Re: kmap_atomic and preemption

[Peter Zijlstra]

On Wed, May 04, 2016 at 04:07:40PM +0530, Vineet Gupta wrote:
> Hi,
> 
> I was staring at some recent ARC highmem crashes and see that kmap_atomic()
> disables preemption even when page is in lowmem and call returns right away.
> This seems to be true for other arches as well.
> 
> arch/arc/mm/highmem.c:
> 
> void *kmap_atomic(struct page *page)
> {
> 	int idx, cpu_idx;
> 	unsigned long vaddr;
> 
> 	preempt_disable();
> 	pagefault_disable();
> 	if (!PageHighMem(page))
> 		return page_address(page);
> 
>         /* do the highmem foo ... */
> ..
> }
> 
> I would really like to implement a inline fastpath for !PageHighMem(page) case and
> do the highmem foo out-of-line.
> 
> Is preemption disabling a requirement of kmap_atomic() callers independent of
> where page is or is it only needed when page is in highmem and can trigger page
> faults or TLB Misses between kmap_atomic() and kunmap_atomic and wants protection
> against reschedules etc.

Traditionally kmap_atomic() disables preemption; and the reason is that
the returned pointer must stay valid. This had a side effect in that it
also disabled pagefaults.

We've since de-coupled the pagefault from the preemption thing, so you
could disable pagefaults while leaving preemption enabled.

Now, I've also done preemptible kmap_atomic() on -rt; which appears to
work, suggesting nothing relies on it disabling preemption (on -rt).

So sure; you can try and leave preemption enabled for lowmem pages, see
what comes apart -- if anything. It gives weird semantics for
kmap_atomic() though, and I'm not sure the cost of doing that
preempt_disable/preempt_enable() is worth the pain.

If you want a fast-slow path splt, you can easily do something like:


static inline void *kmap_atomic(struct page *page)
{
	preempt_disable();
	pagefault_disable();
	if (!PageHighMem(page))
		return page_address(page);

	return __kmap_atomic(page);
}