DFU compatibility for a mesh example

[DeveloperZ]

Hi,

I am interested to know how to make a mesh node (for example, a light server) compatible to perform a DFU.

I have gone through the DFU page and I noticed that the DFU is done over the Serial port. I understand that the dfu firmware is being replaced with blinky firmware. Correct me if I am wrong.

Now, what changes should I make to the commands inorder to make the light_server example node to be DFU compatible/capable over the air?

Would replacing the application hex file on step 8 to light_server instead of dfu solve it? or should I add any other commands to have light_server firmware first?

Thank you.

[Edvin]

Hello,

well. The short answer is that you need the mesh_evt_handler() to handle DFU packets, like it does in the DFU example. In addition, if you want the device to handle Serial DFU (the one who receives the packets from nrfutil), you need it to support serial DFU as well. 

This is done in nrf_mesh_serial_init(NULL); in mesh_init(void).

Inside this call, serial_handler_dfu_init() is called, which handles the DFU serial messages.

When you try to include the DFU packets (over Mesh) you will probably have to include everything you need around as well. Give it a go, and let me know if you get stuck.

BR,

Edvin

[DeveloperZ]

Hi,

Edvin said:

The short answer is that you need the mesh_evt_handler() to handle DFU packets, like it does in the DFU example.

I notice that nrf_mesh_evt_handler_t is used in only few examples like EnOcean, light switch provisioner, Lpn examples. 

So, in order to make other examples DFU capable, I need to add the nrf_mesh_evt_handler while mesh initilization.

Edvin said:

Give it a go, and let me know if you get stuck.

If I am trying to push the dfu packets over the air but not serial, I should push it from the nrf connect app right? or can we do it from the mesh app? Sorry but I have no idea about how to push it OTA!!

Thank you.

[DeveloperZ]

Yes, There is a Mesh assert while loading the new firmware!!

The logs were:

<t:      15314>, main.c, 1112, Initializing and adding models

<t:      20203>, app_error_weak.c,  109, Mesh assert at 0x0002F696 (:0)

After the assert is occurred, I was trying to program the boards with the same firmware without erasing the board using Segger and it was showing the same error!!

When I erase the board, I can program the firmware successfully. 

Similar issue persists with all the examples to which I added the DFU capability. I can push the Blinky successfully but not the customized mesh examples :( 

To add, I used the DFU example/example to which I added DFU capability and pushed light switch server onto the board and it was pushed successfully.

Good news is that I traced out where the problem is occurring... 

It is happening in app_flash_init() function... and the exact location is shown in the figure below in the flash_manager.c file:

The second nrf_mesh_assert() is causing the error when it is checking if the flash area is valid.

We are storing some information in the flash like enocean data and few others. Is that causing this error? 

[Edvin]

DeveloperZ said:

We are storing some information in the flash like enocean data and few others. Is that causing this error? 

 Well. If you try to modify an example to not store this information, and that works, then yes, that is probably causing the error. How (and if you know, where) are you storing the enocean data?

And since you add this data somewhere, have you modified any defines in sdk_config.h or nrf_mesh_config_app.h to make sure that the rest of the application handles this data properly? (i.e. doesn't try to use that flash area for something else?)

Best regards,

Edvin

[DeveloperZ]

Edvin said:

How (and if you know, where) are you storing the enocean data?

I am using the stock enocean example. I don't exactly know the location where the data is being stored. I am pretty sure that the enocean switch information isn't causing any problem as I added DFU functionality to the enocean example and pushed the dfu packet successfully.

Regarding the other data that is being stored, we are storing them using random entry handles like 7EFF and 5EFF(which we assumed are the memory locations on flash). 

Edvin said:

And since you add this data somewhere, have you modified any defines in sdk_config.h or nrf_mesh_config_app.h to make sure that the rest of the application handles this data properly?

No, we didn't make any changes to those files. 

Where is the DFU data being stored on the flash? I mean, are there any specific locations defined to store the DFU data or is the data that we are storing interfering with the DFU data? 

Should we increase the flash page count on nrf_mesh_config_app.h to resolve the issue?

[Edvin]

DeveloperZ said:

Regarding the other data that is being stored, we are storing them using random entry handles like 7EFF and 5EFF(which we assumed are the memory locations on flash). 

 what do you mean by "entry handles"? Are you using FDS to store this data? Or perhaps fstorage?

[DeveloperZ]

Edvin said:

what do you mean by "entry handles"

It is the same way that you guys store the Enocean data in the enocean example. Create an entry handle and use app_flash_store, app_flash_data_store, app_flash_clear functions to write into and erase the flash.

[Edvin]

Ok. I see.

You need to tell the bootloader that you are using some of the flash area for application data, so that the bootloader or flash manager doesn't try to use this for something else.

Please take a look at what my collague, Hung writes in this post.

Basically, this define, FLASH_MANAGER_RECOVERY_PAGE_OFFSET_PAGES, will determine the flash that is reserved in between your bootloader and the flash_manager_recovery_page. Make sure that you store all your application data in this space.

Best regards,

Edvin

[DeveloperZ]

Edvin said:

Basically, this define, FLASH_MANAGER_RECOVERY_PAGE_OFFSET_PAGES, will determine the flash that is reserved in between your bootloader and the flash_manager_recovery_page.

I see that this particular define is initialized as 0. I changed it to 1.

The Enocean entry handle is specified as 00001. So, I had two variables that I am storing in flash. I assigned entry handles as 00002 and 00003. 

The DFU fails again. Should I also increase the APP_FLASH_PAGE_COUNT value which is set to 1 initially?

If I increase the count, how can the chip understand which flash page should it save my variables into?

I can't see anything in the app_flash_data_store, load and clear functions which specifies the flash page.

I am unable to figure out the changes needed to make the DFU work when I am storing data in the flash using app_flash_data_store, load and clear functions. If I can get the exact info on all the changes needed, that would be a great help.

Thank you.

[Edvin]

Just to verify that this is actually the issue. Can you try to perform the DFU without storing the custom data. Does the DFU pass through then?

It is not easy to tell "the exact info on all the changes needed" when I don't know your project. 

The assert you get, that is in the application after you performed the DFU, right? Does it happen if you just flash that image as well? Remember to erase the flash completely before you program your application if you have modified the flash setup, using e.g. "nrfjprog -e".

[DeveloperZ]

Edvin said:

Can you try to perform the DFU without storing the custom data.

When I don't store the data in the flash, DFU process is successful.

Edvin said:

The assert you get, that is in the application after you performed the DFU, right? Does it happen if you just flash that image as well?

When I flash the example with data stored in flash after erasing the board, it doesn't give back any assert. 

Now, I am pretty sure that the assert is occurring while initializing the flash (app_flash_init() function) while loading the new firmware.

Edvin said:

Please take a look at what my collague, Hung writes in this post.

I did go through the post and it says that the address for storing MBR and Softdevice starts at 0x00000 and is of size 0x26000. If you can see the enocean example, in enocean_switch_example.h file, the Entry handle for storing the enocean commissioning data is 0x00001.

Edvin said:

You need to tell the bootloader that you are using some of the flash area for application data, so that the bootloader or flash manager doesn't try to use this for something else.

I don't know how the boot loader understands that that particular location stores enocean data and can't be used for anything else. Where does the application specify it? Because I tried storing them in 0x00002 and 0x00003 locations and it fails.

I am trying to find other ways to store the custom data into the flash by either using ads or fstorage but it would be great if I get it working with this app_flash_init().

I can send you a simple example(zip file) which is similar to what I am trying and I would be glad if you can find some time to check what's going wrong and suggest the changes needed. :)

[Edvin]

If you store data in a new area in the flash, the flash manager needs to be aware. Changing this with a FW update, either a DFU or by manually programming the device will not erase the flash part that is used for application data storage, and hence the flash manager is still initialized with the old settings. So it is not trivial to change the size that the flash manager will use after it has been initialized.

I don't think this is actually DFU related at all. You need your initial FW to use the same flash manager settings as the one you intend to use in the end.