Petersson, Mats
2006-Feb-02 16:24 UTC
RE: [Xen-users] Xen 3.0.0 32bit-pae (testing changeset 8270) crashes(pgtable.c:284, kernel bug?)
> -----Original Message----- > From: xen-users-bounces@lists.xensource.com > [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of Matt Ayres > Sent: 02 February 2006 16:06 > To: Ian Pratt > Cc: Keir Fraser; ian.pratt@cl.cam.ac.uk; Ralph Passgang; > xen-users@lists.xensource.com > Subject: Re: [Xen-users] Xen 3.0.0 32bit-pae (testing > changeset 8270) crashes(pgtable.c:284, kernel bug?) > > Ian, > > I''ll buy that the 3w-xxxx driver sucks and it''s best to have > a controller than that can do DMA above 4GB. The problem is > the 3w-xxxx series has been one of the best RAID controllers > on the market for the past few years and many, many people > have them. > Even if it''s through a bunch hacks, hardware that can''t do > DMA above 4GB needs to work reliably. This is crucial for > Xen to be "Enterprise-Grade Open Source Virtualization". > > (just a casual observation from someone who spends day in and > day out working with Xen)This is why Mark Langsdorf (and probably others) have been working on an IOMMU driver for Xen, to at least support AMD64 solutions for this type of problem [unfortunately, the AMD IOMMU solution doesn''t work on Intel machines due to different hardware architecture :-( ]. And it''s not just 3Ware that has this problem. Many, many PCI DMA-engines are not designed to access memory above 4GB. I''d say almost all DMA cards that haven''t had the actual PCI interface redesigned in the last two years or so, with a few exceptions on high-end SCSI and Ethernet controllers, would have a 32-bit limit for it''s physical address. The PCI spec supports both 32- and 64-bit addressing, but it''s optional for the chip designer to implement more than 32 bits of addressing. -- Mats> > Thank you, > Matt Ayres > > There''s all manner of crappy hardware that can''t do DMA > above 4GB. The > > other problem that caused you to upgrade to 3.0.1 was 3ware > specific, > > a workaround for the driver being lazy. > > > > I believe some newer 3ware controllers are capable of DMA > above 4GB, > > though I don''t know specific model numbers. 3ware tech support are > > actually pretty responsive, so I''m sure they can help you. > > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Matt Ayres
2006-Feb-02 16:30 UTC
Re: [Xen-users] Xen 3.0.0 32bit-pae (testing changeset 8270) crashes(pgtable.c:284, kernel bug?)
Petersson, Mats wrote:>> > This is why Mark Langsdorf (and probably others) have been working on an > IOMMU driver for Xen, to at least support AMD64 solutions for this type > of problem [unfortunately, the AMD IOMMU solution doesn''t work on Intel > machines due to different hardware architecture :-( ]. >I''ll definitely be on the lookout for that, I use Opterons (but in 32-bit mode, of course). Thank you, Matt Ayres _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Javier Guerra
2006-Feb-02 16:54 UTC
Re: [Xen-users] Xen 3.0.0 32bit-pae (testing changeset 8270) crashes(pgtable.c:284, kernel bug?)
On Thursday 02 February 2006 11:24 am, Petersson, Mats wrote:> This is why Mark Langsdorf (and probably others) have been working on an > IOMMU driver for Xen, to at least support AMD64 solutions for this type > of problem [unfortunately, the AMD IOMMU solution doesn''t work on Intel > machines due to different hardware architecture :-( ].is this part of Pacifica? or should current opterons work with that? -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Muli Ben-Yehuda
2006-Feb-02 20:28 UTC
Re: [Xen-users] Xen 3.0.0 32bit-pae (testing changeset 8270) crashes(pgtable.c:284, kernel bug?)
On Thu, Feb 02, 2006 at 11:54:30AM -0500, Javier Guerra wrote:> On Thursday 02 February 2006 11:24 am, Petersson, Mats wrote: > > This is why Mark Langsdorf (and probably others) have been working on an > > IOMMU driver for Xen, to at least support AMD64 solutions for this type > > of problem [unfortunately, the AMD IOMMU solution doesn''t work on Intel > > machines due to different hardware architecture :-( ]. > > is this part of Pacifica? or should current opterons work with that?Current Opterons use the gart for IOMMU support in mainline Linux, but only if the kernel is compiled for x86-64. Mark posted a patch to utilize the gart in Xen as well, but it hasn''t been applied yet. So yes, this should work with current Opterons in Linux todya, and in Xen once Mark''s patch is applied. Cheers, Muli -- Muli Ben-Yehuda http://www.mulix.org | http://mulix.livejournal.com/ _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users