With the introduction of 52-bit virtual addressing for userspace, we are
now in a position where the virtual addressing capability of userspace
may exceed that of the kernel. Consequently, the VA_BITS definition
cannot be used blindly, since it reflects only the size of kernel
virtual addresses.

This patch introduces MAX_USER_VA_BITS which is either VA_BITS or 52
depending on whether 52-bit virtual addressing has been configured at
build time, removing a few places where the 52 is open-coded based on
explicit CONFIG_ guards.
Signed-off-by: Will Deacon <will.deacon@arm.com>

In some randconfig builds, the new CONFIG_ARM64_USER_VA_BITS_52
triggered a build failure:

arch/arm64/mm/proc.S:287: Error: immediate out of range

As it turns out, we were incorrectly setting PGTABLE_LEVELS here,
lacking any other default value.
This fixes the calculation of CONFIG_PGTABLE_LEVELS to consider
all combinations again.

Fixes: 68d23da4 ("arm64: Kconfig: Re-jig CONFIG options for 52-bit VA")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Commit 39624469 ("arm64: preempt: Provide our own implementation of
asm/preempt.h") extended the preempt count field in struct thread_info
to 64 bits, so that it consists of a 32-bit count plus a 32-bit flag
indicating whether or not the current task needs rescheduling.

Whilst the asm-offsets definition of TSK_TI_PREEMPT was updated to point
to this new field, the assembly usage was left untouched meaning that a
32-bit load from TSK_TI_PREEMPT on a big-endian machine actually returns
the reschedule flag instead of the count.

Whilst we could fix this by pointing TSK_TI_PREEMPT at the count field,
we're actually better off reworking the two assembly users so that they
operate on the whole 64-bit value in favour of inspecting the thread
flags separately in order to determine whether a reschedule is needed.
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reported-by: "kernelci.org bot" <bot@kernelci.org>
Tested-by: Kevin Hilman <khilman@baylibre.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This is needed for compilation in some configurations that don't
include it implicitly:

arch/arm64/kernel/machine_kexec_file.c: In function 'arch_kimage_file_post_load_cleanup':
arch/arm64/kernel/machine_kexec_file.c:37:2: error: implicit declaration of function 'vfree'; did you mean 'kvfree'? [-Werror=implicit-function-declaration]

Fixes: 52b2a8af ("arm64: kexec_file: load initrd and device-tree")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>

TASK_SIZE is defined using the vabits_user variable for 64-bit tasks,
so ensure that this variable is exported to modules to avoid the
following build breakage with allmodconfig:

 | ERROR: "vabits_user" [lib/test_user_copy.ko] undefined!
 | ERROR: "vabits_user" [drivers/misc/lkdtm/lkdtm.ko] undefined!
 | ERROR: "vabits_user" [drivers/infiniband/hw/mlx5/mlx5_ib.ko] undefined!
Signed-off-by: Will Deacon <will.deacon@arm.com>

Merge in kexec_file_load() support from Akashi Takahiro.

Pull in KVM workaround for A76 erratum #116522.

Conflicts:
	arch/arm64/include/asm/cpucaps.h

The __cpu_up() routine ignores the errors reported by the firmware
for a CPU bringup operation and looks for the error status set by the
booting CPU. If the CPU never entered the kernel, we could end up
in assuming stale error status, which otherwise would have been
set/cleared appropriately by the booting CPU.
Reported-by: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Rather than add additional variables to detect specific early feature
mismatches with secondary CPUs, we can instead dedicate the upper bits
of the CPU boot status word to flag specific mismatches.

This allows us to communicate both granule and VA-size mismatches back
to the primary CPU without the need for additional book-keeping.
Tested-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Enabling 52-bit VAs for userspace is pretty confusing, since it requires
you to select "48-bit" virtual addressing in the Kconfig.

Rework the logic so that 52-bit user virtual addressing is advertised in
the "Virtual address space size" choice, along with some help text to
describe its interaction with Pointer Authentication. The EXPERT-only
option to force all user mappings to the 52-bit range is then made
available immediately below the VA size selection.
Signed-off-by: Will Deacon <will.deacon@arm.com>

On arm64 52-bit VAs are provided to userspace when a hint is supplied to
mmap. This helps maintain compatibility with software that expects at
most 48-bit VAs to be returned.

In order to help identify software that has 48-bit VA assumptions, this
patch allows one to compile a kernel where 52-bit VAs are returned by
default on HW that supports it.

