Clear the on-stack STACK_FRAME_REGS_MARKER on exception exit in order
to avoid confusing stacktrace like the one below.

  Call Trace:
  [c0e9dca0] [c01c42a0] print_address_description+0x64/0x2bc (unreliable)
  [c0e9dcd0] [c01c4684] kasan_report+0xfc/0x180
  [c0e9dd10] [c0895130] memchr+0x24/0x74
  [c0e9dd30] [c00a9e38] msg_print_text+0x124/0x574
  [c0e9dde0] [c00ab710] console_unlock+0x114/0x4f8
  [c0e9de40] [c00adc60] vprintk_emit+0x188/0x1c4
  --- interrupt: c0e9df00 at 0x400f330
      LR = init_stack+0x1f00/0x2000
  [c0e9de80] [c00ae3c4] printk+0xa8/0xcc (unreliable)
  [c0e9df20] [c0c27e44] early_irq_init+0x38/0x108
  [c0e9df50] [c0c15434] start_kernel+0x310/0x488
  [c0e9dff0] [00003484] 0x3484

With this patch the trace becomes:

  Call Trace:
  [c0e9dca0] [c01c42c0] print_address_description+0x64/0x2bc (unreliable)
  [c0e9dcd0] [c01c46a4] kasan_report+0xfc/0x180
  [c0e9dd10] [c0895150] memchr+0x24/0x74
  [c0e9dd30] [c00a9e58] msg_print_text+0x124/0x574
  [c0e9dde0] [c00ab730] console_unlock+0x114/0x4f8
  [c0e9de40] [c00adc80] vprintk_emit+0x188/0x1c4
  [c0e9de80] [c00ae3e4] printk+0xa8/0xcc
  [c0e9df20] [c0c27e44] early_irq_init+0x38/0x108
  [c0e9df50] [c0c15434] start_kernel+0x310/0x488
  [c0e9dff0] [00003484] 0x3484

Cc: stable@vger.kernel.org
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

As tglx points out, there are no in-tree module users of
save_stack_trace_tsk_reliable() and its x86 counterpart is not
exported, so remove the powerpc symbol export.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joe Lawrence <joe.lawrence@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The commit 24b6d416 ("mm: pass the vmem_altmap to vmemmap_free")
removed a line in vmemmap_free(),

altmap = to_vmem_altmap((unsigned long) section_base);

but left a variable no longer used.

arch/powerpc/mm/init_64.c: In function 'vmemmap_free':
arch/powerpc/mm/init_64.c:277:16: error: variable 'section_base' set but
not used [-Werror=unused-but-set-variable]
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Fix compiler warning:
  arch/powerpc/mm/hugetlbpage-hash64.c: In function '__hash_page_huge':
  arch/powerpc/mm/hugetlbpage-hash64.c:29:28: warning: variable 'sz' set
  but not used [-Wunused-but-set-variable]

mpe: The last usage of sz was removed in 0895ecda ("powerpc/mm:
Bring hugepage PTE accessor functions back into sync with normal
accessors").
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

We were always calling base_hpte_find() with primary = true,
even when we wanted to check the secondary table.

mpe: I broke this when refactoring Rashmica's original patch.

Fixes: 1515ab93 ("powerpc/mm: Dump hash table")
Signed-off-by: Rashmica Gupta <rashmica.g@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The __SYSCALL macro's arguments are system call number,
system call entry name and number of arguments for the
system call.

Argument- nargs in __SYSCALL(nr, entry, nargs) is neither
calculated nor used anywhere. So it would be better to
keep the implementaion as  __SYSCALL(nr, entry). This will
unifies the implementation with some other architetures
too.
Signed-off-by: Firoz Khan <firoz.khan@linaro.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Merge another commit in the topic/ppc-kvm branch we're sharing with
kvm-ppc.

The recent commit got this test wrong, it declared the assembler
symbols the wrong way, and also used the wrong symbol name
(xxx_start rather than start_xxx, see asm/head-64.h).

Fixes: ccd47702 ("powerpc/64s: Fix HV NMI vs HV interrupt recoverability test")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

We store 2 multilevel tables in iommu_table - one for the hardware and
one with the corresponding userspace addresses. Before allocating
the tables, the iommu_table_group_ops::get_table_size() hook returns
the combined size of the two and VFIO SPAPR TCE IOMMU driver adjusts
the locked_vm counter correctly. When the table is actually allocated,
the amount of allocated memory is stored in iommu_table::it_allocated_size
and used to decrement the locked_vm counter when we release the memory
used by the table; .get_table_size() and .create_table() calculate it
independently but the result is expected to be the same.

