Add cpu_hwcap bit for keeping track of the support for 32bit EL0.
Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

On ARMv8 support for AArch32 state is optional. Hence it is
not safe to check the AArch32 ID registers for sanity, which
could lead to false warnings. This patch makes sure that the
AArch32 state is implemented before we keep track of the 32bit
ID registers.

As per ARM ARM (D.1.21.2 - Support for Exception Levels and
Execution States, DDI0487A.h), checking the support for AArch32
at EL0 is good enough to check the support for AArch32 (i.e,
AArch32 at EL1 => AArch32 at EL0, but not vice versa).
Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Adds a helper to extract the support for AArch32 at EL0
Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

In order to handle systems which do not support 32bit at EL0,
split the COMPAT HWCAP entries into a separate table which can
be processed, only if the support is available.
Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

We use hwcaps for referring to ELF hwcaps capability information.
However this can be confusing with 'cpu_hwcaps' which stands for the
CPU capability bit field. This patch cleans up the names to make it
a bit more readable.
Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

We haven't used the push/pop macros for a while now, as it's typically
better to use immediate offsets for batches of accesses to the stack, as
we now do in the entry assembly for the kernel and hyp code.

Remove the unused macros.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

HAVE_ARCH_TRANSPARENT_HUGEPAGE has been defined in arch/Kconfig already,
the ARM64 version is identical with it and the default value is Y. So remove
the redundant definition and just select it under CONFIG_ARM64.
Signed-off-by: Yang Shi <yang.shi@linaro.org>
[will: sort into alphabetical order whilst I'm resolving conflicts]
Signed-off-by: Will Deacon <will.deacon@arm.com>

Show the bss segment information as with text and data in Virtual
memory kernel layout.
Acked-by: James Morse <james.morse@arm.com>
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Each line with single pr_cont() in Virtual kernel memory layout,
or the dump of the kernel memory layout in dmesg is not aligned
when PRINTK_TIME enabled, due to the missing time stamps.
Tested-by: James Morse <james.morse@arm.com>
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

memblock_remove() takes a phys_addr_t, which may be narrower than 64 bits,
causing a harmless warning:

drivers/firmware/efi/arm-init.c: In function 'reserve_regions':
include/linux/kernel.h:29:20: error: large integer implicitly truncated to unsigned type [-Werror=overflow]
 #define ULLONG_MAX (~0ULL)
                    ^
drivers/firmware/efi/arm-init.c:152:21: note: in expansion of macro 'ULLONG_MAX'
  memblock_remove(0, ULLONG_MAX);

This adds an explicit typecast to avoid the warning

Fixes: 500899c2 ("efi: ARM/arm64: ignore DT memory nodes instead of removing them")
Acked-by Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>

When CPUs are stopped during an abnormal operation like panic
for each CPU a line is printed and the stack trace is dumped.

This information is only interesting for the aborting CPU
and on systems with many CPUs it only makes it harder to
debug if after the aborting CPU the log is flooded with data
about all other CPUs too.

Therefore remove the stack dump and printk of other CPUs
and only print a single line that the other CPUs are going to be
stopped and, in case any CPUs remain online list them.
Signed-off-by: Jan Glauber <jglauber@cavium.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

We already re-enable interrupts where necessary in the entry code, so
there is no need to do it again in do_page fault. This patch removes
the redundant code.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Huang Shijie <shijie.huang@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

When hardware updates of the access and dirty states are enabled, the
default ptep_set_access_flags() implementation based on calling
set_pte_at() directly is potentially racy. This triggers the "racy dirty
state clearing" warning in set_pte_at() because an existing writable PTE
is overridden with a clean entry.

There are two main scenarios for this situation:

1. The CPU getting an access fault does not support hardware updates of
   the access/dirty flags. However, a different agent in the system
   (e.g. SMMU) can do this, therefore overriding a writable entry with a
   clean one could potentially lose the automatically updated dirty
   status

2. A more complex situation is possible when all CPUs support hardware
   AF/DBM:

   a) Initial state: shareable + writable vma and pte_none(pte)
   b) Read fault taken by two threads of the same process on different
      CPUs
   c) CPU0 takes the mmap_sem and proceeds to handling the fault. It
      eventually reaches do_set_pte() which sets a writable + clean pte.
      CPU0 releases the mmap_sem
   d) CPU1 acquires the mmap_sem and proceeds to handle_pte_fault(). The
      pte entry it reads is present, writable and clean and it continues
      to pte_mkyoung()
   e) CPU1 calls ptep_set_access_flags()

   If between (d) and (e) the hardware (another CPU) updates the dirty
   state (clears PTE_RDONLY), CPU1 will override the PTR_RDONLY bit
   marking the entry clean again.

