Re: [PATCH for-4.5 v6 00/16] Xen VMware tools support

[Don Slutz <dslutz@verizon.com>]

On 09/23/14 08:30, Ian Campbell wrote:
> On Mon, 2014-09-22 at 13:19 -0400, Don Slutz wrote:

[snip]

>>> I was only responding to the part of your comment in parentheses. :-)
>>>
>>> I suppose in large part it would depend on what the hypercalls were
>>> actually doing; I'd have to go back and look at them to say if they
>>> need to be in Xen or whether they could be passed on to qemu.
>>>
>> Clearly it is possible to pass the VCPU registers to QEMU, but that is
>> currently not done.
> I think there's an existing hypercall to get/set the state for a vcpu,
> perhaps it is too heavy weight to be used here though.

Yes, very heavy weight

> An alternative would be a semantically higher level I/O req which took a
> guest pointer to a key and a guest pointer to the buffer etc, without
> needing the registers themselves.

I am looking at adding a new I/O req type for this.  It turns out that
for vmware_port you need to pass 6 32bit values both ways.  And
I can overlap the .addr, .data, .count and .size for this.  The other
option is to increase the size of struct ioreq, which I am assuming
is not the way to go since it would reduce the max number of vcpus
as long as "struct shared_iopage" is limited to 1 page.

"guest pointer to a key and a guest pointer to the buffer" is not how
this works.  The data is all passed by upto 4 bytes at each IN.  A string
(which is how guestinfo access looks like) is passed as a length, and
then each 4 bytes of the string. (I am not trying to say this is good.)



>>    So a new
>> version of QEMU would also be needed to go this way.  None the the
>> proposed features need
>> any data from QEMU, so I do not think this make sense.
> The concern is that it is adding a load of complex looking string and
> pointer manipulation stuff to the hypervisor, the sort of thing which
> often leads to security vulnerabilities.
>
> So that would be better done outside of Xen itself if possible, if a
> qemu update is the price for that then it doesn't seem so bad to me.

I have yet to come up with a good reason why not to move the
VMware port RPC code into QEMU.  I will be looking to do that for
Xen 4.6 & QEMU 2.3


Related to that, the code to connect Xen to QEMU so that Xen can
use any VMware support in QEMU is not that complex.  So added
the xen part in place of patches 8, 9, 10, 11, 12, 14, 15 and 16
looks doable.  This would allow X to use the VMware mouse
code (which is in both qemu-xen and qemu-xen-traditional).  I have
found this to be a great improvement in using a GUI in a guest
where the network speeds are not that fast.  I had planned
on adjusting the Xen to QEMU connector code for 4.6

Also there is a good chance that the QEMU part could be up streamed
to QEMU 2.2 (and backported to Xen's QEMU tree) for 4.5

Now since I did not include this code sooner, would I need a release
exception to include the Xen to QEMU connector code?


One thing related to this is, should I also change qemu-xen-traditional
to handle the new new I/O req type, or to only send it if using qemu-xen.

It is simple to allow a new QEMU to build with pre-4.5 Xen and post-4.5
Xen.  No idea of a good way to check that a QEMU binary has this
support.  However I can say that enabling vmware_port does require
a QEMU with this support in the docs.


     -Don Slutz

> Ian.
>