Date: Thu, 6 Nov 2003 08:25:41 -0500 (EST) From: Andrew Gallatin <gallatin@cs.duke.edu> To: Fred Clift <fclift@verio.net> Cc: freebsd-alpha@freebsd.org Subject: Re: Miata MX5 and 64 bit gigabit ethernet Message-ID: <16298.19413.772353.515698@grasshopper.cs.duke.edu> In-Reply-To: <20031105103201.B6336@irfcn.qzm.berz.irevb.arg> References: <20031102022925.W28113@loafier.com> <20031104155314.W3281@irfcn.qzm.berz.irevb.arg> <16297.2035.310311.968335@grasshopper.cs.duke.edu> <20031105103201.B6336@irfcn.qzm.berz.irevb.arg>
next in thread | previous in thread | raw e-mail | index | archive | help
Fred Clift writes: > On Wed, 5 Nov 2003, Andrew Gallatin wrote: > > > > > Fred Clift writes: > > > > > If the card does dma, then it will be unlikley to work. I dont know that > > > Gigabit ethernet existed when miatas were new, and hence, DEC wouldn't > > > > Please get a clue or stop spreading FUD. > > My mistake -- my only experience with 64-bit slot incompatabilities is > with 32-bit pci video cards, where using the 64-bit slots definitely > affect performance (ie occasional machine checks) - at least with 3 > different maxtrox cards I tried (Millenium, M-II, and Mystique) FWIW, all "normal" video cards (eg, those without in-kernel drivers like some Nvidaa cards) use memory mapped IO rather than DMA. Eg, the X server writes to / reads from PCI addresses mapped into its address space. Early alphas (like miatas) are very sensative to loads and stores from PCI addresses which have no physical device backing them. Other platforms, such as x86 PCs and ev6 (and later) alphas are much more forgiving, and return -1 for loads and ignore stores. Early alphas deliver a machine check and crash the machine. I'm not sure how a 64 bit slot could affect this, however its conceivable that there is some bug in how we mmap device memory, and the different addresses that the SRM assigns to the memory regions of the cards when they are in 64-bit slots tickle the bug. Drew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?16298.19413.772353.515698>