[RFC 0/2] Reenable might_sleep() checks for might_fault() when atomic

[Christian Borntraeger]

Am 26.11.2014 um 16:37 schrieb Michael S. Tsirkin:
> On Wed, Nov 26, 2014 at 04:30:32PM +0100, Christian Borntraeger wrote:
>> Am 26.11.2014 um 16:17 schrieb Michael S. Tsirkin:
>>> On Wed, Nov 26, 2014 at 11:05:04AM +0100, David Hildenbrand wrote:
>>>>> What's the path you are trying to debug?
>>>>
>>>> Well, we had a problem where we held a spin_lock and called
>>>> copy_(from|to)_user(). We experienced very random deadlocks that took some guy
>>>> almost a week to debug. The simple might_sleep() check would have showed this
>>>> error immediately.
>>>
>>
>>> This must have been a very old kernel.
>>> A modern kernel will return an error from copy_to_user.
>>
>> I disagree. copy_to_user will not return while holding a spinlock, because it does not know! How should it?
>> See: spin_lock will call preempt_disable, but thats a no-op for a non-preempt kernel. So the mere fact that we hold a spin_lock is not known by any user access function. (or others). No?
>>
>> Christian
>>
>>
> 
> Well might_sleep() merely checks preempt count and irqs_disabled too.
> If you want debugging things to trigger, you need to enable
> a bunch of config options.  That's not new.

You miss the point of the whole thread: The problem is that even with debug options enabled, holding a spinlock would not trigger a bug on copy_to_user. So the problem is not the good path, the problem is that a debugging aid for detecting a broken case was lost. Even with all kernel debugging enabled.

That is because CONFIG_DEBUG_ATOMIC_SLEEP selects PREEMPT_COUNT. That means: spin_lock will then be considered as in_atomic and no message comes. Without CONFIG_DEBUG_ATOMIC_SLEEP spin_lock will not touch the preempt_count but we also dont see a message because might_fault is now a nop

I understand that you dont like Davids changes due to other side effects that you have mentioned. So lets focus on how we can fix the debug option. Ok?

Christian