From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Steven J. Magnani" Subject: tcp_recvmesg hang under NET_DMA Date: Tue, 09 Mar 2010 14:44:59 -0600 Message-ID: <1268167499.2985.32.camel@iscandar.digidescorp.com> Reply-To: steve@digidescorp.com Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Cc: linux-kernel@vger.kernel.org, "David S. Miller" To: netdev@vger.kernel.org Return-path: Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Hi, I've run into a case where using NET_DMA seems to cause a PUT to a modified ftpd to hang. AFAICT this is because the NET_DMA design queues up lots of memcpy operations, but doesn't issue or wait for them (and thus free the associated skbs) until it is time for tcp_recvmesg() to return. In a case where userland requests a large read (as is the case in my modified ftpd), packet transfer can grind to a halt because so many packets are buffered up and there is no event that causes them to be freed. I can submit a patch that would issue pending memcpys and clean up completed ones around tcp_recvmsg's call to sk_wait_data(), but first I wanted to understand the design better. Were there specific reasons for deferring the async operations so long? Thanks, ------------------------------------------------------------------------ Steven J. Magnani "I claim this network for MARS! www.digidescorp.com Earthling, return my space modulator!" #include