[Intel-gfx] [PATCH] mm: Report attempts to overwrite PTE from remap_pfn_range()

[Kirill A. Shutemov]

Chris Wilson wrote:
> On Thu, Jun 19, 2014 at 03:57:46PM +0300, Kirill A. Shutemov wrote:
> > One possible option is to create a variant of remap_pfn_range() which will
> > return how many PTEs it was able to setup, before hitting the !pte_none().
> > Caller will decide what to do with partially filled range.
> 
> Looked at just returning the address remap_pfn_range() got up to, which is
> easy enough, but I think given that remap_pfn_range() will clean up
> correctly after a failed remap, any EBUSY from partway through would be
> a pathological driver error. 

I would prefer keep remap_pfn_range() interface intact with BUG_ON() on
unexpected !pte_none() and introduce new function with more flexible
behaviour (sharing underlying infrastructure).
This way we can avoid changing every remap_pfn_range() caller.

-- 
 Kirill A. Shutemov