[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: 32MB BAR Limitation with NT?
- To: Mailing List Recipients <pci-sig-request@znyx.com>
- Subject: Re: 32MB BAR Limitation with NT?
- From: "David O'Shea" <daveo@corollary.com>
- Date: Mon, 27 Jan 1997 21:46:23 GMT
- Old-Return-Path: <daveo@corollary.com>
- Resent-Date: Mon, 27 Jan 1997 21:46:23 GMT
- Resent-From: pci-sig-request@znyx.com
- Resent-Message-Id: <"0NiCM3.0.Bg.PLJxo"@dart>
- Resent-Sender: pci-sig-request@znyx.com
At 01:07 PM 1/27/97 -0500, Bob Goudreau wrote:
>Dave O'Shea wrote:
>
>> But, as I showed in the calculations above, the OS designers were not all
>> that "weird" thinking in choossing 32MB as a restriction if in fact it is
>> a restriction. 256MB for each device is way to much. There is just not
>> that much memory space around for people to design devices with huge
>> memory interfaces because it is either easier, or faster.
>
>But one could legitimately object to the inflexibility of the practice
>of having a *per-BAR* limit at all, versus an overall limit on the
>total amount of PCI Memory Space used by all the devices in the system
>together. It seems kind of a shame to "reserve" 32 MB for each
>Memory BAR and end up disqualifying a 64 MB BAR even though most or
>all of its peers probably use relatively tiny ranges of 1 MB or less
>(sometimes *much* less).
Well, it would appear that I was somewhat stale in my answer anyway.
The actual reason that NT was refusing the request had nothing to do
with memory limitations, but did have to do with PDE & PTE table entry
allocations, which was the second thing that I guessed. But the limit
was not on a per driver basis, but was instead caused because the system
was out of enough PDE's & PTE's to satisfy any request of that size at
all.
Fortunately, as someone pointed out, the number of PDE's & PTE's can be
increased in NT by using a REGISTRY key value override. So, NT does
not actually impose any limits on the allocations.
Personally, I think it might be wise to impose some limits, so as to
constrain designers of boards from going crazy with memory space allocations.
In the specific case being discussed, the 256MB of space was actually for
several different devices. It was video oriented, which was also one of
the only reasons that I can think of to have a very large slave driven
interface on PCI adapters.
As an aside though. Most of what I said in my earlier mail still goes.
Systems are being created with very large numbers of PCI slots. One that
Corollary has in the works has about 20 slots. Other manufacturers create
add-in expansion boxes that allow for even more slots. It will not be
uncommon for PCI machines to have 20-30 devices installed. As some of
my earlier calculations showed, you can only fit so many 64MB spaces into
a 2GB address space.
People should be keeping this in mind when they design their hardware.
Regards,
David O'Shea
daveo@corollary.com
Corollary Inc.
. < )