On Thu, Dec 25, 2025 at 10:55:36PM +0700, Bui Quang Minh wrote: > On 12/24/25 23:49, Bui Quang Minh wrote: > > On 12/24/25 08:47, Michael S. Tsirkin wrote: > > > On Wed, Dec 24, 2025 at 09:37:14AM +0800, Xuan Zhuo wrote: > > > > Hi Jason, > > > > > > > > I'm wondering why we even need this refill work. Why not simply > > > > let NAPI retry > > > > the refill on its next run if the refill fails? That would seem > > > > much simpler. > > > > This refill work complicates maintenance and often introduces a lot of > > > > concurrency issues and races. > > > > > > > > Thanks. > > > refill work can refill from GFP_KERNEL, napi only from ATOMIC. > > > > > > And if GFP_ATOMIC failed, aggressively retrying might not be a great > > > idea. > > > > > > Not saying refill work is a great hack, but that is the reason for it. > > > > In case no allocated received buffer and NAPI refill fails, the host > > will not send any packets. If there is no busy polling loop either, the > > RX will be stuck. That's also the reason why we need refill work. Is it > > correct? > > I've just looked at mlx5e_napi_poll which is mentioned by Jason. So if we > want to retry refilling in the next NAPI, we can set a bool (e.g. > retry_refill) in virtnet_receive, then in virtnet_poll, we don't call > virtqueue_napi_complete. As a result, our napi poll is still in the > softirq's poll list, so we don't need a new host packet to trigger > virtqueue's callback which calls napi_schedule again. > > > > Thanks, > > Quang Minh. > > >
yes yes. but aggressively retrying GFP_ATOMIC until it works is not the thing to do. -- MST
