On 10/05/16 15:57, Eric Dumazet wrote: > On Tue, 2016-05-10 at 07:55 -0700, Eric Dumazet wrote: >> If you care only of the initial pages allocated with GFP_KERNEL at >> device start, really that is a small detail as they should be consumed >> and replaced quite fast.
This patch was addressing just that situation. >> If you worry that "wrong" pages would be reused over and over, >> you could make sure that efx_reuse_page() wont reuse a page on the wrong >> node. >> >> page_to_nid(page) != numa_mem_id() Yes, that would solve the general problem and your patch looks correct for the code as it is right now. However, we have some pending changes to our rx recycle ring that haven't been submitted upstream yet which will alter the way that pages are reused, so the solution may no longer be so simple. I think we avoided making the simple change of checking page_to_nid() right now because of these future changes to rx page recycling. So I think that the original patch in this submission is valuable in addressing the initial startup case - that won't be changed by the other rx page recycling changes. I have no preference on whether we take Eric's follow-up patch to check page_to_nid() right now or wait until the other rx changes are submitted and then fix it up as required. Daniel
