Since commit f2c3c67f (merge commit that adds commit "ARM: mvebu:
completely disable hardware I/O coherency"), we disable I/O coherency
on Armada EBU platforms.

However, we continue to initialize the coherency fabric, because this
coherency fabric is needed on Armada XP for inter-CPU
coherency. Unfortunately, due to this, we also continued to execute
the coherency fabric initialization code for Armada 375/38x, which
switched the PL310 into I/O coherent mode. This has the effect of
disabling the outer cache sync operation: this is needed when I/O
coherency is enabled to work around a PCIe/L2 deadlock. But obviously,
when I/O coherency is disabled, having the outer cache sync operation
is crucial.

Therefore, this commit fixes the armada_375_380_coherency_init() so
that the PL310 is switched to I/O coherent mode only if I/O coherency
is enabled.

Without this fix, all devices using DMA are broken on Armada 375/38x.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Tested-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Signed-off-by: NAndrew Lunn <andrew@lunn.ch>
Cc: <stable@vger.kernel.org> # v3.8+

Now that we have enabled automatic I/O synchronization barriers, we no
longer need any explicit barriers. We can therefore simplify
arch/arm/mach-mvebu/coherency.c by using the existing
arm_coherent_dma_ops instead of our custom mvebu_hwcc_dma_ops, and
re-enable hardware I/O coherency support.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
[Andrew Lunn <andrew@lunn.ch>: Remove forgotten comment]
Signed-off-by: NAndrew Lunn <andrew@lunn.ch>

Merge " mvebu fixes for 3.19-rc (part #3)" from Andrew Lunn:

mvebu: completely disable hardware I/O coherency

* tag 'mvebu-fixes-3.19-3' of git://git.infradead.org/linux-mvebu:
  ARM: mvebu: completely disable hardware I/O coherency
Signed-off-by: NOlof Johansson <olof@lixom.net>

The current hardware I/O coherency is known to cause problems with DMA
coherent buffers, as it still requires explicit I/O synchronization
barriers, which is not compatible with the semantics expected by the
Linux DMA coherent buffers API.

So, in order to have enough time to validate a new solution based on
automatic I/O synchronization barriers, this commit disables hardware
I/O coherency entirely. Future patches will re-enable it.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: <stable@vger.kernel.org> # v3.8+
Signed-off-by: NAndrew Lunn <andrew@lunn.ch>

This patch removes the unneeded include of the armada-370-xp.h header.

It also moves some declarations from this file into more accurate
places.

Finally, it also adds a comment explaining that we can't remove yet the
smp field in the dt machine struct due to backward compatibly of the
device tree.

In a few releases, when the old device tree will be obsolete, we will be
able to remove the smp field and then the armada-370-xp.h header.
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Tested-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Reviewed-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1414669184-16785-2-git-send-email-gregory.clement@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The coherency.c top-level comment mentions that it supports the
coherency fabric for Armada 370 and XP, but it also supports the
coherency fabric on Armada 375 and 38x, so this commit updates the
comment accordingly.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1415871540-20302-6-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

This reverts commit 5ab5afd8 ("ARM: mvebu: implement Armada 375
coherency workaround"), since we are removing the support for the very
early Z1 revision of the Armada 375 SoC.

This commit is an exact revert, with two exceptions:

 - minor adaptations needed due to other changes that have taken place
   in coherency.c since the original commit

 - keep the definition of pr_fmt. This shouldn't originally have been
   part of the Armada 375 Z1 workaround commit since it had nothing to
   do with it.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1415871540-20302-5-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Since commit b21dcafe ("arm: mvebu: remove dependency of SMP init
on static I/O mapping"), the COHERENCY_FABRIC_CFG_OFFSET register
offset definition is no longer used, so this commit removes it.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1415871540-20302-4-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Enabling the hardware I/O coherency on Armada 370, Armada 375, Armada
38x and Armada XP requires a certain number of conditions:

 - On Armada 370, the cache policy must be set to write-allocate.

