On 02/05/2019 07:20, Razvan Cojocaru wrote:
> On 5/2/19 2:52 AM, Tamas K Lengyel wrote:
>> Currently the gs_shadow value is only cached when the vCPU is being scheduled
>> out by Xen. Reporting this (usually) stale value through vm_event is
>> incorrect,
>> since it doesn't represent the actual state of the vCPU at the time the event
>> was recorded. This prevents vm_event subscribers from correctly finding
>> kernel
>> structures in the guest when it is trapped while in ring3.
>>
>> Refresh shadow_gs value when the context being saved is for the current vCPU.
>>
>> Signed-off-by: Tamas K Lengyel <ta...@tklengyel.com>
>> Cc: Razvan Cojocaru <rcojoc...@bitdefender.com>
>> Cc: Jun Nakajima <jun.nakaj...@intel.com>
>> Cc: Kevin Tian <kevin.t...@intel.com>
>> Cc: Jan Beulich <jbeul...@suse.com>
>> Cc: Andrew Cooper <andrew.coop...@citrix.com>
>> Cc: Wei Liu <wei.l...@citrix.com>
>> Cc: Roger Pau Monne <roger....@citrix.com>
>> ---
>> v2: move fix to hvm so vm_event doesn't have to know specifics
>> ---
>> xen/arch/x86/hvm/vmx/vmx.c | 5 +++++
>> 1 file changed, 5 insertions(+)
>>
>> diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
>> index 283eb7b34d..5154ecc2a8 100644
>> --- a/xen/arch/x86/hvm/vmx/vmx.c
>> +++ b/xen/arch/x86/hvm/vmx/vmx.c
>> @@ -779,12 +779,17 @@ static void vmx_load_cpu_state(struct vcpu *v, struct
>> hvm_hw_cpu *data)
>>
>> static void vmx_save_vmcs_ctxt(struct vcpu *v, struct hvm_hw_cpu *ctxt)
>> {
>> + if ( v == current )
>> + vmx_save_guest_msrs(v);
> vmx_save_guest_msrs() is simple enough that the if is probably not
> necessary here (we can just call the function directly, regardless of
> what v holds).
Why? Doing that would fully corrupt v's state when called in remote
context, as you'd clobber v's gs_shadow which whatever the value was
from the current vcpu.
~Andrew
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
