Re: [virtio-dev] [PATCH v8 2/2] virtio-fs: add DAX window

[Halil Pasic <pasic@linux.ibm.com>]

On Tue, 10 Sep 2019 14:09:20 +0100
"Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:

> * Halil Pasic (pasic@linux.ibm.com) wrote:
> > On Thu, 29 Aug 2019 14:52:06 +0100
> > Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > 
> > > Describe how shared memory region ID 0 is the DAX window and how
> > > FUSE_SETUPMAPPING maps file ranges into the window.
> > > 
> > > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > ---
> > > The FUSE_SETUPMAPPING message is part of the virtio-fs Linux patches:
> > > https://gitlab.com/virtio-fs/linux/blob/virtio-fs/include/uapi/linux/fuse.h
> > > 
> > > v8:
> > >  * Make language about using both FUSE_READ/FUSE_WRITE and the DAX
> > >    Window clearer [Cornelia]
> > > v7:
> > >  * Clarify that the DAX Window is optional and can be used together with
> > >    FUSE_READ/FUSE_WRITE requests [Cornelia]
> > > v6:
> > >  * Document timing side-channel attacks [Michael]
> > > ---
> > >  virtio-fs.tex | 66 +++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 66 insertions(+)
> > > 
> > > diff --git a/virtio-fs.tex b/virtio-fs.tex
> > > index 1ae17f8..158d066 100644
> > > --- a/virtio-fs.tex
> > > +++ b/virtio-fs.tex
> > > @@ -179,6 +179,62 @@ \subsubsection{Device Operation: High Priority Queue}\label{sec:Device Types / F
> > >  
> > >  The driver MUST anticipate that request queues are processed concurrently with the hiprio queue.
> > >  
> > > +\subsubsection{Device Operation: DAX Window}\label{sec:Device Types / File System Device / Device Operation / Device Operation: DAX Window}
> > > +
> > > +FUSE\_READ and FUSE\_WRITE requests transfer file contents between the
> > > +driver-provided buffer and the device.  In cases where data transfer is
> > > +undesirable, the device can map file contents into the DAX window shared memory
> > > +region.  The driver then accesses file contents directly in device-owned memory
> > > +without a data transfer.
> > > +
> > > +The DAX Window is an alternative mechanism for accessing file contents.
> > > +FUSE\_READ/FUSE\_WRITE requests and DAX Window accesses are possible at the
> > > +same time.  Providing the DAX Window is optional for devices.  Using the DAX
> > > +Window is optional for drivers.
> > > +
> > > +Shared memory region ID 0 is called the DAX window.  Drivers map this shared
> > > +memory region with writeback caching as if it were regular RAM.  The contents
> > > +of the DAX window are undefined unless a mapping exists for that range.
> > 
> > This last paragraph is a bit concerning form s390x perspective. In case
> > of a PCI transport the shared memory region is a chunk of PCI memory (and
> > must be contained within the declared bar, as mandated by commit
> > 855ad7af2bd6).
> > 
> > The PCI architecture on s390x is at the moment such, that PCI memory
> > *can't be accessed like regular RAM* but specialized instructions have
> > to be used. I've tried to rise concern about this multiple times. Thus
> > the virtio spec would contradict itself a little (at least on s390x).
> > 
> > Of course for virtual zPCI devices we can make this work. But including
> > this paragraph in the VIRTIO specification would mean if one were to
> > implement this in HW it would not work for s390.
> > 
> > I don't have a anything better to propose, so I intend to vote yes
> > for this. I just wanted to make sure, we all are aware of the
> > consequences.
> 
> Thanks.
> 
> Note this is just specifying the way virtiofs uses the existing
> (accepted) shared memory region spec.  You can add a CCW transport of
> that spec to make it appropriate for 390 if needed.
> 

On s390x we have both CCW and PCI transport. And that makes things even
more complicated.

IMHO specifying that virtiofs uses the existing shared memory
specification like regular RAM conflicts with what is architecturally
possible on s390x when the transport is PCI.

Because the fact that this is memory exposed by a PCI device and
contained within a bar with the current s390 architecture implies that
this memory can not be used as regular RAM but needs to be accessed via
specialized instructions (PCI LOAD, PCI STORE). @Pierre: please confirm
or disprove me.

Of course both simply not doing DAX window on s390 if transport PCI,
or conceptually extending the architecture (for virtual systems) and
making it work in the non s390 way is an option.

And yes for the CCW transport we can do whatever we want. And I think
we do want regular RAM for CCW transport, because architecturally there
is no way a CCW device can expose memory. So we would/will need to build
something virtual. And if we do, we should do it the way it suits us
best.

Regards,
Halil