However the allocator does not add the userspace table size to
.it_allocated_size so when we destroy the table because of VFIO PCI
unplug (i.e. VFIO container is gone but the userspace keeps running),
we decrement locked_vm by just a half of size of memory we are
releasing.

To make things worse, since we enabled on-demand allocation of
indirect levels, it_allocated_size contains only the amount of memory
actually allocated at the table creation time which can just be a
fraction. It is not a problem with incrementing locked_vm (as
get_table_size() value is used) but it is with decrementing.

As the result, we leak locked_vm and may not be able to allocate more
IOMMU tables after few iterations of hotplug/unplug.

This sets it_allocated_size in the pnv_pci_ioda2_ops::create_table()
hook to what pnv_pci_ioda2_get_table_size() returns so from now on we
have a single place which calculates the maximum memory a table can
occupy. The original meaning of it_allocated_size is somewhat lost now
though.

We do not ditch it_allocated_size whatsoever here and we do not call
get_table_size() from vfio_iommu_spapr_tce.c when decrementing
locked_vm as we may have multiple IOMMU groups per container and even
though they all are supposed to have the same get_table_size()
implementation, there is a small chance for failure or confusion.

Fixes: 090bad39 ("powerpc/powernv: Add indirect levels to it_userspace")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The commit identified below adds MC_BTB_FLUSH macro only when
CONFIG_PPC_FSL_BOOK3E is defined. This results in the following error
on some configs (seen several times with kisskb randconfig_defconfig)

arch/powerpc/kernel/exceptions-64e.S:576: Error: Unrecognized opcode: `mc_btb_flush'
make[3]: *** [scripts/Makefile.build:367: arch/powerpc/kernel/exceptions-64e.o] Error 1
make[2]: *** [scripts/Makefile.build:492: arch/powerpc/kernel] Error 2
make[1]: *** [Makefile:1043: arch/powerpc] Error 2
make: *** [Makefile:152: sub-make] Error 2

This patch adds a blank definition of MC_BTB_FLUSH for other cases.

Fixes: 10c5e83a ("powerpc/fsl: Flush the branch predictor at each kernel entry (64bit)")
Cc: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Currently the opal log is globally readable. It is kernel policy to
limit the visibility of physical addresses / kernel pointers to root.
Given this and the fact the opal log may contain this information it
would be better to limit the readability to root.

Fixes: bfc36894 ("powerpc/powernv: Add OPAL message log interface")
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Reviewed-by: Stewart Smith <stewart@linux.ibm.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When building with -Wsometimes-uninitialized, Clang warns:

  arch/powerpc/xmon/ppc-dis.c:157:7: warning: variable 'opcode' is used
  uninitialized whenever 'if' condition is false
  [-Wsometimes-uninitialized]
    if (cpu_has_feature(CPU_FTRS_POWER9))
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  arch/powerpc/xmon/ppc-dis.c:167:7: note: uninitialized use occurs here
    if (opcode == NULL)
        ^~~~~~
  arch/powerpc/xmon/ppc-dis.c:157:3: note: remove the 'if' if its
  condition is always true
    if (cpu_has_feature(CPU_FTRS_POWER9))
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  arch/powerpc/xmon/ppc-dis.c:132:38: note: initialize the variable
  'opcode' to silence this warning
    const struct powerpc_opcode *opcode;
                                       ^
                                        = NULL
  1 warning generated.

This warning seems to make no sense on the surface because opcode is set
to NULL right below this statement. However, there is a comma instead of
semicolon to end the dialect assignment, meaning that the opcode
assignment only happens in the if statement. Properly terminate that
line so that Clang no longer warns.

Fixes: 5b102782 ("powerpc/xmon: Enable disassembly files (compilation changes)")
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The OPAL call wrapper gets interrupt disabling wrong. It disables
interrupts just by clearing MSR[EE], which has two problems:

- It doesn't call into the IRQ tracing subsystem, which means tracing
  across OPAL calls does not always notice IRQs have been disabled.

- It doesn't go through the IRQ soft-mask code, which causes a minor
  bug. MSR[EE] can not be restored by saving the MSR then clearing
  MSR[EE], because a racing interrupt while soft-masked could clear
  MSR[EE] between the two steps. This can cause MSR[EE] to be
  incorrectly enabled when the OPAL call returns. Fortunately that
  should only result in another masked interrupt being taken to
  disable MSR[EE] again, but it's a bit sloppy.

