Date: Sun, 15 Aug 1999 14:35:08 -0600 (MDT) From: "Justin T. Gibbs" <gibbs@narnia.plutotech.com> To: Keith Mitchell <kmitch@weenix.guru.org> Cc: scsi@FreeBSD.org Subject: Re: CAM (driver devel) question Message-ID: <199908152035.OAA47448@narnia.plutotech.com> In-Reply-To: <19990811201425.A89846@weenix.guru.org> <199908120100.TAA51068@panzer.kdm.org> <19990811223315.A93634@weenix.guru.org>
next in thread | previous in thread | raw e-mail | index | archive | help
In article <19990811223315.A93634@weenix.guru.org> you wrote: > On Wed, Aug 11, 1999 at 07:00:44PM -0600, Kenneth D. Merry wrote: >> Not all of the drivers support S/G. And you're right, it isn't used in any >> of the upper level code at the moment. > > All of the drivers I looked at (aic7xxx, dpt, buslogic) all supported > scatter/gather. But it looked like they only supported scatter/gather if > they were passed a scatter/gather buffer from the upper layer (and since it > doesn't look like the upper layer sends down a scatter gather list then > I am guessing that scatter/gather never occurs). > > It looks like from the drivers they take whatever buffer they get passed > from the upper layer and pass it down to the card (if its a physical address > or maps it into DMAable memory via bus_dmamap_load() if its a virtual > address). So my initial conclusion was the drivers were architected so > that when the upper layer supported S/G they would too without modification. > but I wanted to be sure before I coded everything. When you perform a bus_dmamap_load(), you are returned a 'generic' S/G list corresponding to the virtual buffer address/size provided. The list conforms to all of the constraints listed in the DMA tag for the map. This is how 99.9% of all S/G will occur for your driver. Some applications may require operations on discontiguous chunks of 'bus addressed' buffers. In this case, the application will provide an S/G list that needs no further interpretation by the driver. I would not expect users of FreeBSD as a 'standard UNIX' to ever exercise this API, but it certainly has many uses in embedded environments. As for your other question about transaction descriptors that are DMA'ed to the card, these are usually handled by allocating DMA safe memory up front. Most of the drivers do this already. At some point in the future, we will likely add some additional calls to the bus dma API that allow you to explicitly sync DMA safe memory after you touch it. For most architectures this would be a no-op, but for those architectures that don't have bus snooping, these calls would flush the data cache. -- Justin To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199908152035.OAA47448>