From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: Mystery packet killing tg3 Date: 5 May 2005 20:01:06 +0200 Message-ID: <20050505180106.GA24386@muc.de> References: <20050502162405.65dfb4a9@localhost.localdomain> <20050502200251.38271b61.davem@davemloft.net> <42791825.2080204@pantasys.com> <20050505114327.GA51761@muc.de> <427A47C9.6040302@osdl.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Peter Buckingham , "David S. Miller" , jgarzik@pobox.com, netdev@oss.sgi.com Return-path: Date: Thu, 5 May 2005 20:01:06 +0200 To: Stephen Hemminger Content-Disposition: inline In-Reply-To: <427A47C9.6040302@osdl.org> Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Thu, May 05, 2005 at 09:20:25AM -0700, Stephen Hemminger wrote: > Andi Kleen wrote: > > >>>IOMMU code on x86-64 should be never active unless Stephen > >>>used IOMMU_DEBUG or iommu=force. THat is because the tg3 > >>>is a 64bit capable device and should always use bypass. > >>> > >>> > >>how is this detected? we have a board that is using a 32bit e1000 and we > >>have to use iommu=force otherwise it doesn't do the right thing with > > >>4GB of RAM. > >> > >> > > > >"32bit e1000"? How did you get such a beast? AFAIK all e1000s are 64bit > >address capable. Please supply a full boot log without iommu=force > >and describe what happens exactly. > > > > > One of the osdl e1000's in the lab is an IBM rebranded card that although > it uses a 64bit slot, is really only 32bit Ok, but it still supports DAC. The PCI bus width is a different thing from the size of the address registers. What we are talking about here is just if the card can directly address physical memory above 4GB. What width the bus is is a hardware internal thing that we do not really care about in the IOMMU code. -Andi