PCI driver module unload race?

[Russell King <rmk@arm.linux.org.uk>]

Hi,

What prevents the following scenario from happening?  It's purely
theoretical - I haven't seen this occuring.

- Load PCI driver.

- PCI driver registers using pci_module_init(), and adds itself to sysfs.

- Hot-plugin a PCI device which uses this driver.  sysfs matches the PCI
  driver, and calls the PCI drivers probe function.

- The probe function calls kmalloc or some other function which sleeps
  (or gets preempted, if CONFIG_PREEMPT is enabled.)

- We switch to another thread, which happens to be rmmod for this PCI
  driver.  We remove the driver since it has a use count of zero.

- We switch back to the PCI driver.  Oops.

I've probably missed something, but I don't think so.  I suspect we need
struct device_driver to include a struct module pointer which sysfs can
take before calling any driver functions.

Comments?

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html