Date: Thu, 06 Sep 2001 15:48:33 +0900 From: Hiroharu Tamaru <tamaru@myn.rcast.u-tokyo.ac.jp> To: FreeBSD-mobile@FreeBSD.ORG Cc: Warner Losh <imp@harmony.village.org> Subject: Re: Toshiba Libretto M3 with Planex FNW-3600-T Message-ID: <sa6elpk95mm.wl@ring.myn.rcast.u-tokyo.ac.jp> In-Reply-To: <200109052145.f85Ljsh48156@harmony.village.org> References: <200109052029.f85KTwh47645@harmony.village.org> <sa6elpmywdf.wl@ring.myn.rcast.u-tokyo.ac.jp> <sa6heui5m2q.wl@ring.myn.rcast.u-tokyo.ac.jp> <sa6itey6deu.wl@ring.myn.rcast.u-tokyo.ac.jp> <200109042357.f84Nvgh39637@harmony.village.org> <200109050545.f855jsh41602@harmony.village.org> <200109051647.f85Glgh45792@harmony.village.org> <200109052145.f85Ljsh48156@harmony.village.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Warner!
At Wed, 05 Sep 2001 10:47:42 -0600,
Warner Losh wrote:
> Just so I'm sure I have things correct. You use PCI interrupt
> routing, you insert the card, you get this "wedge". Is that right?
> And it only happens on this Planex card, right? I recall you saying
> you'd try other ethernet cards, but I can't seem to find the report.
>
> The wedge says that something isn't clearing interrupts. This may at
> the bridge level (eg in the ToPIC) or at the card level. Given that
> you say that other cards work, I have a slight bias towards a small
> problem with the ed driver. But that's a very small chance...
>
> I saw your register dump and will look at it when I arrive at the
> office and can consult my datasheets there. Maybe my boss will have
> his laptop in the office today, so I can try it there too.
At Wed, 05 Sep 2001 15:45:54 -0600,
Warner Losh wrote:
>
> In message <200109052029.f85KTwh47645@harmony.village.org> Warner Losh writes:
> : I did some poking into this crash hang. I found that you were lucky.
> : Sometimes I could eject the card and get a system back, while others
> : times I couldn't.
> :
> : It does look for all the world like the ed driver isn't properly
> : handling interrupts (as in clearing all interrupt sources) for this
> : card. I suspect that there's an interrupt bit in the mii goo that
> : isn't being cleared properly, so we're in infinite interrupt. I know
> : exactly 0 about ne2000 hardware, so I'll have to punt on this.
>
> I found if I put a sleep 15 at the start of /etc/pccard_ether, then
> I've been able to use my Corega card, where before it would hang every
> time.
Bingo! This exactly solved my problem too.
I put a sleep 5 in /etc/start_if.ed0 (I changed to ed0 after kernel
rebuild) and every thing works perfectly now.
Thank you so much!
So, it was more of the ed driver problem than pcic problem.. I
noticed that sleep 4 is enough after the boot phase is over, but 5 is
necessary if I have the card in while it boots; looks very likely a
network chip init problem, doesn't it?
Should I take this to freebsd-net? for ed folks, maybe?
For what it worth, I was able to test 3Com EtherLinkIII (3C589C) card.
Sorry, but it was busy and not available to me until now.
This is an "ep" card and there was no problem at all with this card.
By the way, I found a minor performance issue.
I connected my ethernet cards to a 100/10 full/half duplex switch,
which has a connection to a desk top machine with 100 full duplex link
(fxp0). I fired a tcpblast from Libretto to this machine several
times and saw how much speed I can get. The desktop machine and the
switch are not busy at all, and I tried this several times, so I
believe the results are not meaningless.
I tried with July 10 kernel, 44RC kernel with ISA routing, 44RC kernel
with PCI routing.
44RC-PCI 44RC-ISA 43STABLE
ep0( 10Mbps half dpx) 7.4 7.7 7.5
ed0(100Mbps full dpx) 8.8 9.1 9.4
figures are in units of Mbps.
I am aware that *PCMCIA* 100Base/TX cards aren't that much better than
10Base cards, and for tcpblast half/full duplex doesn't matter so
much, so those aren't my point.
I am curious why should it get slower as we evolve from
ISA based, polling -> PCI based, polling -> PCI based, use intr ?
I thought those changes should increase preformance, since it utilizes
more of the recent speeded up hardware..
May be I am not understanding the changes from 4.3-stable to this new
PCI code at all, but I mention just because I noticed it.
Well, anyway, thanks for all the work.
Now my Libretto is ready for 4.4-RELEASE.. ;-)
--
Hiroharu Tamaru
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-mobile" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?sa6elpk95mm.wl>