flood of downstream messages of type 0x7F (vendor defined)

flood of downstream messages of type 0x7F (vendor defined)

We develop an ExpressCard to PCI expansion product. Several of our customers have found a performance issue when using our solution with platforms using the 82801G (ICH7 Family) chipset. Comparing the commonalities of several different issues it appeared as if the problem was related to interrupts. A customer provided us with a card that interrupted at a known rate, and an application that serviced the interrupt and compared the time between interrupts to the known rate. OnIntel platforms that displayed the problem the application would fail when an interrupt was not serviced within a specific time. We used an Agilent PCI-Express analyzer to capture the PCI Express traffic while the application ran. What we found was that on the platforms without the problem, only the expected TLPs were found (the interrupt, ack and a few reads and writes to the board). However on the problematic platforms in addition to the expected TLPs there was a flood of downstream messages of type 0x7F (vendor defined) that we have no explanation for. This flood of TLPs is causing huge latency on mainly Dell platforms.The interrupts do eventually get cleared, but by this time it is too late.I have attached an excel file containing the trace.

Thanks in advance for any information.

--Julia

AttachmentSize
Download tlp.zip667.1 KB
1 post / 0 new
For more complete information about compiler optimizations, see our Optimization Notice.