Subject: Re: Patch for two mbuf dma optimizations
To: Jason Thorpe <firstname.lastname@example.org>
From: None <email@example.com>
Date: 03/30/2003 13:06:25
> On Saturday, March 29, 2003, at 02:08 PM, Ignatios Souvatzis wrote:
> > Isn't the VAX page size 1k?
> The VAX hardware page size is actually 512 bytes, but the port presents
> a 4K page size to the OS.
Historical note for curious people: > 20 years ago when the new VM
with virtual memory support was written (in 3BSD) something called
"page clustering" was invented; this was the ability for the "MI"
code to handle the underlying hardware with a larger virtual page
size than it had originally. VAX used 1K logical pages then (CLSIZE==2)
and it was "inherited" into NetBSD when I started working with it.
The per-logical-page data structures in NetBSD is quite large so it
required a large amount of memory (60 bytes per 1k page), and the
binaries had grown large causing unneccessary page faults, so I
changed it to 4K some years ago.