Project

General

Profile

BHA5E dev kit: J10 M.2 devices stuck in Polling.Active (no link)

Added by Brady Heater 2 days ago

Hello,

We’re bringing up PCIe devices on the M.2 slot (J10) of our MitySOM-A5E dev kit (A5ED-B64-144-SRC-X, R0 silicon), using the GTS PCIe root port in the FPGA. The Broadcom 9660-16i HBA previously trained and enumerated on J10 through an M.2-to-PCIe adapter. Now neither an x4 NVMe SSD plugged directly into J10 nor a Renesas uPD720202 USB 3.0 card (SIIG JU-P20612-S1, on an externally powered adapter) will link up.

What we see in the root port’s ss_app_ltssmstate (via our status register):

Empty slot: LTSSM 0x00 (Detect.Quiet).
With the NVMe or the USB card: stuck at 0x02 (Polling.Active), with no link-up and no data-link-up.
So the root port detects the device’s receivers, but the device never sends training sequences back. The Linux altera-pcie probe then fails with -110, because the HIP mailbox doesn’t respond while the link is down.

Already ruled out:

Our FPGA PERST# is released 100 ms after configuration (open-drain on BU109 with R77 pull-up). Driving it push-pull high made no difference.
The FPGA design matches the build where the HBA trained. The refclk input is AT120/AT115 (HCSL).
The PCIe subsystem clocks are back at stock values.
Questions:

What drives the J10 REFCLK pair? Is that clock gated by CLKREQ#, a GPIO or I²C setting, or a jumper that we need to enable?
Under what conditions does U13 hold the M.2 PERST# low?
Have you run a standard x4 NVMe SSD directly in J10 on this kit, and was anything needed to make it work?
Thanks,
Brady