Must call dwt_initialise() before dwt_checkidlerc() to avoid segfault?

Wireless Connectivity Ultra-Wideband
1

The rx examples generally have this initialisation sequence:

  1. Set SPI rate.
  2. Power on/reset IC.
  3. dwt_probe()
  4. Poll dwt_checkidlerc()
  5. dwt_initialise()

See for example, simple_rx_nlos.c or rx_diagnostics.c.

In my experience (stm32) this reliably segfaults due to a read from a NULL pointer. And this seems entirely reasonable, since dwt_checkidlerc() calls this chain:

  • dwt_checkidlerc() in deca_compat.c
    • ioctl(DWT_CHECKIDLERC) in dw3000_device.c
      • ull_checkidlerc()
        • dwt_read16bitoffsetreg()
          • dwt_readfromdevice()
            • dwt_xfer3xxx(DW3000_SPI_RD_BIT)

And there, inside the DW3000_SPI_RD_BIT case, is this line:

if ((LOCAL_DATA(dw)->spicrc == DWT_SPI_CRC_MODE_WRRD) ...

LOCAL_DATA(dw) expands to dw->priv and priv is set to NULL in dwt_probe() when static_dw is assigned to dw. So this bombs out reliably.

The first thing dwt_initialise() does is malloc() this field, so calling it before dwt_checkidlerc() resolves the issue.

This seems pretty fundamental so I’m surprised I can find no discussion or explanation. Is my version of the API hosed? Am I using it wrong? Or is everyone quietly doing this as well?

2

Hello,

The most recent version of the driver is 08.02.02. If you have downloaded the most recent QM33 SDK 1.0.0, then you are using the latest software.

I reviewed the code in this SDK and it is similar to what you describe, dwt_checkidlerc is called before dwt_initialise().

dwt_initialise should be called only once the device has properly booted, and reach idle_rc stage.

The error is likely to be linked to an error in the STM32 porting, and incorrect allocation. I will raise it to the internal teams

1 Like
3

Looks like the same story in 08.02.02. For example, the application entry point in rx_diagnostics.c follows the probe-check-initialise ordering:

dwt_probe((struct dwt_probe_s *)&dw3000_probe_interf);

while (!dwt_checkidlerc()) /* Need to make sure DW IC is in IDLE_RC before proceeding */ { };

if (dwt_initialise(DWT_DW_INIT) == DWT_ERROR)

dw3000_probe_interf is defined in both the nRF52840 and STM_Nucleo projects in deca_probe_interface.c as:

const struct dwt_probe_s dw3000_probe_interf = 
{
    .dw = NULL,
    .spi = (void*)&dw3000_spi_fct,
    .wakeup_device_with_io = wakeup_device_with_io,
    .driver_list = (struct dwt_driver_s **)tmp_ptr,
    .dw_driver_num = 2,
};

That NULL dw gets set to &static_dw in dwt_probe(), as defined in deca_compat.c. static_dw is initialised to all zeros, and dwt_probe() only assigns a few fields - priv is not one of them.

So I can’t see how dw->priv can be anything other than NULL by the time dwt_checkidlerc() runs. I’m obviously missing something fundamental here, so I’m looking forward to your findings!