On Tue, Jan 10, 2012 at 5:30 PM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote: > On Tue, 10 Jan 2012 09:41:01 -0800, Mike Waychison <mikew@xxxxxxxxxx> wrote: >> In preparation for allocating receive buffers in the slow path without >> disabling NAPI, split the allocation and addition of receive buffers >> apart into two separate functions (per receive buffer type). >> >> While here, move the vi->num accounting into the add functions. >> >> Signed-off-by: Mike Waychison <mikew@xxxxxxxxxx> > > Hi Mike... > > This exposes a nasty ugliness in the way virtio_net works. We > allocate an skbuff for the small packet case, and just allocate the > pages for the large packet cases, and alloc the skbuff when we fill the > pages. > > I think all the allocators should return a populated skbuff; > this uses a bit more memory in theory, but should make the code simpler. > As an added bonus, your life should get much simpler for these patches. > > I'll try to create such a patch tonight, but I'm busy finalizing my > linux.conf.au presentation, so it might take longer :( > This seems reasonable for the "big" receive buffers, but I don't think it makes a lot of sense for "mergeable" receive buffers as best I can tell. Hopefully making this work isn't too hard though :) _______________________________________________ Virtualization mailing list Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linuxfoundation.org/mailman/listinfo/virtualization
