Re: [Xen-devel] [PATCH 3/4] x86/hyperv: skeleton for L0 assisted TLB flush

[Wei Liu <wl@xxxxxxx>]

On Wed, Feb 12, 2020 at 06:09:24PM +0100, Roger Pau Monné wrote:
> On Wed, Feb 12, 2020 at 04:09:17PM +0000, Wei Liu wrote:
> > Implement a basic hook for L0 assisted TLB flush. The hook needs to
> > check if prerequisites are met. If they are not met, it returns an error
> > number to fall back to native flushes.
> > 
> > Introduce a new variable to indicate if hypercall page is ready.
> > 
> > Signed-off-by: Wei Liu <liuwe@xxxxxxxxxxxxx>
> > ---
> >  xen/arch/x86/guest/hyperv/Makefile  |  1 +
> >  xen/arch/x86/guest/hyperv/hyperv.c  | 17 ++++++++++++
> >  xen/arch/x86/guest/hyperv/private.h |  4 +++
> >  xen/arch/x86/guest/hyperv/tlb.c     | 41 +++++++++++++++++++++++++++++
> >  4 files changed, 63 insertions(+)
> >  create mode 100644 xen/arch/x86/guest/hyperv/tlb.c
> > 
> > diff --git a/xen/arch/x86/guest/hyperv/Makefile 
> > b/xen/arch/x86/guest/hyperv/Makefile
> > index 68170109a9..18902c33e9 100644
> > --- a/xen/arch/x86/guest/hyperv/Makefile
> > +++ b/xen/arch/x86/guest/hyperv/Makefile
> > @@ -1 +1,2 @@
> >  obj-y += hyperv.o
> > +obj-y += tlb.o
> > diff --git a/xen/arch/x86/guest/hyperv/hyperv.c 
> > b/xen/arch/x86/guest/hyperv/hyperv.c
> > index b7044f7193..1cdc88e27c 100644
> > --- a/xen/arch/x86/guest/hyperv/hyperv.c
> > +++ b/xen/arch/x86/guest/hyperv/hyperv.c
> > @@ -33,6 +33,8 @@ DEFINE_PER_CPU_READ_MOSTLY(void *, hv_input_page);
> >  DEFINE_PER_CPU_READ_MOSTLY(void *, hv_vp_assist);
> >  DEFINE_PER_CPU_READ_MOSTLY(uint32_t, hv_vp_index);
> >  
> > +static bool __read_mostly hv_hcall_page_ready;
> > +
> >  static uint64_t generate_guest_id(void)
> >  {
> >      union hv_guest_os_id id = {};
> > @@ -119,6 +121,8 @@ static void __init setup_hypercall_page(void)
> >      BUG_ON(!hypercall_msr.enable);
> >  
> >      set_fixmap_x(FIX_X_HYPERV_HCALL, mfn << PAGE_SHIFT);
> > +
> > +    hv_hcall_page_ready = true;
> 
> I guess filling the hypercall page in the probe function like it's
> done for Xen is too early for HyperV, and hence you need this
> safeguard?

Yes that's too early.

Keep in mind that Hyper-V's hypercall page is an overlay page which is
not backed by real memory from Xen's PoV. Xen can't fill in a stub
there. Xen needs to wait until other infrastructures to go live before
setting up a hypercall page, while in the mean time, it will / may need
to flush TLB.

> 
> TBH, maybe it would be best (and safer) to prevent using any hooks
> until setup has been called, and hence this check could be pulled into
> the generic hook?

"Prevent" is too vague a term here. We can't stop code from executing in
parallel. In some situation there is no way to fail gracefully.

There is only this hook at the moment that requires such special care
afaict, and this is largely due to Hyper-V's idiosyncrasy. Others are
called only in setup / teardown path which can easily be reasoned about.

Wei.

> 
> Thanks, Roger.

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