This patch implements an arm64-specific ptep_set_access_flags() function
to perform an atomic update of the PTE flags.

Fixes: 2f4b829c ("arm64: Add support for hardware updates of the access and dirty pte bits")
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: Ming Lei <tom.leiming@gmail.com>
Tested-by: Julien Grall <julien.grall@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: <stable@vger.kernel.org> # 4.3+
[will: reworded comment]
Signed-off-by: Will Deacon <will.deacon@arm.com>

Enable NUMA balancing for arm64 platforms.
Add pte, pmd protnone helpers for use by automatic NUMA balancing.
Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Attempt to get the memory and CPU NUMA node via of_numa.  If that
fails, default the dummy NUMA node and map all memory and CPUs to node
0.
Tested-by: Shannon Zhao <shannon.zhao@linaro.org>
Reviewed-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

In order to extract NUMA information from the device tree, we need to
have the tree in its unflattened form.

Move the call to bootmem_init() in the tail of paging_init() into
setup_arch, and adjust header files so that its declaration is
visible.

Move the unflatten_device_tree() call between the calls to
paging_init() and bootmem_init().  Follow on patches add NUMA handling
to bootmem_init().
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Add device tree parsing for NUMA topology using device
"numa-node-id" property in distance-map and cpu nodes.

This is a complete rewrite of a previous patch by:
   Ganapatrao Kulkarni<gkulkarni@caviumnetworks.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Add DT bindings for numa mapping of memory, CPUs and IOs.
Reviewed-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

There are two problems with the UEFI stub DT memory node removal
routine:
- it deletes nodes as it traverses the tree, which happens to work
  but is not supported, as deletion invalidates the node iterator;
- deleting memory nodes entirely may discard annotations in the form
  of additional properties on the nodes.

Since the discovery of DT memory nodes occurs strictly before the
UEFI init sequence, we can simply clear the memblock memory table
before parsing the UEFI memory map. This way, it is no longer
necessary to remove the nodes, so we can remove that logic from the
stub as well.
Reviewed-by: Matt Fleming <matt@codeblueprint.co.uk>
Acked-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

With a VHE capable CPU, kernel can run at EL2 and is a decided at early
boot. If some of the CPUs didn't start it EL2 or doesn't have VHE, we
could have CPUs running at different exception levels, all in the same
kernel! This patch adds an early check for the secondary CPUs to detect
such situations.

For each non-boot CPU add a sanity check to make sure we don't have
different run levels w.r.t the boot CPU. We save the information on
whether the boot CPU is running in hyp mode or not and ensure the
remaining CPUs match it.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
[will: made boot_cpu_hyp_mode static]
Signed-off-by: Will Deacon <will.deacon@arm.com>

During the activation of a secondary CPU, we could report serious
configuration issues and hence request to crash the kernel. We do
this for CPU ASID bit check now. We will need it also for handling
mismatched exception levels for the CPUs with VHE. Hence, add a
helper to do the same for reusability.

Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

With CONFIG_PROVE_LOCKING, CONFIG_DEBUG_LOCKDEP and CONFIG_TRACE_IRQFLAGS
enabled, lockdep will compare current->hardirqs_enabled with the flags from
local_irq_save().

When a debug exception occurs, interrupts are disabled in entry.S, but
lockdep isn't told, resulting in:
DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)
------------[ cut here ]------------
WARNING: at ../kernel/locking/lockdep.c:3523
Modules linked in:
CPU: 3 PID: 1752 Comm: perf Not tainted 4.5.0-rc4+ #2204
Hardware name: ARM Juno development board (r1) (DT)
task: ffffffc974868000 ti: ffffffc975f40000 task.ti: ffffffc975f40000
PC is at check_flags.part.35+0x17c/0x184
LR is at check_flags.part.35+0x17c/0x184
pc : [<ffffff80080fc93c>] lr : [<ffffff80080fc93c>] pstate: 600003c5
[...]
---[ end trace 74631f9305ef5020 ]---
Call trace:
[<ffffff80080fc93c>] check_flags.part.35+0x17c/0x184
[<ffffff80080ffe30>] lock_acquire+0xa8/0xc4
[<ffffff8008093038>] breakpoint_handler+0x118/0x288
[<ffffff8008082434>] do_debug_exception+0x3c/0xa8
[<ffffff80080854b4>] el1_dbg+0x18/0x6c
[<ffffff80081e82f4>] do_filp_open+0x64/0xdc
[<ffffff80081d6e60>] do_sys_open+0x140/0x204
[<ffffff80081d6f58>] SyS_openat+0x10/0x18
[<ffffff8008085d30>] el0_svc_naked+0x24/0x28
possible reason: unannotated irqs-off.
irq event stamp: 65857
hardirqs last  enabled at (65857): [<ffffff80081fb1c0>] lookup_mnt+0xf4/0x1b4
hardirqs last disabled at (65856): [<ffffff80081fb188>] lookup_mnt+0xbc/0x1b4
softirqs last  enabled at (65790): [<ffffff80080bdca4>] __do_softirq+0x1f8/0x290
softirqs last disabled at (65757): [<ffffff80080be038>] irq_exit+0x9c/0xd0

