Re: [RESEND][PATCH-2.6.24-rc8] Fix fakephp deadlock

[Greg KH <gregkh@suse.de>]

On Wed, Jan 23, 2008 at 06:38:22PM +0000, Ian Abbott wrote:
> On 23/01/08 17:46, Greg KH wrote:
>> On Tue, Jan 22, 2008 at 02:28:08PM +0000, Ian Abbott wrote:
>>> #include <linux/init.h>
>>> #include <linux/string.h>
>>> #include <linux/slab.h>
>>> +#include <linux/workqueue.h>
>>> #include "../pci.h"
>>> #if !defined(MODULE)
>>> @@ -63,10 +64,13 @@ struct dummy_slot {
>>> 	struct list_head node;
>>> 	struct hotplug_slot *slot;
>>> 	struct pci_dev *dev;
>>> +	struct work_struct remove_work;
>>> +	unsigned long removed;
>> You are treating "removed" as an atomic value, so why not just make it
>> an atomic_t?
>
> Because I'm using it as a boolean?

Heh, an unsigned long as a boolean?  Come on... :)

>> And what is protecting the fact that the flag could be set right after
>> it gets checked?  I don't see a lock here :)
>
> Okay, it looks like there might be a race condition between enable_slot() 
> and disable_slot() if some other task calls disable_slot() while 
> enable_slot() is between the test_bit() and flush_workqueue() calls.  I can 
> fix that by avoiding the call to flush_workqueue() in enable_slot() and 
> allocating and queueing a work queue item to defer the call to 
> pci_rescan().  And enable_slot() won't then need to check if the slot was 
> marked as removed - it can just go ahead and allocate and queue a work 
> item.

That sounds reasonable.

thanks,

greg k-h