Re: [Xen-devel] [PATCH v3 3/5] x86/hyperv: provide percpu hypercall input page

[Wei Liu <wl@xen.org>]

On Wed, Jan 08, 2020 at 11:55:03AM +0100, Jan Beulich wrote:
> On 07.01.2020 18:27, Wei Liu wrote:
> > On Tue, Jan 07, 2020 at 06:08:19PM +0100, Jan Beulich wrote:
> >> On 07.01.2020 17:33, Wei Liu wrote:
> >>> On Mon, Jan 06, 2020 at 11:27:18AM +0100, Jan Beulich wrote:
> >>>> On 05.01.2020 17:47, Wei Liu wrote:
> >>>>> Hyper-V's input / output argument must be 8 bytes aligned an not cross
> >>>>> page boundary. The easiest way to satisfy those requirements is to use
> >>>>> percpu page.
> >>>>
> >>>> I'm not sure "easiest" is really true here. Others could consider adding
> >>>> __aligned() attributes as easy or even easier (by being even more
> >>>> transparent to use sites). Could we settle on "One way ..."?
> >>>
> >>> Do you mean something like
> >>>
> >>>    struct foo __aligned(8);
> >>
> >> If this is in a header and ...
> >>
> >>>    hv_do_hypercall(OP, virt_to_maddr(&foo), ...);
> >>
> >> ... this in actual code, then yes.
> >>
> >>> ?
> >>>
> >>> I don't think this is transparent to user sites. Plus, foo is on stack
> >>> which is 1) difficult to get its maddr,
> >>
> >> It being on the stack may indeed complicate getting its machine address
> >> (if not now, then down the road) - valid point.
> >>
> >>> 2) may cross page boundary.
> >>
> >> The __aligned() of course needs to be large enough to avoid this
> >> happening.
> > 
> > For this alignment to be large enough, it will need to be of PAGE_SIZE,
> > right? Wouldn't that blow up Xen's stack easily?  Given we only have two
> > pages for that.
> 
> Why PAGE_SIZE? For example, a 24-byte structure won't cross a page
> boundary if aligned to 32 bytes.
> 

You're right.

I said PAGE_SIZE because I was too lazy to calculate the size of every
structures. That's tedious and error prone.

> > In light of these restrictions, the approach I take in the original
> > patch should be okay.
> > 
> > I'm fine with changing the wording to "One way ..." -- if that's the
> > only objection you have after this mail.
> 
> Well, the goal was to (a) check whether alternatives have been considered
> (and if not, to consider them) and then (b) if we stick to your approach,
> slightly change the wording as suggested.

I think the determining factor here is to the difficulty of getting
maddr of a stack variable. I will stick with this approach and change
the wording.

Wei.

> 
> Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel