We are experiencing an intermittent issue with the TMS320C6412 DSP used as a PCI target. Occasionally upon power-on, the DSP's PCI interface does not initialize properly and causes the host PC to lockup upon communication. Here is the scenario:
The PCI bus gets into "permanent retry" mode which locks up the PC (no TRDY#). According to the documentation, the DSP's PCI interface causes a retry if its internal PCI FIFO buffer is full. Internal DSP memory transfers are supposed to continue until the FIFO is empty. Unfortunately, it appears that the FIFO is not being emptied upon getting into a bad state due to some power-on sequence.
We found two ways to improve this problem:
1) holding the power-on hard reset low for 4 seconds after the power supplies have stabilized makes the problem much less frequent.
2) Writing the "Warm Reset" bit in the HDCR register after power-on fixes the problem (without the 4 second wait).
Some questions:
1) Is there any way to detect a bad PCI state from either the DSP or the PCI side?
2) Is there any way to issue a warm reset from the DSP side? The memory-mapped HDCR register is host-write only.
3) Is there any way to soft reset the PCI interface separately from an entire warm reset?
4) Any other thoughts on root cause correction?
Thanks,
Chris
