Re: [virtio-dev] Re: [PATCH v2 1/3] shared memory: Define shared memory regions

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Cornelia Huck (cohuck@redhat.com) wrote:
> On Tue, 26 Feb 2019 15:11:34 +0000
> "Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> 
> > * Cornelia Huck (cohuck@redhat.com) wrote:
> > > On Fri, 22 Feb 2019 11:54:31 +0000
> > > "Dr. David Alan Gilbert (git)" <dgilbert@redhat.com> wrote:
> > >   
> > > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> > > > 
> > > > Define the requirements and idea behind shared memory regions.
> > > > 
> > > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > > > ---
> > > >  content.tex    |  2 ++
> > > >  shared-mem.tex | 33 +++++++++++++++++++++++++++++++++
> > > >  2 files changed, 35 insertions(+)
> > > >  create mode 100644 shared-mem.tex
> > > > 
> > > > diff --git a/content.tex b/content.tex
> > > > index 836ee52..3dd504c 100644
> > > > --- a/content.tex
> > > > +++ b/content.tex
> > > > @@ -371,6 +371,8 @@ making any more buffers available. When VIRTIO_F_NOTIFICATION_DATA
> > > >  has been negotiated, these notifications would then have
> > > >  identical \field{next_off} and \field{next_wrap} values.
> > > >  
> > > > +\input{shared-mem.tex}
> > > > +
> > > >  \chapter{General Initialization And Device Operation}\label{sec:General Initialization And Device Operation}
> > > >  
> > > >  We start with an overview of device initialization, then expand on the
> > > > diff --git a/shared-mem.tex b/shared-mem.tex
> > > > new file mode 100644
> > > > index 0000000..85b0c55
> > > > --- /dev/null
> > > > +++ b/shared-mem.tex
> > > > @@ -0,0 +1,33 @@
> > > > +\section{Shared Memory Regions}\label{sec:Basic Facilities of a Virtio Device / Shared Memory Regions}
> > > > +
> > > > +Shared memory regions are an additional facility
> > > > +available to devices that need a region of memory that's
> > > > +continuously shared between the host and the guest, rather
> > > > +than passed between them in the way virtqueue elements are.
> > > > +
> > > > +Example uses include shared caches and version pools for versioned
> > > > +data structures.
> > > > +
> > > > +The region is chosen by the host and presented to the guest, as
> > > > +such it is useful in situations where the memory is accessed on
> > > > +the host by other libraries that can't safely access guest RAM.  
> > > 
> > > This explanation looks good to me.
> > >   
> > > > +
> > > > +Shared memory regions MUST NOT be used to control the operation
> > > > +of the device, nor to stream data; those should still be performed
> > > > +using virtqueues.  
> > > 
> > > The 'MUST NOT' makes it look like a normative statement; however, this
> > > is more like design advice?  
> > 
> > Hmm I'm not sure - the intention of this line is to try and stop people
> > using it as a hack to avoid standardising protocol for no good reason.
> 
> I've looked through the spec again and it seems the best place for this
> would indeed be a device-normative section for shared regions; it would
> imply that a device needs to comply with this statement, or it is not
> conformant.
> 
> It would need some tweaking, however. Perhaps something like the
> following:
> 
> "The device MUST NOT expose shared memory regions which are used to
> control the operation of the device, or to stream data."

> Not sure where to put the pointer to use virtqueues for that.

Thanks I've turned that into a device normative section and dropped the
explicit mention of virtqueues.

> 
> > 
> > > > +
> > > > +A device may have multiple shared memory regions associated with
> > > > +it.  Each region has a \field{shmid} to identify it, the meaning
> > > > +of which is device-specific.
> > > > +
> > > > +Enumeration and location of shared memory regions is performed
> > > > +using a transport-specific data structure and mechanism.
> > > > +
> > > > +Memory consistency rules vary depending on the region and the
> > > > +device.  Devices MUST define the required behaviour for each
> > > > +region.  
> > > 
> > > Same here.  
> > 
> > OK, so should I reword this?
> 
> Maybe put into the device normative section:
> 
> "A device MUST define the required behaviour for each region."
> 
> But I'm not sure whether this is a good normative statement... it's
> more that the regions have specific requirements, which both the device
> and the driver need to fulfill, and those requirements obviously need
> to be put down somewhere.
> 
> > 
> > > > +
> > > > +The guest physical address and the host virtual address MUST NOT
> > > > +be used to identify structures within the memory regions; all
> > > > +addressing MUST be relative to the start of a particular region.
> > > > +  
> > > 
> > > For that high-level overview, I'm not sure if any normative statements
> > > are needed/wanted, or whether those should be confined to the individual
> > > transport or device type definitions...  
> > 
> > I think this one is stronger than the previous two; if people start
> > passing GPA/HVA in the underlying structures things are going to get
> > messy.
> 
> What about
> 
> "The device MUST handle any access to the memory region as addressed
> relatively to the beginning of the region, regardless whether it
> accesses the region itself or the driver does so."
> 
> "The driver MUST address any particular region relatively to the start
> of that region."
> 
> for device and driver normative statements, respectively. Better
> wordings welcome.

What I've done here is turn this into just a subsection describing
how it should work and stopped using the magic MUST.
It's not really a requirement on the device - it's more an indication
of how to design a device.

Dave

> ---------------------------------------------------------------------
> To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
> For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK