[linux-pm] Freezer: Don't count threads waiting for frozen filesystems.
Alan Stern
stern at rowland.harvard.edu
Thu Oct 30 10:07:27 PDT 2008
On Thu, 30 Oct 2008, Miklos Szeredi wrote:
> On Thu, 30 Oct 2008, Alan Stern wrote:
> > On Thu, 30 Oct 2008, Miklos Szeredi wrote:
> >
> > > On Wed, 29 Oct 2008, Alan Stern wrote:
> > > > I discussed this last summer with Rafael. It's a lot harder than it
> > > > looks, for all sorts of reasons. For example, what about user tasks
> > > > that have access to memory-mapped I/O regions?
> > >
> > > What about them? Freezing doesn't seem to help with that.
> >
> > Sure it does. A frozen process can't touch a memory-mapped I/O region,
> > whereas a non-frozen process can.
>
> But it can be in the middle of I/O by your definition.
True. Yet another problem...
> > Would you like to write a first-pass patch? I don't think it will
> > work.
>
> If somebody doesn't beat me to it, I'll do that (first implemented
> with a global rw-sem).
Converting it to per-CPU counters later on should be fairly easy.
> > Doing that seems like a lot of work, just as modifying every driver
> > does. Changing a few kernel entry points is simpler, but I'm pretty
> > sure it won't work. For instance, tasks can block arbitrarily long on
> > read calls (waiting for data to arrive); you can't allow such things to
> > prevent the system from suspending.
>
> But we already do: either
>
> a) it's in interruptible sleep (I/O on sockets, pipes, etc), and
> freezing simply interrupts it, or
>
> b) it's in uninterruptible sleep and suspend will wait it out (or
> time out).
>
> In the new scheme we could retain that part of the freezer: interrupt
> all tasks which are inside the critical region and wait for them to
> exit the critical region.
>
> To put it in another way: it's still the freezer, it does all the same
> things as the old freezer, except that the condition for freezing is
> not that the task is out of the kernel, rather that it's out of the
> disable_supend - enable_suspend region. As such it's not a big change
> to the whole suspend system, and so there shouldn't be anything big
> going wrong there.
Okay. Don't forget things like ioctl for sockets -- they often involve
doing I/O directly to the network interface device.
What happens to a task accessing a non-regular file on a fuse
filesystem? :-)
Alan Stern
More information about the linux-pm
mailing list