
+ ret = futex_trylock(uaddr, vpid,&uval, true);
Did you actually read what I said? You CANNOT access userspace in a preempt
disabled region without disabling pagefaults and handle the resulting
wreckage yourself.

I think I had missed that comment. My bad:-(

I will fix that with the code changes below. I will also double-check your comments again to see if I miss some others.

Cheers,
Longman


---------------------------[ cut here ]-----------------------------------------

diff --git a/kernel/futex.c b/kernel/futex.c
index bc16eca..132a36d 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -3520,6 +3520,13 @@ static int futex_spin_on_owner(u32 __user *uaddr, u32 vpid,
bool on_owner_pi_list = false;

WRITE_ONCE(state->owner, current);
+retry:
+ /*
+ * The preempt_disable() has similar effect as pagefault_disable().
+ * As a result, we will have to disable page fault as well and handle
+ * the case of faulting in the futex word.
+ */
+ pagefault_disable();
preempt_disable();
for (;; loop--) {
ret = futex_trylock(uaddr, vpid, &uval, true);
@@ -3648,6 +3655,14 @@ static int futex_spin_on_owner(u32 __user *uaddr, u32 vpid,
}
out:
preempt_enable();
+ pagefault_enable();
+
+ if (ret == -EFAULT) {
+ ret = fault_in_user_writeable(uaddr);
+ if (!ret)
+ goto retry;
+ }
+
if (owner_task) {
if (on_owner_pi_list)
task_pi_list_del(owner_task, state, false);