PCI/PM: Fix deadlock when unbinding device if parent in D3cold
authorHuang Ying <ying.huang@intel.com>
Wed, 24 Oct 2012 06:54:13 +0000 (14:54 +0800)
committerBjorn Helgaas <bhelgaas@google.com>
Fri, 2 Nov 2012 16:07:17 +0000 (10:07 -0600)
If a PCI device and its parents are put into D3cold, unbinding the
device will trigger deadlock as follow:

- driver_unbind
  - device_release_driver
    - device_lock(dev) <--- previous lock here
    - __device_release_driver
      - pm_runtime_get_sync
        ...
          - rpm_resume(dev)
            - rpm_resume(dev->parent)
              ...
                - pci_pm_runtime_resume
                  ...
                  - pci_set_power_state
                    - __pci_start_power_transition
                      - pci_wakeup_bus(dev->parent->subordinate)
                        - pci_walk_bus
                          - device_lock(dev) <--- deadlock here

If we do not do device_lock in pci_walk_bus, we can avoid deadlock.
Device_lock in pci_walk_bus is introduced in commit:
d71374dafbba7ec3f67371d3b7e9f6310a588808, corresponding email thread
is: https://lkml.org/lkml/2006/5/26/38.  The patch author Zhang Yanmin
said device_lock is added to pci_walk_bus because:

  Some error handling functions call pci_walk_bus. For example, PCIe
  aer. Here we lock the device, so the driver wouldn't detach from the
  device, as the cb might call driver's callback function.

So I fixed the deadlock as follows:

- remove device_lock from pci_walk_bus
- add device_lock into callback if callback will call driver's callback

I checked pci_walk_bus users one by one, and found only PCIe aer needs
device lock.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org # v3.6+
CC: Zhang Yanmin <yanmin.zhang@intel.com>
drivers/pci/bus.c
drivers/pci/pcie/aer/aerdrv_core.c

index 6241fd0..a543746 100644 (file)
@@ -320,10 +320,7 @@ void pci_walk_bus(struct pci_bus *top, int (*cb)(struct pci_dev *, void *),
                } else
                        next = dev->bus_list.next;
 
-               /* Run device routines with the device locked */
-               device_lock(&dev->dev);
                retval = cb(dev, userdata);
-               device_unlock(&dev->dev);
                if (retval)
                        break;
        }
index 06bad96..af4e31c 100644 (file)
@@ -213,6 +213,7 @@ static int report_error_detected(struct pci_dev *dev, void *data)
        struct aer_broadcast_data *result_data;
        result_data = (struct aer_broadcast_data *) data;
 
+       device_lock(&dev->dev);
        dev->error_state = result_data->state;
 
        if (!dev->driver ||
@@ -231,12 +232,14 @@ static int report_error_detected(struct pci_dev *dev, void *data)
                                   dev->driver ?
                                   "no AER-aware driver" : "no driver");
                }
-               return 0;
+               goto out;
        }
 
        err_handler = dev->driver->err_handler;
        vote = err_handler->error_detected(dev, result_data->state);
        result_data->result = merge_result(result_data->result, vote);
+out:
+       device_unlock(&dev->dev);
        return 0;
 }
 
@@ -247,14 +250,17 @@ static int report_mmio_enabled(struct pci_dev *dev, void *data)
        struct aer_broadcast_data *result_data;
        result_data = (struct aer_broadcast_data *) data;
 
+       device_lock(&dev->dev);
        if (!dev->driver ||
                !dev->driver->err_handler ||
                !dev->driver->err_handler->mmio_enabled)
-               return 0;
+               goto out;
 
        err_handler = dev->driver->err_handler;
        vote = err_handler->mmio_enabled(dev);
        result_data->result = merge_result(result_data->result, vote);
+out:
+       device_unlock(&dev->dev);
        return 0;
 }
 
@@ -265,14 +271,17 @@ static int report_slot_reset(struct pci_dev *dev, void *data)
        struct aer_broadcast_data *result_data;
        result_data = (struct aer_broadcast_data *) data;
 
+       device_lock(&dev->dev);
        if (!dev->driver ||
                !dev->driver->err_handler ||
                !dev->driver->err_handler->slot_reset)
-               return 0;
+               goto out;
 
        err_handler = dev->driver->err_handler;
        vote = err_handler->slot_reset(dev);
        result_data->result = merge_result(result_data->result, vote);
+out:
+       device_unlock(&dev->dev);
        return 0;
 }
 
@@ -280,15 +289,18 @@ static int report_resume(struct pci_dev *dev, void *data)
 {
        const struct pci_error_handlers *err_handler;
 
+       device_lock(&dev->dev);
        dev->error_state = pci_channel_io_normal;
 
        if (!dev->driver ||
                !dev->driver->err_handler ||
                !dev->driver->err_handler->resume)
-               return 0;
+               goto out;
 
        err_handler = dev->driver->err_handler;
        err_handler->resume(dev);
+out:
+       device_unlock(&dev->dev);
        return 0;
 }