[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
WIndowsNT "snoop" cycles before doing DMA
- To: Mailing List Recipients <pci-sig-request@znyx.com>
- Subject: WIndowsNT "snoop" cycles before doing DMA
- From: "Lange, Michael" <lange@dvs.de>
- Date: Thu, 25 Mar 1999 09:02:25 +0100
- Delivered-To: pcisig@teleport.COM
- Resent-Date: Thu, 25 Mar 1999 00:37:57 -0800
- Resent-From: pci-sig-request@znyx.com
- Resent-Message-ID: <"ueDmc3.0.mL6.UsU-s"@electra.znyx.com>
- Resent-Sender: pci-sig-request@znyx.com
> PCI experts,
>
> we have some PCI based plug-in boards which run under WindowsNT4.0
> using a P2 hardware platform (LX,HX,FX and BX INTEL chips sets).
>
> We have 2-3 PCI add-in cards which have directly mapped memory on it
> (up to 32 MB) and we transfer data using a DMA directly between the
> boards
> (NOT going over system memory).
>
> The DMA is started by calling NT to generate all the necessary driver
> calls to do a DMA of (say 256 kByte) directly between 2 PCI boards on
> the same
> bus segment.
>
> Basically this is running.
>
> BUT : we observed that BEFORE the DMA hardware engine is running there
> are single
> cycle Read/Writes every 4 kByte (= page size of Windows NT)
> exactly to the physical
> memory address to be used as the destination address for the
> DMA.
>
> Questions :
>
> - We guess that these cycles are issued by the NT kernel (it seems to
> check the
> directly mapped memory before the DMA hardware engine is started).
>
> - Does somebody else made the same experience as we did ?
>
> OK, this is a question which more related to WindowsNT and not to PCI.
> Sorry for this. But perhaps somebody can help us because these
> "snooping" RW cycles
> produce a performance reduction and leads to a large RW activity (mix
> of DMA cycles and
> "snooping" RW cycles on the PCI bus) with a lot of side effects.
>
>
> Michael Lange, Hardware design enginner, DVS
>