[Qemu-devel] [PATCH] VMRUN should store the RIP after the instruction

[Qemu-devel] [PATCH] VMRUN should store the RIP after the instruction

[Alexander Graf]

[-- Attachment #1: Type: text/plain, Size: 216 bytes --]

VMRUN stores the IP which is used to return to the instruction after the 
VMRUN in case of a VMEXIT. Due to recent changes, this does not happen 
anymore and a VMEXIT simply returns to the VMRUN instruction.

Alex



[-- Attachment #2: qemu-svn-rip.patch --]
[-- Type: text/x-patch, Size: 596 bytes --]

Index: target-i386/op_helper.c
===================================================================
--- target-i386/op_helper.c	(revision 4661)
+++ target-i386/op_helper.c	(working copy)
@@ -4836,7 +4836,7 @@
     svm_save_seg(env->vm_hsave + offsetof(struct vmcb, save.ds), 
                  &env->segs[R_DS]);
 
-    stq_phys(env->vm_hsave + offsetof(struct vmcb, save.rip), EIP);
+    stq_phys(env->vm_hsave + offsetof(struct vmcb, save.rip), EIP + 3);
     stq_phys(env->vm_hsave + offsetof(struct vmcb, save.rsp), ESP);
     stq_phys(env->vm_hsave + offsetof(struct vmcb, save.rax), EAX);
 

Re: [Qemu-devel] [PATCH] VMRUN should store the RIP after the instruction

[Anthony Liguori]

Alexander Graf wrote:
> VMRUN stores the IP which is used to return to the instruction after 
> the VMRUN in case of a VMEXIT. Due to recent changes, this does not 
> happen anymore and a VMEXIT simply returns to the VMRUN instruction.
>
> Alex

The open-coded  + 3 bothers me.  Are there any circumstances that a 
prefix could be used with vmrun that would change the instruction 
length?  vmrun doesn't appear to be affected by prefixes but I don't see 
anything in the AMD manual specifically forbidding it.

Regards,

Anthony Liguori

Re: [Qemu-devel] [PATCH] VMRUN should store the RIP after the instruction

[Alexander Graf]

On Jun 4, 2008, at 6:13 PM, Anthony Liguori wrote:

> Alexander Graf wrote:
>> VMRUN stores the IP which is used to return to the instruction  
>> after the VMRUN in case of a VMEXIT. Due to recent changes, this  
>> does not happen anymore and a VMEXIT simply returns to the VMRUN  
>> instruction.
>>
>> Alex
>
> The open-coded  + 3 bothers me.  Are there any circumstances that a  
> prefix could be used with vmrun that would change the instruction  
> length?  vmrun doesn't appear to be affected by prefixes but I don't  
> see anything in the AMD manual specifically forbidding it.

This is true and it bothers me as well. I am open to suggestions how  
to receive the IP after the instruction that is about to be executed.  
Probably this could be done in translate.c?

Alex

Re: [Qemu-devel] [PATCH] VMRUN should store the RIP after the instruction

[Anthony Liguori]

Alexander Graf wrote:
>
> On Jun 4, 2008, at 6:13 PM, Anthony Liguori wrote:
>
>> Alexander Graf wrote:
>>> VMRUN stores the IP which is used to return to the instruction after 
>>> the VMRUN in case of a VMEXIT. Due to recent changes, this does not 
>>> happen anymore and a VMEXIT simply returns to the VMRUN instruction.
>>>
>>> Alex
>>
>> The open-coded  + 3 bothers me.  Are there any circumstances that a 
>> prefix could be used with vmrun that would change the instruction 
>> length?  vmrun doesn't appear to be affected by prefixes but I don't 
>> see anything in the AMD manual specifically forbidding it.
>
> This is true and it bothers me as well. I am open to suggestions how 
> to receive the IP after the instruction that is about to be executed. 
> Probably this could be done in translate.c?

Looks like Fabrice's commit handles it nicely.

Regards,

Anthony Liguori

> Alex