Skip to contentSkip to conversation

PCI: Initialize spinlock in new_id_store()

PATCH New Zhenzhong Duan 1 participant 1 message Apr 17 PCI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 PCI: Initialize spinlock in new_id_store()

new_id_store() creates a dummy pci_dev structure but doesn't initialize driver_override.lock in device structure. The lock is taken by device_match_driver_override() during PCI ID existence checking, then below lock corruption is reported if spinlock debug is enabled.

  BUG: spinlock bad magic on CPU#74, vfio_pci.sh/4575
   lock: 0xff1100018fd3e158, .magic: 00000000, .owner: <none>/-1, .owner_cpu: 0
  Call Trace:
   <TASK>
   dump_stack_lvl+0x51/0x80
   do_raw_spin_lock+0x64/0xc0
   pci_match_device+0x22/0x180
   new_id_store+0x184/0x1e0
   kernfs_fop_write_iter+0x13a/0x1e0
   vfs_write+0x31b/0x440
   ksys_write+0x67/0xe0
   do_syscall_64+0xa5/0x12c0
   entry_SYSCALL_64_after_hwframe+0x76/0x7e

The call trace is harmless as the whole driver_override structure is initialized to zero.

Fixes 10a4206a2401 ("PCI: use generic driver_override infrastructure")
---
 drivers/pci/pci-driver.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/pci/pci-driver.c +3 -0
index dd9075403987..9c3d4606721a 100644
@@ -213,6 +213,9 @@ static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 		pdev->subsystem_device = subdevice;
 		pdev->class = class;
 
+		/* Initialize driver_override.lock as device_match_driver_override() takes it. */
+		spin_lock_init(&pdev->dev.driver_override.lock);
+
 		if (pci_match_device(pdrv, pdev))
 			retval = -EEXIST;
 
Signature · 1 line
2.47.3