PCI: Avoid pci_dev_lock() AB/BA deadlock with sriov_numvfs_store()
authorYicong Yang <yangyicong@hisilicon.com>
Mon, 4 Apr 2022 06:25:39 +0000 (14:25 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 9 Jun 2022 08:22:42 +0000 (10:22 +0200)
[ Upstream commit a91ee0e9fca9d7501286cfbced9b30a33e52740a ]

The sysfs sriov_numvfs_store() path acquires the device lock before the
config space access lock:

  sriov_numvfs_store
    device_lock                 # A (1) acquire device lock
    sriov_configure
      vfio_pci_sriov_configure  # (for example)
        vfio_pci_core_sriov_configure
          pci_disable_sriov
            sriov_disable
              pci_cfg_access_lock
                pci_wait_cfg    # B (4) wait for dev->block_cfg_access == 0

Previously, pci_dev_lock() acquired the config space access lock before the
device lock:

  pci_dev_lock
    pci_cfg_access_lock
      dev->block_cfg_access = 1 # B (2) set dev->block_cfg_access = 1
    device_lock                 # A (3) wait for device lock

Any path that uses pci_dev_lock(), e.g., pci_reset_function(), may
deadlock with sriov_numvfs_store() if the operations occur in the sequence
(1) (2) (3) (4).

Avoid the deadlock by reversing the order in pci_dev_lock() so it acquires
the device lock before the config space access lock, the same as the
sriov_numvfs_store() path.

[bhelgaas: combined and adapted commit log from Jay Zhou's independent
subsequent posting:
https://lore.kernel.org/r/20220404062539.1710-1-jianjay.zhou@huawei.com]
Link: https://lore.kernel.org/linux-pci/1583489997-17156-1-git-send-email-yangyicong@hisilicon.com/
Also-posted-by: Jay Zhou <jianjay.zhou@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/pci/pci.c

index 0380543..09815cb 100644 (file)
@@ -5079,18 +5079,18 @@ static int pci_reset_bus_function(struct pci_dev *dev, bool probe)
 
 static void pci_dev_lock(struct pci_dev *dev)
 {
-       pci_cfg_access_lock(dev);
        /* block PM suspend, driver probe, etc. */
        device_lock(&dev->dev);
+       pci_cfg_access_lock(dev);
 }
 
 /* Return 1 on successful lock, 0 on contention */
 int pci_dev_trylock(struct pci_dev *dev)
 {
-       if (pci_cfg_access_trylock(dev)) {
-               if (device_trylock(&dev->dev))
+       if (device_trylock(&dev->dev)) {
+               if (pci_cfg_access_trylock(dev))
                        return 1;
-               pci_cfg_access_unlock(dev);
+               device_unlock(&dev->dev);
        }
 
        return 0;
@@ -5099,8 +5099,8 @@ EXPORT_SYMBOL_GPL(pci_dev_trylock);
 
 void pci_dev_unlock(struct pci_dev *dev)
 {
-       device_unlock(&dev->dev);
        pci_cfg_access_unlock(dev);
+       device_unlock(&dev->dev);
 }
 EXPORT_SYMBOL_GPL(pci_dev_unlock);