Francois Romieu [mailto:romieu@xxxxxxxxxxxxx] > Sent: Sunday, November 09, 2014 6:12 AM [...] > The performance explanation leaves me a bit unconvinced. Without any > figure one could simply go for the always locked clear_bit because of: > 1. the "I'm racy" message that the open-coded test + set sends > 2. the extra work needed to avoid 1 (comment, explain, ...). Thanks. I would modify this patch with clear_bit only. > The extra time could thus be used to see what happens when napi is > shoehorned in this tasklet machinery. I'd naively expect it to be > relevant for efficiency. I thought about NAPI, but I gave up. The reasons are 1. I don't sure if it would run when autosuspending. 2. There is no hw interrupt for USB device. And I have no idea about how to check if the USB transfer is completed by polling. 3. I have to control the rx packets numbers in poll(). However, I couldn't control the packets number for each bulk-in transfer. I have to do extra works to deal with the rx flow. 4. I don't find much different between tasklet and NAPI. Best Regards, Hayes -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
