linux-2.6-block.git
5 years agopowerpc/mm/hash: hard disable irq in the SLB insert path
Aneesh Kumar K.V [Fri, 1 Jun 2018 08:24:02 +0000 (13:54 +0530)]
powerpc/mm/hash: hard disable irq in the SLB insert path

When inserting SLB entries for EA above 512TB, we need to hard disable irq.
This will make sure we don't take a PMU interrupt that can possibly touch
user space address via a stack dump. To prevent this, we need to hard disable
the interrupt.

Also add a comment explaining why we don't need context synchronizing isync
with slbmte.

Fixes: f384796c4 ("powerpc/mm: Add support for handling > 512TB address in SLB miss")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm/hugetlb: Update hugetlb related locks
Aneesh Kumar K.V [Fri, 1 Jun 2018 08:24:24 +0000 (13:54 +0530)]
powerpc/mm/hugetlb: Update hugetlb related locks

With split pmd page table lock enabled, we don't use mm->page_table_lock when
updating pmd entries. This patch update hugetlb path to use the right lock
when inserting huge page directory entries into page table.

ex: if we are using hugepd and inserting hugepd entry at the pmd level, we
use pmd_lockptr, which based on config can be split pmd lock.

For update huge page directory entries itself we use mm->page_table_lock. We
do have a helper huge_pte_lockptr() for that.

Fixes: 675d99529 ("powerpc/book3s64: Enable split pmd ptlock")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm/hash: Add missing isync prior to kernel stack SLB switch
Aneesh Kumar K.V [Wed, 30 May 2018 13:18:04 +0000 (18:48 +0530)]
powerpc/mm/hash: Add missing isync prior to kernel stack SLB switch

Currently we do not have an isync, or any other context synchronizing
instruction prior to the slbie/slbmte in _switch() that updates the
SLB entry for the kernel stack.

However that is not correct as outlined in the ISA.

From Power ISA Version 3.0B, Book III, Chapter 11, page 1133:

  "Changing the contents of ... the contents of SLB entries ... can
   have the side effect of altering the context in which data
   addresses and instruction addresses are interpreted, and in which
   instructions are executed and data accesses are performed.
   ...
   These side effects need not occur in program order, and therefore
   may require explicit synchronization by software.
   ...
   The synchronizing instruction before the context-altering
   instruction ensures that all instructions up to and including that
   synchronizing instruction are fetched and executed in the context
   that existed before the alteration."

And page 1136:

  "For data accesses, the context synchronizing instruction before the
   slbie, slbieg, slbia, slbmte, tlbie, or tlbiel instruction ensures
   that all preceding instructions that access data storage have
   completed to a point at which they have reported all exceptions
   they will cause."

We're not aware of any bugs caused by this, but it should be fixed
regardless.

Add the missing isync when updating kernel stack SLB entry.

Cc: stable@vger.kernel.org
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
[mpe: Flesh out change log with more ISA text & explanation]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s: Fix compiler store ordering to SLB shadow area
Nicholas Piggin [Wed, 30 May 2018 10:31:22 +0000 (20:31 +1000)]
powerpc/64s: Fix compiler store ordering to SLB shadow area

The stores to update the SLB shadow area must be made as they appear
in the C code, so that the hypervisor does not see an entry with
mismatched vsid and esid. Use WRITE_ONCE for this.

GCC has been observed to elide the first store to esid in the update,
which means that if the hypervisor interrupts the guest after storing
to vsid, it could see an entry with old esid and new vsid, which may
possibly result in memory corruption.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: flush remote CPUs out of single-threaded mm_cpumask
Nicholas Piggin [Fri, 1 Jun 2018 10:01:21 +0000 (20:01 +1000)]
powerpc/64s/radix: flush remote CPUs out of single-threaded mm_cpumask

When a single-threaded process has a non-local mm_cpumask, try to use
that point to flush the TLBs out of other CPUs in the cpumask.

An IPI is used for clearing remote CPUs for a few reasons:
- An IPI can end lazy TLB use of the mm, which is required to prevent
  TLB entries being created on the remote CPU. The alternative is to
  drop lazy TLB switching completely, which costs 7.5% in a context
  switch ping-pong test betwee a process and kernel idle thread.
- An IPI can have remote CPUs flush the entire PID, but the local CPU
  can flush a specific VA. tlbie would require over-flushing of the
  local CPU (where the process is running).
- A single threaded process that is migrated to a different CPU is
  likely to have a relatively small mm_cpumask, so IPI is reasonable.

No other thread can concurrently switch to this mm, because it must
have been given a reference to mm_users by the current thread before it
can use_mm. mm_users can be asynchronously incremented (by
mm_activate or mmget_not_zero), but those users must use remote mm
access and can't use_mm or access user address space. Existing code
makes the this assumption already, for example sparc64 has reset
mm_cpumask using this condition since the start of history, see
arch/sparc/kernel/smp_64.c.

This reduces tlbies for a kernel compile workload from 0.90M to 0.12M,
tlbiels are increased significantly due to the PID flushing for the
cleaning up remote CPUs, and increased local flushes (PID flushes take
128 tlbiels vs 1 tlbie).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: optimise pte_update
Nicholas Piggin [Fri, 1 Jun 2018 10:01:20 +0000 (20:01 +1000)]
powerpc/64s/radix: optimise pte_update

Implementing pte_update with pte_xchg (which uses cmpxchg) is
inefficient. A single larx/stcx. works fine, no need for the less
efficient cmpxchg sequence.

Then remove the memory barriers from the operation. There is a
requirement for TLB flushing to load mm_cpumask after the store
that reduces pte permissions, which is moved into the TLB flush
code.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: avoid ptesync after set_pte and ptep_set_access_flags
Nicholas Piggin [Fri, 1 Jun 2018 10:01:19 +0000 (20:01 +1000)]
powerpc/64s/radix: avoid ptesync after set_pte and ptep_set_access_flags

The ISA suggests ptesync after setting a pte, to prevent a table walk
initiated by a subsequent access from missing that store and causing a
spurious fault. This is an architectual allowance that allows an
implementation's page table walker to be incoherent with the store
queue.

However there is no correctness problem in taking a spurious fault in
userspace -- the kernel copes with these at any time, so the updated
pte will be found eventually. Spurious kernel faults on vmap memory
must be avoided, so a ptesync is put into flush_cache_vmap.

On POWER9 so far I have not found a measurable window where this can
result in more minor faults, so as an optimisation, remove the costly
ptesync from pte updates. If an implementation benefits from ptesync,
it would be better to add it back in update_mmu_cache, so it's not
done for things like fork(2).

fork --fork --exec benchmark improved 5.2% (12400->13100).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: prefetch user address in update_mmu_cache
Nicholas Piggin [Fri, 1 Jun 2018 10:01:18 +0000 (20:01 +1000)]
powerpc/64s/radix: prefetch user address in update_mmu_cache

Prefetch the faulting address in update_mmu_cache to give the page
table walker perhaps 100 cycles head start as locks are dropped and
the interrupt completed.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: make ptep_get_and_clear_full non-atomic for the full case
Nicholas Piggin [Fri, 1 Jun 2018 10:01:17 +0000 (20:01 +1000)]
powerpc/64s/radix: make ptep_get_and_clear_full non-atomic for the full case

This matches other architectures, when we know there will be no
further accesses to the address (e.g., for teardown), page table
entries can be cleared non-atomically.

The comments about NMMU are bogus: all MMU notifiers (including NMMU)
are released at this point, with their TLBs flushed. An NMMU access at
this point would be a bug.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: do not flush TLB on spurious fault
Nicholas Piggin [Fri, 1 Jun 2018 10:01:16 +0000 (20:01 +1000)]
powerpc/64s/radix: do not flush TLB on spurious fault

In the case of a spurious fault (which can happen due to a race with
another thread that changes the page table), the default Linux mm code
calls flush_tlb_page for that address. This is not required because
the pte will be re-fetched. Hash does not wire this up to a hardware
TLB flush for this reason. This patch avoids the flush for radix.

>From Power ISA v3.0B, p.1090:

    Setting a Reference or Change Bit or Upgrading Access Authority
    (PTE Subject to Atomic Hardware Updates)

    If the only change being made to a valid PTE that is subject to
    atomic hardware updates is to set the Refer- ence or Change bit to
    1 or to add access authorities, a simpler sequence suffices
    because the translation hardware will refetch the PTE if an access
    is attempted for which the only problems were reference and/or
    change bits needing to be set or insufficient access authority.

The nest MMU on POWER9 does not re-fetch the PTE after such an access
attempt before faulting, so address spaces with a coprocessor
attached will continue to flush in these cases.

This reduces tlbies for a kernel compile workload from 0.95M to 0.90M.

fork --fork --exec benchmark improved 0.5% (12300->12400).

Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s/radix: do not flush TLB when relaxing access
Nicholas Piggin [Fri, 1 Jun 2018 10:01:15 +0000 (20:01 +1000)]
powerpc/64s/radix: do not flush TLB when relaxing access

Radix flushes the TLB when updating ptes to increase permissiveness
of protection (increase access authority). Book3S does not require
TLB flushing in this case, and it is not done on hash. This patch
avoids the flush for radix.

>From Power ISA v3.0B, p.1090:

    Setting a Reference or Change Bit or Upgrading Access Authority
    (PTE Subject to Atomic Hardware Updates)

    If the only change being made to a valid PTE that is subject to
    atomic hardware updates is to set the Reference or Change bit to 1
    or to add access authorities, a simpler sequence suffices because
    the translation hardware will refetch the PTE if an access is
    attempted for which the only problems were reference and/or change
    bits needing to be set or insufficient access authority.

The nest MMU on POWER9 does not re-fetch the PTE after such an access
attempt before faulting, so address spaces with a coprocessor
attached will continue to flush in these cases.

This reduces tlbies for a kernel compile workload from 1.28M to 0.95M,
tlbiels from 20.17M 19.68M.

fork --fork --exec benchmark improved 2.77% (12000->12300).

Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm/radix: Change pte relax sequence to handle nest MMU hang
Aneesh Kumar K.V [Tue, 29 May 2018 14:28:41 +0000 (19:58 +0530)]
powerpc/mm/radix: Change pte relax sequence to handle nest MMU hang

When relaxing access (read -> read_write update), pte needs to be marked invalid
to handle a nest MMU bug. We also need to do a tlb flush after the pte is
marked invalid before updating the pte with new access bits.

We also move tlb flush to platform specific __ptep_set_access_flags. This will
help us to gerid of unnecessary tlb flush on BOOK3S 64 later. We don't do that
in this patch. This also helps in avoiding multiple tlbies with coprocessor
attached.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm: Change function prototype
Aneesh Kumar K.V [Tue, 29 May 2018 14:28:40 +0000 (19:58 +0530)]
powerpc/mm: Change function prototype

In later patch, we use the vma and psize to do tlb flush. Do the prototype
update in separate patch to make the review easy.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm/radix: Move function from radix.h to pgtable-radix.c
Aneesh Kumar K.V [Tue, 29 May 2018 14:28:39 +0000 (19:58 +0530)]
powerpc/mm/radix: Move function from radix.h to pgtable-radix.c

In later patch we will update them which require them to be moved
to pgtable-radix.c. Keeping the function in radix.h results in
compile warning as below.

./arch/powerpc/include/asm/book3s/64/radix.h: In function ‘radix__ptep_set_access_flags’:
./arch/powerpc/include/asm/book3s/64/radix.h:196:28: error: dereferencing pointer to incomplete type ‘struct vm_area_struct’
  struct mm_struct *mm = vma->vm_mm;
                            ^~
./arch/powerpc/include/asm/book3s/64/radix.h:204:6: error: implicit declaration of function ‘atomic_read’; did you mean ‘__atomic_load’? [-Werror=implicit-function-declaration]
      atomic_read(&mm->context.copros) > 0) {
      ^~~~~~~~~~~
      __atomic_load
./arch/powerpc/include/asm/book3s/64/radix.h:204:21: error: dereferencing pointer to incomplete type ‘struct mm_struct’
      atomic_read(&mm->context.copros) > 0) {

Instead of fixing header dependencies, we move the function to pgtable-radix.c
Also the function is now large to be a static inline . Doing the
move in separate patch helps in review.

