HardFault_Handler() called when NRF_LOG_ENABLED is disabled

[Prasad]

In my application when I disable NRF_LOG_ENABLED, HardFault_Handler() is getting called. To debug this issue I defined a HardFault_Handler() in my code and inserted a break point in it. Below is the backtrace

#0  HardFault_Handler () at main.c:165
#1  <signal handler called>
#2  0x0002b870 in fs_init () at ../../../../Tools/nrf_SDK/nRF5x/components/libraries/fstorage/fstorage.c:290
#3  0x0002af96 in fds_init () at ../../../../Tools/nrf_SDK/nRF5x/components/libraries/fds/fds.c:1621
#4  0x00023630 in pds_init () at ../../../../Tools/nrf_SDK/nRF5x/components/ble/peer_manager/peer_data_storage.c:334
#5  0x000245e2 in pm_init () at ../../../../Tools/nrf_SDK/nRF5x/components/ble/peer_manager/peer_manager.c:470
#6  0x00030744 in peer_manager_init (erase_bonds=false) at main.c:1453
#7  0x00030b04 in main () at main.c:1626

I guess something is breaking when fs_init() is called. Note that the application runs just fine when NRF_LOG_ENABLED is set to 1. I am clueless about reason that must be triggering this hard fault? Or how to debug this hardware fault?

[Prasad]

Short answer

This is a memory alignment issue and it got fixed by adding . = ALIGN(4) in my linker script.

Long answer

I debugged the code in GDB. Problem gets manifested in the below snippet from fstorage.c

for (uint32_t i = 0; i < total_users; i++)
{
    fs_config_t const * const p_config = FS_SECTION_VARS_GET(i);

    ==>if ((p_config->p_start_addr != NULL) &&
        (p_config->p_end_addr   != NULL))
    {
        configs_to_init--;
    }
}

When I step over from this point I end up in the HardFault_Handler(). After investigation, I found that FS_SECTION_VARS_GET(i) is not returning the correct address of fs_config. Then with objdum I checked the addresses of __start_fs_data and fs_config in the object file.

When NRF_LOG_ENABLED = 1

200029e4 g       .data  00000000 __start_fs_data
200029e4 l     O .data  00000010 fs_config

When NRF_LOG_ENABLED = 0

2000202a g       .data  00000000 __start_fs_data
2000202c l     O .data  00000010 fs_config

You can see that when NRF_LOG_ENABLED = 0, __start_fs_data is not aligned with fs_config and hence the macro FS_SECTION_VARS_GET(i) returns an address where valid data doesn’t exist, causing a hardware fault.

Resolution is to add . = ALIGN(4) in your linker file as show below.

.data : AT(_sidata)
{
	. = ALIGN(4);
	_sdata = .;

	PROVIDE(__data_start__ = _sdata);
	*(.data)
	*(.data*)

==>	. = ALIGN(4);

	PROVIDE(__start_fs_data = .);
	KEEP(*(.fs_data))
	PROVIDE(__stop_fs_data = .);
	
	. = ALIGN(4);
	_edata = .;

	PROVIDE(__data_end__ = _edata);
} > SRAM

I chose to fix it this way, if anyone knows of a better solution then please let me know.