Cleanup the PPC64 PACA structure.  It was previously a big mess of
unecessary fields, overengineered cache layout and uninformative comments. 
This is essentially a rewrite of include/asm-pp64/paca.h with associated
changes elsewhere.  The patch:

- Removes unused PACA fields

- Removes uneeded #includes

- Uses gcc attributes instead of explicit padding to get the desired
  cacheline layout, also rethinks the layout and comments accordingly.

- Better comments where asm or firmware dependencies apply non-obvious
  layout constraints.

- Splits up the pointless STAB structure, letting us move its elements
  independently.

- Uses offsetof instead of hardcoded offset in spinlocks.

- Eradicates xStudlyCaps identifiers

- Replaces PACA guard page with an explicitly defined emergency stack
  (removing more than NR_CPUS pages from the initialized data segment).

Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Somebody back in the mists of time decided that call_prom and rtas_call
should return longs even though both of those bits of firmware run in
32-bit mode and produce a 32-bit result.  To make life more interesting,
the 32-bit result gets zero-extended to 64 bits, which makes checking
for a -1 return value more complicated than it should be.

This patch changes call_prom and rtas_call to return an int, and makes
the corresponding changes to use ints for the variables used to hold
those return values. 

While I was doing this I finally got annoyed enough with the strings of
prom_print() and prom_print_hex() calls that we do to write a simple
prom_printf.  I deliberately didn't use snprintf because the execution
environment is weird at this point - we aren't running at the address we
are linked at just yet - and I didn't want to inflict that on any code
outside this file.  I also did a prom_debug() macro, which eliminated a
few ifdefs. 

There are also a bunch of other minor cleanups.  This patch makes very
few algorithmic changes but does get rid of a lot of casts.  :)

I have been running with this patch for a couple of weeks, and Anton has
tested it too.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

From: Anton Blanchard <anton@samba.org>

Even though we have a variable containing the number of entries in the SLB,
we hardwire the value at boot.  We also dont use the variable in the head.S
fastpath handler.

This patch removes it.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This fixes a couple of bugs in arch/ppc64/kernel/prom.c.  We were
missing a couple of places where we needed to use RELOC().  I added the
RELOC in one case, and in the other, moved the variable that we were
accessing onto the stack (and reduced its size).  (We use the variable
to get a property value, but we aren't interested in the value, just in
whether the property exists or not.  Since we pass the size of the
variable to the OF getprop call, it won't overflow.) The effect of
missing the RELOCs would be that random memory locations get used on IBM
pSeries systems (possibly causing random boot failures). 

The other thing that this does is add a linux,phandle property to each
node, containing the phandle for the node, which is the token that OF
uses to identify the node.  Some nodes reference other nodes by means
of their phandle.  Without the linux,phandle property, userspace code
looking at the OF device-tree image in /proc/device-tree has no way of
knowing which other node is being referenced.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Paul Mackerras <paulus@samba.org>

Early in the boot process on pSeries machines, we look in the Open Firmware
device tree for information about the interrupt assignments, and assign
virtual IRQ numbers for each physical IRQ.  There is currently a couple of
bugs in this code which result in us assigning virtual IRQs for nonexistent
physical IRQs.  This causes problems when we call the firmware to enable or
disable those nonexistent physical IRQs.  Some versions at least of the
firmware will hit an assertion failure and crash the machine when this
happens.

This patch fixes the bugs and ensures that we don't try and use nonexistent
physical IRQ numbers.  One bug was that we were mapping ISA interrupts,
which is unnecessary since virtual IRQ numbers 0 - 15 are reserved for
them.  The other was that when we had a PCI interrupt (which is always in
the range 1 to 4, corresponding to INTA to INTD) which didn't have a
mapping in the PCI host bridge above it, we were just using the original
number (usually 1) rather than ignoring it.

From: Rusty Russell <rusty@rustcorp.com.au>

Doesn't compile with CONFIG_BLK_DEV_INITRD=n.  Be more careful with the
conditionals.

Spotted by Dave Boutcher...

From: Rusty Russell <rusty@rustcorp.com.au>

This patch moves the initrd when "make_room" runs out of space, rather than
stepping over it.  The problem with stepping over it is that it gets copied
with the kernel when this happens, which wastes space.

Also ensures that the initrd isn't where the kernel wants to be moved to: if
it is, it gets moved out the way.

From: Rusty Russell <rusty@rustcorp.com.au>

Clean up initrd handling.

1) Expose initrd_start and initrd_end to prom.c (replacing its local
   initrd_start and initrd_len).

2) Don't hand mem (aka klimit) through functions which don't need it.

3) Add more debugging under DEBUG_PROM in case we broke anything.

From: Rusty Russell <rusty@rustcorp.com.au>

1) Define PROM_ERROR: checks for call_prom() < 0 are bogus (returns ulong)
2) All OF access functions should be marked __init.
3) prom_strtoul isn't used at all.

From: Rusty Russell <rusty@rustcorp.com.au>

Ben Herrenschmidt made this actually boot.

From: Rusty Russell <rusty@rustcorp.com.au>

copy_device_tree calls inspect_node() which recurses to copy the Open Firmware
device tree.  It just copied into memory above the kernel for about 8MB, which
can overwrite the initrd, and/or get truncated on big machines.

The real solution is to allocate memory properly for all the prom_init memory
users.  However, that's quite a change.  This patch introduces a "make_room"
macro which inspect_node uses to get memory.  This does the same as before,
except skips over the initrd if neccessary.

From: Rusty Russell <rusty@rustcorp.com.au>

finish_device_tree simply allocates nodes by incrementing klimit (ie.  using
memory on top of the kernel).  Change it to figure out how much memory it
needs, then use lmb_alloc to allocate that, then fill it in.

This gets rid of the only manipulation of klimit after prom_init.

From: Rusty Russell <rusty@rustcorp.com.au>

Rather than pre-declaring static functions, order them so it's not neccessary.
This makes the code easier to navigate and patch.

From: Rusty Russell <rusty@rustcorp.com.au>

Rather than pre-declaring static functions, order them so it's not neccessary.
This makes the code easier to navigate and patch.

From: Rusty Russell <rusty@rustcorp.com.au>

Rather than pre-declaring static functions, order them so it's not neccessary.
This makes the code easier to navigate and patch.

From: Rusty Russell <rusty@rustcorp.com.au>

Rather than pre-declaring static functions, order them so it's not neccessary.
This makes the code easier to navigate and patch.

From: Rusty Russell <rusty@rustcorp.com.au>

Move all the functions that run normally (ie.  with the kernel in the correct
address) to the bottom of prom.c, and keep all the functions which use RELOC()
to the top, above "prom_init()" which calls them.

Also, make prom_initialize_dart_table and prom_initialize_tce_table static.

This fix comes from ppc32. Always initialise dn->type and dn->name so
that code doesnt have to check for NULL everywhere. There is at least
one bug report where we oopsed because of this.

From: Anton Blanchard <anton@samba.org>

Add smt_snooze_delay cpu sysfs attribute

From: Paul Mackerras <paulus@samba.org>

Create and use irq_offset_up/down, get_irq_desc, for_each_irq

From: Anton Blanchard <anton@samba.org>

Disable SMT snooze by default

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

A recent patch that cleaned up some absolute/virt translation macros forgot
one occurence, thus breaking g5 build with iommu support.

From: Anton Blanchard <anton@samba.org>
      Rusty Russell <rusty@rustcorp.com.au>

The iSeries has an arch-specific mapping from physical <-> absolute
addresses.  Fortunately this is only used in a few places.  However, the
following arch-specific macros/functions are provided in addition to the
standard macros:

	__a2p()
	__a2v()
	__p2a()
	__p2v()
	__v2a()
	__v2p()
	absolute_to_phys()
	phys_to_absolute()
	virt_to_absolute()
	absolute_to_virt()

Reduce them to these, with slightly shorter names, and taking either pointers
or unsigned long (as per __va and __pa) rather than making the caller cast:

	abs_to_phys()
	phys_to_abs()

And helper macros:

	virt_to_abs()
	abs_to_virt()

As is standard, virtual addresses are returned as void *, physical and
absolute as unsigned long.

Note that the change the iSeries_setup is a little subtle: ea is set to
__va(pa) above, so "phys_to_abs(pa)" is the same as "virt_to_abs(ea)".

Also, REALADDR is renamed to ISERIES_HV_ADDR and used in a couple of places
where appropriate.

This increases the DART table to use the full size.  We allocate a full
16MB page anyway, so there's no difference in memory consumption. 

Thanks to Ben for spotting this, it was left over from debugging...

From: Anton Blanchard <anton@samba.org>

Clean up early boot messages.