 - On Armada 375, 38x and XP, the cache policy must be set to
   write-allocate, the pages must be mapped with the shareable
   attribute, and the SMP bit must be set

Currently, on Armada XP, when CONFIG_SMP is enabled, those conditions
are met. However, when Armada XP is used in a !CONFIG_SMP kernel, none
of these conditions are met. With Armada 370, the situation is worse:
since the processor is single core, regardless of whether CONFIG_SMP
or !CONFIG_SMP is used, the cache policy will be set to write-back by
the kernel and not write-allocate.

Since solving this problem turns out to be quite complicated, and we
don't want to let users with a mainline kernel known to have
infrequent but existing data corruptions, this commit proposes to
simply disable hardware I/O coherency in situations where it is known
not to work.

And basically, the is_smp() function of the kernel tells us whether it
is OK to enable hardware I/O coherency or not, so this commit slightly
refactors the coherency_type() function to return
COHERENCY_FABRIC_TYPE_NONE when is_smp() is false, or the appropriate
type of the coherency fabric in the other case.

Thanks to this, the I/O coherency fabric will no longer be used at all
in !CONFIG_SMP configurations. It will continue to be used in
CONFIG_SMP configurations on Armada XP, Armada 375 and Armada 38x
(which are multiple cores processors), but will no longer be used on
Armada 370 (which is a single core processor).

In the process, it simplifies the implementation of the
coherency_type() function, and adds a missing call to of_node_put().
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Fixes: e60304f8 ("arm: mvebu: Add hardware I/O Coherency support")
Cc: <stable@vger.kernel.org> # v3.8+
Acked-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1415871540-20302-3-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

There is a missing of_node_put() to decrement the device_node
reference counter after a of_find_matching_node() in coherency_init().

Fixes: 501f928e ("ARM: mvebu: add a coherency_available() call")
Cc: <stable@vger.kernel.org> # v3.16+
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1414423955-5933-4-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Currently, the coherency fabric support registers two bus notifiers;
one for platform, one for pci bus types, with the same notifier block.
However, this is illegal and can cause serious issues: the notifier
block is also a link in the notifier list and cannot be inserted twice.

This commit fixes this by using different notifier blocks (with the same
notifier callback) to set the platform and pci bus types notifiers.

Fixes: b0063aad ("ARM: mvebu: use hardware I/O coherency also for PCI devices")
Reported-by: NPaolo Pisati <p.pisati@gmail.com>
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Link: https://lkml.kernel.org/r/1404826657-6977-1-git-send-email-ezequiel.garcia@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Commit b0063aad ("ARM: mvebu: use hardware I/O coherency also for
PCI devices") added a reference to the pci_bus_type variable, but this
variable is only available when CONFIG_PCI is enabled. Therefore,
there is now a build failure in !CONFIG_PCI situations.

This commit fixes that by enclosing the entire initcall into a
IS_ENABLED(CONFIG_PCI) condition.
Reported-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1400598783-706-1-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The Marvell Armada 375 and Armada 38x SOCs, which use the Cortex-A9
CPU core, the PL310 cache and the Marvell PCIe hardware block are
affected a L2/PCIe deadlock caused by a system erratum when hardware
I/O coherency is used.

This deadlock can be avoided by mapping the PCIe memory areas as
strongly-ordered (note: MT_UNCACHED is strongly-ordered), and by
removing the outer cache sync done in software. This is implemented in
this patch by:

 * Registering a custom arch_ioremap_caller function that allows to
   make sure PCI memory regions are mapped MT_UNCACHED.

 * Adding at runtime the 'arm,io-coherent' property to the PL310 cache
   controller. This cannot be done permanently in the DT, because the
   hardware I/O coherency can only be enabled when CONFIG_SMP is
   enabled, in the current kernel situation.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1400165974-9059-4-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Since the beginning of the introduction of hardware I/O coherency
support for Armada 370 and Armada XP, the special DMA operations
should have applied to all DMA capable devices. Unfortunately, while
the original code properly took into account platform devices, it
didn't take into account PCI devices, which can also be DMA masters.

