[dpdk-dev] [PATCH v2 4/4] vfio: enable vfio hotplug by req notifier handler

[Jeff Guo <jia.guo@intel.com>]

When device is be hot-unplugged, the vfio kernel module will sent req
notifier to request user space to release the allocated resources at
first. After that, vfio kernel module will detect the device disappear,
and then delete the device in kernel.

This patch aim to add req notifier processing to enable hotplug for vfio.
By enable the req notifier monitoring and register the notifier callback,
when device be hot-unplugged, the hot-unplug handler will be called to
process hotplug for vfio.

Signed-off-by: Jeff Guo <jia.guo@intel.com>
---
v2->v1:
refine some code logic.
---
 drivers/bus/pci/linux/pci_vfio.c | 95 ++++++++++++++++++++++++++++++++++++++++
 drivers/bus/pci/pci_common.c     | 10 +++++
 2 files changed, 105 insertions(+)

diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c
index 686386d..c780860 100644
--- a/drivers/bus/pci/linux/pci_vfio.c
+++ b/drivers/bus/pci/linux/pci_vfio.c
@@ -17,6 +17,8 @@
 #include <rte_eal_memconfig.h>
 #include <rte_malloc.h>
 #include <rte_vfio.h>
+#include <rte_eal.h>
+#include <rte_bus.h>
 
 #include "eal_filesystem.h"
 
@@ -277,6 +279,89 @@ pci_vfio_setup_interrupts(struct rte_pci_device *dev, int vfio_dev_fd)
 	return -1;
 }
 
+static void
+pci_vfio_req_handler(void *param)
+{
+	struct rte_bus *bus;
+	int ret;
+	struct rte_device *device = (struct rte_device *)param;
+
+	bus = rte_bus_find_by_device(device);
+	if (bus == NULL) {
+		RTE_LOG(ERR, EAL, "Cannot find bus for device (%s)\n",
+			device->name);
+		return;
+	}
+
+	/**
+	 * vfio kernel module request user space to release allocated
+	 * resources before device be deleted in kernel, so it can directly
+	 * call the vfio bus hot-unplug handler to process it.
+	 */
+	ret = bus->hot_unplug_handler(device);
+	if (ret)
+		RTE_LOG(ERR, EAL, "Can not handle hot-unplug for "
+			"device (%s)\n", device->name);
+}
+
+/* enable notifier (only enable req now) */
+static int
+pci_vfio_enable_notifier(struct rte_pci_device *dev, int vfio_dev_fd)
+{
+	int ret;
+	int fd = -1;
+
+	/* set up an eventfd for req notifier */
+	fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
+	if (fd < 0) {
+		RTE_LOG(ERR, EAL, "Cannot set up eventfd, "
+				"error %i (%s)\n", errno, strerror(errno));
+		return -1;
+	}
+
+	dev->req_notifier_handler.fd = fd;
+	dev->req_notifier_handler.type = RTE_INTR_HANDLE_VFIO_REQ;
+	dev->req_notifier_handler.vfio_dev_fd = vfio_dev_fd;
+	ret = rte_intr_callback_register(&dev->req_notifier_handler,
+					 pci_vfio_req_handler,
+					 (void *)&dev->device);
+	if (ret) {
+		RTE_LOG(ERR, EAL, "Fail to register req notifier handler.\n");
+		return -1;
+	}
+
+	ret = rte_intr_enable(&dev->req_notifier_handler);
+	if (ret) {
+		RTE_LOG(ERR, EAL, "Fail to enable req notifier.\n");
+		return -1;
+	}
+
+	return 0;
+}
+
+/*disable notifier (only disable req now) */
+static int
+pci_vfio_disable_notifier(struct rte_pci_device *dev)
+{
+	int ret;
+
+	ret = rte_intr_disable(&dev->req_notifier_handler);
+	if (ret) {
+		RTE_LOG(ERR, EAL, "fail to disable req notifier.\n");
+		return -1;
+	}
+
+	ret = rte_intr_callback_unregister(&dev->req_notifier_handler,
+					   pci_vfio_req_handler,
+					   (void *)&dev->device);
+	if (ret) {
+		RTE_LOG(ERR, EAL,
+			 "fail to unregister req notifier handler.\n");
+		return -1;
+	}
+	return 0;
+}
+
 static int
 pci_vfio_is_ioport_bar(int vfio_dev_fd, int bar_index)
 {
@@ -430,6 +515,7 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev)
 	struct pci_map *maps;
 
 	dev->intr_handle.fd = -1;
+	dev->req_notifier_handler.fd = -1;
 
 	/* store PCI address string */
 	snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT,
@@ -521,6 +607,11 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev)
 		goto err_vfio_res;
 	}
 
+	if (pci_vfio_enable_notifier(dev, vfio_dev_fd) != 0) {
+		RTE_LOG(ERR, EAL, "Error setting up notifier!\n");
+		return -1;
+	}
+
 	TAILQ_INSERT_TAIL(vfio_res_list, vfio_res, next);
 
 	return 0;
@@ -546,6 +637,7 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev)
 	struct pci_map *maps;
 
 	dev->intr_handle.fd = -1;
+	dev->req_notifier_handler.fd = -1;
 
 	/* store PCI address string */
 	snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT,
@@ -586,6 +678,7 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev)
 
 	/* we need save vfio_dev_fd, so it can be used during release */
 	dev->intr_handle.vfio_dev_fd = vfio_dev_fd;
+	dev->req_notifier_handler.vfio_dev_fd = vfio_dev_fd;
 
 	return 0;
 err_vfio_dev_fd:
@@ -658,6 +751,8 @@ pci_vfio_unmap_resource_primary(struct rte_pci_device *dev)
 	snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT,
 			loc->domain, loc->bus, loc->devid, loc->function);
 
+	pci_vfio_disable_notifier(dev);
+
 	if (close(dev->intr_handle.fd) < 0) {
 		RTE_LOG(INFO, EAL, "Error when closing eventfd file descriptor for %s\n",
 			pci_addr);
diff --git a/drivers/bus/pci/pci_common.c b/drivers/bus/pci/pci_common.c
index f313fe9..2a8e5e9 100644
--- a/drivers/bus/pci/pci_common.c
+++ b/drivers/bus/pci/pci_common.c
@@ -446,6 +446,16 @@ pci_hot_unplug_handler(struct rte_device *dev)
 		return -1;
 
 	switch (pdev->kdrv) {
+	case RTE_KDRV_VFIO:
+		/**
+		 * vfio kernel module guaranty the pci device would not be
+		 * deleted until the user space release the resource, so no
+		 * need to remap BARs resource here, just directly notify
+		 * the req event to the user space to handle it.
+		 */
+		rte_dev_event_callback_process(dev->name,
+					       RTE_DEV_EVENT_REMOVE);
+		break;
 	case RTE_KDRV_IGB_UIO:
 	case RTE_KDRV_UIO_GENERIC:
 	case RTE_KDRV_NIC_UIO:
-- 
2.7.4