Return-Path: Received: from quartz.orcorp.ca ([184.70.90.242]:48913 "EHLO quartz.orcorp.ca" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755799AbbJASiW (ORCPT ); Thu, 1 Oct 2015 14:38:22 -0400 Date: Thu, 1 Oct 2015 12:38:18 -0600 From: Jason Gunthorpe To: Chuck Lever Cc: linux-rdma , Devesh Sharma , Sagi Grimberg , Linux NFS Mailing List Subject: Re: [PATCH v1 03/18] xprtrdma: Remove completion polling budgets Message-ID: <20151001183818.GA9477@obsidianresearch.com> References: <55FE8C0F.1050706@dev.mellanox.co.il> <0804C887-9E32-4257-96D2-6C1FBC9CB271@oracle.com> <20151001171310.GA8428@obsidianresearch.com> <20151001181548.GA8670@obsidianresearch.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: Sender: linux-nfs-owner@vger.kernel.org List-ID: On Thu, Oct 01, 2015 at 02:31:41PM -0400, Chuck Lever wrote: > Does a ULP want to continue polling if ib_poll_cq{_and_arm) > returns a negative RC? No. We should try and figure out if that can even happen, if not get rid of the possibility. If it can happen, it can only mean the CQ is busted, and needs to be destroyed. Ie the whole ULP probably needs to restart. Jason