This commit fixes that by registering a bus notifier on pci_bus_type,
to register our custom DMA operations, like is already done for
platform devices. While doing this, we also rename
mvebu_hwcc_platform_notifier() to mvebu_hwcc_notifier() and
mvebu_hwcc_platform_nb to mvebu_hwcc_nb because they are no longer
specific to platform devices.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1399997070-11434-1-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The Armada 375 coherency workaround only needs to be applied to the Z1
revision of the SoC. The A0 and later revisions have been fixed, and
no longer need this workaround.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1399302326-6917-6-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

Actually enabling coherency and adding a CPU on a SMP group are two
different operations which can be done separately. This patch splits
this in two functions.

Moreover as they use common pattern, this patch also creates local low
level functions (ll_get_coherency_base and ll_get_cpuid) to be used by
the exposed functions (ll_add_cpu_to_smp_group and
ll_enable_coherency)
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1397488214-20685-6-git-send-email-gregory.clement@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

set_cpu_coherent() took the SMP group ID as parameter. But this
parameter was never used, and the CPU always uses the SMP group 0. So
we can remove this parameter.
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1397488214-20685-5-git-send-email-gregory.clement@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

ll_set_cpu_coherent is always used on the current CPU, so instead of
passing the CPU id as argument, ll_set_cpu_coherent() can find it by
itself.
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1397488214-20685-4-git-send-email-gregory.clement@free-electrons.comAcked-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

In order to be able to deal with the MMU enabled and the MMU disabled
cases, the base address of the coherency registers was passed to the
function. The address by itself was not interesting as it can't change
for a given SoC, the only thing we need is to have a distinction
between the physical or the virtual address.

This patch add a check of the MMU bit to choose the accurate address,
then the calling function doesn't have to pass this information.
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Link: https://lkml.kernel.org/r/1397488214-20685-3-git-send-email-gregory.clement@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The Armada 38x has a coherency unit that is similar to the one of the
Armada 375 SoC, except that it does not have the bug of the Armada 375
coherency unit that requires the XOR based workaround.

This commit therefore extends the Marvell EBU coherency code with a
new compatible string to support the Armada 38x coherency unit.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-9-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The early revisions of Armada 375 SOCs (Z1 stepping) have a bug in the
I/O coherency unit that prevents using the normal method for the I/O
coherency barrier. The recommended workaround is to use a XOR memset
transfer to act as the I/O coherency barrier.

This involves "borrowing" a XOR engine, which gets disabled in the
Device Tree so the normal XOR driver doesn't use it.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-8-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The Armada 375, like the Armada 370 and Armada XP, has a coherency
unit. However, unlike the coherency unit of 370/XP which does both CPU
and I/O coherency, the one on Armada 735 only does I/O
coherency. Therefore, instead of having two sets of registers (the
first one being used mainly to register each CPU in the coherency
fabric, the second one being used for the I/O coherency barrier), it
has only one set of register (for the I/O coherency barrier).

This commit adds a new "marvell,armada-375-coherency-fabric"
compatible string for this variant of the coherency fabric. The custom
DMA operations, and the way of triggering an I/O barrier is the same
as Armada 370/XP, so the code changes are minimal. However, the
set_cpu_coherent() function is not needed on Armada 375 and will not
work.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-7-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

In the mach-mvebu coherency code, instead of using
of_find_matching_node() and then of_match_node(), directly use the
of_find_matching_node_and_match() which does both at once.

We take this opportunity to also simplify the initialization of the
"type" variable.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-5-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

This commit extends the coherency fabric code to provide a
coherency_available()function that the SoC code can call to be told
whether coherency support is available or not. On Armada 370/XP,
coherency support is available as soon as the relevant DT node is
present. On some upcoming SoCs, the DT node needs to be present *and*
the system running with CONFIG_SMP enabled.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-3-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

The code that handles the coherency fabric of Armada 370 and Armada XP
in arch/arm/mach-mvebu/coherency.c made the assumption that there was
only one type of coherency fabric. Unfortunately, it turns out that
upcoming SoCs have a slightly different coherency unit.