No functional change in this patch. Only code movement.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/mm/hugetlb: Update huge_ptep_set_access_flags to call __ptep_set_access_flags...
Aneesh Kumar K.V [Tue, 29 May 2018 14:28:38 +0000 (19:58 +0530)]
powerpc/mm/hugetlb: Update huge_ptep_set_access_flags to call __ptep_set_access_flags directly

In a later patch, we want to update __ptep_set_access_flags take page size
arg. This makes ptep_set_access_flags only work with mmu_virtual_psize.
To simplify the code make huge_ptep_set_access_flags directly call
__ptep_set_access_flags so that we can compute the hugetlb page size in
hugetlb function.

Now that ptep_set_access_flags won't be called for hugetlb remove
the is_vm_hugetlb_page() check and add the assert of pte lock
unconditionally.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoocxl: Document new OCXL IOCTLs
Alastair D'Silva [Fri, 11 May 2018 06:13:03 +0000 (16:13 +1000)]
ocxl: Document new OCXL IOCTLs

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoocxl: Add an IOCTL so userspace knows what OCXL features are available
Alastair D'Silva [Fri, 11 May 2018 06:13:02 +0000 (16:13 +1000)]
ocxl: Add an IOCTL so userspace knows what OCXL features are available

In order for a userspace AFU driver to call the POWER9 specific
OCXL_IOCTL_ENABLE_P9_WAIT, it needs to verify that it can actually
make that call.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoocxl: Expose the thread_id needed for wait on POWER9
Alastair D'Silva [Fri, 11 May 2018 06:13:01 +0000 (16:13 +1000)]
ocxl: Expose the thread_id needed for wait on POWER9

In order to successfully issue as_notify, an AFU needs to know the TID
to notify, which in turn means that this information should be
available in userspace so it can be communicated to the AFU.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoocxl: Rename pnv_ocxl_spa_remove_pe to clarify it's action
Alastair D'Silva [Fri, 11 May 2018 06:13:00 +0000 (16:13 +1000)]
ocxl: Rename pnv_ocxl_spa_remove_pe to clarify it's action

The function removes the process element from NPU cache.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: use task_pid_nr() for TID allocation
Alastair D'Silva [Fri, 11 May 2018 06:12:59 +0000 (16:12 +1000)]
powerpc: use task_pid_nr() for TID allocation

The current implementation of TID allocation, using a global IDR, may
result in an errant process starving the system of available TIDs.
Instead, use task_pid_nr(), as mentioned by the original author. The
scenario described which prevented it's use is not applicable, as
set_thread_tidr can only be called after the task struct has been
populated.

In the unlikely event that 2 threads share the TID and are waiting,
all potential outcomes have been determined safe.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Reviewed-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: Use TIDR CPU feature to control TIDR allocation
Alastair D'Silva [Fri, 11 May 2018 06:12:58 +0000 (16:12 +1000)]
powerpc: Use TIDR CPU feature to control TIDR allocation

Switch the use of TIDR on it's CPU feature, rather than assuming it
is available based on architecture.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Reviewed-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: Add TIDR CPU feature for POWER9
Alastair D'Silva [Fri, 11 May 2018 06:12:57 +0000 (16:12 +1000)]
powerpc: Add TIDR CPU feature for POWER9

This patch adds a CPU feature bit to show whether the CPU has
the TIDR register available, enabling as_notify/wait in userspace.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Reviewed-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/powernv: process all OPAL event interrupts with kopald
Nicholas Piggin [Thu, 10 May 2018 17:20:05 +0000 (03:20 +1000)]
powerpc/powernv: process all OPAL event interrupts with kopald

Using irq_work for processing OPAL event interrupts is not necessary.
irq_work is typically used to schedule work from NMI context, a
softirq may be more appropriate. However OPAL events are not
particularly performance or latency critical, so they can all be
invoked by kopald.

