Re: [Qemu-devel] [RFC PATCH v4 1/3] Mediated device Core driver

[Dong Jia]

On Tue, 7 Jun 2016 22:29:30 -0600
Alex Williamson <address@hidden> wrote:

> On Wed, 8 Jun 2016 11:18:42 +0800
> Dong Jia <address@hidden> wrote:
> 
> > On Tue, 7 Jun 2016 19:39:21 -0600
> > Alex Williamson <address@hidden> wrote:
> > 
> > > On Wed, 8 Jun 2016 01:18:42 +0000
> > > "Tian, Kevin" <address@hidden> wrote:
> > >   
> > > > > From: Alex Williamson [mailto:address@hidden
> > > > > Sent: Wednesday, June 08, 2016 6:42 AM
> > > > > 
> > > > > On Tue, 7 Jun 2016 03:03:32 +0000
> > > > > "Tian, Kevin" <address@hidden> wrote:
> > > > >     
> > > > > > > From: Alex Williamson [mailto:address@hidden
> > > > > > > Sent: Tuesday, June 07, 2016 3:31 AM
> > > > > > >
> > > > > > > On Mon, 6 Jun 2016 10:44:25 -0700
> > > > > > > Neo Jia <address@hidden> wrote:
> > > > > > >    
> > > > > > > > On Mon, Jun 06, 2016 at 04:29:11PM +0800, Dong Jia wrote:    
> > > > > > > > > On Sun, 5 Jun 2016 23:27:42 -0700
> > > > > > > > > Neo Jia <address@hidden> wrote:
> > > > > > > > >
> > > > > > > > > 2. VFIO_DEVICE_CCW_CMD_REQUEST
> > > > > > > > > This intends to handle an intercepted channel I/O 
> > > > > > > > > instruction. It
> > > > > > > > > basically need to do the following thing:    
> > > > > > > >
> > > > > > > > May I ask how and when QEMU knows that he needs to issue such 
> > > > > > > > VFIO ioctl at
> > > > > > > > first place?    
> > > > > > >
> > > > > > > Yep, this is my question as well.  It sounds a bit like there's an
> > > > > > > emulated device in QEMU that's trying to tell the mediated device 
> > > > > > > when
> > > > > > > to start an operation when we probably should be passing through
> > > > > > > whatever i/o operations indicate that status directly to the 
> > > > > > > mediated
> > > > > > > device. Thanks,
> > > > > > >
> > > > > > > Alex    
> > > > > >
> > > > > > Below is copied from Dong's earlier post which said clear that
> > > > > > a guest cmd submission will trigger the whole flow:
> > > > > >
> > > > > > ----
> > > > > > Explanation:
> > > > > > Q1-Q4: Qemu side process.
> > > > > > K1-K6: Kernel side process.
> > > > > >
> > > > > > Q1. Intercept a ssch instruction.
> > > > > > Q2. Translate the guest ccw program to a user space ccw program
> > > > > >     (u_ccwchain).
> > > > > > Q3. Call VFIO_DEVICE_CCW_CMD_REQUEST (u_ccwchain, orb, irb).
> > > > > >     K1. Copy from u_ccwchain to kernel (k_ccwchain).
> > > > > >     K2. Translate the user space ccw program to a kernel space ccw
> > > > > >         program, which becomes runnable for a real device.
> > > > > >     K3. With the necessary information contained in the orb passed 
> > > > > > in
> > > > > >         by Qemu, issue the k_ccwchain to the device, and wait event 
> > > > > > q
> > > > > >         for the I/O result.
> > > > > >     K4. Interrupt handler gets the I/O result, and wakes up the 
> > > > > > wait q.
> > > > > >     K5. CMD_REQUEST ioctl gets the I/O result, and uses the result 
> > > > > > to
> > > > > >         update the user space irb.
> > > > > >     K6. Copy irb and scsw back to user space.
> > > > > > Q4. Update the irb for the guest.
> > > > > > ----    
> > > > > 
> > > > > Right, but this was the pre-mediated device approach, now we no longer
> > > > > need step Q2 so we really only need Q1 and therefore Q3 to exist in
> > > > > QEMU if those are operations that are not visible to the mediated
> > > > > device; which they very well might be, since it's described as an
> > > > > instruction rather than an i/o operation.  It's not terrible if that's
> > > > > the case, vfio-pci has its own ioctl for doing a hot reset.    
> > Dear Alex, Kevin and Neo,
> > 
> > 'ssch' is a privileged I/O instruction, which should be finally issued
> > to the dedicated subchannel of the physical device.
> > 
> > BTW, I did remove step Q2 with all of the user-space translation code,
> > according to your comments in another thread.
> > 
> > > > 
> > > >   
> > > > >     
> > > > > > My understanding is that such thing belongs to how device is 
> > > > > > mediated
> > > > > > (so device driver specific), instead of something to be abstracted 
> > > > > > in
> > > > > > VFIO which manages resource but doesn't care how resource is used.
> > > > > >
> > > > > > Actually we have same requirement in vGPU case, that a guest driver
> > > > > > needs submit GPU commands through some MMIO register. vGPU device
