PCI: hv: Report resources release after stopping the bus
authorVitaly Kuznetsov <vkuznets@redhat.com>
Fri, 29 Apr 2016 09:39:10 +0000 (11:39 +0200)
committerBjorn Helgaas <bhelgaas@google.com>
Mon, 2 May 2016 20:40:42 +0000 (15:40 -0500)
Kernel hang is observed when pci-hyperv module is release with device
drivers still attached.  E.g., when I do 'rmmod pci_hyperv' with BCM5720
device pass-through-ed (tg3 module) I see the following:

 NMI watchdog: BUG: soft lockup - CPU#1 stuck for 22s! [rmmod:2104]
 ...
 Call Trace:
  [<ffffffffa0641487>] tg3_read_mem+0x87/0x100 [tg3]
  [<ffffffffa063f000>] ? 0xffffffffa063f000
  [<ffffffffa0644375>] tg3_poll_fw+0x85/0x150 [tg3]
  [<ffffffffa0649877>] tg3_chip_reset+0x357/0x8c0 [tg3]
  [<ffffffffa064ca8b>] tg3_halt+0x3b/0x190 [tg3]
  [<ffffffffa0657611>] tg3_stop+0x171/0x230 [tg3]
  ...
  [<ffffffffa064c550>] tg3_remove_one+0x90/0x140 [tg3]
  [<ffffffff813bee59>] pci_device_remove+0x39/0xc0
  [<ffffffff814a3201>] __device_release_driver+0xa1/0x160
  [<ffffffff814a32e3>] device_release_driver+0x23/0x30
  [<ffffffff813b794a>] pci_stop_bus_device+0x8a/0xa0
  [<ffffffff813b7ab6>] pci_stop_root_bus+0x36/0x60
  [<ffffffffa02c3f38>] hv_pci_remove+0x238/0x260 [pci_hyperv]

The problem seems to be that we report local resources release before
stopping the bus and removing devices from it and device drivers may try to
perform some operations with these resources on shutdown.  Move resources
release report after we do pci_stop_root_bus().

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Jake Oshins <jakeo@microsoft.com>
drivers/pci/host/pci-hyperv.c

index ed651ba..328c653 100644 (file)
@@ -2268,11 +2268,6 @@ static int hv_pci_remove(struct hv_device *hdev)
 
        hbus = hv_get_drvdata(hdev);
 
-       ret = hv_send_resources_released(hdev);
-       if (ret)
-               dev_err(&hdev->device,
-                       "Couldn't send resources released packet(s)\n");
-
        memset(&pkt.teardown_packet, 0, sizeof(pkt.teardown_packet));
        init_completion(&comp_pkt.host_event);
        pkt.teardown_packet.completion_func = hv_pci_generic_compl;
@@ -2295,6 +2290,11 @@ static int hv_pci_remove(struct hv_device *hdev)
                pci_unlock_rescan_remove();
        }
 
+       ret = hv_send_resources_released(hdev);
+       if (ret)
+               dev_err(&hdev->device,
+                       "Couldn't send resources released packet(s)\n");
+
        vmbus_close(hdev->channel);
 
        /* Delete any children which might still exist. */