This patch removes the irq_work queueing, and instead wakes up
kopald when there is an event to be processed. kopald processes
interrupts individually, enabling irqs and calling cond_resched
between each one to minimise latencies.

Event handlers themselves should still use threaded handlers,
workqueues, etc. as necessary to avoid high interrupts-off latencies
within any single interrupt.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/powernv: call OPAL_QUIESCE before OPAL_SIGNAL_SYSTEM_RESET
Nicholas Piggin [Thu, 10 May 2018 12:21:48 +0000 (22:21 +1000)]
powerpc/powernv: call OPAL_QUIESCE before OPAL_SIGNAL_SYSTEM_RESET

Although it is often possible to recover a CPU that was interrupted
from OPAL with a system reset NMI, it's undesirable to interrupt them
for a few reasons. Firstly because dump/debug code itself needs to
call firmware, so it could hang on a lock or possibly corrupt a
per-cpu data structure if it or another CPU was interrupted from
OPAL. Secondly, the kexec crash dump code will not return from
interrupt to unwind the OPAL call.

Call OPAL_QUIESCE with QUIESCE_HOLD before sending an NMI IPI to
another CPU, which wait for it to leave firmware (or time out) to
avoid this problem in normal conditions. Firmware bugs may still
result in a timeout and interrupting OPAL, but that is the best
option (stops the CPU, and possibly allows firmware to be debugged).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64: change softe to irqmask in show_regs and xmon
Nicholas Piggin [Thu, 10 May 2018 01:04:24 +0000 (11:04 +1000)]
powerpc/64: change softe to irqmask in show_regs and xmon

When the soft enabled flag was changed to a soft disable mask, xmon
and register dump code was not updated to reflect that, which is
confusing ('SOFTE: 1' previously meant interrupts were soft enabled,
currently it means the opposite, the general interrupt type has been
disabled).

Fix this by using the name irqmask, and printing it in hex.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/pmu/fsl: fix is_nmi test for irq mask change
Nicholas Piggin [Thu, 10 May 2018 01:04:23 +0000 (11:04 +1000)]
powerpc/pmu/fsl: fix is_nmi test for irq mask change

When soft enabled was changed to irq disabled mask, this test missed
being converted (although the equivalent book3s test was converted).

The PMU drivers consider it an NMI when they take a PMI while general
interrupts are disabled. This change restores that behaviour.

Fixes: 01417c6cc7 ("powerpc/64: Change soft_enabled from flag to bitmask")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/time: account broadcast timer event interrupts separately
Nicholas Piggin [Fri, 4 May 2018 17:19:35 +0000 (03:19 +1000)]
powerpc/time: account broadcast timer event interrupts separately

These are not local timer interrupts but IPIs. It's good to be able
to see how timer offloading is behaving, so split these out into
their own category.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: move a stray NMI IPI case under NMI_IPI ifdef
Nicholas Piggin [Fri, 4 May 2018 17:19:34 +0000 (03:19 +1000)]
powerpc: move a stray NMI IPI case under NMI_IPI ifdef

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: move timer broadcast code under GENERIC_CLOCKEVENTS_BROADCAST ifdef
Nicholas Piggin [Fri, 4 May 2018 17:19:33 +0000 (03:19 +1000)]
powerpc: move timer broadcast code under GENERIC_CLOCKEVENTS_BROADCAST ifdef

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: allow soft-NMI watchdog to cover timer interrupts with large decrementers
Nicholas Piggin [Fri, 4 May 2018 17:19:32 +0000 (03:19 +1000)]
powerpc: allow soft-NMI watchdog to cover timer interrupts with large decrementers

Large decrementers (e.g., POWER9) can take a very long time to wrap,
so when the timer iterrupt handler sets the decrementer to max so as
to avoid taking another decrementer interrupt when hard enabling
interrupts before running timers, it effectively disables the soft
NMI coverage for timer interrupts.