From: Anton Blanchard <anton@samba.org>

Implement iommu=off for pseries.  We limit memory to 2GB as pmac does,
however the default for < 2GB remains iommu=on unlike pmac.

From: Anton Blanchard <anton@samba.org>

- Fix for machines with 3GB IO holes (eg nighthawk).
- Increase the maximum number of PHBs and warn if we exceed this (we used
  to walk off the end of the array)
- Only allocate an 8MB TCE table on POWER4

From: Anton Blanchard <anton@samba.org>

We have some versions of firmware out there that have huge OF properties.
So huge that we end up overwriting our initrd.

Place a 1MB limit and warn bitterly if its over this.  Also fix a use of
package-to-path where the variable was 64bytes but we would pass in a
length of 255.

From: Linda Xie <lxiep@us.ibm.com>

The changes in this patch are for multifunction cards insertions/removals
and  bug fixes:

1. fix up new nodes' linux_phandle field.
2. new nodes' phb, devfn(and so on) need to be fixed even the nodes don't
	have "interrupts".
3. change of_remove_node to non-recurisve func. The recursions will be done
	by the caller.
4. add a new function -- of_finish_dynamic_node_interrupts()

This patch reworks the early boot calls to OF to initialize displays.

All present displays are now initialized in reverse order so the
OF console stays on the first one. Initializing them all is necessary
for dual head configurations as we need OF driver to properly setup
the secondary TMDS of the video card, XFree isn't able to do that
currently

This makes it possible for people like me with a small-mem G5 to enable
the DART.  I see two reasons for wanting to do so:

1. To debug/test DART/iommu code itself (small audience, including
   myself).
2. To debug drivers on small-mem machines, since bad pci_map*() usage will
   be punished (possibly larger audience).

This allows us to use more than 2Gb of RAM on a PowerMac G5.  You can
also boot with iommu=off to disable it, but then you get back the 2Gb
RAM limitation.

Lots of things renamed, sillicaps killed, stuffs moved around and common
code properly extracted from implementation specific code, new
allocator, etc...  The code is overall a lot simpler, faster, less prone
to fail, and a lot more manageable.  I didn't use "bk mv", there is no
need to keep the old history attached to the new file.

On the larger ppc64 machines we remap the interrupt numbers used by
the hardware/firmware to virtual IRQ numbers < NR_IRQS.  Up until now
we have used an array for the "real" (hardware) -> virtual IRQ number
mapping, but with new machines coming out that will have 24-bit
hardware IRQ numbers, this will break.  However, in fact it is only
the XICS interrupt controller which cares about this mapping.  This
patch moves that side of the mapping (real -> virtual) into the XICS
code and makes it use a radix tree.

On iSeries we have a similar issue, where the "real" IRQ numbers that
we need are in fact an encoding of the bus/device/function address of
the device.  This patch fixes iSeries to use the virt->real IRQ
mapping, allowing us to support larger iSeries machines.  This patch
also gets rid of the temporary hack that Stephen Rothwell submitted.

On machines with OpenPIC (including the G5) the mapping is explicitly
1-1, and that hasn't changed.  For other machines this patch cleans up
and simplifies the code that sets up the virtual->real mapping.

I have tested this code and verified that G5, pSeries and iSeries boot
and run correctly with this patch.

From: Anton Blanchard <anton@samba.org>

There were a number of bugs in our cmd_line handling:

- We were looking at cmd_line after it had been tokenised. Use saved_command
  line instead
- Use strlcpy instead of memcpy/explicit NULL set.
- Remove magic numbers in iseries cmd_line parsing

From: Anton Blanchard <anton@samba.org>

UP compile fixes.

From: Linda Xie <lxiep@us.ibm.com>

set err to -ENODEV when a new node doesn't have "interrupt" property.

From: Rusty Russell <rusty@rustcorp.com.au>

1) Check for cpuids >= NR_CPUS in prom_hold_cpus: we previously
   overflowed arrays when this happened, and checking it here ensures
   that it doesn't happen elsewhere.  Still move processors to
   secondary_hold (out of OF), but we won't release them.

2) smp_iSeries_message_pass and smp_xics_message_pass don't need to
   iterate if given a specific CPU: cleanup and fix.

3) kick_cpu variants don't need to check cpu range (caller is screwed
   anyway if this happened): replaced with BUG_ON for now.

4) smp_openpic_probe and smp_xics_probe can use cpus_weight(): it's
   faster and simpler.

5) User for_each_cpu/for_each_online_cpu as appropriate.