Nothing is using this any more, so kill it off.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This moves the remaining common bits in to pci-lib. Thereby reducing
pci.c/pci-new.c to simple bus fixups and controller registration.

As more platforms are moved over, the old code will disappear completely
and the pci-new bits will be rolled in to pci-lib, eventually replacing
pci.c completely.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Now that the io and mem offsets are tracked accordingly, the pci-new
version of the bus<->resource mappers can be used generically. This
moves them in to pci-lib.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This implements a per-hose offset for I/O and mem resources.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This moves off of the board_pci_channels[] approach for bus registration
and over to a cleaner register_pci_controller(), all derived from the
MIPS code.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Not all PCI channels have non-translatable memory windows, this is a
special property of the on-chip PCIC with its 0xfd00... mapping, handle
this explicitly.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This consolidates the pci_iomap() definitions and reworks how the I/O
port base is handled. PCI channels can register their own I/O map base,
or if none is provided, the system-wide generic I/O base is used instead.

Functionally nothing changes, while this allows us to kill off lots of
I/O address special casing and lookups.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This is left over cruft that hasn't been used by anything in a long time,
kill off bits that weren't purged previously.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This introduces a saner pcibios_align_resource() that can be used
regardless of whether pci-auto or pci-new are being used, and
consolidates it in pci-lib.c.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Derived from the MIPS version, now uses pgprot_noncached().
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This splits off a 'pci-new.c' which is aimed at gradually replacing the
pci-auto backend and the arch/sh/drivers/pci/pci.c core respectively.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

The new PCI code wants its own bus<->resource mappings instead of the
generic equivalents, so drop the asm-generic include in preparation.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This starts moving out the common initialization bits from the various
fixup paths in to the shared init path.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Now that the platform code is a bit leaner, we can start consolidating
the various IRQ routing implementations. There are effectively only 2
variants, and the others can use those directly.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Commit 68b42d1b ("sh: sh7785lcr: Map
whole PCI address space.") changed around the semantics of how various
chip-selects are made accessible to PCI. Now that there is a single
large mapping covering from CS0-CS6, there is no longer any need to
do multi-window mapping. Subsequently, all of the differing
implementations can be consolidated in to pci-sh7780.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This consolidates all of the PCI I/O and memory window definitions across
the pci-sh7780 users in pci-sh7780 itself. No functional changes, in that
every platform had exactly the same implementation.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Presently the I/O port base isn't being set anywhere, which allows things
like generic_inl() to blow up. Fix this up to point at the PCI IO window.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

The SH7780 PCIC contains a read-only cache line size register that we can
derive pci_cache_line_size from. So, make sure that the software idea of
the cache line size actually matches the host controller's idea.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Don't use pci_write_reg() for these, as it defaults to 32-bit. Rather
than using the helper, use __raw_writeb() directly.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This reworks how the host controller is probed, and makes it a bit more
verbose in the event a new type of controller is detected. Additionally,
we also log the revision information.

This now uses the proper access sizes for the vendor/device registers,
rather than relying on a larger access that encapsulated both of them.
Not all devices support 32-bit read cycles for these registers.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

The SH7780 PCI host controller implements a configuration header that
requires a fair bit of hand-holding to initialize properly. By default
it appears as a pre-2.0 host controller given the zeroed out class code,
so fix this up properly.

Some boards that happened to be using the R7780RP version of the PCIC
fixups had set this correctly, but this belongs in the standard
initialization, and is by no means board specific.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

P1SEGADDR is obsolete and will be killed off completely in the future,
so transition off of it and reference P1SEG explicitly.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

These fixups belong in the board INTC setup code, not in the middle of
pci-sh7780.c.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Nothing ended up using this anymore, so just kill it off.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

SE7780 has the same PCIC fixup as SDK7780, and SH7785LCR the same
as R7780RP. Switch to using those, and drop the duplicate code.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

With board_pci_channels now being exported in a single place, update the
boards that duplicated the export.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Instead of sometimes exporting board_pci_channels[] in the board specific
code just export it in one place.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Adds a __get_pci_io_base() function which is used to match a port range
against struct pci_channel. This allows us to detect if a port range is
assigned to pci or happens to be legacy port io. While at it, remove unused
cpu-specific cruft.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This patch changes the code to use __is_pci_memory() instead of
is_pci_memaddr(). __is_pci_memory() loops through all the pci
channels on the system to match memory windows.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Store the io window base address in struct pci_channel and use that one
instead of SH77xx_PCI_IO_BASE.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Store the base address of the pci host controller registers in struct
pci_channel and use the address in pci_read_reg() and pci_write_reg().
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Store a struct pci_channel pointer in bus->sysdata. This makes whatever
struct pci_channel assigned to a bus available for sh4_pci_read() and
sh4_pci_write(). We also modify PCIBIOS_MIN_IO and PCIBIOS_MIN_MEM to
use bus->sysdata - this to gives us support for multiple pci channels.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

Replaces PCIBIOS_MIN_IO and PCIBIOS_MIN_MEM with direct struct
pci_channel access. This allows us to have more than one pci
channel.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

This patch adds an init callback to struct pci_channel and makes sure
it is initialized properly. Code is added to call this init function
from pcibios_init(). Return values are adjusted and a warning is is
printed if init fails.
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

These patches rework the pci code for the sh architecture.

Currently each board implements some kind of ioport to address mapping.
Some boards use generic_io_base others try passing addresses as io ports.
This is the first set of patches that try to unify the pci code as much
as possible to avoid duplicated code. This will in the end lead to fewer
lines board specific code and more generic code.

This patch makes sure a struct pci_channel pointer is passed along to
various pci functions such as pci_read_reg(), pci_write_reg(),
pci_fixup_pcic(), sh7751_pcic_init() and sh7780_pcic_init().
Signed-off-by: NMagnus Damm <damm@igel.co.jp>
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>

It turns out that 'smp_call_function_many()' doesn't work at all like
'smp_call_function_single()', and my change to Andrew's patch to use it
rather than a loop over all CPU's acpi-cpufreq doesn't work.

My bad.

'smp_call_function_many()' has two "features" (aka "documented bugs"):

 (a) it needs to be called with preemption disabled, because it uses
     smp_processor_id() without guarding the CPU lookup with 'get_cpu()'
     and 'put_cpu()' like the 'single' variant does.

 (b) even if the current CPU is part of the CPU mask, it won't do the
     call on that CPU.

Still, we're better off trying to use 'smp_call_function_many()' than
looping over CPU's, since it at least in theory allows us to use a
broadcast IPI and do it all in parallel.  So let's just work around the
silly semantic bugs in that function.
Reported-and-tested-by: NAli Gholami Rudi <ali@rudi.ir>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Dave Jones <davej@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

A randconfig build on powerpc failed with:

dtl.c: In function 'dtl_init':
dtl.c:238: error: implicit declaration of function 'firmware_has_feature'
dtl.c:238: error: 'FW_FEATURE_SPLPAR' undeclared (first use in this function)

- We need firmware.h for these definitions.
Signed-off-by: NSachin Sant <sachinp@in.ibm.com>
Signed-off-by: NJeremy Kerr <jk@ozlabs.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Richard Henderson pointed out that the powerpc __futex_atomic_op has a
bug: it will write the wrong value if the stwcx. fails and it has to
retry the lwarx/stwcx. loop, since 'oparg' will have been overwritten
by the result from the first time around the loop.  This happens
because it uses the same register for 'oparg' (an input) as it uses
for the result.

This fixes it by using separate registers for 'oparg' and 'ret'.

Cc: stable@kernel.org
Signed-off-by: NPaul Mackerras <paulus@samba.org>