Greg and everyone else:
The following patches form a general clean-up and reorganization of the
suspend/resume support in usbcore. I developed them in the course of
working on autosuspend and autoresume. That work isn't finished yet, but
it still seemed like a good idea to submit these first because they are
fairly substantial and it wouldn't hurt to get some feedback on them.
Although several of these changes are of interest, only two of them (7 and
9) should cause any noticeable difference in the core's behavior. The
rest are more along the lines of bug fixes and code re-arrangement.
Here's what the individual patches do:
1. Change the disconnect checking in devio.c (the usbfs code) so
that it will detect cases where the device has been unregistered
from usbfs, even if the device is still connected. Although this
can't happen now, it will be able to occur after patch 4.
2. Fix usb-skeleton so that it won't try to submit URBs after its
disconnect method has returned. It's quite likely that there
are drivers based on usb-skeleton in need of the same fix, but
I don't know which ones they are.
3. Move some driver-related code from usb.c to driver.c. These
routines (concerned with uevent generation and claiming/releasing
interfaces as well as suspend/resume handling) should have gone
into driver.c when it was first split off, but they got left
behind somehow.
4. Make usb_generic into a genuine device driver. This involves
adding a mechanism to tell apart USB interface drivers from USB
device drivers -- of which usb_generic is currently the only one.
Doing this opens the possibility of having alternate USB device
drivers. For example, when a USB device is exported over the
network or to a guest (virtual) operating system, it should be
driven by the other system. The host to which the device is
actually connected would bind a stub driver in place of
usb_generic.
5. Rename usb_suspend_device to usb_port_suspend (and likewise for
_resume). Originally I did this because I wanted to name a
different routine "usb_suspend_device". In the end I chose a
separate name, but it still seemed worthwhile to keep the
"usb_port_suspend" name, for two reasons: The routine isn't any
longer a main entry into the suspend pathway, and it isn't much
more than a wrapper around hub_port_suspend.
6. Split the core suspend and resume routines into two pieces, one
for handling USB devices and the other for handling USB
interfaces. This is cleaner than doing them both in one routine.
7. Remove the recursion in suspend and resume. This has been overdue
for some time. It won't affect system suspends, since the PM core
does its own recursion through the entire device tree. It will
affect only runtime PM done by userspace writing to power/state
attribute files in sysfs. (That whole mechanism is slated to
be overhauled anyway.)
8. General cleanup of the suspend/resume code. Duplicated code,
unnecessary checks, that sort of thing.
9. Tie together suspend/resume operations on a device and on its
interfaces. There doesn't seem to be much need for the peculiar
intermediate state where a USB device is awake but some of its
interfaces are suspended. The patch effectively rules them out;
attempts to suspend or resume an interface won't do anything,
and attempts to suspend or resume a device will affect all its
interfaces as well as the device as a whole.
This stuff works okay in preliminary light testing, but it could use some
more, especially with different sorts of host controllers. I expect a
certain amount of adjustment/cleanup will be needed. Not much, hopefully.
The actual autosuspend changes should require only a pair of additional
patches: one to add the core mechanisms, and one to start using it in
usbfs and the hub driver. They will appear in due course.
Alan Stern
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
