Re: bad lmbench numbers for mmap

V Ganesh (ganesh@vxindia.veritas.com)
Mon, 19 Apr 1999 03:48:54 +0530 (IST)


> From: David Miller <davem@twiddle.net>
>
> I believe they have the same buffercache/pagecache issues we do.
> If you look in the Solaris header files (/usr/include/sys/buf.h to be
> exact) they have a page struct pointer in the buffer structure which
> seems to support my theory. It's just a theory, I could be wrong.

well if it is similar to SVR4 then I don't think so. when we want to
do IO to a page we create buffer heads which are _not_ part of the
buffer cache and point them to the page. likely they do something
similar, and the page struct pointer exists so that lower level routines
can map the page into kernel VM.
the only reason why they might want to do a copy is to write out a
consistent snapshot of the page, because users could happily keep changing
the page as it is being DMA'd or whatever to disk (is this a problem at
all ? I guess not, because mmap semantics promise nothing of the kind - and
for write() you have to do the copy to kernelspace anyway).

ganesh

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/