Since day 1 of the driver, there has been a race between hv_pci_query_relations() and survey_child_resources(): during fast device hotplug, hv_pci_query_relations() may error out due to device-remove and the stack variable 'comp' is no longer valid; however, pci_devices_present_work() -> survey_child_resources() -> complete() may be running on another CPU and accessing the no-longer-valid 'comp'. Fix the race by flushing the workqueue before we exit from hv_pci_query_relations().
Fixes: 4daace0d8ce8 ("PCI: hv: Add paravirtual PCI front-end for Microsoft Hyper-V VMs") Signed-off-by: Dexuan Cui decui@microsoft.com Reviewed-by: Michael Kelley mikelley@microsoft.com Acked-by: Lorenzo Pieralisi lpieralisi@kernel.org Cc: stable@vger.kernel.org ---
v2: Removed the "debug code". No change to the patch body. Added Cc:stable
v3: Added Michael's Reviewed-by.
v4: Provided more details of the issue in the comment and commit log. Added Lorenzo's Acked-by.
drivers/pci/controller/pci-hyperv.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c index bc32662c6bb7f..ea8862e656b68 100644 --- a/drivers/pci/controller/pci-hyperv.c +++ b/drivers/pci/controller/pci-hyperv.c @@ -3401,6 +3401,24 @@ static int hv_pci_query_relations(struct hv_device *hdev) if (!ret) ret = wait_for_response(hdev, &comp);
+ /* + * In the case of fast device addition/removal, it's possible that + * vmbus_sendpacket() or wait_for_response() returns -ENODEV but we + * already got a PCI_BUS_RELATIONS* message from the host and the + * channel callback already scheduled a work to hbus->wq, which can be + * running pci_devices_present_work() -> survey_child_resources() -> + * complete(&hbus->survey_event), even after hv_pci_query_relations() + * exits and the stack variable 'comp' is no longer valid; as a result, + * a hang or a page fault may happen when the complete() calls + * raw_spin_lock_irqsave(). Flush hbus->wq before we exit from + * hv_pci_query_relations() to avoid the issues. Note: if 'ret' is + * -ENODEV, there can't be any more work item scheduled to hbus->wq + * after the flush_workqueue(): see vmbus_onoffer_rescind() -> + * vmbus_reset_channel_cb(), vmbus_rescind_cleanup() -> + * channel->rescind = true. + */ + flush_workqueue(hbus->wq); + return ret; }