Fix this by using the traditional 31-bit value instead, which wraps
after a few seconds. masked interrupt code does the same thing, and
in normal operation neither of these paths would ever wrap even the
31 bit value.

Note: the SMP watchdog should catch timer interrupt lockups, but it
is preferable for the local soft-NMI to catch them, mainly to avoid
the IPI.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: generic clockevents broadcast receiver call tick_receive_broadcast
Nicholas Piggin [Fri, 4 May 2018 17:19:31 +0000 (03:19 +1000)]
powerpc: generic clockevents broadcast receiver call tick_receive_broadcast

The broadcast tick recipient can call tick_receive_broadcast rather
than re-running the full timer interrupt.

It does not have to check for the next event time, because the sender
already determined the timer has expired. It does not have to test
irq_work_pending, because that's a direct decrementer interrupt and
does not go through the clock events subsystem. And it does not have
to read PURR because that was removed with the previous patch.

This results in no code size change, but both the decrementer and
broadcast path lengths are reduced.

Cc: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/pseries: lparcfg calculate PURR on demand
Nicholas Piggin [Fri, 4 May 2018 17:19:30 +0000 (03:19 +1000)]
powerpc/pseries: lparcfg calculate PURR on demand

For SPLPAR, lparcfg provides a sum of PURR registers for all CPUs.
Currently this is done by reading PURR in context switch and timer
interrupt, and storing that into a per-CPU variable. These are summed
to provide the value.

This does not work with all timer schemes (e.g., NO_HZ_FULL), and it
is sub-optimal for performance because it reads the PURR register on
every context switch, although that's been difficult to distinguish
from noise in the contxt_switch microbenchmark.

This patch implements the sum by calling a function on each CPU, to
read and add PURR values of each CPU.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64: remove start_tb and accum_tb from thread_struct
Nicholas Piggin [Fri, 4 May 2018 17:19:29 +0000 (03:19 +1000)]
powerpc/64: remove start_tb and accum_tb from thread_struct

These fields are only written to.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64s: micro-optimise __hard_irq_enable() for mtmsrd L=1 support
Nicholas Piggin [Fri, 4 May 2018 17:19:28 +0000 (03:19 +1000)]
powerpc/64s: micro-optimise __hard_irq_enable() for mtmsrd L=1 support

Book3S minimum supported ISA version now requires mtmsrd L=1. This
instruction does not require bits other than RI and EE to be supplied,
so __hard_irq_enable() and __hard_irq_disable() does not have to read
the kernel_msr from paca.

Interrupt entry code already relies on L=1 support.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/pseries: put cede MSR[EE] check under IRQ_SOFT_MASK_DEBUG
Nicholas Piggin [Fri, 4 May 2018 17:19:26 +0000 (03:19 +1000)]
powerpc/pseries: put cede MSR[EE] check under IRQ_SOFT_MASK_DEBUG

This check does not catch IRQ soft mask bugs, but this option is
slightly more suitable than TRACE_IRQFLAGS.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64: irq_work avoid interrupt when called with hardware irqs enabled
Nicholas Piggin [Fri, 4 May 2018 17:19:25 +0000 (03:19 +1000)]
powerpc/64: irq_work avoid interrupt when called with hardware irqs enabled

irq_work_raise should not cause a decrementer exception unless it is
called from NMI context. Doing so often just results in an immediate
masked decrementer interrupt:

   <...>-550    90d...    4us : update_curr_rt <-dequeue_task_rt
   <...>-550    90d...    5us : dbs_update_util_handler <-update_curr_rt
   <...>-550    90d...    6us : arch_irq_work_raise <-irq_work_queue
   <...>-550    90d...    7us : soft_nmi_interrupt <-soft_nmi_common
   <...>-550    90d...    7us : printk_nmi_enter <-soft_nmi_interrupt
   <...>-550    90d.Z.    8us : rcu_nmi_enter <-soft_nmi_interrupt
   <...>-550    90d.Z.    9us : rcu_nmi_exit <-soft_nmi_interrupt
   <...>-550    90d...    9us : printk_nmi_exit <-soft_nmi_interrupt
   <...>-550    90d...   10us : cpuacct_charge <-update_curr_rt

The soft_nmi_interrupt here is the call into the watchdog, due to the
decrementer interrupt firing with irqs soft-disabled. This is
harmless, but sub-optimal.

When it's not called from NMI context or with interrupts enabled, mark
the decrementer pending in the irq_happened mask directly, rather than
having the masked decrementer interupt handler do it. This will be
replayed at the next local_irq_enable. See the comment for details.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/powernv/ioda2: Remove redundant free of TCE pages
Alexey Kardashevskiy [Wed, 30 May 2018 09:22:50 +0000 (19:22 +1000)]
powerpc/powernv/ioda2: Remove redundant free of TCE pages

When IODA2 creates a PE, it creates an IOMMU table with it_ops::free
set to pnv_ioda2_table_free() which calls pnv_pci_ioda2_table_free_pages().

Since iommu_tce_table_put() calls it_ops::free when the last reference
to the table is released, explicit call to pnv_pci_ioda2_table_free_pages()
is not needed so let's remove it.

This should fix double free in the case of PCI hotuplug as
pnv_pci_ioda2_table_free_pages() does not reset neither
iommu_table::it_base nor ::it_size.

This was not exposed by SRIOV as it uses different code path via
pnv_pcibios_sriov_disable().

IODA1 does not inialize it_ops::free so it does not have this issue.

Fixes: c5f7700bbd2e ("powerpc/powernv: Dynamically release PE")
Cc: stable@vger.kernel.org # v4.8+
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/xmon: use match_string() helper
Yisheng Xie [Thu, 31 May 2018 11:11:25 +0000 (19:11 +0800)]
powerpc/xmon: use match_string() helper

match_string() returns the index of an array for a matching string,
which can be used instead of open coded variant.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc: Fix build by disabling attribute-alias warning for SYSCALL_DEFINEx
Christophe Leroy [Tue, 29 May 2018 16:06:41 +0000 (16:06 +0000)]
powerpc: Fix build by disabling attribute-alias warning for SYSCALL_DEFINEx

GCC 8.1 emits warnings such as the following. As arch/powerpc code is
built with -Werror, this breaks the build with GCC 8.1.

  In file included from arch/powerpc/kernel/pci_64.c:23:
  ./include/linux/syscalls.h:233:18: error: 'sys_pciconfig_iobase' alias
  between functions of incompatible types 'long int(long int, long
  unsigned int, long unsigned int)' and 'long int(long int, long int,
  long int)' [-Werror=attribute-alias]
    asmlinkage long sys##name(__MAP(x,__SC_DECL,__VA_ARGS__)) \
                    ^~~
  ./include/linux/syscalls.h:222:2: note: in expansion of macro '__SYSCALL_DEFINEx'
    __SYSCALL_DEFINEx(x, sname, __VA_ARGS__)

This patch inhibits those warnings.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Trim change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/64: Fix strncpy() related build failures with GCC 8.1
Christophe Leroy [Tue, 29 May 2018 06:03:53 +0000 (06:03 +0000)]
powerpc/64: Fix strncpy() related build failures with GCC 8.1

GCC 8.1 warns about possible string truncation:

  arch/powerpc/kernel/nvram_64.c:1042:2: error: 'strncpy' specified
  bound 12 equals destination size [-Werror=stringop-truncation]
    strncpy(new_part->header.name, name, 12);

  arch/powerpc/platforms/ps3/repository.c:106:2: error: 'strncpy'
  output truncated before terminating nul copying 8 bytes from a
  string of the same length [-Werror=stringop-truncation]
    strncpy((char *)&n, text, 8);

Fix it by using memcpy(). To make that safe we need to ensure the
destination is pre-zeroed. Use kzalloc() in the nvram code and
initialise the u64 to zero in the ps3 code.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Use kzalloc() in the nvram code, flesh out change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoMerge branch 'topic/pkey' into next
Michael Ellerman [Sun, 3 Jun 2018 10:35:27 +0000 (20:35 +1000)]
Merge branch 'topic/pkey' into next

This is a branch with a mixture of mm, x86 and powerpc commits all
relating to some minor cross-arch pkeys consolidation. The x86/mm
changes have been reviewed by Ingo & Dave Hansen and the tree has been
in linux-next for some weeks without issue.

