[plug] [OT] Server Hardware "shops"

[Andrew Cooks]

On Tue, Mar 20, 2012 at 5:29 AM, Tim White <weirdit at gmail.com> wrote:

> In the near future I'm looking at building my own server that will
> eventually take upto maybe 8 HDDs for storage requirements, as well as
> being a server for a number of functions.
>

Resurrecting the dead, but this is a cautionary tale:
If the other functions you mention involve virtual machines with VGA
pass-though, like for a media centre for example, Marvell SATA controllers
will cause problems and you'll have to disable either the SATA
controller or VT-d (and therefore any device pass-though). There has been
some discussion about this on LKML and it seems like a pci quirk patch
could be created, but I haven't found one yet. This will also affect PCIe
cards based on the same controller, afaik.

The Marvell 88SE91xx series of chips initiate DMA from a different PCI
address from the one they expose to the kernel.  lspci reports three
controllers:
05:00.0 SATA controller: Marvell Technology Group Ltd. 88SE9172 SATA 6Gb/s
Controller (rev 11)
06:00.0 SATA controller: Marvell Technology Group Ltd. 88SE9172 SATA 6Gb/s
Controller (rev 11)
0b:00.0 SATA controller: Marvell Technology Group Ltd. 88SE9172 SATA 6Gb/s
Controller (rev 11)

But DMA write comes from 05:00.1, 06:00.1 and 0b:00.1 which causes a DMA
remapping fault. dmesg says:
DRHD: handling fault status reg 2
DMAR:[DMA Write] Request device [05:00.1] fault addr fffe0000
DMAR:[fault reason 02] Present bit in context entry is clear

So in my case, on a GA-X79-UD5 motherboard, I lose 6 SATA ports, or I don't
get PCI pass-through.

Cheers!

A.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.plug.org.au/pipermail/plug/attachments/20120409/41f741b1/attachment.html>