Re: [PATCH RFC 00/15] decouple pagefault_disable() from preempt_disable()

[Christian Borntraeger]

Am 07.05.2015 um 11:48 schrieb Ingo Molnar:
> 
> * Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
>> On Wed,  6 May 2015 19:50:24 +0200 David Hildenbrand <dahi@xxxxxxxxxxxxxxxxxx> wrote:
>>
>>> As Peter asked me to also do the decoupling in one shot, this is
>>> the new series.
>>>
>>> I recently discovered that might_fault() doesn't call might_sleep()
>>> anymore. Therefore bugs like:
>>>
>>>   spin_lock(&lock);
>>>   rc = copy_to_user(...);
>>>   spin_unlock(&lock);
>>>
>>> would not be detected with CONFIG_DEBUG_ATOMIC_SLEEP. The code was
>>> changed to disable false positives for code like:
>>>
>>>   pagefault_disable();
>>>   rc = copy_to_user(...);
>>>   pagefault_enable();
>>>
>>> Whereby the caller wants do deal with failures.
>>
>> hm, that was a significant screwup.  I wonder how many bugs we
>> subsequently added.
> 
> So I'm wondering what the motivation was to allow things like:
> 
>    pagefault_disable();
>    rc = copy_to_user(...);
>    pagefault_enable();
> 
> and to declare it a false positive?
> 
> AFAICS most uses are indeed atomic:
> 
>         pagefault_disable();
>         ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
>         pagefault_enable();
> 
> so why not make it explicitly atomic again?

Hmm, I am probably misreading that, but it sound as you suggest to go back
to Davids first proposal
https://lkml.org/lkml/2014/11/25/436
which makes might_fault to also contain might_sleep. Correct?

Christian

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/