Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 23 Aug 1997 10:18:16 +0200
From:      Stefan Esser <se@FreeBSD.ORG>
To:        Terry Lambert <terry@lambert.org>
Cc:        Satoshi Asami <asami@cs.berkeley.edu>, Shimon@i-connect.net, tom@sdf.com, hackers@FreeBSD.ORG, mrcpu@cdsnet.net, nate@mt.sri.com
Subject:   Re: Final request for help with release.  (DPT boot floppy)
Message-ID:  <19970823101816.47978@mi.uni-koeln.de>
In-Reply-To: <199708221807.LAA26407@phaeton.artisoft.com>; from Terry Lambert on Fri, Aug 22, 1997 at 11:07:13AM -0700
References:  <199708221157.EAA00987@silvia.HIP.Berkeley.EDU> <199708221807.LAA26407@phaeton.artisoft.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Aug 22, Terry Lambert <terry@lambert.org> wrote:
> > I have also seen mysterious lockups (for a few seconds) with Solaris
> > x86 when I bombarded more than two twin-channel controllers (Adaptec
> > 3940W/UW) with lots of requests.  FreeBSD had no problems whatsoever.
> 
> This is probably a PCI issue.  Most Intel PCI bridges do not
> correctly arbitrate between more than two bus mastering PCI
> devices simultaneously.
> 
> Stefan posted about this a while back.

Ummm, I don't remember doing so, actually ...

I sent out a reminder about the problem early 
chip-sets had with multiple PCI bus-masters.

The Triton-I seems to be OK, and the later ones 
too, but everything before may fail under high 
load.

There are also some design flaws in PCI 2.0,
which have been resolved in 2.1, and which can
lead to a bus lockup (or delivery of wrong data)
in systems with bus-masters on the PCI bus and
PCI to PCI bridges.

Don't use early PCI to PCI bridges in heavily
loaded systems, or you may suffer from crashes
or system hangs.

> I think there's a newer chipset that actually supports any number
> of devices (up to the number of slots, anyway), but I don't remember
> the name of the thing off the top of my head.

The PCI bus arbiter limits the number of bus-
master PCI devices on a single bus. Since an
arbiter comes with each host to PCI bridge and
PCI to PCI bridge, most motherboard and card 
designs use those, and thus there is a chip-set
specific (or PPB specific) limit on the number
of bus-master slots (often 4).

Since there are typically 4 or more slots, and
the PCI to ISA bridge counts as a PCI bus-master,
you can not expect that 4 bus-master cards will 
work on the motherboard ...

(Early PCI motherboards often had dedicated bus-
master and slave PCI slots, but this is no longer 
the case. The bus-master slots are dynamically 
assigned, but still limited in number.)

But one is a VGA card, usually, and thus there
can be only 3 others in a 4 slot system, and you
will never be hurt by the limit :)

Regards, STefan



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970823101816.47978>