Intel® Quartus® Prime Standard Edition User Guide: Partial Reconfiguration

ID 683499
Date 9/24/2018
Public
Document Table of Contents

1.6.5. Partial Reconfiguration Cycle Waveform

When you are using the Altera PR IP in the internal host mode, all the timing relations between various interface signals are met by default, and you can skip reading this section. If you are using PR with an external host or implementing your own custom PR internal host logic, pay attention to these timing relationships when designing your logic. The PR host initiates the PR request, transfers the data to the FPGA device when it is ready, and monitors the PR process for any errors or until it is done.

A PR cycle is initiated by the host (internal or external) by asserting the PR_REQUEST signal high. When the FPGA device is ready to begin partial reconfiguration, it responds by asserting the PR_READY signal high. The PR host responds by sending configuration data on DATA [15:0]. The data is sent synchronous to PR_CLK. When the FPGA device receives all PR data successfully, it asserts the PR_DONE high, and de-asserts PR_READY to indicate the completion of the PR cycle. The PR host must monitor the PR process until either the successful completion of PR (indicated by PR_DONE), or an error condition is asserted.

Figure 13. Partial Reconfiguration Timing Diagram

If there is an error encountered during partial reconfiguration, the FPGA device asserts the PR_ERROR signal high and de-asserts the PR_READY signal low.

Whenever either of these two signals are asserted, the host must de-assert PR_REQUEST within eight PR_CLK cycles. As a response to PR_ERROR error, the host can optionally request another partial reconfiguration or perform a full FPGA configuration.

To prevent incorrect behavior, the PR_CLK signal must be active a minimum of twenty clock cycles after PR_DONE or PR_ERROR signal is asserted high. Once PR_DONE is asserted, PR_REQUEST must be de-asserted within eight clock cycles. PR_DONE is de-asserted by the device within twenty PR_CLK cycles. The host can assert PR_REQUEST again after the 20 clocks after PR_DONE is de-asserted.

Table 4.  Partial Reconfiguration Clock RequirementsSignal timing requirements for partial reconfiguration.

Timing Parameters

Value (clock cycles)

PR_READY to first data

4 (exact)

PR_ERROR to last clock

20 (minimum)

PR_DONE to last clock

20 (minimum)

DONE_to_REQ_low

8 (maximum)

Compressed PR_READY to first data

4 (exact)

Encrypted PR_READY to first data (when using double PR)

8 (exact)

Encrypted and Compressed PR_READY to first data (when using double PR)

12 (exact)

At any time during partial reconfiguration, to pause sending PR_DATA, the PR host can stop toggling PR_CLK. The clock can be stopped either high or low.

At any time during partial reconfiguration, the PR host can terminate the process by de-asserting the PR request. A partially completed PR process results in a PR error. You can have the PR host restart the PR process after a failed process by sending out a new PR request 20 cycles later.

If you terminate a PR process before completion, and follow it up with a full FPGA configuration by asserting nConfig, then you must toggle PR_CLK for an additional 20 clock cycles prior to asserting nConfig to flush the PR_CONTROL_BLOCK and avoid lock up.

During these steps, the PR control block might assert a PR_ERROR or a CRC_ERROR signal to indicate that there was an error during the partial reconfiguration process. Assertion of PR_ERROR indicates that the PR bitstream data was corrupt, and the assertion of CRC error indicates a CRAM CRC error either during or after completion of PR process. If the PR_ERROR or CRC_ERROR signals are asserted, you must plan whether to reconfigure the PR region or reconfigure the whole FPGA, or leave it unconfigured.

Important: The PR_CLK signal has different a nominal maximum frequency for each device. Most Stratix V devices have a nominal maximum frequency of at least 62.5 MHz.