Muli Ben-Yehuda's journal

January 18, 2006

Filed under: Uncategorized — Muli Ben-Yehuda @ 2:15 PM

Slides from the Xen Winter Summit at Austin are starting to show up. Our Xen/IOMMU Integration should be there soon.

7 Comments »

  1. interesting slides !
    I saw in your slides the term DAC.
    What is DAC ?
    DMAC stands for Direct Memory Access Controller (DMAC).
    Does DAC has to to with DMA/DMAC?

    Comment by Anonymous — January 19, 2006 @ 7:16 AM | Reply

    • Re: interesting slides !
      Not Muli, but still an answer: AFAIK DAC stands for Double Address Cycle, meaninig sending an address in a transaction (like DMA) in two bus transfers, not one. Usefull when the possible size of the address is bigger then the bus width.
      Gilad

      Comment by Anonymous — January 19, 2006 @ 9:34 AM | Reply

    • Re: interesting slides !
      DAC is double address cycle – a PCI mechanism by which a device with 32 address lines pushes out a 64 bit address in two cycles, low half first and then high half.

      Comment by mulix — January 19, 2006 @ 12:19 PM | Reply

  2. are you coming to the OS lecutre in the technion or not?
    ?

    Comment by Anonymous — January 25, 2006 @ 7:11 PM | Reply

    • Re: are you coming to the OS lecutre in the technion or not?
      Not this semester, no. Too many other obligations in the next month and a half.

      Comment by mulix — January 25, 2006 @ 8:55 PM | Reply

  3. 2 questions about DMA and Xen
    – thanks for that link.
    – I am a DMA ignorant.
    2 questions (if I may: little late, but better than never):
    1) is the DMA problem ** in XEN ** restricted to 64 bit processors ?
    will 32 bit processors work in Xen with no fear that one domain will DMA into memory of a second domain ?
    2) “The Xen unstable tree does not support any HW
    IOMMUs.”
    I am also IOMMU ignorant.
    Does the linux kernel for x86 ,IA-64 and x86_64 includes
    support for HW IOMMU ? in case it does – is it builtin and
    activated by default , or you should somehow activate (or patch the kernel
    ,etc)
    you say in the slides:
    “swiotlb and grant tables can be considered SW
    implementations of IOMMU functionality.”
    If I am not wrong,in IA64 (I am speaking about ordinary linux ,not Xen) swiotlb is used as an implementation
    of IOMMU.
    But: Can grant tables of Xen , with their shared pages mechanism, solve the problem of two domains
    which access **through DMA** the same address ?
    is such solution better than swiotlb in any aspect ? (which has it’s cost)
    DMA&IOMMU ignorant

    Comment by Anonymous — January 27, 2006 @ 7:54 PM | Reply

    • Re: 2 questions about DMA and Xen
      > 1) is the DMA problem ** in XEN ** restricted to 64 bit processors ?
      >
      > will 32 bit processors work in Xen with no fear that one domain will DMA
      > into memory of a second domain ?
      No, this is an architectural issue unrelated to the native word size. Note that both 32 and 64 bits processors suffer from this problem if granting *untrusted* domains direct HW access. Since Xen at the moment only grants trusted domains direct HW access, this is not a problem.
      > Does the linux kernel for x86 ,IA-64 and x86_64 includes
      > support for HW IOMMU ?
      x86 – no, IA-64 – no, x86-64 – yes. Again, this is architectural – there are no x86 and IA-64 machines with HW IOMMUs. There are x86-64 machines (AMD Opterons, for example) with a HW IOMMU.
      > in case it does – is it builtin and
      > activated by default , or you should somehow activate (or patch the kernel
      > ,etc)
      On x86-64 Opteron machines it is activated by default if necessary (e.g., you have more than 4GB of memory).
      > “swiotlb and grant tables can be considered SW implementations of IOMMU
      > functionality.”
      > If I am not wrong,in IA64 (I am speaking about ordinary linux ,not Xen)
      > wiotlb is used as an implementation of IOMMU.
      Right.
      > But: Can grant tables of Xen , with their shared pages mechanism, solve the
      > problem of two domains which access **through DMA** the same address ?
      > is such solution better than swiotlb in any aspect ? (which has it’s cost)
      grant tables attempt to address the isolation requirement, that is, letting one domain trigger a DMA into another domain’s memory, while doing it “securely”. swiotlb addresses different things (4GB DMA limited devices in Linux, same + non-machine-contigous DMAs in Xen).

      Comment by mulix — January 28, 2006 @ 5:31 PM | Reply


RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in:

Gravatar
WordPress.com Logo

You are commenting using your WordPress.com account. ( Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. ( Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. ( Log Out /  Change )

Cancel

Connecting to %s

Blog at WordPress.com.

%d bloggers like this: