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

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Thu, Nov 27, 2014 at 09:03:01AM +0100, David Hildenbrand wrote:
> > Code like
> > 	spin_lock(&lock);
> > 	if (copy_to_user(...))
> > 		rc = ...
> > 	spin_unlock(&lock);
> > really *should* generate warnings like it did before.
> > 
> > And *only* code like
> > 	spin_lock(&lock);
> 
> Is only code like this valid or also with the spin_lock() dropped?
> (e.g. the access in patch1 if I remember correctly)
> 
> So should page_fault_disable() increment the pagefault counter and the preempt
> counter or only the first one?

Given that a sequence like

	page_fault_disable();
 	if (copy_to_user(...))
 		rc = ...
 	page_fault_enable();

is correct code right now I think page_fault_disable() should increase both.
No need for surprising semantic changes.

> So we would have pagefault code rely on:
> 
> in_disabled_pagefault() ( pagefault_disabled() ... whatever ) instead of
> in_atomic().

No, let's be more defensive: the page fault handler should do nothing if
in_atomic() just like now. But it could have a quick check and emit a one
time warning if page faults aren't disabled in addition.
That might help debugging but keeps the system more likely alive.

might_fault() however should call might_sleep() if page faults aren't
disabled, but that's what you proposed anyway I think.