Date: Sun, 19 Sep 1999 12:22:50 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Poul-Henning Kamp <phk@critter.freebsd.dk> Cc: Matthew Jacob <mjacob@feral.com>, dg@root.com, Greg Lehey <grog@lemis.com>, cvs-committers@FreeBSD.ORG, cvs-all@FreeBSD.ORG Subject: Re: User block device access Message-ID: <199909191922.MAA74009@apollo.backplane.com> References: <17519.937768020@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
:In message <199909191900.MAA73792@apollo.backplane.com>, Matthew Dillon writes:
:
:> In fact, a memory-mappable buffered block device with write-through would
:> be much, much more useful to a database then a character device, and I
:> think it's only a two line patch to make mmap() work, and probably a
:> four line patch to implement write-through. It would be virtually
:> unbeatable... use of mmap() removes the extra copy overhead, read-caching
:> takes the burden off the application, optional write-through gives you
:> instant feedback *AND* reblocking. I would even be willing to make the
:> write-through the default.
:>
:> That would give us an extremely powerful and useful buffered block device
:> implementation.
:
:So you are saying that we could basically leave read/write as they
:are for cdevs and provide the buffer/cache mechanisms with mmap ?
:
:--
:Poul-Henning Kamp FreeBSD coreteam member
:phk@FreeBSD.ORG "Real hackers run -current on their laptop."
We cannot really implement mmap for cdev's that aren't direct memory
maps (such as video frame buffers), at least not without breaking
cache coherency.
We *can* implement mmap for bdev's and we *can* change the way writes
to bdev's work.
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199909191922.MAA74009>