Re: [Qemu-devel] [PATCH] memory: Don't update all memory region when ioeventfd changed

[Gonglei (Arei)]

Hi,

> -----Original Message-----
> From: Paolo Bonzini [mailto:address@hidden
> Sent: Thursday, May 08, 2014 8:49 PM
> To: Gonglei (Arei); address@hidden
> Cc: address@hidden; address@hidden; address@hidden;
> address@hidden; Huangweidong (C); Herongguang (Stephen)
> Subject: Re: [PATCH] memory: Don't update all memory region when ioeventfd
> changed
> 
> Il 08/05/2014 05:47, address@hidden ha scritto:
> > From: Gonglei <address@hidden>
> >
> > memory mappings don't rely on ioeventfds, there is no need
> > to destroy and rebuild them when manipulating ioeventfds,
> > otherwise it scarifies performance.
> >
> > according to testing result, each ioeventfd deleing needs
> > about 5ms, within which memory mapping rebuilding needs
> > about 4ms. With many Nics and vmchannel in a VM doing migrating,
> > there can be many ioeventfds deleting which increasing
> > downtime remarkably.
> >
> > Signed-off-by: Gonglei <address@hidden>
> > Signed-off-by: Herongguang <address@hidden>
> > ---
> >  memory.c | 33 +++++++++++++++++++++++----------
> >  1 file changed, 23 insertions(+), 10 deletions(-)
> >
> > diff --git a/memory.c b/memory.c
> > index 3f1df23..2c783f6 100644
> > --- a/memory.c
> > +++ b/memory.c
> > @@ -28,6 +28,7 @@
> >
> >  static unsigned memory_region_transaction_depth;
> >  static bool memory_region_update_pending;
> > +static bool ioeventfd_update_pending;
> >  static bool global_dirty_log = false;
> >
> >  /* flat_view_mutex is taken around reading as->current_map; the critical
> > @@ -786,22 +787,34 @@ void memory_region_transaction_begin(void)
> >      ++memory_region_transaction_depth;
> >  }
> >
> > +static void memory_region_clear_pending(void)
> > +{
> > +    memory_region_update_pending = false;
> > +    ioeventfd_update_pending = false;
> > +}
> > +
> >  void memory_region_transaction_commit(void)
> >  {
> >      AddressSpace *as;
> >
> >      assert(memory_region_transaction_depth);
> >      --memory_region_transaction_depth;
> > -    if (!memory_region_transaction_depth &&
> memory_region_update_pending) {
> > -        memory_region_update_pending = false;
> > -        MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
> > +    if (!memory_region_transaction_depth) {
> > +        if (memory_region_update_pending) {
> > +            MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
> >
> > -        QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
> > -            address_space_update_topology(as);
> > -        }
> > +            QTAILQ_FOREACH(as, &address_spaces,
> address_spaces_link) {
> > +                address_space_update_topology(as);
> > +            }
> >
> > -        MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
> > -    }
> > +            MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
> > +        } else if (ioeventfd_update_pending) {
> > +            QTAILQ_FOREACH(as, &address_spaces,
> address_spaces_link) {
> > +                address_space_update_ioeventfds(as);
> > +            }
> > +        }
> > +        memory_region_clear_pending();
> > +   }
> >  }
> >
> >  static void memory_region_destructor_none(MemoryRegion *mr)
> > @@ -1373,7 +1386,7 @@ void memory_region_add_eventfd(MemoryRegion
> *mr,
> >      memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
> >              sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
> >      mr->ioeventfds[i] = mrfd;
> > -    memory_region_update_pending |= mr->enabled;
> > +    ioeventfd_update_pending |= mr->enabled;
> >      memory_region_transaction_commit();
> >  }
> >
> > @@ -1406,7 +1419,7 @@ void memory_region_del_eventfd(MemoryRegion
> *mr,
> >      --mr->ioeventfd_nb;
> >      mr->ioeventfds = g_realloc(mr->ioeventfds,
> >
> sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
> > -    memory_region_update_pending |= mr->enabled;
> > +    ioeventfd_update_pending |= mr->enabled;
> >      memory_region_transaction_commit();
> >  }
> >
> >
> 
> Applied for 2.1, thanks -- but IIUC Chris's patch (if it worked) would
> improve performance even more.  Is this correct?
> 
Yep. In theory, Chri's patch make all the eventfds do a single 
commit at last. It will reduce the time of update ioeventfds.


Best regards,
-Gonglei