This feature is intended to be for development systems only.
Signed-off-by: Steve Capper <steve.capper@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

On arm64 there is optional support for a 52-bit virtual address space.
To exploit this one has to be running with a 64KB page size and be
running on hardware that supports this.

For an arm64 kernel supporting a 48 bit VA with a 64KB page size,
some changes are needed to support a 52-bit userspace:
 * TCR_EL1.T0SZ needs to be 12 instead of 16,
 * TASK_SIZE needs to reflect the new size.

This patch implements the above when the support for 52-bit VAs is
detected at early boot time.

On arm64 userspace addresses translation is controlled by TTBR0_EL1. As
well as userspace, TTBR0_EL1 controls:
 * The identity mapping,
 * EFI runtime code.

It is possible to run a kernel with an identity mapping that has a
larger VA size than userspace (and for this case __cpu_set_tcr_t0sz()
would set TCR_EL1.T0SZ as appropriate). However, when the conditions for
52-bit userspace are met; it is possible to keep TCR_EL1.T0SZ fixed at
12. Thus in this patch, the TCR_EL1.T0SZ size changing logic is
disabled.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For cases where there is a mismatch in ARMv8.2-LVA support between CPUs
we have to be careful in allowing secondary CPUs to boot if 52-bit
virtual addresses have already been enabled on the boot CPU.

This patch adds code to the secondary startup path. If the boot CPU has
enabled 52-bit VAs then ID_AA64MMFR2_EL1 is checked to see if the
secondary can also enable 52-bit support. If not, the secondary is
prevented from booting and an error message is displayed indicating why.

Technically this patch could be implemented using the cpufeature code
when considering 52-bit userspace support. However, we employ low level
checks here as the cpufeature code won't be able to run if we have
mismatched 52-bit kernel va support.
Signed-off-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Enabling 52-bit VAs on arm64 requires that the PGD table expands from 64
entries (for the 48-bit case) to 1024 entries. This quantity,
PTRS_PER_PGD is used as follows to compute which PGD entry corresponds
to a given virtual address, addr:

pgd_index(addr) -> (addr >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)

Userspace addresses are prefixed by 0's, so for a 48-bit userspace
address, uva, the following is true:
(uva >> PGDIR_SHIFT) & (1024 - 1) == (uva >> PGDIR_SHIFT) & (64 - 1)

In other words, a 48-bit userspace address will have the same pgd_index
when using PTRS_PER_PGD = 64 and 1024.

Kernel addresses are prefixed by 1's so, given a 48-bit kernel address,
kva, we have the following inequality:
(kva >> PGDIR_SHIFT) & (1024 - 1) != (kva >> PGDIR_SHIFT) & (64 - 1)

In other words a 48-bit kernel virtual address will have a different
pgd_index when using PTRS_PER_PGD = 64 and 1024.

If, however, we note that:
kva = 0xFFFF << 48 + lower (where lower[63:48] == 0b)
and, PGDIR_SHIFT = 42 (as we are dealing with 64KB PAGE_SIZE)

We can consider:
(kva >> PGDIR_SHIFT) & (1024 - 1) - (kva >> PGDIR_SHIFT) & (64 - 1)
 = (0xFFFF << 6) & 0x3FF - (0xFFFF << 6) & 0x3F	// "lower" cancels out
 = 0x3C0

In other words, one can switch PTRS_PER_PGD to the 52-bit value globally
provided that they increment ttbr1_el1 by 0x3C0 * 8 = 0x1E00 bytes when
running with 48-bit kernel VAs (TCR_EL1.T1SZ = 16).

For kernel configuration where 52-bit userspace VAs are possible, this
patch offsets ttbr1_el1 and sets PTRS_PER_PGD corresponding to the
52-bit value.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Suggested-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Steve Capper <steve.capper@arm.com>
[will: added comment to TTBR1_BADDR_4852_OFFSET calculation]
Signed-off-by: Will Deacon <will.deacon@arm.com>

Now that we have DEFAULT_MAP_WINDOW defined, we can arch_get_mmap_end
and arch_get_mmap_base helpers to allow for high addresses in mmap.
Signed-off-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

We wish to introduce a 52-bit virtual address space for userspace but
maintain compatibility with software that assumes the maximum VA space
size is 48 bit.

In order to achieve this, on 52-bit VA systems, we make mmap behave as
if it were running on a 48-bit VA system (unless userspace explicitly
requests a VA where addr[51:48] != 0).