In preparation to the introduction of the coherency support for more
SoCs, this commit:

 * Introduces a data associated to the compatible string in the
   compatible string match table, so that the code can differantiate
   the variant of coherency unit being used.

 * Separates the coherency unit initialization code into its own
   function.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Link: https://lkml.kernel.org/r/1397483228-25625-2-git-send-email-thomas.petazzoni@free-electrons.comSigned-off-by: NJason Cooper <jason@lakedaemon.net>

This patch fixes conflicting types for 'set_cpu_coherent' and fixes the
following sparse warnings.

arch/arm/mach-mvebu/system-controller.c:42:38:
warning: symbol 'armada_370_xp_system_controller' was not declared. Should it be static?
arch/arm/mach-mvebu/system-controller.c:49:38:
warning: symbol 'orion_system_controller' was not declared. Should it be static?
arch/arm/mach-mvebu/system-controller.c:67:6:
warning: symbol 'mvebu_restart' was not declared. Should it be static?
arch/arm/mach-mvebu/coherency.c:31:15:
warning: symbol 'coherency_phys_base' was not declared. Should it be static?
arch/arm/mach-mvebu/coherency.c:48:5:
warning: symbol 'set_cpu_coherent' was not declared. Should it be static?
arch/arm/mach-mvebu/coherency.c:123:12:
warning: symbol 'coherency_init' was not declared. Should it be static?
arch/arm/mach-mvebu/pmsu.c:38:5: warning:
symbol 'armada_xp_boot_cpu' was not declared. Should it be static?
arch/arm/mach-mvebu/pmsu.c:61:12: warning:
symbol 'armada_370_xp_pmsu_init' was not declared. Should it be static?
arch/arm/mach-mvebu/platsmp.c:49:13: warning:
symbol 'set_secondary_cpus_clock' was not declared. Should it be static?
arch/arm/mach-mvebu/platsmp.c:97:13: warning:
symbol 'armada_xp_smp_prepare_cpus' was not declared. Should it be static?
arch/arm/mach-mvebu/hotplug.c:24:12: warning:
symbol 'armada_xp_cpu_die' was not declared. Should it be static?
Signed-off-by: NJisheng Zhang <jszhang@marvell.com>
Acked-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

Add of_node_put to properly decrement the refcount when we are
done using a given node.
Signed-off-by: NJisheng Zhang <jszhang@marvell.com>
Reviewed-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

Conflicts:
	arch/arm/mach-mvebu/armada-370-xp.c
	arch/arm/mach-mvebu/platsmp.c

The __cpuinit type of throwaway sections might have made sense
some time ago when RAM was more constrained, but now the savings
do not offset the cost and complications.  For example, the fix in
commit 5e427ec2 ("x86: Fix bit corruption at CPU resume time")
is a good example of the nasty type of bugs that can be created
with improper use of the various __init prefixes.

After a discussion on LKML[1] it was decided that cpuinit should go
the way of devinit and be phased out.  Once all the users are gone,
we can then finally remove the macros themselves from linux/init.h.

Note that some harmless section mismatch warnings may result, since
notify_cpu_starting() and cpu_up() are arch independent (kernel/cpu.c)
and are flagged as __cpuinit  -- so if we remove the __cpuinit from
the arch specific callers, we will also get section mismatch warnings.
As an intermediate step, we intend to turn the linux/init.h cpuinit
related content into no-ops as early as possible, since that will get
rid of these warnings.  In any case, they are temporary and harmless.

This removes all the ARM uses of the __cpuinit macros from C code,
and all __CPUINIT from assembly code.  It also had two ".previous"
section statements that were paired off against __CPUINIT
(aka .section ".cpuinit.text") that also get removed here.

