Xen devel - Aug 2006 - Is hypercall_hvm_op in changeset 11060 effective?

hypercall_hvm_op in changeset 11060 seem not effective, should we use
vmmcall instead since hvmloader is running in none-root vmcs mode?
Thanks
-Xin

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> hypercall_hvm_op in changeset 11060 seem not effective, should we use
> vmmcall instead since hvmloader is running in none-root vmcs mode?
It should be possible for a non-root vmcs to make hypercalls due to
changesets 10883 and 10911.  Is this not working?

Steven.


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 12/8/06 5:30 am, "Li, Xin B" <xin.b.li@intel.com> wrote:
> hypercall_hvm_op in changeset 11060 seem not effective, should we use
> vmmcall instead since hvmloader is running in none-root vmcs mode?
What do you think is the problem? Init_hypercall() sets up a hypercall
transfer page suitable for the architecture you are running on (so it uses
vmmcall when running on VT-x). There are a few example hypercalls at the end
of that function which do work properly.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>> hypercall_hvm_op in changeset 11060 seem not effective, should we use
>> vmmcall instead since hvmloader is running in none-root vmcs mode?
>
>What do you think is the problem? Init_hypercall() sets up a hypercall
>transfer page suitable for the architecture you are running on 
>(so it uses
>vmmcall when running on VT-x). There are a few example 
>hypercalls at the end
>of that function which do work properly.
>
Just found the code in hvmloader.c seems not work.
	hvm_param.domid = DOMID_SELF;
	hvm_param.index = HVM_PARAM_APIC_ENABLED;
	if (!hypercall_hvm_op(HVMOP_get_param, &hvm_param) &&
hvm_param.value)
 		create_mp_tables();

Need more investigation.
Thanks
-Xin

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 12/8/06 4:20 pm, "Li, Xin B" <xin.b.li@intel.com> wrote:

> Just found the code in hvmloader.c seems not work.
> hvm_param.domid = DOMID_SELF;
> hvm_param.index = HVM_PARAM_APIC_ENABLED;
> if (!hypercall_hvm_op(HVMOP_get_param, &hvm_param) &&
> hvm_param.value)
> create_mp_tables();
> 
> Need more investigation.
The use of HVMOP_get_param in hvmloader was inconsistent with its
implementation within Xen. This is now fixed in c/s 11070. Thanks for
pointing out the problem!

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>> Just found the code in hvmloader.c seems not work.
>> hvm_param.domid = DOMID_SELF;
>> hvm_param.index = HVM_PARAM_APIC_ENABLED;
>> if (!hypercall_hvm_op(HVMOP_get_param, &hvm_param) &&
>> hvm_param.value)
>> create_mp_tables();
>> 
>> Need more investigation.
>
>The use of HVMOP_get_param in hvmloader was inconsistent with its
>implementation within Xen. This is now fixed in c/s 11070. Thanks for
>pointing out the problem!
>
Still can not work, I got:
(XEN) (GUEST: 6) HVM Loader
(XEN) (GUEST: 6) Detected Xen v3.0
(XEN) (GUEST: 6) Loading ROMBIOS ...
(XEN) (file=hvm.c, line=596) Bad HVM op 1222980.
(XEN) (GUEST: 6) hypercall_hvm_op returned error: 38
(XEN) (GUEST: 6) Loading Cirrus VGABIOS ...
(XEN) (GUEST: 6) Loading VMXAssist ...
(XEN) (GUEST: 6) VMX go ...

Maybe caused by the hypercall page 
-Xin

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Looks like it could be a 32-on-64 issue if you''re running 64-bit Xen.
Perhaps the current test for this situation, in hvm_do_hypercall() in
arch/x86/hvm/hvm.c, is broken. It just tests guest_paging_levels, but that
may not work if paging is disabled. That''s probably where you should
look if
you''re trying to debug this.

 -- Keir

On 14/8/06 11:17 am, "Li, Xin B" <xin.b.li@intel.com> wrote:
>> The use of HVMOP_get_param in hvmloader was inconsistent with its
>> implementation within Xen. This is now fixed in c/s 11070. Thanks for
>> pointing out the problem!
>> 
> 
> Still can not work, I got:
> (XEN) (GUEST: 6) HVM Loader
> (XEN) (GUEST: 6) Detected Xen v3.0
> (XEN) (GUEST: 6) Loading ROMBIOS ...
> (XEN) (file=hvm.c, line=596) Bad HVM op 1222980.
> (XEN) (GUEST: 6) hypercall_hvm_op returned error: 38
> (XEN) (GUEST: 6) Loading Cirrus VGABIOS ...
> (XEN) (GUEST: 6) Loading VMXAssist ...
> (XEN) (GUEST: 6) VMX go ...
> 
> Maybe caused by the hypercall page 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>
>Looks like it could be a 32-on-64 issue if you''re running 64-bit
Xen.
>Perhaps the current test for this situation, in hvm_do_hypercall() in
>arch/x86/hvm/hvm.c, is broken. It just tests 
>guest_paging_levels, but that
>may not work if paging is disabled. That''s probably where you 
>should look if
>you''re trying to debug this.
This patch fixes the HVM hypercall issue when paging is not enabled yet
on 64bit host.

Signed-off-by: Xin Li <xin.b.li@intel.com>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel