Equivalent transformation:
	call of pi_do_claimed() in run_fsm() expanded; recursive
	call of run_fsm() replaced with loop.

Equivalent transformation:
	pi_unclaim() moved from phase functions to run_fsm().

Equivalent transformation:
	* calls of pi_disconnect() in phase functions expanded.
	* since both do_pd_read_start() and do_pd_write_start()
	  are always called claimed (as the matter of fact,
	  all phase functions except do_pd_io() are), we can
	  replace call of pi_connect() with direct method call -
	  pi_claim() is no-op there.

Equivalent transformation:
	* new type - enum action.  phase functions return it now.
	  run_fsm() looks at the result of phase() and does the
	  corresponding action - next_request(), pi_do_claimed()
	  or ps_set_intr().

Equivalent transformation:
	* ps_continuation renamed - it's "phase" now.
	* run_fsm() does what ps_tq_int() used to do
	* calls of pi_do_claimed(..., func) replaced with
	  phase = func; pi_do_claimed(..., run_fsm);
	Now all phase functions are called from run_fsm() and we
	can start pulling the common code into it.

Equivalent transformation:
	do_pd_reqeust1() calls lambda-expanded, function killed.
	resulting call of ps_set_intr() in next_request() has been
	moved past dropping pd_lock.

Equivalent transformation:
	instead of having do_pd_request1() calling do_pd_io() via
	pi_do_claimed() and do_pd_io() calling do_pd_io_start() via
	ps_set_intr(), do it the other way round.

Equivalent transformation:
	Handling of bogus requests taken into do_pd_io_start() - we
	call next_request(0) as we would for any other error.  We
	need to call pi_unclaim() before that - on other paths that
	would happen as a side effect of pi_disconnect().
	Setting variables moved into do_pd_io_start() - they don't
	need pd_lock being held, so we can shift that outside.

Equivalent transformation:
	* new function: do_pd_io_start().  Checks pd_cmd and calls
	  do_pd_read_start() or do_pd_write_start().
	* do_pd_read() and do_pd_write() merged into do_pd_io(), which
	  schedules execution of do_pd_io_start().

pd_busy gone - we can check for pd_req != NULL instead.

Logics around "if not ready, schedule the same step again" taken
out of ps_tq_intr() and into the steps that need it.

Removed junk (see the analysis in the previous patch), removed
ps_set_intr() arguments.

Preparation to crapectomy - expanded include of pseudo.h, split
the part of pd_do_request() into pd_do_request1(), leaving in pd_do_request()
only the check that FSM is not running and call of pd_do_request1().
Added locking analysis - it's in drivers/block/paride/Transition-notes
and it will be used on the next step.  Even though the subsequent steps are
small, they rely on properties of the damn thing and proof of these properties
is hairy.  Amazing perversions people manage to produce when writing a trivial
FSM - the code is very convoluted for no good reason...

* new helper: pi_schedule_claimed() - pi_do_claimed() sans the call
  of continuation if we don't have to wait; returns 1 if we don't
  have to wait and 0 otherwise.

	daisy.c used to access the topology list with no locking whatsoever.
Protected by spinlock, fixed numerous bugs in handling of the single-linked
list (I'm not kidding - just take a look at the old parport_daisy_fini()
and weep; it is supposed to go through a simple list and remove all entries
that satisfy a condition).

	bw-qcam.c turned into proper parport driver.  Instead of (racy)
scanning the list of ports we use ->attach() and ->detach().

	parport_gsc.c turned into proper parisc driver; instead of scanning
the list of ports upon rmmod in search of ones that had been created by us,
we do cleanup where it belongs - in parisc driver ->remove().

	A bunch of parport_enumerate() users was duplicating parport_find_...()
without proper locking.  Replaced with use of appropriate helpers, races closed.

	We use a new mutex to protect all additions/removals of drivers and
ports.  That cures a lot of insanity:
	* driver removals can't hit us in the middle of attach_driver_chain().
Old code simply dies on that.
	* port removals can't hit us in the middle of driver registration.
Again, old code dies on that.
	* driver ->detach() is allowed to block now.
	* we are guaranteed that by the time when parport_unregister_driver()
returns, all ->detach() calls are finished.  Old code did _not_ guarantee
that (read: was inherently racy since rmmod of driver could race with port
removal and get driver->detach(port) called after the module was gone).
	* we are guaranteed that driver->attach(port) won't be called
more than once.  With the old code that was a matter of luck.
	* removed piles and piles of braindead code.

	Current tree has all allocated ports on portlist.  However,
most of the portlist users assume that we only have announced ports
there and break badly if they happen to see the list after port driver
has created a port (parport_register_port()) but before it finishes
the setup (and calls parport_announce_port()).
	The only place that wants to see all allocated port is
parport_register_port() itself and only to pick the first unused
port number.
	We add a new list (all_ports) and put ports there when allocated;
that list is kept ordered by port->number.  Ports are placed on portlist
only by parport_announce_port().
	Gobs of shite in parport_register_port() removed, pile of races
closed...

into home.osdl.org:/home/torvalds/v2.5/linux

into redhat.com:/spare/repo/libata-2.5

into redhat.com:/spare/repo/net-drivers-2.5

	Add PCI device support - there are atmel_pci.ko and atmel_cs.ko
	modules and a library module called atmel.ko

	Tweak the PCMCIA card -> firmware table for new cards.

	Fix workarounds for uniquely broken 3com cards, which were
	rendered unuable by the 0.9 changes.

	Bump version to 0.91

Hi Jes.  I really do have a Rev. 5 card, and it works much better treated
as a Tigon I than as a Tigon II.

I send these a while ago but didn't get a reply.
Can you apply these patches for the hamradio drivers?

details:

+ scc.c usage of rtnl lock and register_netdev.
+ scc.c properly initialize timers.
+ remove cli() call in 6pack.c
+ fix init of reused devices and remove cli()
+ scc.c statistics

Hi !

I didn't include that with the bitkeeper based PowerMac updates so Jeff
can assume his role as maintainer/filter of network driver updates :)

Here's my latest update to this pmac-only driver, please apply if you
are ok with it. The driver now uses the macio infrastructure for
detection & power management notifications, getting into sysfs at
the same time.

Ben.


# This is a BitKeeper generated diff -Nru style patch.
#
# ChangeSet
#   2004/01/23 09:38:35+11:00 benh@kernel.crashing.org
#   Update PowerMac "bmac" driver to the "macio" device so it fits in sysfs
#   and gets proper power management ordering
#
# drivers/net/bmac.c
#   2004/01/23 09:38:22+11:00 benh@kernel.crashing.org +185 -203
#   Update PowerMac "bmac" driver to the "macio" device so it fits in sysfs
#   and gets proper power management ordering
#

This is a multi-part message in MIME format.

into redhat.com:/spare/repo/net-drivers-2.5

into redhat.com:/spare/repo/net-drivers-2.5

Now that Al Viro fixed cramfs, it works beautifully as an initrd
filesystem.

So finally plumb it in.

into kernel.bkbits.net:/home/davem/sparc-2.6

o Cleanup reduced and faster code

o gcc-3.4 warning fix.

o Deleted obsolute header include

into home.osdl.org:/home/torvalds/v2.5/linux

into home.osdl.org:/home/torvalds/v2.5/linux

I've been chasing a weird SELinux bug which shows up mostly when doing
installs of a dev-* rpm (ie. creating and overwriting lots of block
device inodes), but which I've also seen when doing mkinitrd.

It turned out not to be an SELinux problem at all, but a core VFS
S_ISBLK bug.  It seems that SELinux simply widens the race window.

The code at fault is fs/fs-writeback.c:__mark_inode_dirty():

		/*
		 * Only add valid (hashed) inodes to the superblock's
		 * dirty list.  Add blockdev inodes as well.
		 */
		if (!S_ISBLK(inode->i_mode)) {
			if (hlist_unhashed(&inode->i_hash))
				goto out;
			if (inode->i_state & (I_FREEING|I_CLEAR))
				goto out;
		}

The "I_FREEING|I_CLEAR" condition was added after the ISBLK/unhashed
tests were already in the source, but I can't see any reason why we'd
want the I_FREEING test not to apply to block devices.  And indeed, this
results in all sorts of inode list corruptions.  Simply moving the
I_FREEING|I_CLEAR test out of the protection of the S_ISBLK() condition
fixes things entirely.

The existing 2.6 kernel will reliably fail on me in about 2 seconds once
"rpm -Uvh --force dev*.rpm" starts its actual installation of the new
inodes.  With the patch below I can't reproduce it at all.