This patch adds the annotations to do_debug_exception(), while trying not
to call trace_hardirqs_off() if el1_dbg() interrupted a task that already
had irqs disabled.
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Since commit 1cf4f629 ("cpu/hotplug: Move online calls to
hotplugged cpu") it is ensured that callbacks of CPU_ONLINE and
CPU_DOWN_PREPARE are processed on the hotplugged CPU. Due to this SMP
function calls are no longer required.

Replace smp_call_function_single() with a direct call of
hw_breakpoint_reset(). To keep the calling convention, interrupts are
explicitly disabled around the call.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Since commit 1cf4f629 ("cpu/hotplug: Move online calls to
hotplugged cpu") it is ensured that callbacks of CPU_ONLINE and
CPU_DOWN_PREPARE are processed on the hotplugged CPU. Due to this SMP
function calls are no longer required.

Replace smp_call_function_single() with a direct call to
clear_os_lock(). The function writes the OSLAR register to clear OS
locking. This does not require to be called with interrupts disabled,
therefore the smp_call_function_single() calling convention is not
preserved.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The mapping of the kernel consist of four segments, each of which is mapped
with different permission attributes and/or lifetimes. To optimize the TLB
and translation table footprint, we define various opaque constants in the
linker script that resolve to different aligment values depending on the
page size and whether CONFIG_DEBUG_ALIGN_RODATA is set.

Considering that
- a 4 KB granule kernel benefits from a 64 KB segment alignment (due to
  the fact that it allows the use of the contiguous bit),
- the minimum alignment of the .data segment is THREAD_SIZE already, not
  PAGE_SIZE (i.e., we already have padding between _data and the start of
  the .data payload in many cases),
- 2 MB is a suitable alignment value on all granule sizes, either for
  mapping directly (level 2 on 4 KB), or via the contiguous bit (level 3 on
  16 KB and 64 KB),
- anything beyond 2 MB exceeds the minimum alignment mandated by the boot
  protocol, and can only be mapped efficiently if the physical alignment
  happens to be the same,

we can simplify this by standardizing on 64 KB (or 2 MB) explicitly, i.e.,
regardless of granule size, all segments are aligned either to 64 KB, or to
2 MB if CONFIG_DEBUG_ALIGN_RODATA=y. This also means we can drop the Kconfig
dependency of CONFIG_DEBUG_ALIGN_RODATA on CONFIG_ARM64_4K_PAGES.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Keeping .head.text out of the .text mapping buys us very little: its actual
payload is only 4 KB, most of which is padding, but the page alignment may
add up to 2 MB (in case of CONFIG_DEBUG_ALIGN_RODATA=y) of additional
padding to the uncompressed kernel Image.

Also, on 4 KB granule kernels, the 4 KB misalignment of .text forces us to
map the adjacent 56 KB of code without the PTE_CONT attribute, and since
this region contains things like the vector table and the GIC interrupt
handling entry point, this region is likely to benefit from the reduced TLB
pressure that results from PTE_CONT mappings.

So remove the alignment between the .head.text and .text sections, and use
the [_text, _etext) rather than the [_stext, _etext) interval for mapping
the .text segment.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Apart from the arm64/linux and EFI header data structures, there is nothing
in the .head.text section that must reside at the beginning of the Image.
So let's move it to the .init section where it belongs.

