On Wed, 19 Mar 2008, Rafael J. Wysocki wrote:

quoted text
> In the meantime I had an IRC chat with Ben, who wanted a couple of quite
> substantial changes to be made to this patch.
> 
> First, Ben thinks that ->prepare() should be called in a separate loop for
> all devices, before any of them is suspended, so that drivers can assume the
> availability of the other devices during ->prepare() (for example, so that they
> can use GFP_KERNEL memory allocations).  Accordingly, ->complete() would
> be called in a separate loop after calling ->resume() for all devices.

In order to make this work, you would have to prevent new children from
being registered starting from the time just after prepare() returns,
rather than from the time just before suspend() is called.  Nothing is
wrong with that, but it requires a redesign of the new flags.  (It's 
worth mentioning that drivers will want to have a flag that gets set 
just _before_ prepare() -- just _after_ is too late.)

In the other direction, it ought to be okay to allow new children to 
be registered during resume().  There's no need to wait for complete().

quoted text
> Next, he wants the number of noirq callbacks to be reduced.

I agree.  For example, there isn't much point in having prepare() and
complete() entries for noirq.

quoted text
> Both of the above things make sense, so I'll rework the patch (or maybe even
> all three patches) to implement them and we'll see how they will look like.

Alan Stern

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