We have not really encountered any case where we had to force retraining in factory.
Is that what is required on non-x86 platforms?
I think we should enable this on all platforms with training data cache.
The concern is few ODMs are using copy machine to preflash - and they do it like
1. Install the images on one DUT.
2. Verify the DUT is working all fine (and boots).
3. Unmount the SPI flash / eMMC and use that as golden sample of copy source.
So it's always better to ensure DRAM cache is cleared.
It's also very helpful to debug memory issues with one simple command.
Can the saved training data be simply erased by factory scripts?
It is possible, but we think it may be easier to have a unified way across different platforms, especially in firmware.
For example, on x86 there may be different cache FMAP names. Even on ARM, different SOCs may have different designs (e.g. kukui does not have RO but trogdor may have).
If we have to make this a script in the end, can we have it on all images and not just factory?
To view, visit change 40623. To unsubscribe, or for help writing mail filters, visit settings.