* Robert Lipe <[EMAIL PROTECTED]> [010122 12:50] wrote:
> Alfred Perlstein wrote:
>
> > > So if I sandbagged the allocs to be *larger* the KMA would behave more
> > > consistently with what I'd expect becuase it would then reclaim? I
> > > didn't see that one coming. :-)
> >
> > but... this is a terrible workaround, I'm not sure it would work
> > and you shouldn't do it. :)
>
> Good. I didn't actully *want* to do that. I still have to respect
> myself in the morning. :-)
>
> > > Is there an s/g memory interface in FreeBSD? This was my first choice,
> >
> > See the busdma stuff,
>
> Yes, this looks to be much closer to the interface I really wanted
> anyway. I see no man pages for them. Is there any doc anywhere? "Read
> the source and look at existing examples" will do if it must but any
> pointers to better doc are appreciated.
I know of no docs, I don't write drivers. (lucky me)
> > there's a problem because there's no busdma
> > for mbufs (well actually there is but it fails on really small
> > unaligned blocks which basically breaks it for mbufs).
>
> Can you tell me more about this hazard? I can't control the alignment
> that's asked for by the driver, but I can certainly fudge the size and
> alignment of the request by the time it gets to this interface. Just
> tell me the rules. :-)
I don't know, Bill Paul explained that the normal busdma stuff is
pretty broken for chunks too small. Basically, disks work, network
cards won't because mbufs are too small.
> > ack, I'd review patches if you wanted to write/port busdma for mbufs..
>
> Thanx for the offer, but I'm hoping to avoid that right now.
>
> It's very much a goal of mine (esp. now that we're in the final days
> of the NDA requirement) to stamp out something that works on shipping
> FreeBSD well enough for this crowd to see what remains to be done and
> see where we are. Then we can figure out how to improve interfaces or
> otherwise enhance the base OS if we want to. So given a choice between,
> say, incurring an extra copy in a data path and porting busdma for mbufs
> I'll take the former for now.
With mbufs you're stuck. I have some code that needs to send > PAGE_SIZE
chunks of data down the network stack, my current "hack" is that I
prebuild the mbuf packets and split them on page boundries, then just
call m_copym on the chain, but I doubt that'll help you.
--
-Alfred Perlstein - [[EMAIL PROTECTED]|[EMAIL PROTECTED]]
"I have the heart of a child; I keep it in a jar on my desk."
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
