Commit 301c5aee authored by Jin Yu's avatar Jin Yu Committed by Tomasz Zawadzki
Browse files

pci: fix the hotplug issue 



DPDK intr thread is designed that it can't unregister the src
callback in this callback handler. So I think we can't detach
the PCI device in the hotremove callback as it needs to unregister
the VFIO notification callback which will be not successful
but it still can free the device. So at the next req notification
in the handler function, we meet the freed device.

Fix #994

Change-Id: Id4b45a2d0fe6b45b132355d59471bc80240fad70
Signed-off-by: default avatarJin Yu <jin.yu@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/473176


Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: default avatarChangpeng Liu <changpeng.liu@intel.com>
Reviewed-by: default avatarShuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
parent c330fbe6

lib/env_dpdk/pci.c

+17 −3
Original line number Diff line number Diff line
@@ -176,6 +176,12 @@ spdk_pci_driver_register(struct spdk_pci_driver *driver) 
}



#if RTE_VERSION >= RTE_VERSION_NUM(18, 5, 0, 0)

static void

spdk_pci_device_rte_hotremove_cb(void *dev)

{

	spdk_detach_rte((struct spdk_pci_device *)dev);

}



static void

spdk_pci_device_rte_hotremove(const char *device_name,

			      enum rte_dev_event_type event,
@@ -191,7 +197,6 @@ spdk_pci_device_rte_hotremove(const char *device_name, 
	pthread_mutex_lock(&g_pci_mutex);

	TAILQ_FOREACH(dev, &g_pci_devices, internal.tailq) {

		struct rte_pci_device *rte_dev = dev->dev_handle;



		if (strcmp(rte_dev->name, device_name) == 0 &&

		    !dev->internal.pending_removal) {

			can_detach = !dev->internal.attached;
@@ -203,8 +208,17 @@ spdk_pci_device_rte_hotremove(const char *device_name, 
	pthread_mutex_unlock(&g_pci_mutex);



	if (dev != NULL && can_detach) {

		/* if device is not attached, we can remove it right away. */

		spdk_detach_rte(dev);

		/* If device is not attached, we can remove it right away.

		 *

		 * Because the user's callback is invoked in eal interrupt

		 * callback, the interrupt callback need to be finished before

		 * it can be unregistered when detaching device. So finish

		 * callback soon and use a deferred removal to detach device

		 * is need. It is a workaround, once the device detaching be

		 * moved into the eal in the future, the deferred removal could

		 * be deleted.

		 */

		rte_eal_alarm_set(1, spdk_pci_device_rte_hotremove_cb, dev);

	}

}

#endif