[PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Jan Beulich]

[-- Attachment #1: Type: text/plain, Size: 852 bytes --]

PVH guests accessing I/O ports via string ops is not supported yet.

Reported-by: Roger Pau Monné<roger.pau@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
v2: handle_pio() is already safe (pointed out by Mukesh), so only refuse
    entering handle_mmio().
Note: Only compile tested so far.

--- unstable.orig/xen/arch/x86/hvm/vmx/vmx.c	2014-11-27 09:37:27.000000000 +0100
+++ unstable/xen/arch/x86/hvm/vmx/vmx.c	2014-12-08 10:04:27.000000000 +0100
@@ -3086,7 +3086,8 @@ void vmx_vmexit_handler(struct cpu_user_
         if ( exit_qualification & 0x10 )
         {
             /* INS, OUTS */
-            if ( !handle_mmio() )
+            if ( unlikely(is_pvh_vcpu(v)) /* PVH fixme */ ||
+                 !handle_mmio() )
                 hvm_inject_hw_exception(TRAP_gp_fault, 0);
         }
         else



[-- Attachment #2: VMX-PVH-no-emulated-IO-string-ops.patch --]
[-- Type: text/plain, Size: 848 bytes --]

PVH guests accessing I/O ports via string ops is not supported yet.

Reported-by: Roger Pau Monné<roger.pau@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
v2: handle_pio() is already safe (pointed out by Mukesh), so only refuse
    entering handle_mmio().
Note: Only compile tested so far.

--- unstable.orig/xen/arch/x86/hvm/vmx/vmx.c	2014-11-27 09:37:27.000000000 +0100
+++ unstable/xen/arch/x86/hvm/vmx/vmx.c	2014-12-08 10:04:27.000000000 +0100
@@ -3086,7 +3086,8 @@ void vmx_vmexit_handler(struct cpu_user_
         if ( exit_qualification & 0x10 )
         {
             /* INS, OUTS */
-            if ( !handle_mmio() )
+            if ( unlikely(is_pvh_vcpu(v)) /* PVH fixme */ ||
+                 !handle_mmio() )
                 hvm_inject_hw_exception(TRAP_gp_fault, 0);
         }
         else

[-- Attachment #3: Type: text/plain, Size: 126 bytes --]

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

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Mukesh Rathor]

On Mon, 08 Dec 2014 09:12:37 +0000
"Jan Beulich" <JBeulich@suse.com> wrote:

> PVH guests accessing I/O ports via string ops is not supported yet.
> 
> Reported-by: Roger Pau Monné<roger.pau@citrix.com>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> ---
> v2: handle_pio() is already safe (pointed out by Mukesh), so only
> refuse entering handle_mmio().
> Note: Only compile tested so far.
> 
> --- unstable.orig/xen/arch/x86/hvm/vmx/vmx.c	2014-11-27
> 09:37:27.000000000 +0100 +++
> unstable/xen/arch/x86/hvm/vmx/vmx.c	2014-12-08
> 10:04:27.000000000 +0100 @@ -3086,7 +3086,8 @@ void
> vmx_vmexit_handler(struct cpu_user_ if ( exit_qualification & 0x10 ) {
>              /* INS, OUTS */
> -            if ( !handle_mmio() )
> +            if ( unlikely(is_pvh_vcpu(v)) /* PVH fixme */ ||
> +                 !handle_mmio() )
>                  hvm_inject_hw_exception(TRAP_gp_fault, 0);
>          }
>          else
> 

Good idea, and if needed:

Acked-by: Mukesh Rathor <mukesh.rathor@oracle.com>


thanks
Mukesh


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

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Roger Pau Monné]

El 08/12/14 a les 10.12, Jan Beulich ha escrit:
> PVH guests accessing I/O ports via string ops is not supported yet.
> 
> Reported-by: Roger Pau Monné<roger.pau@citrix.com>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>

This looks fine to me (at least it doesn't break the existing IN/OUT
users), and seems the best solution 4.5 wise:

Acked-by: Roger Pau Monné <roger.pau@citrix.com>

For 4.6 I think we need to start using a different hvm_io_bitmap for PVH
Dom0 that allows direct access to the IO ports, bypassing the vmexit and
simplifying the code in Xen (this would also fix INS/OUTS). Still not
sure what should be done for PVH DomUs, specially when PVH gains support
for pci-passthrough.

Roger.


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

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Jan Beulich]

>>> On 09.12.14 at 18:01, <roger.pau@citrix.com> wrote:
> For 4.6 I think we need to start using a different hvm_io_bitmap for PVH
> Dom0 that allows direct access to the IO ports, bypassing the vmexit and
> simplifying the code in Xen (this would also fix INS/OUTS). Still not
> sure what should be done for PVH DomUs, specially when PVH gains support
> for pci-passthrough.

