ADXL FIFO Example Watermark

[JonasD27]

Hi,
I have a question on the following section of the adxl38x fifo example under
https://github.com/analogdevicesinc/no-OS/blob/main/projects/eval-adxl38x/src/examples/fifo_example/fifo_example_main.c

while (true) {
		// Read status to assert if FIFO_WATERMARK bit set
		ret = adxl38x_read_device_data(adxl38x_desc, ADXL38X_STATUS0, 1, &status0);
		if (ret)
			goto error;
		pr_info("Status 0: %d\n", status0);
		ret = adxl38x_read_device_data(adxl38x_desc, ADXL38X_FIFO_STATUS0, 2,
					       fifo_status);
		if (ret)
			goto error;
		fifo_entries = no_os_get_unaligned_le16(fifo_status);
		fifo_entries = fifo_entries & 0x01ff;


		// Read FIFO status and data if FIFO_WATERMARK is set
		if (status0 & NO_OS_BIT(3)) {
			pr_info(" FIFO_WATERMARK is set. Total fifo entries =  %d\n", fifo_entries);
			if (fifo_entries < set_fifo_entries)
				goto unmatch_error;

This loop basically checks for the WATERMARK Bit in the status register. Why is there an unmatch error check after detecting the set WATERMARK Bit? By definition, the watermark bit should only be set when the desired amount of samples is reached in the fifo.
Why is this check needed?

In Addition:
In my implementation I noted, that after detecting the watermark bit as set, the number of fifo entries given by the FIFO_STATUS0 register is sometimes one sample behind.

For example:
Watermark threshold is set to 9 entries. After detecting that the watermark bit is set I immediately read the number of fifo entries from FIFO_STATUS0 register --> this yields only 6 samples.

Thank you

[RaduSabau1]

Hi, @JonasD27

Thank you for the question.

The unmatch_error check if (status0 & NO_OS_BIT(3)) exists as a check to guard against a race condition in the polling logic.

The code performs two separate register reads: STATUS0 to check the WATERMARK flag, followed by FIFO_STATUS0 to get the entry count. Since these are independent bus transactions, there is potential latency between them. During this window, the FIFO state can change, which may cause FIFO_STATUS0 to report a count that differs from what was present when the WATERMARK flag was set.

This is why the check on fifo_entries if (fifo_entries < set_fifo_entries) is necessary - it ensures we don't proceed with reading the FIFO if the reported count is less than expected, avoiding potential data inconsistencies.

Hope this helps.

Regards,
Radu