NCS: SRAM usage overflow with data length extension support on hci_rpmsg

[Adrian Eggenberger]

Hi

I'm playing around with the configuration of the hci_rpmsg as child image in the NCS 1.5.1 release. I've added a hci_rpmsg.conf file in a "child_image" folder and added there my configuration extension for the hci_rpmsg project. It worked properly, but when I tried to enable the data length extension I've struggled over a massive SRAM usage for this feature.

When I compile it without any configuration change I get the following memory footprint for the network CPU:

Memory region Used Size Region Size %age Used
FLASH: 179084 B 256 KB 68.32%
SRAM: 45836 B 64 KB 69.94%
IDT_LIST: 168 B 2 KB 8.20%

When I try to enable the data length extension to the max supported size of 251 the memory already overflows:

Memory region Used Size Region Size %age Used
FLASH: 179084 B 256 KB 68.32%
SRAM: 67340 B 64 KB 102.75%
IDT_LIST: 168 B 2 KB 8.20%

If we divide the difference of SRAM usage and the increase of the buffer we should get the number of buffers that are using this "size", right?

BufferCount = (67340 - 45836) / (251 - 27) = 96

Why is this buffer used 96 times? Is there some other configuration wrong? What would be a good configuration for maximal data length extention?

Regards Adrian

[Sigurd]

If we divide the difference of SRAM usage and the increase of the buffer we should get the number of buffers that are using this "size", right?

 or the number of bytes RAM required for each Data Length Extension byte increase? But the more links you have, the more it will cost.

Try to set CONFIG_BT_MAX_CONN=2 , or 1, as we have done in e.g. the throughput example, https://github.com/nrfconnect/sdk-nrf/blob/v1.5.1/samples/bluetooth/throughput/child_image/hci_rpmsg.conf

What would be a good configuration for maximal data length extention?

Take a look at the throughput sample config:

https://github.com/nrfconnect/sdk-nrf/blob/v1.5.1/samples/bluetooth/throughput/prj_nrf5340dk_nrf5340_cpuapp.conf

https://github.com/nrfconnect/sdk-nrf/blob/v1.5.1/samples/bluetooth/throughput/child_image/hci_rpmsg.conf

[Adrian Eggenberger]

Hi Sigurd

That was the right hint. I didn't think about the connection count that also multiplies with the buffer count. By adding the CONFIG_BT_MAX_CONN I was able to make it compile and flash it.

After this was fixed I tested the data length extension in the NCS 1.5.1. On my tests with nRF5340 DK (and even a nRF52840 DK) with a hci_usb example. I'm able to trigger the data length extension process, but I don't get the response from the peer device. The only way to make it work again is to switch back to the Zephyr LL layer by setting the CONFIG_BT_LL_SW_SPLIT option to 'y'. Is it possible that the data length implementation is somehow broken in the softdevice controller?

Regards Adrian 

[Sigurd]

Hmm, maybe you can try setting 

CONFIG_BT_USER_DATA_LEN_UPDATE=y

http://developer.nordicsemi.com/nRF_Connect_SDK/doc/1.5.1/kconfig/CONFIG_BT_USER_DATA_LEN_UPDATE.html

I assume you are calling bt_conn_le_data_len_update(), and that you have registered a callback le_data_length_updated as e.g. show-cased in the throughput sample.

If setting CONFIG_BT_USER_DATA_LEN_UPDATE=y , does not help, then please upload a sniffer trace that shows the issue.