Re: [PATCH v5] mm/gup: disallow GUP writing to file-backed mappings by default

[John Hubbard]

On 4/28/23 10:29, David Hildenbrand wrote:
...
>> Is that an issue with interrupts disabled though? Will block page tables being
>> removed and as Kirill says (sorry I maybe misinterpreted you) we should be ok.
> 
> Let's rule out page table freeing. If our VMA only spans a single page 
> and falls into the same PMD as another VMA, an munmap() would not even 
> free a single page table.
> 
> However, if unmapping a page (flushing the TLB) would imply an IPI as 
> Kirill said, we'd be fine. I recall that that's not the case for all 
> architectures, but I might be just wrong.
> 

Some architectures use IPIs, and others use an RCU-like approach instead.

thanks,
-- 
John Hubbard
NVIDIA