[vbox-dev] HGCM question...guest memory mapping

[Ramshankar]

On 17/07/12 00:17, Jaikumar G wrote:
> Hi Ram,
>
> Thanks that makes a lot of sense.
>
> Regarding allocation of memory in guest space:
>
> "QEMU has a feature where it can translate a guest virtual address in
> user space directly into the corresponding physical/backing page in RAM
> without the need for allocating memory in Guest space."
>
> Is is something that can be done in virtual box ? Is there any
> significant performance differences ?
>

I don't know the context of that quote or what feature it refers to. As 
far as VirtualBox is concerned, we could classify this into 2 cases for 
the sake of simplicity.

1. Without Nested Paging (NP) / Extended Page Tables (EPT): the 
hypervisor traps #PFs in the guest and syncs its shadow page tables. 
Host-side translation of guest addresses is done by traversing the 
shadow page tables.

2. With NP/EPT: #PFs in the guest are handled without causing VM-exits 
as the hardware supports an extra indirection of address translation. 
The host TLBs can cache guest linear addresses to host physical 
addresses. VirtualBox fully supports NP/EPT if the hardware has support 
for it.

Regards,
Ram.