On Wed, 22 Jan 2025 15:25:46 +0100 Eric Dumazet wrote: > napi_schedule() can run from arbitrary contexts though... > > BH protection seems strange to me, but this is orthogonal to your fix. Right, it doesn't need the BH "protection", it's just what we do to make sure there is a softirq hook point after we call it. Since local_irq_restore() does not call softirqs. I didn't know how to express that in a way that would be understandable for most :S Thanks for catching the other bug!
