Specific data send to Bigfoot Killer 2100 memory gets dropped by MPC8314E's CSB PCIe Bridge

[DvdBoon]

Specific data send by the 68K to Prometheus over the PCIe bridge to the PCIe Bigfoot card gets dropped.

This is not address specific but data specific and 100% reproducable.

The following error bit is flagged: PEX_CSB_IBSTAT_BRGER = 1

For byte transfers, the following data is lost and the MPC8314E indicates an error on CSB inbound transfer (tested from $00-$C0, i guess I should test till $FF):

$31, $32, $34, $37, $38, $3B, $3D, $3E

Word transfers and longword transfers are harder to predict which ones get dropped, but at least a very vital longword gets dropped which is used in the reset procedure:

$52535445

Old data remains in the memory location the write was supposed to go. It does not lock the card and data can be continued to be send.

Another example; At a location the inital data is 0,0,0,0 in longwords. Writing the following longwords:

$806001E4, $38630001, $906001E4, $4BFFFFF4

Gives the following result: $806001E4, $00000000, $906001E4, $4BFFFFF4

With further tests just writing $38630001 anywhere in the address range, the results are always that this longword gets dropped with a CSB inbound error.

[DvdBoon]

I am pretty sure this is a HW issue. But just putting it here just in case.

[DvdBoon]

Data is not dropped when using Xilinx chips with vhdl firmware. Checking to see if we can update/fix the Altera firmware.

Apparently also the pci variant of M1/K1 work with the Xilinx firmware without crashing in Warp3D

[DvdBoon]

E3B will be providing the source of the Altera firmware. Hopefully a solution will be found