Doubts regarding Shadow Page Table Management in KVM

Doubts regarding Shadow Page Table Management in KVM

[Sukanto Ghosh]

Hi all,

I was reading the "The shadowy depths of the KVM MMU" presentation by 
Avi Kivity.

 From my understanding of shadow page-table,
Guest Page Table :     maps GV -> GP 
Shadow Page Table :  maps GV -> HP

What is meant by shadowed guest page ? (Does it refer to : the guest PT 
pages that have a corresponding shadow PT page)


I has been mentioned that: "The same page may already be mapped by the 
guest in multiple locations. So, we track writable mappings of every 
guest page ".
I can't understand why do we have to keep track of these reverse 
mappings ? And what are these reverse mappings ? (from HP -> GP or HP -> 
GV or GP -> GV)

Let's consider the following scenario:
GVA1 -> GPA1 -> HPA1
GVA2 -> GPA1 -> HPA1

Also, let GPA1 be part of a guest PT page.

i)  The guest modifies a PTE in GPA1 by doing "mov GVA2"
ii) The host traps this as the page HPA1 is write-protected, and does 
the corresponding modification to the guest PT page and its 
corresponding shadow PT page.
(where do we need to use a reverse mapping)


I would be glad if you can point out where I am making the mistake.


--
Thanks & Regards,
Sukanto Ghosh



NB:    GVA = guest virtual address      GPA = guest physical address   
   HPA = host physical address

Re: Doubts regarding Shadow Page Table Management in KVM

[Sukanto Ghosh]

I think I have cleared the issue regarding the reverse mapping.

I missed the fact, that the write permission is associated with the PTE. 
I thought that it was associated only with the kernel structures 
maintaining the physical page-frame.

But I want to confirm whether a shadowed guest page refers to "the guest 
PT page that has a corresponding shadow PT page. And the corresponding 
shadow PT page is called the shadow of the former.  Am I right ?


Thanks and regards
Sukanto


---
* PT = page table
---

Re: Doubts regarding Shadow Page Table Management in KVM

[Avi Kivity]

Sukanto Ghosh wrote:
>
> But I want to confirm whether a shadowed guest page refers to "the 
> guest PT page that has a corresponding shadow PT page. And the 
> corresponding shadow PT page is called the shadow of the former.  Am I 
> right ?

Yes.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.