sd_flash_page_erase() time execution on nRF52840

RE: sd_flash_page_erase() time execution on nRF52840

era — Thu, 16 Jul 2020 11:02:17 GMT

Thank you for your help! :)

RE: sd_flash_page_erase() time execution on nRF52840

Jørgen Holmefjord — Thu, 16 Jul 2020 10:59:17 GMT

I checked with the softdevice developers. The softdevice does not support partial page erase, and will disable all interrupts during the erase (up to 90 ms). You can implement partial erase in timeslots yourself.

RE: sd_flash_page_erase() time execution on nRF52840

era — Thu, 16 Jul 2020 09:32:50 GMT

OK. But my main problem is that after scheduling page erase with SD api, I get very high priority interrupt, which takes up to 90 ms. I try to find a workaround, because it is not acceptable for me to stop my application in random moments for so long. So to sum up:

- I have softdevice enabled (and I want it to be enabled all the time)

- I need to erase flash page from time to time

- I can't let my application to be stopped for tenths of milliseconds

Is there any workaround for this? Would erasing flash page by chunks in timeslots using NVMC api help?

RE: sd_flash_page_erase() time execution on nRF52840

Turbo J — Thu, 16 Jul 2020 09:27:55 GMT

I believe the SD already knows how to partially erase a flash page, otherwise it would be impossible in most BTLE timing setups - especially during a connection (for peer manager).

Edit: Wrong assumtion, see below.

RE: sd_flash_page_erase() time execution on nRF52840

era — Thu, 16 Jul 2020 09:21:31 GMT

But does it work with softdevice enabled? I think the only way is to erase page in chunks using timeslots, right? SD has no api for partial erase.

RE: sd_flash_page_erase() time execution on nRF52840

Turbo J — Thu, 16 Jul 2020 09:20:01 GMT

The nRF52840 has partial flash erase, around 1ms each time.

RE: sd_flash_page_erase() time execution on nRF52840

era — Wed, 15 Jul 2020 19:03:37 GMT

Thanks. After page erase scheduling, the erase is performed from priority 0, as Flash erase/write operations are defined as softdevice timing-critical. So this means that I cannot preempt this process and it takes up to 90 ms. Can anything be done about this? Or do I need to be prepared that after page erase scheduling my code can be completely halted for so long?

RE: sd_flash_page_erase() time execution on nRF52840

Jørgen Holmefjord — Wed, 15 Jul 2020 15:59:43 GMT

Hi,

See the following sections in the softdevice specifications:

The API call itself is not expected to take a long time, but it can get interrupted by a timing-critical softdevice interrupt, which may take up to 250 us. See Bluetooth Low Energy processor usage patterns for more details.

Best regards,
Jørgen

RE: sd_flash_page_erase() time execution on nRF52840

Turbo J — Wed, 15 Jul 2020 15:27:51 GMT

[quote userid="76835" url="~/f/nordic-q-a/63810/sd_flash_page_erase-time-execution-on-nrf52840"]We measured time execution of this function and it usually takes about 50 us to be executed.[/quote]

Yes, this function does not actually erase a page. It just schedules an erase during the next available radio time slot.

Which may be much later than anticipated in situations with very high radio usage. The application will be informed (via SD event) when the page erase transaction finshed. This can take significantly more time than 85 ms.