Re: [Xen-devel] [PATCH v12 2/3] x86/mem_sharing: reset a fork

[Tamas K Lengyel <tamas@tklengyel.com>]

On Wed, Mar 25, 2020 at 9:52 AM Roger Pau Monné <roger.pau@citrix.com> wrote:
>
> On Mon, Mar 23, 2020 at 10:04:36AM -0700, Tamas K Lengyel wrote:
> > Implement hypercall that allows a fork to shed all memory that got allocated
> > for it during its execution and re-load its vCPU context from the parent VM.
> > This allows the forked VM to reset into the same state the parent VM is in a
> > faster way then creating a new fork would be. Measurements show about a 2x
> > speedup during normal fuzzing operations. Performance may vary depending how
> > much memory got allocated for the forked VM. If it has been completely
> > deduplicated from the parent VM then creating a new fork would likely be more
> > performant.
> >
> > Signed-off-by: Tamas K Lengyel <tamas.lengyel@intel.com>
>
> LGTM:
>
> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
>
> One minor nit below.
>
> > ---
> > v12: remove continuation & add comment back
> >      address style issues pointed out by Jan
> > ---
> >  xen/arch/x86/mm/mem_sharing.c | 77 +++++++++++++++++++++++++++++++++++
> >  xen/include/public/memory.h   |  1 +
> >  2 files changed, 78 insertions(+)
> >
> > diff --git a/xen/arch/x86/mm/mem_sharing.c b/xen/arch/x86/mm/mem_sharing.c
> > index 23deeddff2..930a5f58ef 100644
> > --- a/xen/arch/x86/mm/mem_sharing.c
> > +++ b/xen/arch/x86/mm/mem_sharing.c
> > @@ -1775,6 +1775,60 @@ static int fork(struct domain *cd, struct domain *d)
> >      return rc;
> >  }
> >
> > +/*
> > + * The fork reset operation is intended to be used on short-lived forks only.
> > + * There is no hypercall continuation operation implemented for this reason.
> > + * For forks that obtain a larger memory footprint it is likely going to be
> > + * more performant to create a new fork instead of resetting an existing one.
> > + *
> > + * TODO: In case this hypercall would become useful on forks with larger memory
> > + * footprints the hypercall continuation should be implemented (or if this
> > + * feature needs to be become "stable").
> > + */
> > +static int mem_sharing_fork_reset(struct domain *d, struct domain *pd)
> > +{
> > +    int rc;
> > +    struct p2m_domain *p2m = p2m_get_hostp2m(d);
> > +    struct page_info *page, *tmp;
> > +
> > +    spin_lock(&d->page_alloc_lock);
> > +    domain_pause(d);
> > +
> > +    page_list_for_each_safe(page, tmp, &d->page_list)
> > +    {
> > +        p2m_type_t p2mt;
> > +        p2m_access_t p2ma;
> > +        mfn_t mfn = page_to_mfn(page);
> > +        gfn_t gfn = mfn_to_gfn(d, mfn);
> > +
> > +        mfn = __get_gfn_type_access(p2m, gfn_x(gfn), &p2mt, &p2ma,
> > +                                    0, NULL, false);
> > +
> > +        /* only reset pages that are sharable */
> > +        if ( !p2m_is_sharable(p2mt) )
> > +            continue;
> > +
> > +        /* take an extra reference or just skip if can't for whatever reason */
> > +        if ( !get_page(page, d) )
> > +            continue;
>
> You can join both conditions above into a single one, if both just
> need to perform a continue.

We could but I think it's easier to read it this way. So I prefer to
keep it separate.

Thanks for the review!
Tamas