The scheduler is completed b0rked on x86_64, and I finally found out
why.  sched_clock() always returned 0, because rdtscll() always returned
0.  The 'a' in the macro doesn't agree with the 'a' in the function,
yippe :-)

This is a show stopper for x86_64.

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

The pid_alive() check within the loop is incorrect.  If we are within
the tasklist lock and the thread group leader is valid then the thread
chain will be fully intact.

Instead, the check should be _outside_ the loop, since if the group
leader no longer exists, the whole list is gone and we must not try
to access it.

Move the check around, and add comment.

Bug-hunting and fix by Srivatsa Vaddagiri

When doing sync wakeups we must not skip the notification of other cpus
if the task is not on this runqueue.

We need to initialize st->state in tcp_seq_start().  Otherwise
tcp_seq_stop() is run with previous st->state, and it calls the unneeded
unlock etc, causing a kernel crash.

I've got a lot more changes than what's included here.  I've put this
down to the bear minimum to get things working sanely.

Mainly, I just want to get all the people hit by this a chance to use
2.6.0 without having to get our tree. Changes itemized:

 - Fix deadlock possibility in csr.c:read_maps()
 - Fix kmalloc to use ATOMIC in highlevel.c.
 - s/in_interrupt/irqs_disabled/ in ieee1394_transactions.c to fix
   warnings when transactions occured.
 - Introduce a release callback for the host driver and use it correctly.
 - Reorganize the nodemgr probe so we do an initial scan to discover
   devices, check IRM/CycleMaster, then do a final full probe when things
   are kosher. Fixes a problem where device registration and hotplug
   would cause some serious problems when a bus reset was forced in the
   middle of the probe.

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

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

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

There are /proc handles there setup by proc_tty_register_driver, but there is
no module ownership association, so anything that reads after module unload
will blow.

The trivial fix is to propagate the owner of tty_driver to proc entry.

This fixes a small, but important bug in smsc-ircc2.c.

Without this the driver tries release the same region twice.

      Tracked down by Rohit Seth.

Do not special case ENOMEM as an error code, we should free up
the packet for all errors.

Most of the breakage came from cset 1.838.10.3

The br_netfilter_{init,fini}() calls need to be guarded by
BRIDGE_NETFILTER, from Michal Janusz Miroslaw.

When bringing an interface down do not forget to clear the
IF_RS_SENT and IF_RA_RCVD flags.

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

This fixes an oops that can occur if JFS is used as the root filesystem.

Writes to a device node may cause a ->write_inode to be called during a
read-only mount.  JFS needs to check for NULL log in jfs_flush_journal. 

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

Also, make sure NET_RX_DROP is returned if we did not accept the
packet.

In implementations that use no socket locking, such as RAW sockets,
once we queue the SKB to the socket another cpu can remove the SKB
from the socket queue and free up the SKB making the skb->len access
touch freed memory.

Based upon a report from Burton Windle, kernel bugzilla #937

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

Also add some ServerWorks-specific tweaks.

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

All the users of this function in the SCSI tree call it with the host
lock held.  With the new list traversal code, it was trying to take
the lock again to traverse the list.

Fix it to use the unlocked version of list traversal and modify the
header comments to make it clear that the lock is expected to be held
on calling it.

Noted by Harald Welte (laforge@netfilter.org)

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

I've been looking at enforcing lifetime phases for SCSI devices
(primarily to try to get the mid layer to offload as much of the device
creation and hotplug pieces as it can).

I've hijacked the sdev_state field of the struct scsi_device (formerly
this was a bitmap, now it becomes an enumerated state).

I've also begun adding references sdev_gendev into the code to pin the
scsi_device---initially in the queue function, but possibly this should
also be done in the scsi_command_get/put, the idea being to prevent
scsi_device freeing while there's still device activity.

The object phases I identified are:

1. SDEV_CREATED - we've just allocated the device.  It may respond to
internally generated commands, but not to user ones (the user should
actually have no way to access a device in this state, but just in
case).

2. SDEV_RUNNING - the device is fully operational

3. SDEV_CANCEL - The device is cleanly shutting down.  It may respond to
internally generated commands (for cancellation/recovery) only; all user
commands are errored unless they have already been queued (QUEUE_FULL
handling and the like).

4. SDEV_DEL - The device is gone. *all* commands are errored out.

Ordinarily, the device should move through all four phases from creation
to destruction, but moving SDEV_RUNNING->SDEV_DEL because of surprise
ejection should work.

It's starting to look like the online flag should be absorbed into this
(offlined devices move essentially to SDEV_CANCEL and could be
reactivated by moving to SDEV_RUNNING).

I haven't altered the similar bitmap model that scsi_host has, although
this too should probably move to an enumerated state model.

I've tested this by physically yanking a module out from underneath a
running filesystem with no ill effects (other than a slew of I/O
errors).

The obvious problem is that this kills possible user error handling, but
we don't do any of that yet.