[BUG] completely bonkers use of set_need_resched + VM_FAULT_NOPAGE

[Thomas Hellstrom]

On 09/12/2013 10:39 PM, Thomas Gleixner wrote:
> On Thu, 12 Sep 2013, Daniel Vetter wrote:
>
>> On Thu, Sep 12, 2013 at 10:20 PM, Thomas Gleixner <tglx at linutronix.de> wrote:
>>>> I think for ttm drivers it's just execbuf being exploitable. But on
>>>> drm/i915 we've
>>>> had the same issue with the pwrite/pread ioctls, so a simple
>>>> glBufferData(glMap) kind of recursion from gl clients blew the kernel
>>>> to pieces ...
>>> And the only answer you folks came up with is set_need_resched() and
>>> yield()? Oh well....
>> The yield was for a different lifelock, and that one is also fixed by
>> now. The fault handler deadlock was fixed in the usual "drop locks and
>> jump into slowpath" fasion, at least in drm/i915.
> So we can remove that whole yield/set_need_resched() mess completely ?
>
> Thanks,
>
> 	tglx
No.

The while(trylock) is there to address a potential locking inversion 
deadlock. If the trylock fails, the code returns to user-space which 
retries the fault. This code needs to stay until we can come up with 
either a way to drop the mmap_sem and sleep before returning to 
user-space, or a bunch of code is fixed with a different locking order.

The set_need_resched() can (and should according to Peter) go.

/Thomas