Date: Fri, 31 Jan 1997 06:20:35 -0500 (EST) From: Peter Dufault <dufault@hda.com> To: dg@root.com Cc: joerg_wunsch@uriah.heep.sax.de, freebsd-scsi@freebsd.org Subject: Re: XXXminpys question Message-ID: <199701311120.GAA12564@hda.hda.com> In-Reply-To: <199701310339.TAA27523@root.com> from David Greenman at "Jan 30, 97 07:39:22 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
> >I remember that David Greenman once said that the main reason for the > >existing minphys was the limitation of the SCSI adapters. > > Right, it has to do with the number of scatter-gather entries that the > controller can have per-DMA. Given that the memory isn't locked down yet and may not be mapped to physical memory, and that the maximum size of the scatter-gather list is dependent on the fragmentation of the physical memory, this problem is a bit thornier than at first glance. Is there a problem with considering minphys to be a guarantee to do the I/O so that a side effect of minphys is to lock down the pages and build the scatter-gather list using some published entry points? Then you can do system minphys to limit to the system max, driver minphys which pokes in its resources, locks down, and builds the scatter-gather and other parts of the pending transaction, then continue on through the physio lockdown which will either trivially succeed or figure out that it was already done, and finally do the I/O. Peter -- Peter Dufault (dufault@hda.com) Realtime Machine Control and Simulation HD Associates, Inc. Voice: 508 433 6936
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199701311120.GAA12564>