Re: BIOS Bus Master Option

[Tony Clark <clark@MGI.com>]

I've been told that the BIOS folks disabled bus mastering during a soft
reset as it is the correct thing to do according to the white paper.  They
then had the problem that lots of stuff didn't work because the drivers did
not turn bus mastering back on.  To work around that, the added the options
in the BIOS to allow you to force bus mastering enable.

Now the problem:  Under windows NT you normally do not deal directly with
the hardware because NT is designed to hide all of that.  We currently use
the BIOS to force the bus master bit to be enabled.  Anybody have an opinion
on what we should be doing?  Perhaps NT has a function that we do not know
about that someone could clue us in on.

Thanks in advance,

Tony Clark

		-----Original Message-----
		From:	Olaf Birkeland [mailto:Olaf.Birkeland@fast.no]
<mailto:[mailto:Olaf.Birkeland@fast.no]> 
		Sent:	Wednesday, December 08, 1999 6:01 AM
		To:	Mailing List Recipients
		Cc:	Lloyd Bircher
		Subject:	RE: BIOS Bus Master Option



		> -----Original Message-----
		> From: Lloyd Bircher [mailto:nicad2@yahoo.com]
<mailto:[mailto:nicad2@yahoo.com]> 
		> Sent: 7 December, 1999 17:22
		> To: Mailing List Recipients
		> Subject: BIOS Bus Master Option
		>
		>
		> I am having an issue with a legacy PCI configuration
		> in an x86 PC. There is an option in many BIOS setups
		> for disabling Bus Mastering for ALL PCI devices in a
		> system. The original purpose of this option was to
		> allow broken Bus Masters to function. However, this
		> setting does not selectively disable problem adapters,
		> it disables Bus Mastering on ALL PCI devices.
		>
		> Does anyone know a good reason why this option should
		> exist in a modern PCI system? Are there many cards
		> that have broken BM capabilty?
		>

		I don't quite understand the issue on broken bus masters,
but this setting
		might be related to a PCI SIG white paper on "soft reset"
problems in PCI
		systems.

		Since soft reset does not trigger a PCI reset in all
systems, bus master
		disabling is required to prevent a bus master with a pending
access from
		before the reset to start an unintended operation after the
reset (since the
		soft reset was not observed by the bus master device). If
the BIOS assert
		PCIRST# also on soft reset, then explicit bus master
disabling is not
		required as this is part of the PCI reset. In both cases,
all PCI bus
		masters will and should be disabled when BIOS POST
completes.

		Likewise, expansion ROM/device driver/operating system
should always assume
		that bus mastering is initially disabled, and enable each
through the PCI
		header after proper initialization.

		Regards,
		- Olaf