Stratix V Avalon-MM Interface for PCIe Solutions: User Guide

ID 683411
Date 5/21/2017
Public
Document Table of Contents

12.2.1. Debugging Link that Fails To Reach L0

The following table describes possible causes of the failure to reach L0.

Table 75.  Link Training Fails to Reach L0

Possible Causes

Symptoms and Root Causes

Workarounds and Solutions

Link fails the Receiver Detect sequence.

LTSSM toggles between Detect.Quiet(0) and Detect.Active(1) states

Check the following termination settings:

  • For Gen1 and Gen2, the PCI Express Base Specification, Rev 3.0. states a range of 0.075 µF–0.265 µF for on‑chip termination (OCT).
  • For Gen3, the PCI Express Base Specification, Rev 3.0 states a range of 0.176 µF–0.265 µF for OCT.
  • Intel uses 0.22 µF terminations to ensure compliance across all data rates.
  • Link partner RX pins must also have appropriate values for terminations.

Link fails with LTSSM stuck in Detect.Active state (1)

This behavior may be caused by a PMA issue if the host interrupts the Electrical Idle state as indicated by high to low transitions on the RxElecIdle (rxelecidle)signal when TxDetectRx=0 (txdetectrx0) at PIPE interface. Check if OCT is turned off by a Quartus Settings File (.qsf) command. PCIe requires that OCT must be used for proper Receiver Detect with a value of 100 Ohm. You can debug this issue using SignalTap II and oscilloscope.

For Stratix V devices, a workaround is implemented in the reset sequence.

Link fails with the LTSSM toggling between: Detect.Quiet (0), Detect.Active (1), and Polling.Active (2),

or:

Detect.Quiet (0), Detect.Active (1), and Polling.Configuration (4)

On the PIPE interface extracted from the test_out bus, confirm that the Hard IP for PCI Express IP Core is transmitting valid TS1 in the Polling.Active(2) state or TS1 and TS2 in the Polling.Configuration (4) state on txdata0. The Root Port should be sending either the TS1 Ordered Set or a compliance pattern as seen on rxdata0. These symptoms indicate that the Root Port did not receive the valid training Ordered Set from Endpoint because the Endpoint transmitted corrupted data on the link. You can debug this issue using SignalTap II. Refer to PIPE Interface Signals for a list of the test_out bus signals. The following are some of the reasons the Endpoint might send corrupted data:
  • Signal integrity issues. Measure the TX eye and check it against the eye opening requirements in the PCI Express Base Specification, Rev 3.0. Adjust the transceiver pre-emphasis and equalization settings to open the eye.
  • Bypass the Transceiver Reconfiguration Controller IP Core to see if the link comes up at the expected data rate without this component. If it does, make sure the connection to Transceiver Reconfig Controller IP Core is correct.

Link fails due to unstable rx_signaldetect

Confirm that rx_signaldetect bus of the active lanes is all 1’s. If all active lanes are driving all 1’s, the LTSSM state machine toggles between Detect.Quiet(0), Detect.Active(1), and Polling.Active(2) states.

This issue may be caused by mismatches between the expected power supply to RX side of the receiver and the actual voltage supplied to the FPGA from your boards. If your PCB drives VCCT/VCCR with 1.0 V, you must apply the following command to both P and N pins of each active channel to override the default setting of 0.85 V.

set_instance_assignment -name XCVR_VCCR_VCCT_VOLTAGE 1_0V –to “pin”

Substitute the pin names from your design for “pin”.

Link fails because the LTSSM state machine enters Compliance

Confirm that the LTSSM state machine is in Polling.Compliance(3) using SignalTap II.

Possible causes include the following:

  • Setting test_in[6]=1 forces entry to Compliance mode when a timeout is reached in the Polling.Active state.
  • Differential pairs are incorrectly connected to the pins of the device. For example, the Endpoint’s TX signals are connected to the RX pins and the Endpoint’s RX signals are to the TX pins.