Re: [XenAccess-devel] Towards HVM support

[Steve B. <st...@at...>]

> The error that you are getting is (most likely) because XenAccess
> cannot find the System map file.  Looking at the xml output that you
> provided, it looks as if libvirt doesn't have guest kernel
> information for an HVM domain.  Normally, XenAccess would use this
> information to guess the location of the System map file.  However,
> without this information, it won't find the System map file.     
> 
> One question, for HVM domains do you still boot off a kernel stored
> in dom0?  If so, I wonder why the kernel info isn't included in
> libvirt's output.  If not, then we may have to manually install the
> System map file into dom0 and find some way to point XenAccess to the
> appropriate file.    

This seems to be the crux.  HVM domains do NOT boot off a kernel stored in
dom0.  This is why you can run a Windows guest for example.  It acts more
like VMware: all the files are in the guest's image, including the /boot
directory.  The image is partitioned as if it were a separate hard drive,
with separate /boot and / partitions and file systems.  

I now see that XenAccess opens the System map file on the host, and I have
no idea how to have it grab a file from inside the guest.  So I copied the
guest's System.map onto the host and with some hacking of your code and some
extra printf's I've gotten this far:

# ./run ./module_list 16
dom!=NULL
kernel_name=hvm
sysmap=/boot/System.map-2.6.17
swapper_pg_dir=e000
XA_PAGE_OFFSET=0
memory=NULL
failed to init XenAccess library: Bad address

These the last few debug statements are from helper_init() inside of
xa_core.c.  I looked up swapper_pg_dir in the System.map and it's at
0xc044e000.  What fails is the call to linux_access_physical_address().

> Once we have the System map file, we can see where we are at :-)
> But, I'm hopeful that the critical functions that I use from Xen will
> behave the same in HVM domains.  If that turns out to be the case,
> then we are very close to getting this working.  

As you say, where are we at now?  I don't know if swapper_pg_dir=e000 is
correct of if it should have been c044e000.  I also don't know if the
XA_PAGE_OFFSET should be 0 or not.  It seems that either: 

(1) these values are correct and the method you/libvirt use to get memory
from a guest only works for paravirtualized VMs, or
(2) these values are incorrect, and the memory access method may or may not
work for HVMs.

Either way, where do you think I should go from here?

Thanks,

 - Steve