Subject: USB drives, contiguous kernel memory.
To: None <netbsd-help@netbsd.org>
From: Richard Rauch <rkr@olib.org>
List: netbsd-help
Date: 07/05/2005 16:30:34
Hi.
As many, I've occasionally wanted to hook up a USB disk-like device to
NetBSD's umass driver. Generally, this has been pretty rare, so I've
been able to use the nasty hack of just rebooting the machine with the
drive attached.
However, some time back (6 months? a year? more?) someone posted that
the problem was due to kernel memory fragmentation. The USB umass
driver wants a fairly small (compared to half-gig or so memory spaces)
chunk of contiguous kernel memory. And that much memory is hard to
find after the kernel has been up for a while.
I recall someone had a little program that could cause the kernel
to drop unused buffers, effectively freeing enough contiguous space
(maybe not with 100% guarantees, but at least with high probability)
for a USB drive to be attached.
I've spent a little while looking for it, first via the mailing
list search, then by simply walking up and down the lists using
links's search to find promising subject lines (anything mentioning
USB). I've found several descriptions of the problem, but have
yet to rediscover the workaround to force the kernel to make
buffers available.
Can anyone point it out to me?
I seem to recall that someone had a way to do it just via dd from/to
a suitable source, or creating a large enough file and doing something
trivial with it. But that could be a bogus memory.
(Aside: Once again, it would be *really* *nice* if the old mbox
archives were once more downloadable. If I didn't have to use the
cumbersome web-interface, I could probably find what I'm looking
for without asking. (^& But I've looked for 15 or 30 minutes in
the web interface, crawling up and down links, and for all I know a
sloppy keypress has already made me step past what I want...)
--
"I probably don't know what I'm talking about." http://www.olib.org/~rkr/