> > > > > > model will intercept the submission request (in its own way), do its
> > > > > > necessary scan/audit to ensure correctness/security, and then submit
> > > > > > to physical GPU through vendor specific interface.
> > > > > >
> > > > > > No difference with channel I/O here.    
> > > > > 
> > > > > Well, if the GPU command is submitted through an MMIO register, is 
> > > > > that
> > > > > MMIO register part of the mediated device?  If so, could the mediated
> > > > > device recognize the command and do the scan/audit itself?  QEMU must
> > > > > not be the point at which mediation occurs for security purposes, QEMU
> > > > > is userspace and userspace is not to be trusted.  I'm still open to
> > > > > ioctls where it makes sense, as above, we have PCI specific ioctls and
> > > > > already, but we need to evaluate each one, why it needs to exist, and
> > > > > whether we can skip it if the mediated device can trigger the action 
> > > > > on
> > > > > its own.  After all, that's why we're using the vfio api, so we can
> > > > > re-use much of the existing infrastructure, especially for a vGPU that
> > > > > exposes itself as a PCI device.  Thanks,
> > > > >     
> > > > 
> > > > My point is that a guest submission on vGPU is just a normal trapped 
> > > > register write, which is forwarded from Qemu to VFIO through pwrite 
> > > > interface and then hit mediated vGPU device. The mediated device
> > > > will recognize this register write as a submission request and then do
> > > > necessary scan (looks we are saying same thing) and then submit to
> > > > physical device driver. If loading ccw cmds on channel i/o are also 
> > > > through some I/O registers, it can be implemented same way w/o
> > > > introducing new ioctl.  
> > We are different here. The target of an I/O instruction is the
> > subchannel. CCW devices don't have these kind of registers. The mediated
> > ccw device can not recognize such an submission by its own capbilities. 
> > 
> > A CCW device does not have such registers in both the physical and the
> > mediated devices to sense or recognize the submission request. It's the
> > CPU that recognizes the submission by intercepting the guest ssch
> > instruction.
> > 
> > CPU can not tell if it is issued from a passed thru device driver or a
> > virtio device driver from the guest. So it has to exit to QEMU, and let
> > QEMU take over.
> > 
> > Once QEMU identifies the target subchannel is serving a passed thru
> > device, it uses the ioctl to pass the instruction parameters into the
> > kernel all the way along the mediated driver to the physical driver to
> > the subchannel to perform the I/O operation.
> > 
> > > > The r/w handler of mediated device can figure
> > > > out whether it's a ccw submission or not. But my understanding might 
> > > > be wrong here.  
> > We don't have registers to sense an instruction or operation.
> 
> Ok, so it seems we need to create some sort of interface to initiate
> the ccw program, but I suppose I'm not yet convinced that it needs a
> new ioctl.  For instance if you only need to "kick" the device to tell
> it when to begin translation and execution, we could create a virtual
> interrupt into the mediated device with an irqfd.  QEMU writes to the
> irqfd (eventfd), the mediated driver receives this kick and begins
> processing.  Another virtual interrupt out to the user might indicate
> completion. On the other hand if the ioctl was intended to write the
> ccw program itself to the device, we have vfio device regions that can
> do this.  Simply define within the vfio-ccw API that one of the regions
> is a virtual program buffer and define the API between the mediated
> driver and user the sequence of writes that load the program state,
> initiate the program, and return the result.
> 
> The vfio API already has a very extensible mechanism for communicating
> with a device through regions and interrupts, not all of which
> necessarily need to match physical attributes of the device.  ioctls
> can be added, but lets exhaust the mechanisms we already have through
> the vfio api first.  Thanks,
Dear Alex and Neo,

I tried as what you suggested - add an MMIO region to the device, and
it works fine. It's an interesting and elegant way. I like it. :>

So indeed, we neither need to introduce a new ioctl command, nor the
ioctl callback on phy_device_ops.

Thanks!

> 
> Alex
> 



--------
Dong Jia