The existing code also saves MSR to PACA, which is not re-entrant if
there is a nested OPAL call from different MSR contexts, which can
happen these days with SRESET interrupts on bare metal.

To fix these issues, move the tracing and IRQ handling code to C, and
call into asm just for the low level call when everything is ready to
go. Save the MSR on stack rather than PACA.

Performance cost is kept to a minimum with a few optimisations:

- The endian switch upon return is combined with the MSR restore,
  which avoids an expensive context synchronizing operation for LE
  kernels. This makes up for the additional mtmsrd to enable
  interrupts with local_irq_enable().

- blr is now used to return from the opal_* functions that are called
  as C functions, to avoid link stack corruption. This requires a
  skiboot fix as well to keep the call stack balanced.

A NULL call is more costly after this, (410ns->430ns on POWER9), but
OPAL calls are generally not performance critical at this scale.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Handlers for interrupts that set DAR / DSISR, set MSR[RI] before those
SPRs are read. If a d-side machine check hits in this window, DAR /
DSISR will be clobbered silently, leading to random corruption.

Fix this by having handlers save those registers before setting MSR[RI].
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

A subsequent fix for data interrupts (those that set DAR / DSISR)
requires some interrupt macros to be open-coded, and also requires
the 0x300 interrupt handler to be moved out-of-line.

This patch does that without changing behaviour, which makes the later
fix a smaller change.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Code that uses HSRR registers is not required to clear MSR[RI] by
convention, however the system reset NMI itself may use HSRR
registers (e.g., to call OPAL) and clobber them.

Rather than introduce the requirement to clear RI in order to use
HSRRs, have system reset interrupt save and restore HSRRs.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

HV interrupts that use HSRR registers do not enter with MSR[RI] clear,
but their entry code is not recoverable vs NMI, due to shared use of
HSPRG1 as a scratch register to save r13.

This means that a system reset or machine check that hits in HSRR
interrupt entry can cause r13 to be silently corrupted.

Fix this by marking NMIs non-recoverable if they land in HV interrupt
ranges.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When doing top-down search the low_limit is not PAGE_SIZE but rather
max(PAGE_SIZE, mmap_min_addr). This handle cases in which mmap_min_addr >
PAGE_SIZE.

Fixes: fba2369e ("mm: use vm_unmapped_area() on powerpc architecture")
Reviewed-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

After we ALIGN up the address we need to make sure we didn't overflow
and resulted in zero address. In that case, we need to make sure that
the returned address is greater than mmap_min_addr.

This fixes selftest va_128TBswitch --run-hugetlb reporting failures when
run as non root user for

mmap(-1, MAP_HUGETLB)

The bug is that a non-root user requesting address -1 will be given address 0
which will then fail, whereas they should have been given something else that
would have succeeded.

We also avoid the first mmap(-1, MAP_HUGETLB) returning NULL address as mmap address
with this change. So we think this is not a security issue, because it only affects
whether we choose an address below mmap_min_addr, not whether we
actually allow that address to be mapped. ie. there are existing capability
checks to prevent a user mapping below mmap_min_addr and those will still be
honoured even without this fix.

Fixes: 48483760 ("powerpc/mm: Add radix support for hugetlb")
Reviewed-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Remove duplicate headers which are included twice.
Signed-off-by: Sabyasachi Gupta <sabyasachi.linux@gmail.com>
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Modulo Signed Doubleword (modsd)
  * Modulo Unsigned Doubleword (modud)
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Modulo Signed Word (modsw)
  * Modulo Unsigned Word (moduw)
Signed-off-by: PrasannaKumar Muralidharan <prasannatsmkumar@gmail.com>
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Extend-Sign Word and Shift Left Immediate (extswsli[.])
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Count Trailing Zeros Word (cnttzw[.])
  * Count Trailing Zeros Doubleword (cnttzd[.])
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Deliver A Random Number (darn)

As suggested by Michael, this uses a raw .long for specifying the
instruction word when using inline assembly to retain compatibility
with older binutils.
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This adds emulation support for the following integer instructions:
  * Multiply-Add High Doubleword (maddhd)
  * Multiply-Add High Doubleword Unsigned (maddhdu)
  * Multiply-Add Low Doubleword (maddld)

As suggested by Michael, this uses a raw .long for specifying the
instruction word when using inline assembly to retain compatibility
with older binutils.
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Some stack pointers used to also be thread_info pointers
and were called tp. Now that they are only stack pointers,
rename them sp.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Now that current_thread_info is located at the beginning of 'current'
task struct, CURRENT_THREAD_INFO macro is not really needed any more.

