Eric Dumazet [mailto:eric.duma...@gmail.com] > Sent: Friday, August 16, 2019 5:27 PM [...] > Maybe you would avoid messing with a tasklet (we really try to get rid > of tasklets in general) using two NAPI, one for TX, one for RX. > > Some drivers already use two NAPI, it is fine. > > This might avoid the ugly dance in r8152_poll(), > calling napi_schedule(napi) after napi_complete_done() !
The reason is that the USB device couldn't control the interrupt of USB controller. That is, I couldn't disable the interrupt before napi_schedule and enable it after napi_complete_done. If the callback function occurs during the following timing, it is possible no one would schedule the napi again. static int r8152_poll(struct napi_struct *napi, int budget) { struct r8152 *tp = container_of(napi, struct r8152, napi); int work_done; work_done = rx_bottom(tp, budget); bottom_half(tp); --> callback occurs here and try to call napi_schedule napi_complete_done(napi, work_done) That is, no tx or rx could be handled unless something trigger napi_schedule. Best Regards, Hayes