Fix the longstanding race between hv_pci_query_relations() and
survey_child_resources() 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@xxxxxxxxxxxxx>
---
drivers/pci/controller/pci-hyperv.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
With the below debug code:
@@ -2103,6 +2103,8 @@ static void survey_child_resources(struct hv_pcibus_device *hbus)
}
spin_unlock_irqrestore(&hbus->device_list_lock, flags);
+ ssleep(15);
+ printk("%s: completing %px\n", __func__, event);
complete(event);
}
@@ -3305,8 +3307,12 @@ static int hv_pci_query_relations(struct hv_device *hdev)
ret = vmbus_sendpacket(hdev->channel, &message, sizeof(message),
0, VM_PKT_DATA_INBAND, 0);
- if (!ret)
+ if (!ret) {
+ ssleep(10); // unassign the PCI device on the host during the 10s
ret = wait_for_response(hdev, &comp);
+ printk("%s: comp=%px is becoming invalid! ret=%d\n",
+ __func__, &comp, ret);
+ }
return ret;
}
@@ -3635,6 +3641,8 @@ static int hv_pci_probe(struct hv_device *hdev,
retry:
ret = hv_pci_query_relations(hdev);
+ printk("hv_pci_query_relations() exited\n");
+
if (ret)
goto free_irq_domain;
I'm able to repro the below hang issue:
[ 74.544744] hv_pci b92a0085-468b-407a-a88a-d33fac8edc75: PCI VMBus probing: Using version 0x10004
[ 76.886944] hv_netvsc 818fe754-b912-4445-af51-1f584812e3c9 eth0: VF slot 1 removed
[ 84.788266] hv_pci b92a0085-468b-407a-a88a-d33fac8edc75: The device is gone.
[ 84.792586] hv_pci_query_relations: comp=ffffa7504012fb58 is becoming invalid! ret=-19
[ 84.797505] hv_pci_query_relations() exited
[ 89.652268] survey_child_resources: completing ffffa7504012fb58
[ 150.392242] rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
[ 150.398447] rcu: 15-...0: (2 ticks this GP) idle=867c/1/0x4000000000000000 softirq=947/947 fqs=5234
[ 150.405851] rcu: (detected by 14, t=15004 jiffies, g=2553, q=4833 ncpus=16)
[ 150.410870] Sending NMI from CPU 14 to CPUs 15:
[ 150.414836] NMI backtrace for cpu 15
[ 150.414840] CPU: 15 PID: 10 Comm: kworker/u32:0 Tainted: G W E 6.3.0-rc3-decui-dirty #34
...
[ 150.414849] Workqueue: hv_pci_468b pci_devices_present_work [pci_hyperv]
[ 150.414866] RIP: 0010:__pv_queued_spin_lock_slowpath+0x10f/0x3c0
...
[ 150.414905] Call Trace:
[ 150.414907] <TASK>
[ 150.414911] _raw_spin_lock_irqsave+0x40/0x50
[ 150.414917] complete+0x1d/0x60
[ 150.414924] pci_devices_present_work+0x5dd/0x680 [pci_hyperv]
[ 150.414946] process_one_work+0x21f/0x430
[ 150.414952] worker_thread+0x4a/0x3c0
With this patch, the hang issue goes away:
[ 186.143612] hv_pci b92a0085-468b-407a-a88a-d33fac8edc75: The device is gone.
[ 186.148034] hv_pci_query_relations: comp=ffffa7cfd0aa3b50 is becoming invalid! ret=-19
[ 191.263611] survey_child_resources: completing ffffa7cfd0aa3b50
[ 191.267732] hv_pci_query_relations() exited
diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index f33370b75628..b82c7cde19e6 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c
@@ -3308,6 +3308,19 @@ 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 survey_child_resources() -> complete(&hbus->survey_event),
+ * even after hv_pci_query_relations() exits and the stack variable
+ * 'comp' is no longer valid. This can cause a strange hang issue
+ * or sometimes a page fault. Flush hbus->wq before we exit from
+ * hv_pci_query_relations() to avoid the issues.
+ */
+ flush_workqueue(hbus->wq);
+
return ret;
}
--
2.25.1
