Re: can device drivers return non-ram via vm_ops->nopage?

[Andrea Arcangeli <andrea@suse.de>]

On Sat, Mar 20, 2004 at 09:39:51AM -0800, Linus Torvalds wrote:
> 
> 
> On Sat, 20 Mar 2004, Andrea Arcangeli wrote:
> >
> > The only bugreport I've got so far for the latest anon_vma code is from
> > Jens, and it's a device driver bug in my opinion, but I'd like to have a
> > definitive confirmation from you about the ->nopage API.
> 
> I'd say that this is definitely a driver bug.
> 
> If a driver wants to map non-RAM pages, that's perfectly ok, but it MUST 
> NOT happen through "nopage()". The driver should map them with 
> "remap_page_range()", and thus never take a page fault for such pages at 
> all.
> 
> There is no reason to ever lazily map non-RAM pages - clearly they aren't 
> using any "real memory", so there is no reason to not fill the page tables 
> at mmap() time.
> 
> In other words, the driver is horribly broken.

thanks for the clarification.

At the moment I'm not sure anymore if this was non-ram or a
VM_FAULT_SIGBUS because I noticed I was doing BUG_ON(!pfn_valid)
_before_ checking new_page == VM_FAULT_SIGBUS. Though my theory about
do_no_page working fine with non-ram page_t with >=128m machines up to
2.6.4 still holds, and it's not obvious that Jens triggered a SIGBUS
either.