With the difficulty being that for PV the hypervisor intentionally
intercepts accesses to some of the ports, so we can't blindly
allow PVH access to all the ports its allowed access to.

Jan

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Roger Pau Monné]

El 09/12/14 a les 18.19, Jan Beulich ha escrit:
>>>> On 09.12.14 at 18:01, <roger.pau@citrix.com> wrote:
>> For 4.6 I think we need to start using a different hvm_io_bitmap for PVH
>> Dom0 that allows direct access to the IO ports, bypassing the vmexit and
>> simplifying the code in Xen (this would also fix INS/OUTS). Still not
>> sure what should be done for PVH DomUs, specially when PVH gains support
>> for pci-passthrough.
> 
> With the difficulty being that for PV the hypervisor intentionally
> intercepts accesses to some of the ports, so we can't blindly
> allow PVH access to all the ports its allowed access to.

I assume this is mainly for DomUs but not for Dom0? Or should PVH Dom0
access to the IO space also be filtered and emulated for some ports?

Roger.

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Jan Beulich]

>>> On 09.12.14 at 18:49, <roger.pau@citrix.com> wrote:
> El 09/12/14 a les 18.19, Jan Beulich ha escrit:
>>>>> On 09.12.14 at 18:01, <roger.pau@citrix.com> wrote:
>>> For 4.6 I think we need to start using a different hvm_io_bitmap for PVH
>>> Dom0 that allows direct access to the IO ports, bypassing the vmexit and
>>> simplifying the code in Xen (this would also fix INS/OUTS). Still not
>>> sure what should be done for PVH DomUs, specially when PVH gains support
>>> for pci-passthrough.
>> 
>> With the difficulty being that for PV the hypervisor intentionally
>> intercepts accesses to some of the ports, so we can't blindly
>> allow PVH access to all the ports its allowed access to.
> 
> I assume this is mainly for DomUs but not for Dom0? Or should PVH Dom0
> access to the IO space also be filtered and emulated for some ports?

It should indeed - just look at guest_io_read() and guest_io_write().
I.e. an eventual PVH specific bitmap would need to be populated
based on what admin_io_okay() returns (instead of
ioports_access_permitted()), carefully taking into consideration the
possible access widths.

Jan

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Konrad Rzeszutek Wilk]

On Tue, Dec 09, 2014 at 06:01:54PM +0100, Roger Pau Monné wrote:
> El 08/12/14 a les 10.12, Jan Beulich ha escrit:
> > PVH guests accessing I/O ports via string ops is not supported yet.
> > 
> > Reported-by: Roger Pau Monné<roger.pau@citrix.com>
> > Signed-off-by: Jan Beulich <jbeulich@suse.com>
> 
> This looks fine to me (at least it doesn't break the existing IN/OUT
> users), and seems the best solution 4.5 wise:
> 
> Acked-by: Roger Pau Monné <roger.pau@citrix.com>

Release-Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

> 
> For 4.6 I think we need to start using a different hvm_io_bitmap for PVH
> Dom0 that allows direct access to the IO ports, bypassing the vmexit and
> simplifying the code in Xen (this would also fix INS/OUTS). Still not
> sure what should be done for PVH DomUs, specially when PVH gains support
> for pci-passthrough.
> 
> Roger.
> 

Re: [PATCH v2] VMX: don't allow PVH to reach handle_mmio()

[Tian, Kevin]

> From: Jan Beulich [mailto:JBeulich@suse.com]
> Sent: Monday, December 08, 2014 5:13 PM
> 
> PVH guests accessing I/O ports via string ops is not supported yet.
> 
> Reported-by: Roger Pau Monné<roger.pau@citrix.com>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>

Acked-by: Kevin Tian <kevin.tian@intel.com>

> ---
> v2: handle_pio() is already safe (pointed out by Mukesh), so only refuse
>     entering handle_mmio().
> Note: Only compile tested so far.
> 
> --- unstable.orig/xen/arch/x86/hvm/vmx/vmx.c	2014-11-27
> 09:37:27.000000000 +0100
> +++ unstable/xen/arch/x86/hvm/vmx/vmx.c	2014-12-08
> 10:04:27.000000000 +0100
> @@ -3086,7 +3086,8 @@ void vmx_vmexit_handler(struct cpu_user_
>          if ( exit_qualification & 0x10 )
>          {
>              /* INS, OUTS */
> -            if ( !handle_mmio() )
> +            if ( unlikely(is_pvh_vcpu(v)) /* PVH fixme */ ||
> +                 !handle_mmio() )
>                  hvm_inject_hw_exception(TRAP_gp_fault, 0);
>          }
>          else
> 

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