On Fri, 8 Dec 2017, Julien Grall wrote:
> On 8 Dec 2017 22:26, "Stefano Stabellini" <sstabellini@kernel.org> wrote:
>       On Fri, 8 Dec 2017, Julien Grall wrote:
>       > On 06/12/17 12:27, Julien Grall wrote:
>       > > On 12/06/2017 01:26 AM, Stefano Stabellini wrote:
>       > > > On Thu, 23 Nov 2017, Julien Grall wrote:
>       > > > > Hi Andrew,
>       > > > >
>       > > > > On 23/11/17 18:49, Andrew Cooper wrote:
>       > > > > > On 23/11/17 18:32, Julien Grall wrote:
>       > > > > > > This new function will be used in a follow-up patch to copy data to
>       > > > > > > the
>       > > > > > > guest
>       > > > > > > using the IPA (aka guest physical address) and then clean the cache.
>       > > > > > >
>       > > > > > > Signed-off-by: Julien Grall <julien.grall@linaro.org>
>       > > > > > > ---
>       > > > > > >    xen/arch/arm/guestcopy.c           | 10 ++++++++++
>       > > > > > >    xen/include/asm-arm/guest_access.h |  6 ++++++
>       > > > > > >    2 files changed, 16 insertions(+)
>       > > > > > >
>       > > > > > > diff --git a/xen/arch/arm/guestcopy.c b/xen/arch/arm/guestcopy.c
>       > > > > > > index be53bee559..7958663970 100644
>       > > > > > > --- a/xen/arch/arm/guestcopy.c
>       > > > > > > +++ b/xen/arch/arm/guestcopy.c
>       > > > > > > @@ -110,6 +110,16 @@ unsigned long raw_copy_from_guest(void *to,
>       > > > > > > const
>       > > > > > > void __user *from, unsigned le
>       > > > > > >                          COPY_from_guest | COPY_linear);
>       > > > > > >    }
>       > > > > > >    +unsigned long copy_to_guest_phys_flush_dcache(struct domain *d,
>       > > > > > > +                                              paddr_t gpa,
>       > > > > > > +                                              void *buf,
>       > > > > > > +                                              unsigned int len)
>       > > > > > > +{
>       > > > > > > +    /* P2M is shared between all vCPUs, so the vCPU used does not
>       > > > > > > matter.
>       > > > > > > */
>       > > > > >
>       > > > > > Be very careful with this line of thinking.  It is only works after
>       > > > > > DOMCTL_max_vcpus has succeeded, and before that point, it is a latent
>       > > > > > NULL pointer dereference.
>       > > > >
>       > > > > I really don't expect that function been used before DOMCT_max_vcpus is
>       > > > > set.
>       > > > > It is only used for hardware emulation or Xen loading image into the
>       > > > > hardware
>       > > > > domain memory. I could add a check d->vcpus to be safe.
>       > > > >
>       > > > > >
>       > > > > > Also, what about vcpus configured with alternative views?
>       > > > >
>       > > > > It is not important because the underlying call is get_page_from_gfn
>       > > > > that does
>       > > > > not care about the alternative view (that function take a domain in
>       > > > > parameter). I can update the comment.
>       > > > Since this is a new function, would it make sense to take a struct
>       > > > vcpu* as parameter, instead of a struct domain* ?
>       > >
>       > > Well, I suggested this patch this way because likely everyone will use with
>       > > d->vcpus[0]. And then you would have to wonder why d->vcpus[0] and not
>       > > d->vcpus[1]...
>       >
>       > Thinking a bit more to this, it might be better/safer to pass either a domain
>       > or a vCPU to copy_guest. I can see 2 solutions:
>       >       1# Introduce a union that use the same parameter:
>       >               union
>       >               {
>       >                       struct
>       >                       {
>       >                               struct domain *d;
>       >                       } ipa;
>       >                       struct
>       >                       {
>       >                               struct vcpu *v;
>       >                       } gva;
>       >               }
>       >         The structure here would be to ensure that it is clear that only
>       > domain (resp. vcpu) should be used with ipa (resp. gva).
>       >
>       >       2# Have 2 parameters, vcpu and domain.
>       >
>       > Any opinions?
> 
> I think that would be clearer. You could also add a paddr_t/vaddr_t
> correspondingly inside the union maybe.
> 
> 
> Well, you will have nameclash happening I think.
> 
> 
> And vaddr_t and paddr_t are confusing because you don't know which address you speak about (guest vs hypervisor). At least ipa/gpa, gva are known naming.

That's not what I meant. ipa and gva are good names.

I was suggesting to put an additional address field inside the union to
avoid the issue with paddr_t and vaddr_t discussed elsewhere
(alpine.DEB.2.10.1712081313070.8052@sstabellini-ThinkPad-X260).

I am happy either way, it was just a suggestion.