Sparc Ultra60 U2P chip bug

[wen-king@myri.com (Wen-King Su)]

Hello all,

I have looked through sun web site but I couldn't figure out how to report
a hardware bug with the U2P chip.  I figure I may get in touch with
somebody in sun through this list. 

I have an ultra-60 clone from Tatung we received last week as part of our
64-bit PCI design effort.  A 64-bit device samples the REQ64- signal on
the cycle PCI_RST- becomes deasserted.  If REQ64- is found to be asserted,
the device assumes it is in a 64-bit slot.  If REQ64- is found to be
deasserted, the device assumes it is in a 32-bit slot.  I expect to see
something like this in a 64-bit slot:

   PCI_RST-:   0000001111111111111111
   REQ64-  :   0000000111111111111111

This is what is I saw instead: 

   PCI_RST-:   0000001000011111111111
   REQ64-  :   0000000111111111111111

Consequently, a spec conforming 64-bit interface in a 64-bit ultra60 slot
will decide it is sitting on a 32-bit bus, and will not generate 64-bit
DMA transactions.  Since both signals are generated from the U2P chip,
the problem most likely is in Sun's U2P chip, even if the machine itself
is a clone.

We are revising our design to add a software override register to force
the interface into 64-bit mode.  We would like to know if the latter
versions of the U2P chip, or the ones used in higher end ultrasparcs have
the same problem.

								- Wen