- device_sem is added to make the global list walks easier (for power mgmt and shutdown), so
  they don't have to take and drop device_lock a lot. 

- Change list modifications to take device_sem instead of device_lock.

- Ditto for device refcount modifications.

- Kill get_device_locked(), as all the users are now gone. 

- make sure we check what we get back from get_bus()
- do down_read() instead of down_write(), so multiple people can do iterations
  concurrently.
- Note that devices and drivers are removed from these while doing a down_write(), 
  so all removals will be stalled until the iterators are done.
 

Drivers must belong to bus, and each bus has an rwsem. Instead of mucking with
the device_lock spinlock, and dropping it on each iteration of the loop, we take 
the bus's lock (read, so multiple drivers can access their list at the same time)
around the entire walk of the list.

->shutdown() is added to be explicitly called during reboot transitions by 
device_shutdown(). Drivers implementing this method should use it solely to quiesce
the device. Drivers do not have to free memory allocated by the driver, as it is 
implied that the system is rebooting; and, they must not fuss with the reference
counts. Devices are guaranteed not to disappear during the call, though their 
validity may be checked using device_present().

This is similar to struct bus_type's rwsem, though classes aren't doing 
anything nearly as fancy with it. We just make sure to take it when ever
we add or remove any devices or drivers. 

when removing system and platform devices.

- move list walking and matching to bus.c (since it's a function of the bus driver)
- do specialized walks of the bus's lists when binding; no more callbacks being passed
  to bus_for_each_*.
- take rwsem when adding and removing both devices and drivers. lists of each are now
  fully protected by that rwsem. It's also taken before we walk each list.
- move calls of device_{de,at}tach() to bus_{add,remove}_device() and calls of 
  driver_{de,at}tach() to bus_{add,remove}_driver().

There is a chance devclass_add_device() could get passed a device with no driver, so
we need to check that.

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-core

Synchronize all walks of the device and driver lists of a bus with an rwsem wrapped
around the entire iterator, instead of using device_lock and dropping it after we 
grabbed each node. 

Note this also prevents deadlock when walking the list of drivers and calling 
get_driver(), since get_driver() tries to take device_lock while we already have it
held.

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into tp1.ruhr-uni-bochum.de:/home/kai/src/kernel/v2.5/linux-2.5.make

into tp1.ruhr-uni-bochum.de:/home/kai/src/kernel/v2.5/linux-2.5.isdn

plus other small cleanups, in particular getting rid of
isdn_ppp_{receive,send}_ccp() and calling into the CCP code directly.

No (well, hardly any ;) code changes, only moving all /dev/ipppX related
code next to each other and some indenting changes.

o Now that all the infrastructure is in place, struct ipppd can
  easily be allocated when /dev/ipppX is opened and freed
  in the destructor.
o Separate the mix of state and flags in struct ipppd::state.
  We only have three states, open, assigned and connected, and
  additionally two flags to jump out of poll(), returning POLLHUP /
  POLLIN.

Instead of just recording the slot number and then retrieving
the ipppd via that, we can now just keep a pointer and get a
reference which makes sure that it does not go away until
we drop the reference.

We had introduced ipppd_get/put() earlier, so now add a real reference
count.

Basically, all what was missing was a common way to xmit frames
given the isdn_net_dev / isdn_net_local. We have this now by means
of function pointers (still room for improvement there, though), so
now let's do things from isdn_ppp_ccp.c.

Let isdn_ppp_ccp.c take care of keeping state / flags by itself.

Use a shared function for generating the PPP header, use the
{put,get}_u{8,16,32} helpers for endian-safe access.

I cannot see now reason why ISDN PPP devices should have a fake
::header routine which reserves a couple of bytes just to reclaim them
later during hard_start_xmit(), and since it works just as well without
it...

Fix missed STAT_BSENT acks and don't use skb_clone() for the skb which is
sent back, the network layer gets confused by that.

Create the new files isdn_ppp_ccp.[hc] which deal with PPP CCP
(compression control protocol) related stuff. Move most of the CCP
related stuff from isdn_ppp.c into isdn_ppp_ccp.c, putting CCP
state into a new "struct ippp_ccp".

A lot of the state in struct ipppd actually belongs to
isdn_net_dev or isdn_net_local, making it more easily accessible at a
most places, and also removing the ambiguity whether it's link-
or bundle-specific.

and move the declaration from include/linux/isdn_ppp.h
into drivers/isdn/i4l/isdn_ppp.c

Make sync PPP compile with the latest changes to the ISDN net device
layer - PPP negotiations work again after this patch.

Definitions which is only needed in one source file should go there and not
in a common header.

Last step in separating generic / interface-type specific code,
the former is now all in isdn_net_lib.c, the specific code in
isdn_net.c / isdn_ppp.c / isdn_concap.c

Instead of using locking per-channel, just lock the entire ISDN net work
interface as an entity, getting rid of weird locking.
  
Basically, ::hard_start_xmit() is already serialized by the network core,
so we could just rely on that. However, we want to send keep alive
frames, PPP messages from ipppd and such things directly without going
through the network stack, so we additionally lock
isdn_net_local->xmit_lock in hard_start_xmit(), which is taken in
the other paths dealing with transmitting frames as well.

into tp1.ruhr-uni-bochum.de:/home/kai/src/kernel/v2.5/linux-2.5.make

device classes join the club of devices, buses, and drivers. They get a 
->present flag, which is set on registration and cleared on unregistration.
They also get get_devclass() and put_devclass(), which, you guessed it, 
bump the reference count. get_...() of course checks the present flag and
returns NULL if clear. 

In the spirit of devices and buses, change driver refcounting model to 
match the way that devices and buses are done.

struct device_driver gets a ->present field, which is set on registration 
and cleared in driver_unregister(). get_device() checks the state of this
flag and returns NULL if cleared. 

Note that the horribly wrong remove_driver() is deprecated and simply BUG()s
when called. Please convert callers to use driver_unregister(). Updates to 
callers will be coming soon.

Note also that this still doesn't fix the race in which a driver module can
be removed while the refcount on a driver > 1. Near future work should help
to remedy it, but no solutions are guaranteed..

Based on recent changes to device refcounting/unregistration.

struct bus_type gets a ->present field, which is set when the bus is 
registered. bus_unregister() is added, which only clears ->present and 
calls put_bus(). get_bus() checks the ->present and returns NULL if clear.

This allows all current references to buses to keep valid structures, while
preventing any new references to the bus structure.

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-core

There are races WRT walking the lists of the buses and classes and the 
device being unregistered while it's the current node we're accessing. If 
the node was deleted, ->next would point to itself, and we'd sit in an
inifinite loop (until that memory was freed from underneath us and we Oops'd).

This changes device_unregister() to only mark the device as not present. put_device()
deletes the device from all the lists it belongs to, while holding device_lock,
then drops it to call device_del().

This should guarantee that everything in the device is valid until the last reference
goes away, regardless of whether it's still physically present or not. 

If you compile the kernel with -ffunction-sections, each function gets
put in a section .text.function_name.  This collides with our current use
of .text.init.  So here's a patch which converts x86 to use .init.text
instead.

I've tested it on x86 and it still frees 120k of ram, so it seems to work.
Other architectures will need to change their vmlinux.lds appropriately,
and may need other changes (arm, m68k seem to use .text.init verbatim).

into home.transmeta.com:/home/torvalds/v2.5/linux