Re: [RFC PATCH 3/3] KVM: MMU: Optimize guest page table walk

[Takuya Yoshikawa <takuya.yoshikawa@gmail.com>]

Andi Kleen <andi@firstfloor.org> wrote:

> > Only I can guess for that reason is the reduction of some function calls
> > by inlining some functions.
> 
> Yes once at a time cfu was inline too and just checked for the right
> sizes and the used g*u, but it got lost in the "icache over everything else" mania which is unfortunately en vogue for quite some time in kernel land (aka 
> measuring patches only based on their impact on the .text
> size, not the actual performance)
> 
> But you might getter better gains by fixing this general
> c*u() regression.
> 

What I mentioned was about not only cfu but 3 functions.

Originally, we were doing the following function calls:

walk_addr_generic()              ---1
  kvm_read_guest_page_mmu()      ---2
    kvm_read_guest_page()        ---3
      copy_from_user()           ---4

And now, with my patch already applied, we are not using
generic kvm_read_guest_page_mmu() and some address
calculations are done in walk_addr_generic():

walk_addr_generic()              ---1'
  get_user()                     ---2'

The length is passed in from walk_addr_generic().

Do you think the following case would not differ so much
from (1' 2') ?

walk_addr_generic()              ---1''
  copy_from_user()               ---2''

This can satisfy your "assuming you pass in a constant length"
and almost same as get_user() ?

Takuya