Dear Lukasz,
In message 20140515090904.32f1d13d@amdc2363 you wrote:
What I complained about is the change in behaviour. I asked to make the existing behaviour the default, so unaware users will not be affected. Only if you intentionally want some other behaviour you can then enable this by setting the env variable.
Well, looking at mainline usage of DFU, Lukasz is speaking for about half of the users / implementors. Since Denx is working with the other half, can you shed some light on actual use vs theoretical possibilities?
I don't want to urge anybody on making any conclusion here :-), but I would be very grateful if we could come up with an agreement.
As I've stated previously, my opinion is similar to the one presented by Tom in this message.
For me it would be best to not calculate any checksum on default and only enable it when needed.
I asked Heiko to run some actual tests on the boards where he has to maintain DFU for. For a 288 MiB image he did not measure any difference - with your patch applied, both with and without CRC enabled, we would get the same (slow) 1:54 minutes download time.
This reinforces my speculation that you are actually addressing the wrong problem. Instead of adding new code and environment variables and making the system even more complex, we should just leave everything as is, and you should try to find out why the CRC calculation is so low for you. Checking if caches are enabled is probably among the things that should be done first.
Regarding the checksumming topic in general: the fact that the DFU standard defines a method to verify the checksum of the image (dwCRC field in the DFU File Suffix), but does not transmit this vital data to the target so the actual file download and storage procedure on the target is completely unprotected is IMO a serious design flaw of the DFU protocl. Do you think it would be possible to have this augmented / fixed?
Best regards,
Wolfgang Denk