On 10/29/18 1:10 PM, Johannes Berg wrote: >>> I'd argue for trying it though as it makes the code MUCH simpler. >> I totally agree that it will be simpler to do the work directly from >> the cfg context rather than scheduling it. >> On a cortex A5 MPU and SPI bus running on 48Mhz, this takes 20ms in >> the idle case, and 100 to 300ms in case of data being transferred in >> parallel. > That does sound pretty slow ... but OTOH, mostly wpa_s etc. would expect > this to actually complete before the next step, so if this is slow and > wpa_s/hostapd is much faster, it might send you a lot of things to do > and so you'd end up being slow anyway? Correct. The speed of executing the work will be the same in both cases (maybe a little slower in case of deferring the work) The intuition here was to do minimum work in the cfg's context, but since this isn't a concern, I can skip deferring the work.
Thanks, Adham