On a system running a 52-bit userspace we need TASK_SIZE to represent
the 52-bit limit as it is used in various places to distinguish between
kernelspace and userspace addresses.

Thus we need a new limit for mmap, stack, ELF loader and EFI (which uses
TTBR0) to represent the non-extended VA space.

This patch introduces DEFAULT_MAP_WINDOW and DEFAULT_MAP_WINDOW_64 and
switches the appropriate logic to use that instead of TASK_SIZE.
Signed-off-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

This patch adds support for "high" userspace addresses that are
optionally supported on the system and have to be requested via a hint
mechanism ("high" addr parameter to mmap).

Architectures such as powerpc and x86 achieve this by making changes to
their architectural versions of arch_get_unmapped_* functions. However,
on arm64 we use the generic versions of these functions.

Rather than duplicate the generic arch_get_unmapped_* implementations
for arm64, this patch instead introduces two architectural helper macros
and applies them to arch_get_unmapped_*:
 arch_get_mmap_end(addr) - get mmap upper limit depending on addr hint
 arch_get_mmap_base(addr, base) - get mmap_base depending on addr hint

If these macros are not defined in architectural code then they default
to (TASK_SIZE) and (base) so should not introduce any behavioural
changes to architectures that do not define them.
Signed-off-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>

If the kernel is configured with KASAN_EXTRA, the stack size is
increased significantly due to setting the GCC -fstack-reuse option to
"none" [1]. As a result, it can trigger a stack overrun quite often with
32k stack size compiled using GCC 8. For example, this reproducer

  https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/syscalls/madvise/madvise06.c

can trigger a "corrupted stack end detected inside scheduler" very
reliably with CONFIG_SCHED_STACK_END_CHECK enabled. There are other
reports at:

  https://lore.kernel.org/lkml/1542144497.12945.29.camel@gmx.us/
  https://lore.kernel.org/lkml/721E7B42-2D55-4866-9C1A-3E8D64F33F9C@gmx.us/

There are just too many functions that could have a large stack with
KASAN_EXTRA due to large local variables that have been called over and
over again without being able to reuse the stacks. Some noticiable ones
are,

size
7536 shrink_inactive_list
7440 shrink_page_list
6560 fscache_stats_show
3920 jbd2_journal_commit_transaction
3216 try_to_unmap_one
3072 migrate_page_move_mapping
3584 migrate_misplaced_transhuge_page
3920 ip_vs_lblcr_schedule
4304 lpfc_nvme_info_show
3888 lpfc_debugfs_nvmestat_data.constprop

There are other 49 functions over 2k in size while compiling kernel with
"-Wframe-larger-than=" on this machine. Hence, it is too much work to
change Makefiles for each object to compile without
-fsanitize-address-use-after-scope individually.

[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81715#c23Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The dcache_by_line_op macro suffers from a couple of small problems:

First, the GAS directives that are currently being used rely on
assembler behavior that is not documented, and probably not guaranteed
to produce the correct behavior going forward. As a result, we end up
with some undefined symbols in cache.o:

$ nm arch/arm64/mm/cache.o
         ...
         U civac
         ...
         U cvac
         U cvap
         U cvau

This is due to the fact that the comparisons used to select the
operation type in the dcache_by_line_op macro are comparing symbols
not strings, and even though it seems that GAS is doing the right
thing here (undefined symbols by the same name are equal to each
other), it seems unwise to rely on this.

Second, when patching in a DC CVAP instruction on CPUs that support it,
the fallback path consists of a DC CVAU instruction which may be
affected by CPU errata that require ARM64_WORKAROUND_CLEAN_CACHE.

Solve these issues by unrolling the various maintenance routines and
using the conditional directives that are documented as operating on
strings. To avoid the complexity of nested alternatives, we move the
DC CVAP patching to __clean_dcache_area_pop, falling back to a branch
to __clean_dcache_area_poc if DCPOP is not supported by the CPU.
Reported-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Now that the infrastructure to handle erratum 1165522 is in place,
let's make it a selectable option and add the required documentation.
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

In order to avoid TLB corruption whilst invalidating TLBs on CPUs
affected by erratum 1165522, we need to prevent S1 page tables
from being usable.

For this, we set the EL1 S1 MMU on, and also disable the page table
walker (by setting the TCR_EL1.EPD* bits to 1).

This ensures that once we switch to the EL1/EL0 translation regime,
speculated AT instructions won't be able to parse the page tables.
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

In order to ensure that slipping HCR_EL2.TGE is done at the right
time when switching translation regime, let insert the required ISBs
that will be patched in when erratum 1165522 is detected.

