sys_quotacl tries to do a get_super on a struct block_device * to which
it doesn't hold a reference (nor does it actually have to be non-NULL).

As lookup bdev by name is a rather common operation I splitted out a new
helper, lookup_bdev() that does this out of open_bdev_excl and switched
quota.c to use it.  lookup_bdev() holds a proper reference that needs
to be dropped by bdput(), and it's well documented.

The use of umode_t instead of devfs-specific char vs block #defines
in Adam's mini-devfs patch makes sense independant of whether his patch
should get merged.  While reviewing his changes I also notices that
most of the number allocation functionality in devfs has no business
beeing exported.  In addition I cleaned up devfs_alloc_devnum/
devfs_dealloc_devnum a bit.

It's simply not needed anymore in 2.5

Have I already mentioned that the intermezzo code isn't exactly nicely
readable? ..

- fix cpufreq drivers compilation on not-bleeding-edge-gcc's (Adrian Bunk)
- gx-suspmod.c hasn't had a name yet

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

It's annoying having to touch 20+ arch vmlinux.lds.S file for every
new section introduced, just because they all duplicate the same
statements. Since we preprocess vmlinux.lds.S anyway, let's
#include <asm-generic/vmlinux.lds.h> and share the common statements.

This is a first step in consolidating most of the read-only sections.

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

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

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

Adding a check for allocation failure was a good idea, it just needs
checking the right variable...

Now it actually even compiles.

From Jeff.Wiedemeier@hp.com:

Bump non-legacy start addr to 16mb to accomodate new larger
SRM console footprint.

Adds the summit subarch hook to the config file, and updates various things
all inside the summit subarch directories (ie this can't possibly break
anyone else ;-)). The Summit's subarch had got out of sync in a few places.

Patch from James Cleverdon & John Stultz

Changes IO_APIC_MAX_ID to depend on the APIC type we're using.
The Summit machines have to use a larger set of bits in the apic registers,
we enlarge under ifdef for Summit only. We enlarge MAX_APICS for summit
as well as NUMA-Q (it would be nice to move this to subarch, but it creates
circular dependency problems ... I'll fix this up later).

Adds a check for the newer Summit boxes with a different name.

Patch from James Cleverdon & John Stultz

The IRQ balancing code currently assumes that the logical apicid is
always '1 << cpu', which is not true for the larger platforms.
We express this as an abstracted macro instead, and move the
cpu_to_logical_apicid definition to subarch, so we can make it exactly
"1 << cpu" for normal machines - maximum speed, minimum change risk.

A couple of things are abstracted from the smp_boot_cpus loop in order
to enable us to use the bios_cpu_apicid array to boot cpus from without
disturbing the code path of current machines.

Patch from James Cleverdon & John Stultz

This adds machine a type detection hook to the acpi code, and renames
raw_phys_apicid to bios_cpu_apicid (it's an array of apicid's to boot,
indexed by the bios' cpu numbering), and I other large machines will
need to use it later ... not necessarily using physical interrupts.

From Jeff.Wiedemeier@hp.com:

Misc alpha smp updates for 2.5 tree.

Patch from James Cleverdon & John Stultz

Currently the naming for the IO-APIC fields is very confused, we assign
dest_LowestPrio to delivery_mode, and INT_DELIVERY_MODE to dest_mode.
The values are correct, but the naming is wrong - this patch corrects
that confusion. It also moves the definitions of those settings into
subarch, where they belong (we have to use fixed delivery mode for Summit
due to what seems to be an Intel IO-APIC bug with P4 clustered mode).

From Jeff.Wiedemeier@hp.com:

numa mm update including moving alpha numa support into 
machine vector so a generic numa kernel can be used.

From Jeff.Wiedemeier@hp.com:

On NUMA alpha systems, attempt to allocate scatter-gather
tables local to IO processor.  If that doesn't work, then
allocate anywhere in the system.

From Jeff.Wiedemeier@hp.com:

This adds the 32GB limit to setup.c.  (It actually hits the first
2 nodes on Marvel, but that's ok, where we really run into a big
problem is if we go past 4, then we hit a much larger hole.)

From Jeff.Wiedemeier@hp.com:

Make the max IDE HWIFS configurable on alpha (default to
previous hardwired value of 4).

From Jeff.Wiedemeier@hp.com:

Add open_console / close_console callback definitions.

From Jeff.Wiedemeier@hp.com:

Adjust virt_to_phys / phys_to_virt functions to follow
EV6/7 PA sign extension to properly convert between 43-bit
superpage I/O addresses and physical addresses.
This change is backwards compatible with all previous Alphas
as they implemented fewer than 41 bits of physical address.

From Jeff.Wiedemeier@hp.com:

Remove redundant lock in osf_shmat (sys_shmat locks already);
redundant lock has been seen to cause livelock in some workloads.

From Jeff.Wiedemeier@hp.com:

Adjust kernel layout format to match other architectures and
prevent reording of the first entry in a section with the
section start label.

From Jeff.Wiedemeier@hp.com:

Adjust Alpha HARDIRQ_BITS check to make sure there is enough
room for each IPL, not each interrupt (Marvel can have
too many unique device interrupts for that, and it really
only needs to cover potential nesting of interrupts, which covering
the IPLs does)

From Jeff.Wiedemeier@hp.com:

Two stage timeout in alpha call_function_on_cpu. If the
primary timeout expires with no response, log a message and
start secondary timeout. If reponse is received log how
far into secondary timeout. If no response is received,
crash.

From Jeff.Wiedemeier@hp.com:

Fix alpha Makefiles for bootpfile target.

From Jeff.Wiedemeier@hp.com:

Export proper functions when debugging is enabled.

The warning

	/lockd/clntXxXxXxXx RPC: Couldn't create pipefs entry

is due to the lockd process starting RPC clients as an unprivileged
user, causing path_walk() to fail.  The following patch fixes it.

I have more carefully tested the proposed removal of the NT flag
clearing on lcall entry.

The question I wanted to answer is: is it necessary to clear NT in the
sysenter entry path as implemented for lcall7/lcall27 or is it possible
to remove the flag manipulation from do_lcall?

Doing it only for one and not the other looks wrong since several return
paths are shared, especially the ones which end up in iret, the only
instruction which is affected by the NT flag.

The conclusion is that 2.5 is NT safe (had to dig out an old P5-133 which
I could crash without fear of data loss, so I have only tested on 1
machine). The reason this cleanup works is that now (since Jan 5th) flags
are saved and restored in switch_to() to keep IOPL private to a process
even when using sysenter/sysexit.

The side effect of that patch is that NT becomes also process-private
instead of infecting all processes and triggering a killfest of all user
mode processes, including init (AFAICT kernel threads survived, but I
did not have any debug tools enabled in the kernel).

The only addition to the preceding version is that interrupts are
reenabled in the iret fixup path because it seems that do_exit() might
otherwise spend quite some time with interrupts disabled.