On di, 07 jan 2020 10:58:26 -0600, Elenita Hinds wrote: > Hi, > I've been testing version 4 of this patch series on our product and it > looks good so far (only 1 packet loss at the beginning of each tests > after hours of run, which is insignificant). On my end, the 64 message > objects plus setting the TI chip's CPU frequency scaling/governor to > 'performance' (from ondemand which apparently had issues) made a huge > difference. Thanks for sharing your results. It confirms that it really improved. > > I'm assuming there is no other way to make the RX/TX split configurable? I understood from Marc on 17 Dec that one would want to avoid large HW queues. Making it configurable from Kconfig seems not appropriate. >From device-tree seems possible. I considered that 8 TX queue is good enough, and assign the remaining HW space to recv. I honestly see no reason anymore to make it adustable, other than that I choose a bad fixed value. How would you want to make the tx/rx split? Kind regards, Kurt