Note that this involves some minor tweaking of the EFI header, primarily
because the address of 'stext' no longer coincides with the start of the
.text section. It also requires a couple of relocated symbol references
to be slightly rewritten or their definition moved to the linker script.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Replace the poorly defined term chunk with segment, which is a term that is
already used by the ELF spec to describe contiguous mappings with the same
permission attributes of statically allocated ranges of an executable.
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Now that the vmemmap region has been redefined to cover the linear region
rather than the entire physical address space, we no longer need to
perform a virtual-to-physical translation in the implementaion of
virt_to_page(). This restricts virt_to_page() translations to the linear
region, so redefine virt_addr_valid() as well.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This moves the vmemmap region right below PAGE_OFFSET, aka the start
of the linear region, and redefines its size to be a power of two.
Due to the placement of PAGE_OFFSET in the middle of the address space,
whose size is a power of two as well, this guarantees that virt to
page conversions and vice versa can be implemented efficiently, by
masking and shifting rather than ordinary arithmetic.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Before restricting virt_to_page() to the linear mapping, ensure that
the text patching code does not use it to resolve references into the
core kernel text, which is mapped in the vmalloc area.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The zero page is statically allocated, so grab its struct page pointer
without using virt_to_page(), which will be restricted to the linear
mapping later.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The implementation of free_initmem_default() expects __init_begin
and __init_end to be covered by the linear mapping, which is no
longer the case. So open code it instead, using addresses that are
explicitly translated from kernel virtual to linear virtual.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The translation performed by virt_to_page() is only valid for linear
addresses, and kernel symbols are no longer in the linear mapping.
So perform the __pa() translation explicitly, which does the right
thing in either case, and only then translate to a struct page offset.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This removes the relocate_initrd() implementation and invocation, which are
no longer needed now that the placement of the initrd is guaranteed to be
covered by the linear mapping.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Instead of going out of our way to relocate the initrd if it turns out
to occupy memory that is not covered by the linear mapping, just add the
initrd to the linear mapping. This puts the burden on the bootloader to
pass initrd= and mem= options that are mutually consistent.

Note that, since the placement of the linear region in the PA space is
also dependent on the placement of the kernel Image, which may reside
anywhere in memory, we may still end up with a situation where the initrd
and the kernel Image are simply too far apart to be covered by the linear
region.

Since we now leave it up to the bootloader to pass the initrd in memory
that is guaranteed to be accessible by the kernel, add a mention of this to
the arm64 boot protocol specification as well.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This reverts commit 36e5cd6b, since the
section alignment is now guaranteed by construction when choosing the
value of memstart_addr.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This redefines ARM64_MEMSTART_ALIGN in terms of the minimal alignment
required by sparsemem vmemmap. This comes down to using 1 GB for all
translation granules if CONFIG_SPARSEMEM_VMEMMAP is enabled.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

After choosing memstart_addr to be the highest multiple of
ARM64_MEMSTART_ALIGN less than or equal to the first usable physical memory
address, we clip the memblocks to the maximum size of the linear region.
Since the kernel may be high up in memory, we take care not to clip the
kernel itself, which means we have to clip some memory from the bottom if
this occurs, to ensure that the distance between the first and the last
usable physical memory address can be covered by the linear region.

However, we fail to update memstart_addr if this clipping from the bottom
occurs, which means that we may still end up with virtual addresses that
wrap into the userland range. So increment memstart_addr as appropriate to
prevent this from happening.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Currently, we check two pointers: cpu_ops and cpu_suspend on every idle
state entry. These pointers check can be avoided:

If cpu_ops has not been registered, arm_cpuidle_init() will return
-EOPNOTSUPP, so arm_cpuidle_suspend() will never have chance to
run. In other word, the cpu_ops check can be avoid.

Similarly, the cpu_suspend check could be avoided in this hot path by
moving it into arm_cpuidle_init().

I measured the 4096 * time from arm_cpuidle_suspend entry point to the
cpu_psci_cpu_suspend entry point. HW platform is Marvell BG4CT STB
board.

1. only one shell, no other process, hot-unplug secondary cpus, execute
the following cmd

while true
do
	sleep 0.2
done

before the patch: 1581220ns

after the patch: 1579630ns

reduced by 0.1%

2. only one shell, no other process, hot-unplug secondary cpus, execute
the following cmd

while true
do
	md5sum /tmp/testfile
	sleep 0.2
done

NOTE: the testfile size should be larger than L1+L2 cache size

before the patch: 1961960ns
after the patch: 1912500ns

reduced by 2.5%

So the more complex the system load, the bigger the improvement.
Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Acked-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>