PCI: pciehp: Use per-slot workqueues to avoid deadlock

When we have a hotplug-capable PCIe port with a second hotplug-capable
PCIe port below it, removing the device below the upstream port causes
a deadlock.

The deadlock happens because we use the pciehp_wq workqueue to run
pciehp_power_thread(), which uses pciehp_disable_slot() to remove devices
below the upstream port.  When we remove the downstream PCIe port, we call
pciehp_remove(), the pciehp driver's .remove() method.  That calls
flush_workqueue(pciehp_wq), which deadlocks because the
pciehp_power_thread() work item is still running.

This patch avoids the deadlock by creating a workqueue for every PCIe port
and removing the single shared workqueue.

Here's the call path that leads to the deadlock:

  pciehp_queue_pushbutton_work
    queue_work(pciehp_wq)                   # queue pciehp_power_thread
    ...

  pciehp_power_thread
    pciehp_disable_slot
      remove_board
	pciehp_unconfigure_device
	  pci_stop_and_remove_bus_device
	    ...
	      pciehp_remove                 # pciehp driver .remove method
		pciehp_release_ctrl
		  pcie_cleanup_slot
		    flush_workqueue(pciehp_wq)

This is fairly urgent because it can be caused by simply unplugging a
Thunderbolt adapter, as reported by Daniel below.

[bhelgaas: changelog]
Reference: http://lkml.kernel.org/r/CAMVG2ssiRgcTD1bej2tkUUfsWmpL5eNtPcNif9va2-Gzb2u8nQ@mail.gmail.com
Reported-and-tested-by: Daniel J Blueman <daniel@quora.org>
Reviewed-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Yijing Wang <wangyijing@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org

1 files changed, 10 insertions, 1 deletions

diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
index 13b2eaf7ba43..5127f3f41821 100644
--- a/drivers/pci/hotplug/pciehp_hpc.c
+++ b/drivers/pci/hotplug/pciehp_hpc.c
@@ -773,23 +773,32 @@ static void pcie_shutdown_notification(struct controller *ctrl)
 static int pcie_init_slot(struct controller *ctrl)
 {
 	struct slot *slot;
+	char name[32];
 
 	slot = kzalloc(sizeof(*slot), GFP_KERNEL);
 	if (!slot)
 		return -ENOMEM;
 
+	snprintf(name, sizeof(name), "pciehp-%u", PSN(ctrl));
+	slot->wq = alloc_workqueue(name, 0, 0);
+	if (!slot->wq)
+		goto abort;
+
 	slot->ctrl = ctrl;
 	mutex_init(&slot->lock);
 	INIT_DELAYED_WORK(&slot->work, pciehp_queue_pushbutton_work);
 	ctrl->slot = slot;
 	return 0;
+abort:
+	kfree(slot);
+	return -ENOMEM;
 }
 
 static void pcie_cleanup_slot(struct controller *ctrl)
 {
 	struct slot *slot = ctrl->slot;
 	cancel_delayed_work(&slot->work);
-	flush_workqueue(pciehp_wq);
+	destroy_workqueue(slot->wq);
 	kfree(slot);
 }