On Thu, Aug 22, 2019 at 08:00:15PM +0300, Jarkko Sakkinen wrote: > On Thu, 2019-08-22 at 07:50 -0700, Sean Christopherson wrote: > > On Thu, Aug 22, 2019 at 05:37:18PM +0300, Jarkko Sakkinen wrote: > > > On Mon, 2019-08-12 at 18:12 -0700, Sean Christopherson wrote: > > > > Invoke EADD with the userspace source address instead of first copying > > > > the data to a kernel page to avoid the overhead of alloc_page() and > > > > copy_from_user(). > > > > > > > > Suggested-by: Andy Lutomirski <luto@xxxxxxxxxx> > > > > Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx> > > > > > > NAK because takes away TCS validation and the commit message > > > does not give any reasoning for doing that. > > > > Doh, I have a thorough explanation, but apparently it never made it from > > my head to the changelog. I'll send v2 as a standalone patch. > > Yeah, w/o explanation I won't just take away functionality :-) I came to realize that also from security perspective it might be helpful to EADD, not from a copy of the source, but from the actual source. So yes, I'm for not supporting copy approach at all. I think this viewpoint is important to note in addition to the performance perspective. /Jarkko
