Re: [PATCH 0/2] exclude hyperv synic sections from vhost

[Michael S. Tsirkin]

On Thu, Jan 09, 2020 at 12:02:16PM +0000, Dr. David Alan Gilbert wrote:
> * Jason Wang (address@hidden) wrote:
> > 
> > On 2020/1/8 下午9:53, Dr. David Alan Gilbert (git) wrote:
> > > From: "Dr. David Alan Gilbert" <address@hidden>
> > > 
> > > Hyperv's synic (that we emulate) is a feature that allows the guest
> > > to place some magic (4k) pages of RAM anywhere it likes in GPA.
> > > This confuses vhost's RAM section merging when these pages
> > > land over the top of hugepages.
> > 
> > 
> > Hi David:
> > 
> > A silly question, is this because the alignment when adding sections? If
> > yes, what's the reason for doing alignment which is not a must for vhost
> > memory table.
> 
> Page alignment is a bit odd with vhost-user - it ends up having to mmap
> each of the sections itself; and still has to map them as hugepages
> to be able to mmap - in the old world you could sometimes have
> the daemon mmaping the same chunk of memory twice into the vhost-user
> process; without the aggregation  you'd get a hugepage mapping for the
> 0-2MB chunk for the 0-512K mapping, and then maybe another 0-2MB chunk
> for some of the other parts over 512K.
> With postcopy we can't have the multiple mappings of the same part of
> guest memory; we need to have one mapping for userfault.
> 
> Also, given the 16 separate synic regions, you'd probably end up having
> a lot of wasted vhost-sections.
> 
> Dave

So I'd worry that this is more an abuse of an interface.
E.g. this means it's skipped from dumps, which is not nice.


And for vhost I worry these patches will break pass-through of
PCI attached memory.


> 
> 
> 
> > Thanks
> > 
> > 
> > > 
> > > Since they're not normal RAM, and they shouldn't have vhost DMAing
> > > into them, exclude them from the vhost set.
> > > 
> > > I do that by marking them as device-ram and then excluding device-ram
> > > from vhost.
> > > 
> > > bz: https://bugzilla.redhat.com/show_bug.cgi?id=1779041
> > > 
> > > Dr. David Alan Gilbert (2):
> > >    vhost: Don't pass ram device sections
> > >    hyperv/synic: Allocate as ram_device
> > > 
> > >   hw/hyperv/hyperv.c | 14 ++++++++------
> > >   hw/virtio/vhost.c  |  1 +
> > >   2 files changed, 9 insertions(+), 6 deletions(-)
> > > 
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK