Re: [PATCH resend mm,net-next 2/3] mm: Add vm_insert_pages().

[Arjun Roy <arjunroy@google.com>]

On Thu, Feb 13, 2020 at 1:54 PM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Mon, Jan 27, 2020 at 06:59:57PM -0800, Arjun Roy wrote:
> >  int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
> > +int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
> > +                     struct page **pages, unsigned long *num);
> >  int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
> >                       unsigned long pfn);
>
> Sorry I didn't notice these patches earlier.  I'm not thrilled about
> the addition of a new vm_insert_* operation; we're moving towards a
> vmf_insert_* API.  There are almost no users left of vm_insert_page
> (10, at a quick count).  Once they're all gone, we can switch the
> underlying primitives over to a vm_fault_t return type and get rid of the
> errno-to-vm-fault translation step that currently goes on.
>
> So ... is this called in the fault path?  Do you have a struct vm_fault
> around?  Can you handle a vm_fault_t return value instead of an errno?

This is not a page fault, really. This customer of vm_insert_page() is
the TCP receive zerocopy code, which is remapping pages from the NIC
into the userspace process (in lieu of sys_recvmsg()'s copy). See:
tcp_zerocopy_receive() in net/ipv4/tcp.c .

I took a peek at vmf_insert_page(). I think that hides the presence of
EBUSY, which would be a necessary signal for us. If that was exposed I
think vm_fault_t could be fine, *but* I shall defer to Eric for
actually deciding on it.

-Arjun