Re: [PATCH v4 1/5] KVM: MMU: fix release noslot pfn

[Marcelo Tosatti <mtosatti@redhat.com>]

On Fri, Oct 12, 2012 at 05:49:30PM +0800, Xiao Guangrong wrote:
> On 10/11/2012 10:31 PM, Marcelo Tosatti wrote:
> > On Thu, Oct 11, 2012 at 09:06:12PM +0800, Xiao Guangrong wrote:
> >> On 10/10/2012 11:11 PM, Marcelo Tosatti wrote:
> >>
> >>>
> >>> Why does is_error_pfn() return true for mmio spte? Its not an "error",
> >>> after all. 
> >>>
> >>> Please kill is_invalid_pfn and use
> >>>
> >>> -> is_error_pfn for checking for errors (mmio spte is not an error pfn,
> >>> its a special pfn)
> >>>
> >>> -> add explicit is_noslot_pfn checks where necessary in the code
> >>> (say to avoid interpreting a noslot_pfn's pfn "address" bits).
> >>>
> >>> (should have noticed this earlier, sorry).
> >>
> >> Never mind, your comments are always appreciated! ;)
> >>
> >> Marcelo, is it good to you?
> >> (will post it after your check and full test)
> > 
> > Yes, this works (please check the validity of each case in addition to
> > testing, haven't done it).
> > 
> > Also add a oneline comment on top of each
> > is_error_pfn,is_noslot_pfn,is_error_noslot_pfn
> > 
> > /* is_noslot_pfn: userspace translation valid but no memory slot */
> > /* is_error_pfn: ... */
> > 
> > etc.
> > 
> 
> Marcelo, i think this fix should be backport and your idea can be a
> separate patchset. Yes?

The current invalid/is_error/noslot_pfn separation is confusing, leading 
to one immediate bug and IMO more future bugs.

The proposed patch you sent is quite small, why is it troublesome to
backport? (and i am just asking one line of comment, summing to 3 total
of lines of comments).

Can't see the advantage of a special easily backportable fix?