Matthew Wilcox wrote:
On Sun, Oct 26, 2008 at 08:51:25PM +0800, Yu Zhao wrote:
This reminds me of other problems of PCI search functions.
The 'dev_start' is passed to bus_find_device(), and its 'knode_bus'
reference count is decreased by klist_iter_init_node() in that function.
The problem is the reference count may be already decrease to 0 because
the PCI device 'from' is hot-plugged off (e.g., pci_remove_bus) when the
search goes. A warning is fired when klist_iter_init_node() detects the
reference count becomes 0.
Some code uses pci_find_device() in a way that is not safe with the
hotplug, because a device may be destroyed after bus_find_device()
returns it and before it's held by pci_dev_get() in the next round.
Following is an example from a random grep:
Yes, that's why pci_find_device() is deprecated. But it doesn't also
need to be buggy ;-)
How about pci_get_bus_and_slot()? People would meet the problem with it
anyway.
Thanks,
Yu
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