[1] https://lkml.org/lkml/2013/5/20/589

Cc: Russell King <linux@arm.linux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>

As noticed by Arnaud Patard (Rtp) <arnaud.patard@rtp-net.org>, commit
865e0527 ('arm: mvebu: avoid hardcoded virtual address in
coherency code') added a postcore_initcall() to register the bus
notifier that the mvebu code needs to apply correct DMA operations on
its platform devices breaks the multiplatform boot on other platforms,
because the bus notifier registration is unconditional.

This commit fixes that by registering the bus notifier only if we have
the mvebu coherency unit described in the Device Tree. The conditional
used is exactly the same in which the bus_register_notifier() call was
originally enclosed before 865e0527 ('arm: mvebu: avoid hardcoded
virtual address in coherency code').
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Reported-by: NArnaud Patard (Rtp) <arnaud.patard@rtp-net.org>
Acked-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

Now that the coherency_init() function is called a bit earlier, we can
actually read the physical address of the coherency unit registers
from the Device Tree, and communicate that to the headsmp.S code,
which avoids hardcoding a physical address.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: NArnd Bergmann <arnd@arndb.de>
Reviewed-by: NWill Deacon <will.deacon@arm.com>
Acked-by: NNicolas Pitre <nico@linaro.org>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

Now that the coherency_get_cpu_count() function no longer requires a
very early mapping of the coherency unit registers, we can avoid the
hardcoded virtual address in coherency.c. However, the coherency
features are still used quite early, so we need to do the of_iomap()
early enough, at the ->init_timer() level, so we have the call of
coherency_init() at this point.

Unfortunately, at ->init_timer() time, it is not possible to register
a bus notifier, so we add a separate coherency_late_init() function
that gets called as as postcore_initcall(), when bus notifiers are
available.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

The ->smp_init_cpus() function is called very early during boot, at a
point where dynamic I/O mappings are not yet possible. However, in the
Armada 370/XP implementation of this function, we have to get the
number of CPUs. We used to do that by accessing a hardware register,
which requires relying on a static I/O mapping set up by
->map_io(). Not only this requires hardcoding a virtual address, but
it also prevents us from removing the static I/O mapping.

So this commit changes the way used to get the number of CPUs: we now
use the Device Tree, which is a representation of the hardware, and
provides us the number of available CPUs. This is also more accurate,
because it potentially allows to boot the Linux kernel on only a
number of CPUs given by the Device Tree, instead of unconditionally on
all CPUs.

As a consequence, the coherency_get_cpu_count() function becomes no
longer used, so we remove it.
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NJason Cooper <jason@lakedaemon.net>

Armada 370 and XP come with an unit called coherency fabric. This unit
allows to use the Armada 370/XP as a nearly coherent architecture. The
coherency mechanism uses snoop filters to ensure the coherency between
caches, DRAM and devices. This mechanism needs a synchronization
barrier which guarantees that all the memory writes initiated by the
devices have reached their target and do not reside in intermediate
write buffers. That's why the architecture is not totally coherent and
we need to provide our own functions for some DMA operations.

Beside the use of the coherency fabric, the device units will have to
set the attribute flag of the decoding address window to select the
accurate coherency process for the memory transaction. This is done
each device driver programs the DRAM address windows. The value of the
attribute set by the driver is retrieved through the
orion_addr_map_cfg struct filled during the early initialization of
the platform.
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Reviewed-by: NYehuda Yitschak <yehuday@marvell.com>
Acked-by: NMarek Szyprowski <m.szyprowski@samsung.com>

The Armada 370 and Armada XP SOCs have a coherency fabric unit which
is responsible for ensuring hardware coherency between all CPUs and
between CPUs and I/O masters. This patch provides the basic support
needed for SMP.
Signed-off-by: NYehuda Yitschak <yehuday@marvell.com>
Signed-off-by: NGregory CLEMENT <gregory.clement@free-electrons.com>
Reviewed-by: NWill Deacon <will.deacon@arm.com>