On di, 08 okt 2019 10:32:18 +0200, Marc Kleine-Budde wrote: > On 10/8/19 10:24 AM, Kurt Van Dijck wrote: > >>>> taking up Kurt's work. I've cleaned up the rx-offload and c_can patches > >>>> a bit. Untested as I don't have any hardware at hand. > >>> > >>> I had created equivalent code (skb_queue in isr, skb_dequeue in napi > >>> handler) running on a 4.9 kernel since some days now. I didn't observe > >>> any problems yet. > >> > >> This is based on the patches you send around. > >> Anyways can you send me your currently working version? > > > > I know. My first attempt was to backport rx-offload, but this was more > > work than expected, so I created this patch, doing the skb_queue inside > > c_can driver directly. I wrote that patch with the latest rx-offload.c > > side-by-side. > > > > I just wrote it this way quickly so I could go ahead and test the mower, > > upgrading to a more recent kernel is scheduled within a few months or > > so. > > Ok, so which variant should be integrated into the kernel? The maintenance of a linear skb_queue by itself felt not that complicated. The napi part however requires more attention. My ad-hoc skb_queue implementation feels dirty given the rx-offload work. The risk of having duplicated napi-handlers convinced me to use rx-offload, and I'm still convinced of that. So I vote for the rx-offload variant. As said, I created my ad-hoc skb_queue because rx-offload wasn't in v4.9. Kurt
