[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Dom0 Virtual Memory

  • To: "Petersson, Mats" <Mats.Petersson@xxxxxxx>
  • From: "Eric Benton" <benton71@xxxxxxxxx>
  • Date: Mon, 23 Oct 2006 15:09:44 +0200
  • Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
  • Delivery-date: Mon, 23 Oct 2006 06:10:13 -0700
  • Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:references; b=t0p+FX5AkGJOBa/FqeU78ysGoRzP/3otKE7KTpBw+qvBAZPpV6qs0iJdhTd3rnfTKW6zkD1+OVmyr+/77CJkyuvmibMutar8qAG9EyxUvjWwX1J3SQHAnjE0oWBnwtf+KRnb5byq/+Ee73gsQAAsF2tgAuxQdweTEYYYaCrJaqc=
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>

Thank you for the quick replay.

Sorry, I got a bit confused. It's obvious that each OS have a virutal memory manger to translate virtual addresses to physical addresses. I didn't know what shadow mode means and I saw some CR3 handling / a lot of page-tables structures.

What are all those l1 l2 l3 page tables in page.h? are they being used only for shadowed guests?

What I'm trying to achieve is the ability to write to dom0 virtual memory from Xen (translate them to machine addresses). Is there an easy method to do that? (except from using the event channel and do that from dom0 itself). Are there any genric functions (in page.h) to handle that? I also saw that dom0 pfns corresponds to the real mfns in order to create a 1:1 match for DMA to work. Btw, does xen keeps track of the real hardware resources? or is it just letting the operating system handle this?


On 10/23/06, Petersson, Mats < Mats.Petersson@xxxxxxx> wrote:
I'm not sure if this is answering the right or wrong question, but I think you're missing a bit somewhere.
Xen is managing the MACHINE PHYSICAL memory and the GUEST is believing that the GUEST PHYSICAL is "real" physical memory. The next level above, which is the OS's handling of virtual to physical and physical to virtual will just handle this as if GUEST PHYSICAL really was the true picture of the world. There are however some places where the difference is "real" - drivers that use direct memory access (bus mastering) will need to have the bus-address, and until IOMMU is a reality for most people, the bus-address is MACHINE PHYSICAL. So there's special code inside the device-drivers to handle the translation of the GUEST PHYSICAL to MACHNE PHYSICAL when translating the virtual address given by the user-mode application into a PHYSICAL address.
The tables used to manage virtual to guest-physical is the page-tables. Obviously, for ease of use, the OS will also implement some sort of virtual to physical and physical to virtual tables to translate between the two as necessary.
In HVM (or any other case of "shadow mode"), you can use shadow_gva_to_[gfn,gpa] to get the guest-physical address, and then sh_vcp_gfn_to_mfn() to translate that to a machine frame. mfn_x appears to translate that to a "real address" - although my 2 minutes of grepping didn't find it...

From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Eric Benton
Sent: 23 October 2006 11:42
To: xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-devel] Dom0 Virtual Memory


I looked at the sources and figured out all about the P2M, M2P tables and mappings,
Now I'm trying to figure out how dom0 virtual memory is implemented, and I have some questions.

The platform I'm working on is x86 with Intel VT enabled.

1. Is dom0 virtual memory managed differently than the other domains? what about hvms? and domUs?
2. What tables are used to manage the virtual memory?
3. What is the easiest method to do a dom0 virtual address to machine address translation (from Xen itself)?

Thanks for your help in advance!


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.