From mboxrd@z Thu Jan 1 00:00:00 1970 From: Greg KH Subject: Re: [PATCH v3 19/23] staging/rdma/hfi: modify workqueue for parallelism Date: Tue, 27 Oct 2015 17:44:36 +0900 Message-ID: <20151027084436.GA9810@kroah.com> References: <1445869729-7507-1-git-send-email-ira.weiny@intel.com> <1445869729-7507-20-git-send-email-ira.weiny@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline In-Reply-To: <1445869729-7507-20-git-send-email-ira.weiny@intel.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: driverdev-devel-bounces@linuxdriverproject.org Sender: "devel" To: ira.weiny@intel.com Cc: devel@driverdev.osuosl.org, linux-rdma@vger.kernel.org, dledford@redhat.com, dennis.dalessandro@intel.com List-Id: linux-rdma@vger.kernel.org On Mon, Oct 26, 2015 at 10:28:45AM -0400, ira.weiny@intel.com wrote: > From: Mike Marciniszyn > > The workqueue is currently single threaded per port which for a small number of > SDMA engines is ok. > > For hfi1, the there are up to 16 SDMA engines that can be fed descriptors in > parallel. > > This patch: > - Converts to use alloc_workqueue > - Changes the workqueue limit from 1 to num_sdma > - Makes the queue WQ_CPU_INTENSIVE and WQ_HIGHPRI > - The sdma_engine now has a cpu that is initialized > as the MSI-X vectors are setup > - Adjusts the post send logic to call a new scheduler > that doesn't get the s_lock > - The new and old workqueue schedule now pass a > cpu > - post send now uses the new scheduler > - RC/UC QPs now pre-compute the sc, sde > - The sde wq is eliminated since the new hfi1_wq is > multi-threaded When you have to start enumerating all of the different things that your patch does, that's a huge hint that you need to break it up into smaller pieces. Please break this up, it's not acceptable as-is. thanks, greg k-h