From mboxrd@z Thu Jan 1 00:00:00 1970 From: Thomas Monjalon Subject: Re: [RFC] ethdev: complete closing to free all resources Date: Wed, 10 Oct 2018 00:00:07 +0200 Message-ID: <2115860.EbHCjSuZbr@xps> References: <20180907233929.21950-1-thomas@monjalon.net> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7Bit Cc: dev@dpdk.org, arybchenko@solarflare.com To: Ferruh Yigit Return-path: Received: from out3-smtp.messagingengine.com (out3-smtp.messagingengine.com [66.111.4.27]) by dpdk.org (Postfix) with ESMTP id 29CB91B513 for ; Wed, 10 Oct 2018 00:02:55 +0200 (CEST) In-Reply-To: List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" 28/09/2018 14:46, Ferruh Yigit: > On 9/8/2018 12:39 AM, Thomas Monjalon wrote: > > After closing a port, it cannot be restarted. > > So there is no reason to not free all associated resources. > > > > The last step was done with rte_eth_dev_detach() which is deprecated. > > Instead of removing the associated rte_device, the driver should check > > if no more port (ethdev, cryptodev, etc) is still open for the device. > > Then the device resources can be freed by the driver inside the > > dev_close() driver callback operation. > > > > The last ethdev freeing (dev_private and final release), which were done > > by rte_eth_dev_detach(), are now done at the end of rte_eth_dev_close(). > > > > Signed-off-by: Thomas Monjalon > > --- > > This patch contains only the change in the close function as RFC. > > > > This idea was presented at Dublin during the "hotplug talk". > > --- > > @@ -1372,6 +1373,10 @@ rte_eth_dev_close(uint16_t port_id) > > dev->data->nb_tx_queues = 0; > > rte_free(dev->data->tx_queues); > > dev->data->tx_queues = NULL; > > + > > + rte_free(dev->data->dev_private); > > + > > + rte_eth_dev_release_port(dev); > > These already done in: > rte_eth_dev_pci_generic_remove() > rte_eth_dev_pci_release() > > Perhaps all content of rte_eth_dev_pci_release(), including above updates, > should move to rte_eth_dev_close() and rte_eth_dev_pci_generic_remove() call > rte_eth_dev_close() directly. Yes I think it is the way to go: - when removing a rte_device, close all associated ports. and the reverse can be done in addition: - when closing the last port of a rte_device, remove it. > Just thinking aloud, > > driver->probe() called when a new device added. > Application startup can be thought as all devices added one by one. [Perhaps > this can be change in the future to add devices only when explicitly stated] > > driver->remove() called when device removed. > When application terminated this path not called at all, perhaps we need a way > to remove all devices one by one on exit. > > eth_dev_close(), when eth_dev removed in ethdev layer but device is not removed > in eal level, > > I think it make sense for eth_dev_close(): > - It does more cleanup, free resources and port_id > - but it may need to do more clean up, like call uninit() and do more driver > internal clean up too, and clean up the hw. > - so call stack can be: > driver->remove() > rte_eth_dev_pci_generic_remove() > eth_dev_close() > dev_uninit() [driver unint function ] > > > Another question, after eth_dev_close() ethdev is not usable and not able to > restart it back. So why we need eth_dev_close() in addition to dev remove, why > not directly call rte_eth_dev_detach()? rte_eth_dev_detach is assuming a 1:1 mapping between EAL rte_device and rte_eth_dev port. That's why it is deprecated and going to be removed.