Subject: Re: M_EXT_PAGES
To: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
From: Jonathan Stone <jonathan@dsg.stanford.edu>
List: tech-kern
Date: 06/05/2004 15:38:50
In message <1086421637.804118.18498.nullmailer@yamt.dyndns.org>YAMAMOTO Takashi
writes
>--NextPart-20040605163633-1799000
>Content-Type: Text/Plain; charset=us-ascii
>
>hi,
>
>currently, ext_pgs is merely used as a hint for sodopendfreel().
>i'd like to change it to be used by bus_dmamap_load_mbuf() as well.
>(x86 diff is attached)
Uh, wait a minute: I seem to recall that short-circuiting pa lookups
in bus_dmamap_load_mbuf() was the reason Jason added ext_pgs in the
first place? Did that last part never quite happen?
>while i think that the change itsself has its own benefits,
>eventual goals are:
> - delay pmap_kenter_pa() until it's really needed.
> (likely on the first mtod()?)
> if you're lucky enough, no pmap_kenter_pa() is needed at all.
> - use mbufs (or something similar) for disk i/o.
>
>comments?
mbufs for disk I/O? You mean something like what OSF1/Digital
Unix/Tru64 called UMC? "Yes, please". That can help a *lot* for
high-performance NFS, or for sendfile().