[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

SubVendorID, take 2

In the next version of Windows 95 (aka memphis), we now use both
SubVendorId and revision has part of the devnode ID name. This is so
that if the user remove a card made by ABC and replace it (in the same
slot) with a card by DEF, both using XYZ chipset, we do notice the swap.
We wouldn't in windows 95 even with SubVenDorId, as it was just use for
the driver selection and device name, not for the instance creation.

However, we are seeing a number of PCI cards having a writeable
SubVendorID and being written by the BIOS of those cards. This create a
severe problem in memphis:

A secondary video card is normally not posted by BIOS.
We read the SubVendorId to create the devnode: at this point it is 0.
We realize the device is new, so we wait until GUI and start the
installation process.
We search the INFs (and prompt for floppy) based on hardware IDs that
have a 0 SubVendorId. So we install driver based on just the
The driver tells VDD whether mode switch requires int 10h or not.
If the driver require int 10h, we do some magic to enable the card and
post the ROM in V86 mode.
	that ROM then populate the SubVendorId, about a minute too
	On a subsequent reenumeration (user hit refresh in device
manager, machine docks/undock/suspend, user install a printer, etc)
CONFIGMG gets told to hard remove
PCI\VEN_{XYZ}&DEV_{XYZ'}&SUBSYS_000000000&REV_{XYZ"}\... and create a
brand spanking new devnode with ID
PCI\VEN_{XYZ}&DEV_{XYZ'}&SUBSYS_{DEF}&REV_{XYZ"}\... That second devnode
gets installed, most likely surprising the user in the process, and from
then on, when the machine boot we load one instance of the driver and
replace it with another instance when reenumeration of the PCI bus
occur.This leads to many problems.

The solution for memphis is to have a list of chipsets that exhibit this
behavior (SubVendorId being written by ROM post code). For these cards,
I will make an INF entry to override the use of SubVendorId, considering
it to always be 0. I encourage you to send hardware to Nick Dimmitt so
that we can test cards with this behavior.

Please reply to me with the Vendor ID/Device ID of cards that have the
SubVendorId being written by ROM post code, so that I can create this

Pierre-Yves Santerre
Win 9x PnP Dude

Nick Dimmitt
Building 27 South, #3080
1 Microsoft Way,
Redmond, WA