This patch replaces it by loads of the value at PACA_THREAD_INFO(r13).
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Add PACA_THREAD_INFO rather than using PACACURRENT]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Now that thread_info is similar to task_struct, its address is in r2
so CURRENT_THREAD_INFO() macro is useless. This patch removes it.

This patch also moves the 'tovirt(r2, r2)' down just before the
reactivation of MMU translation, so that we keep the physical address
of 'current' in r2 until then. It avoids a few calls to tophys().

At the same time, as the 'cpu' field is not anymore in thread_info,
TI_CPU is renamed TASK_CPU by this patch.

It also allows to get rid of a couple of
'#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE' as ACCOUNT_CPU_USER_ENTRY()
and ACCOUNT_CPU_USER_EXIT() are empty when
CONFIG_VIRT_CPU_ACCOUNTING_NATIVE is not defined.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Fix a missed conversion of TI_CPU idle_6xx.S]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

The table of pointers 'current_set' has been used for retrieving
the stack and current. They used to be thread_info pointers as
they were pointing to the stack and current was taken from the
'task' field of the thread_info.

Now, the pointers of 'current_set' table are now both pointers
to task_struct and pointers to thread_info.

As they are used to get current, and the stack pointer is
retrieved from current's stack field, this patch changes
their type to task_struct, and renames secondary_ti to
secondary_current.
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

thread_info is not anymore in the stack, so the entire stack
can now be used.

There is also no risk anymore of corrupting task_cpu(p) with a
stack overflow so the patch removes the test.

When doing this, an explicit test for NULL stack pointer is
needed in validate_sp() as it is not anymore implicitely covered
by the sizeof(thread_info) gap.

In the meantime, with the previous patch all pointers to the stacks
are not anymore pointers to thread_info so this patch changes them
to void*
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

This patch activates CONFIG_THREAD_INFO_IN_TASK which
moves the thread_info into task_struct.

Moving thread_info into task_struct has the following advantages:
  - It protects thread_info from corruption in the case of stack
    overflows.
  - Its address is harder to determine if stack addresses are leaked,
    making a number of attacks more difficult.

This has the following consequences:
  - thread_info is now located at the beginning of task_struct.
  - The 'cpu' field is now in task_struct, and only exists when
    CONFIG_SMP is active.
  - thread_info doesn't have anymore the 'task' field.

This patch:
  - Removes all recopy of thread_info struct when the stack changes.
  - Changes the CURRENT_THREAD_INFO() macro to point to current.
  - Selects CONFIG_THREAD_INFO_IN_TASK.
  - Modifies raw_smp_processor_id() to get ->cpu from current without
    including linux/sched.h to avoid circular inclusion and without
    including asm/asm-offsets.h to avoid symbol names duplication
    between ASM constants and C constants.
  - Modifies klp_init_thread_info() to take a task_struct pointer
    argument.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Add task_stack.h to livepatch.h to fix build fails]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Make sure CURRENT_THREAD_INFO() is used with r1 which is the virtual
address of the stack, in order to ease the switch to r2 when we enable
THREAD_INFO_IN_TASK, as we have no register having the phys address of
current.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Change current_pt_regs() to use task_stack_page() rather than
current_thread_info() so that it keeps working once we enable
THREAD_INFO_IN_TASK.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Split out of large patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

When we enable THREAD_INFO_IN_TASK we will remove our definition of
current_thread_info(). Instead it will come from linux/thread_info.h

So switch processor.h to include the latter, so that it can continue
to find current_thread_info().
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Currently INIT_SP_LIMIT uses sizeof(init_thread_info), but that symbol
won't exist when we enable THREAD_INFO_IN_TASK. So just use the sizeof
the type which is the same value but will continue to work.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Rather than using the thread info use task_stack_page() to initialise
paca->kstack, that way it will work with THREAD_INFO_IN_TASK.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Update a few comments that talk about current_thread_info() in
preparation for THREAD_INFO_IN_TASK.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

We have a few places that use current_thread_info()->task to access
current. This won't work with THREAD_INFO_IN_TASK so fix them now.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

A few places use CURRENT_THREAD_INFO, or the C version, to find the
stack. This will no longer work with THREAD_INFO_IN_TASK so change
them to find the stack in other ways.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>