From Benjamin Herrenschmidt <benh@kernel.crashing.org>.

- avoid device list corruption and scheduling at interrupt time
  (more work needed on proper locking)
- make sure "hold" flag and ide_dma_queued_* ops are properly
  transferred from old to new interface

From Andrey Borzenkov <arvidjaar@mail.ru>:

Apparently drive->driver is never NULL now but defaults to the default driver.

fixes kobject_register() oops

Now that we properly track queue references, that showed a bug in
scsi_ioctl.c, where it was dropping queues that it had never properly
gotten in the first place.

To be able to properly be able to keep references to block queues,
we make blk_init_queue() return the queue that it initialized, and
let it be independently allocated and then cleaned up on the last
reference.

I have grepped high and low, and there really shouldn't be any broken
uses of blk_init_queue() in the kernel drivers left.  The added bonus
being blk_init_queue() error checking is explicit now, most of the
drivers were broken in this regard (even IDE/SCSI).

No drivers have embedded request queue structures. Drivers that don't
use blk_init_queue() but blk_queue_make_request(), should allocate the
queue with blk_alloc_queue(gfp_mask). I've converted all of them to do
that, too. They can call blk_cleanup_queue() now too, using the define
blk_put_queue() is probably cleaner though.

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

This patch fixes the BIO_RW_AHEAD being tested at the wrong end
in the bio layer.

The problem with PF_READAHEAD is that if someone does a non-GFP_ATOMIC memory
allocation we can enter page reclaim and then call writepage, while
PF_READAHEAD is set.  The block layer then drops writes or the wrong reads on
the floor.  It can cause data loss.

A fix is complex (well, intrusive).  Given that the readahead code is now
skipping the entire readahead attempt if the queue is congested, the setting
of PF_READAHEAD probably is not doing anything useful anyway, so simply
remove it.

There seemed to be increase in Athlon users reporting MCEs in 2.6 that
they never saw in 2.4 and I didn't buy the "2.6 pushes hardware harder"
for a second given folks are running 2.4+preempt+O(1) etc patchkits
without problems.

So I did a mini-audit of the mcheck code.

How embarressing. Another fix that went into 2.4 that never made
it forward. Once I'm done with various Red Hat administrivia
I'll see if I can devote some time to going through the changesets
mailing list. I wonder just how many other really stupid bugs
like this got fixed and not brought forward.

into kernel.bkbits.net:/home/davem/net-2.5

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

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

In order to properly track down who needs to program the IRQ MUX
register, add the subsystem vendor and device IDs to the kernel
message indicating discovery of the cardbus bridge.

The IRQ steering code operates too early at present, and actually
prevents us detecting ISA interrupts.

We should not touch the IRQ MUX register on TI bridges - only the
machine itself knows the right value for this.  The kernel doesn't
have the knowledge to know what function the cardbus controllers
multi-function pins have been assigned by the hardware manufacturer.

Currently, yenta does not try to clean up after an error occurs while
initialising a cardbus socket.  This cset ensures that we release
resources.  We also claim the cardbus MMIO memory resource.

pnp_x_valid returns 1 if valid.  Therefore we should be using
!pnp_port_valid.  Also cleans up some formatting issues.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

Convert X.25 async driver to have dynamic net_device's.
This driver is a lot like SLIP so the code changes were similar.  
        - Added similar locking to SLIP
        - replaced code that snooped for MTU changes with the
          net_device change mtu callback.
        - cleaned up the statistics by using the net_device_stats structure.

Patch is against 2.6.0-test2.

Not sure if anyone ever uses this.  I tested by bringing up an x.25 async
line using a modified version of slattach.

The conversion from cli/sti to locking in X.25 must not have been tested
on a real SMP with memory debugging enabled.  It OOPS right away if
I do:
        modprobe x25; ifconfig -a

The problem is that it dereferences the socket after it has already been
freed.  The fix for this is to make the call to sock_put, later in
x25_destroy_socket do the free.  Also, need a go to avoid references
in x25_release.

This patch is against 2.6.0-test2.  The problem is that the ax25_destroy_socket
function frees the socket buffer, but then ax25_release dereferences this causing
an OOPS.  To reproduce:
        modprobe ax25; ifconfig -a

Replaced sk_free with sock_put which will free if this is the last reference.

The existing ROSE /proc interface has no module owner, and doesn't check for
bounds overflow.  Easier to just convert it to the seq_file wrapper functions.

This patch is against 2.6.0-test2 (offsets assume earlier patch).

This patch changes the ROSE protocol to allocate an array of pointers and each network device
separately.  This sets up later change where network_device object's are released on last use
which may be after the module is unloaded.

The patch is against 2.6.0-test2 (though this code hasn't changed in a long time).


Allocation is done via alloc_netdev so the dev->priv area is already reserved and
doesn't need to be allocated separately. 

When a dead detached thread has been temporarily zombified because it's
ptraced and its tracer tries to reap it, it deadlocks on SMP.

Here's the fix.

A recent patch changed the rpc/*/channel files to behave
differently depending on whether they were open for read or not.

This hadn't been tested thoroughly.  The cache_reader structure
that was now only allocated when opening for read, had a field that
was iused when writing.

This patch removes that field and takes a different approach to solving
the issue it addressed.

And it has been tested a bit better.

A dying initial thread (thread group leader) sends SIGCHLD when it exits,
but it ought to wait until all other threads exit as well.  The cases of
secondary threads exitting first were handled properly, but not this one.

This exit.c patch fixes that test case, and I think catches the other
potential bugs of this kind as well.  The signal.c change adds some bug
catchers, the second of which will trip on the test case in the absence
of the exit.c fix.

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

This patch removes the pnp name database code.  Most buses, including
pnp, will be using userspace to name devices in the near future.  Also
dev->name will be removed from the driver model soon.

This patch moves the low level bios calls to a separate file, 
"bioscalls.c".  It is a cleanup that will improve organization of the
pnpbios driver code.

This patch is needed in order to avoid a potential oops.  It is
similiar to the changes made to pci.

This patch moves the resource parsing functions from support.c to the
pnpbios driver.  Originally these functions were intended for other
pnp protocols but in reality they are only used by the PnPBIOS driver.
This patch greatly cleans up the code in both the parsing functions
and their connection with the pnpbios driver.  Also note that
pnpbios.h has been added for local pnpbios functions.

This is not needed.

  This patch is similar to the disabled resource patch in that it
  avoids direct numeric comparisons with data in unset resource
  structures.

the entry that we got through the lookup_mnt().

Noted by Dick Streefland.

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