7-series SRAM load: the board's own QSPI boot wins the race after JPROGRAM, and flip-flops in the loaded design do not get their INIT values

Open
#741 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
58/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
cpp
Domain
embedded-iot

Research direction

Start with src/xilinx.cpp and program_mem(), especially the JPROGRAM/INIT_B polling and the Run-Test/Idle clock wait described here. Check how the existing 7-series sequence handles INIT_B and when it enters CFG_IN; the reproduction compares INIT values and MIG calibration across repeated loads. Done means SRAM loading reliably applies flip-flop INIT values and calibrates the MIG on the reported master-SPI setup, without regressing other configuration modes.

Written by the indexing model from the issue text.

Description

Board / setup

  • Puzhi PA35T-Starlite (XC7A35T-2FGG484), onboard FT232H JTAG (-c digilent_hs2), Debian 13.
  • The mode pins are hard-wired to master SPI, and the QSPI flash holds a valid image (the vendor's
    LED demo).
  • openFPGALoader: Debian v0.13.1, and master bf2b529 (v1.1.1). Both load to SRAM only.

Symptom

  • After openFPGALoader -c digilent_hs2 design.bit, DONE=1 and the new design runs.
  • However, its flip-flops do not start at their INIT values. Some keep the state of the previously
    running design; others are garbage.
  • Logic that resets itself after configuration works. Anything that relies on INIT values breaks:
    • A Xilinx MIG DDR3 controller never calibrates, in the vendor's Microblaze designs too.
    • FSMs that are only reset by a push-button start in illegal states.
  • The same .bit loaded with Vivado (program_hw_devices) or xsdb (fpga -file) always works.

Reproduction. The test design has a UART status reporter whose 32-bit report counter T has
INIT 0, plus the MIG's init_calib_complete. Before each run a different design is loaded, so
retained state cannot fake a pass. Results over 10 loads:

loader MIG calibrated T starts at 0/1 (INIT applied)
Vivado 2026.1 program_hw_devices always always
Debian v0.13.1 1/10 0/10 (e.g. T = 0xF4846F)
master bf2b529 0/10 0/10
master + patch below 10/10 10/10

Reloading the same bitstream over a running copy with the unpatched loader, T just continues
counting (0x88D72 → 0x88D73 …). The flip-flops are never re-initialised.

What happens

  • After JPROGRAM this board reconfigures itself from flash within milliseconds. After
    openFPGALoader -r, Vivado reads CONFIG_STATUS DONE_PIN = 1 and BOOT_STATUS = valid.
  • program_mem() polls INIT_B and then clocks 10,000 × 12 TCK in Run-Test/Idle (20 ms at 6 MHz)
    before loading CFG_IN, following UG470's table.
  • In master-SPI mode with a valid flash image, the device has already booted the flash image by
    then. The JTAG bitstream is written into a running device, the frames are replaced, and the GSR
    initialisation does not happen.

Patch (against bf2b529, src/xilinx.cpp, program_mem()): wait for INIT_B to go low
(housecleaning), then high, then load CFG_IN immediately.

 	_jtag->shiftIR(get_ircode(_ircode_map, "JPROGRAM"), NULL, _irlen);
-	/* Poll INIT_B (bit 4 of IR capture) until config memory is cleared */
+	/* Wait for INIT_B to go low (housecleaning), then high, and load CFG_IN at once: with
+	 * master-SPI mode pins and a valid flash image, any delay lets the device boot the flash
+	 * image first, and the JTAG bitstream then lands in a running device. */
 	tx_buf = get_ircode(_ircode_map, "BYPASS");
+	int guard = 0;
+	do {
+		_jtag->shiftIR(tx_buf, rx_buf, _irlen);
+	} while ((rx_buf[0] & 0x10) && ++guard < 100000);
 	do {
 		_jtag->shiftIR(tx_buf, rx_buf, _irlen);
 	} while (!(rx_buf[0] & 0x10));
-	/*
-	 * 8: Move into the RTI state.                        X     0   10,000(1)
-	 */
-	_jtag->set_state(Jtag::RUN_TEST_IDLE);
-	_jtag->toggleClk(10000*12);

Notes

  • Removing only the 120,000-clock wait (no wait for INIT_B low) calibrated 10/10 when a different
    design ran before. But T sometimes started at 0x17 or 0x1DC, and one run without that scrub
    failed outright. Waiting for the low phase as well made every run clean.
  • I have not checked whether UG470's 10,000 TCK are needed on boards strapped for JTAG or slave
    modes. If they are, the wait could be made conditional on the mode pins, or skipped only when
    INIT_B has been seen low and high.
  • v0.13.1 polled INIT_B on bit 0 of the IR capture, which is constant 1 on 7-series. Master
    already tests bit 4.
Dominant language
C++
Stars
1.7k
Forks
396
Avg merge
2d 4h
Merged PRs (30d)
15

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from trabucayre/openFPGALoader

All issues in trabucayre/openFPGALoader

Similar issues

More C++ issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.