Re: apm emulation driver broken ?

[Rafael J. Wysocki]

On Friday, 16 of November 2007, Franck Bui-Huu wrote:
> On Nov 16, 2007 5:20 PM, Rafael J. Wysocki <rjw@xxxxxxx> wrote:
> >
> > The freezer doesn't regard the current task as freezable.
> >
> 
> 
> Hmm, I don't get your point.
> 
> If I understood this driver correctly, several processes can be
> waiting for a suspend event by reading /dev/apm_bios, apmd (the _user_
> space daemon) can be one of them.
> 
> Then another process asks to suspend the system by calling 'apm -s',
> which results in a apm_ioctl() call. This process will basically
> execute:
> 
> 	err = queue_suspend_event(APM_USER_SUSPEND, as);
> 	flags = current->flags;
> 	wait_event_interruptible(apm_suspend_waitqueue,
> 			as->suspend_state == SUSPEND_DONE);
> 
> It's basically waiting for the waiters to ack the event. But it won't
> be the process that is going to suspend the system, right ?
> 
> So now all waiting processes are waken up and need to acknolwedge the
> event for the system to actually suspend. So they need to call
> apm_ioctl(). They'll basically do:
> 
> 	flags = current->flags;
> 	wait_event(apm_suspend_waitqueue,
> 			as->suspend_state == SUSPEND_DONE);
> 
> Except for the last acknowledging process which will do instead:
> 
> 	apm_suspend();
> 
> It's a call to pm_suspend().
> 
> So you can see that the process which initiates the suspend, the one
> that calls 'apm -s', is not the current process but is going to be
> waken up by the fake signal sent by freeze_task().
> 
> One of the consequence I can see is at this time 'as->result' won't be
> setup, so the return value of apm_ioctl() may be wrong.

Ah, that.  Yes, I see your point.

However, using PF_NOFREEZE to prevent this from happening doesn't seem to be
a good idea.

I'd probably use wait_event_freezable() (defined in include/linux/freezer.h)
for that.

> As I said, I'm not familiar with this code, so please correct me if
> I'm wrong.

No, you're not wrong and I have overlooked the problem.

> BTW, how does try_to_freeze_tasks() deal with user land thread waiting
> in the UNINTERRUPTIBLE state ?

It tries to send them fake signals and waits for them to freeze.  If they don't
freeze within the timeout, it fails and clears their TIF_FREEZE bits.

Greetings,
Rafael
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/