Date: Tue, 27 Aug 1996 10:30:27 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: hal@post.vale.com (Hal Snyder) Cc: koshy@india.hp.com, terry@lambert.org, freebsd-hackers@freebsd.org Subject: Re: SCSI vs IDE (was: Multiple swaps...) Message-ID: <199608271730.KAA25132@phaeton.artisoft.com> In-Reply-To: <01BB940E.2E62F1F0@jaguar> from "Hal Snyder" at Aug 27, 96 11:52:10 am
next in thread | previous in thread | raw e-mail | index | archive | help
> What about DMA on working IDE controllers? Is it the same kind of > DMA as on high end SCSI controllers? If by "working IDE controllers", you mean the 40% of them out there that don't have the "silently trash data" bug, or the 20% that don't have any known bugs at all, then yes, this is exactly what it is, and is exactly why it's desirable. "I'd like my hardware to work really fast and validate my decision to buy it" is a common complaint, from cars to toasters. 8-). > Correct me if I'm wrong here (have been away from kernel and driver > hacking a couple years), but Adaptec uses "First Party" or "Bus > Mastering" DMA - which is allegedly better than "Third Party" DMA > going through the DMA controller on the motherboard. Yes. And it is only better if the OS is capable of having multiple outstanding requests going simultaneously. So most DOS-only vendors are only now having NT drivers, and a smaller percentage of Windows95 drivers, shoved in their face over it. > Not all SCSI controllers are alike. I wrote a unix device driver for > the old Seagate/Future Domain SCSI controllers. They were basically > a parallel port, so you had to do most of the SCSI protocol in software. > I never did get "asynchronous" mode to work reliably, i.e. start an > i/o transfer on one device, disconnect, start another transfer on a > second device, etc. I suspect that the SCSI controllers that appeared > glued onto sound boards a couple years ago were equally nasty. Most of them are AIC 6360's -- Adaptec 1520/1522 equivalents, with no BIOS, so you can't boot from them. > [I changed the subject of the thread. A defect of the mailing list system > is that the title propagates forever while the discussion wanders from > memory wait states to demon plushies...] It's a problem with synchronizing the contents with the label in the thread participants minds. Like early Intel PCI chipsets, there is no cache invalidate pin on humans. 8-). Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199608271730.KAA25132>