Take this opportunity to add the missing include of asm/alternative.h
which was getting there by pure luck.
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

In order to easily mitigate ARM erratum 1165522, we need to force
affected CPUs to run in VHE mode if using KVM.
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

We are soon going to play with TCR_EL1.EPD{0,1}, so let's add the
relevant definitions.
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

It is a bit odd that we only install stage-2 translation after having
cleared HCR_EL2.TGE, which means that there is a window during which
AT requests could fail as stage-2 is not configured yet.

Let's move stage-2 configuration before we clear TGE, making the
guest entry sequence clearer: we first configure all the guest stuff,
then only switch to the guest translation regime.

While we're at it, do the same thing for !VHE. It doesn't hurt,
and keeps things symmetric.
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

An SVE system is so far the only case where we mandate VHE. As we're
starting to grow this requirements, let's slightly rework the way we
deal with that situation, allowing for easy extension of this check.
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Contrary to the non-VHE version of the TLB invalidation helpers, the VHE
code  has interrupts enabled, meaning that we can take an interrupt in
the middle of such a sequence, and start running something else with
HCR_EL2.TGE cleared.

That's really not a good idea.

Take the heavy-handed option and disable interrupts in
__tlb_switch_to_guest_vhe, restoring them in __tlb_switch_to_host_vhe.
The latter also gain an ISB in order to make sure that TGE really has
taken effect.

Cc: stable@vger.kernel.org
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Now that arm64ksyms.c has been reduced to a stub, let's remove it
entirely. New exports should be associated with their function
definition.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the ftrace exports
to the assembly files the functions are defined in.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the string routine
exports to the assembly files the functions are defined in. Routines
which should only be exported for !KASAN builds are exported using the
EXPORT_SYMBOL_NOKASAN() helper.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the uaccess exports
to the assembly files the functions are defined in.  As we have to
include <asm/assembler.h>, the existing includes are fixed to follow the
usual ordering conventions.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the copy_page and
clear_page exports to the assembly files the functions are defined in.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the SMCCC exports to
the assembly file the functions are defined in.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

For a while now it's been possible to use EXPORT_SYMBOL() in assembly
files, which allows us to place exports immediately after assembly
functions, as we do for C functions.

As a step towards removing arm64ksyms.c, let's move the tishift exports
to the assembly file the functions are defined in.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

So that we can export symbols directly from assembly files, let's make
use of the generic <asm/export.h>. We have a few symbols that we'll want
to conditionally export for !KASAN kernel builds, so we add a helper for
that in <asm/assembler.h>.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Since we define memstart_addr in a C file, we can have the export
immediately after the definition of the symbol, as we do elsewhere.

As a step towards removing arm64ksyms.c, move the export of
memstart_addr to init.c, where the symbol is defined.

There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Now that the arm64 bitops are inlines built atop of the regular atomics,
we don't need to export anything.

Remove the redundant exports.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

The "L" AArch64 machine constraint, which we use for the "old" value in
an LL/SC cmpxchg(), generates an immediate that is suitable for a 64-bit
logical instruction. However, for cmpxchg() operations on types smaller
than 64 bits, this constraint can result in an invalid instruction which
is correctly rejected by GAS, such as EOR W1, W1, #0xffffffff.

Whilst we could special-case the constraint based on the cmpxchg size,
it's far easier to change the constraint to "K" and put up with using
a register for large 64-bit immediates. For out-of-line LL/SC atomics,
this is all moot anyway.
Reported-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

Our percpu code is a bit of an inconsistent mess:

  * It rolls its own xchg(), but reuses cmpxchg_local()
  * It uses various different flavours of preempt_{enable,disable}()
  * It returns values even for the non-returning RmW operations
  * It makes no use of LSE atomics outside of the cmpxchg() ops
  * There are individual macros for different sizes of access, but these
    are all funneled through a switch statement rather than dispatched
    directly to the relevant case

This patch rewrites the per-cpu operations to address these shortcomings.
Whilst the new code is a lot cleaner, the big advantage is that we can
use the non-returning ST- atomic instructions when we have LSE.
Signed-off-by: Will Deacon <will.deacon@arm.com>

The CAS instructions implicitly access only the relevant bits of the "old"
argument, so there is no need for explicit masking via type-casting as
there is in the LL/SC implementation.

Move the casting into the LL/SC code and remove it altogether for the LSE
implementation.
Signed-off-by: Will Deacon <will.deacon@arm.com>