lkml.org 
[lkml]   [2004]   [Aug]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC] Fix Device Power Management States

On Tue, 10 Aug 2004, Pavel Machek wrote:

> Can you explain why this class-based quiescing is good idea? It seems
> to me that "quiesce this tree" is pretty much same as "suspend this
> tree", and can be handled in the same way.
>
> Nigel wanted to do class-based quiescing, but if we make quiescing
> fast enough, it should be okay to do whole tree, always. (And I
> believe quiescing *can* be fast enough).

It has nothing to do with speed and everything to do with domains of
responsibility. Each device belongs to 1 device class. That class and the
assoicated class device represent the functional interface(s) to the
physical device. While the physical device controls physical attributes,
like actual I/O and power states, the class controls the logical
interface.

When we quiesce/stop a device, we need to make sure that the device is
going to either finish or cancel its current I/O transactions. If we do
that at a hardware, the higher levels may think the device has gone awry
and completely disable it. If we do it at a higher level, we can make
sure that the request is gracefully delayed. We can also make sure that
there are no new connnections to the device in a saner manner (by simply
removing the class device from the list of advertised interfaces), rather
than adding logic to every low-level driver to stop/restart I/O.

Plus, and not that I'm a feature-monger, the code is useful in other
cases. To throw a buzzword, it's the same as effectively 'taking the
device offline', since it implies the hardware is claimed by a driver but
not accepting connections and effectively idle.

AFAICT, it should also provide a necessary piece of infrastructure for
some upcoming resource management patches (from Adam Belay) that rely on
userspace being able to disable/enable a device, change its resources, and
restart it (ideally with transparency).

And, if it turns out to be too hard, then we can rethink it and start
over. :)


Pat
-
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/

\
 
 \ /
  Last update: 2005-03-22 14:05    [from the cache]
©2003-2011 Jasper Spaans