From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sujeev Dias Subject: Reclaiming memory for network interface Date: Tue, 16 Oct 2018 09:36:08 -0700 Message-ID: Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Cc: netdev@vger.kernel.org, Tony Truong To: unlisted-recipients:; (no To-header on input) Return-path: Received: from smtp.codeaurora.org ([198.145.29.96]:36362 "EHLO smtp.codeaurora.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726778AbeJQA1Z (ORCPT ); Tue, 16 Oct 2018 20:27:25 -0400 Content-Language: en-US Sender: netdev-owner@vger.kernel.org List-ID: Hi Setup: sdm845 connected to external modem over pcie interface During a data call, we found out we spend more than 25% of cpu for memory ops with io coherency.  That include allocation, freeing, dma mapping, and unmapping.  As we pushing to higher data rate (beyond 7 Gbps), the time we spend in memory operation is significant. So, we're looking into ways we can reclaim this memory. One of idea we're thinking is: 1. allocate pages 2. Increment reference count of page 3. allocate skb, and assign page into paged data portion 4. Assign cb function to skb->destructor 5. once destructor get called, move the page to a new skb Sound simple enough, but we couldn't find anyone actually doing this way.  Anything to be concern with above proposal? We see some example of using destructor to do deferred unmap but didn't see any example of re-using the buffer. Also, couldn't find any meaningful discussion about reclaiming memory for network data. Any thoughts on how we should solve this issue?  Any comment is welcome, thanks. Sincerely Sujeev -- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project