5 years agoMerge branch 'fixes' into next
Michael Ellerman [Sun, 3 Jun 2018 10:32:02 +0000 (20:32 +1000)]
Merge branch 'fixes' into next

We ended up with an ugly conflict between fixes and next in ftrace.h
involving multiple nested ifdefs, and the automatic resolution is
wrong. So merge fixes into next so we can fix it up.

5 years agoMerge branch 'topic/kbuild' into next
Michael Ellerman [Sun, 3 Jun 2018 10:24:15 +0000 (20:24 +1000)]
Merge branch 'topic/kbuild' into next

Merge in some commits we're sharing with the kbuild tree.

5 years agoMerge branch 'topic/ppc-kvm' into next
Michael Ellerman [Sun, 3 Jun 2018 10:23:54 +0000 (20:23 +1000)]
Merge branch 'topic/ppc-kvm' into next

Merge in some commits we're sharing with the kvm-ppc tree.

5 years agopowerpc/mm: Fix kernel crash on page table free
Aneesh Kumar K.V [Wed, 30 May 2018 12:32:25 +0000 (18:02 +0530)]
powerpc/mm: Fix kernel crash on page table free

Fix the below crash on Book3E 64. pgtable_page_dtor expects struct
page *arg.

Also call the destructor on non book3s platforms correctly. This frees
up the split PTL locks correctly if we had allocated them before.

Call Trace:
  .kmem_cache_free+0x9c/0x44c (unreliable)
  .ptlock_free+0x1c/0x30
  .tlb_remove_table+0xdc/0x224
  .free_pgd_range+0x298/0x500
  .shift_arg_pages+0x10c/0x1e0
  .setup_arg_pages+0x200/0x25c
  .load_elf_binary+0x450/0x16c8
  .search_binary_handler.part.11+0x9c/0x248
  .do_execveat_common.isra.13+0x868/0xc18
  .run_init_process+0x34/0x4c
  .try_to_run_init_process+0x1c/0x68
  .kernel_init+0xdc/0x130
  .ret_from_kernel_thread+0x58/0x7c

Fixes: 702346768 ("powerpc/mm/nohash: Remove pte fragment dependency from nohash")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/prom: Fix %u/%llx usage since prom_printf() change
Mathieu Malaterre [Tue, 29 May 2018 19:20:01 +0000 (21:20 +0200)]
powerpc/prom: Fix %u/%llx usage since prom_printf() change

In commit eae5f709a4d7 ("powerpc: Add __printf verification to
prom_printf") __printf attribute was added to prom_printf(), which
means GCC started warning about type/format mismatches. As part of
that commit we changed some "%lx" formats to "%llx" where the type is
actually unsigned long long.

Unfortunately prom_printf() doesn't know how to print "%llx", it just
prints a literal "lx", eg:

  reserved memory map:
    lx - lx
    lx - lx

prom_printf() also doesn't know how to print "%u" (only "%lu"), it
just prints a literal "u", eg:

  Max number of cores passed to firmware: u (NR_CPUS = 2048)

Instead of:

  Max number of cores passed to firmware: 2048 (NR_CPUS = 2048)

This commit adds support for the missing formatters.

Fixes: eae5f709a4d7 ("powerpc: Add __printf verification to prom_printf")
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Tested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agocxl: Configure PSL to not use APC virtual machines
Vaibhav Jain [Tue, 17 Apr 2018 05:11:02 +0000 (10:41 +0530)]
cxl: Configure PSL to not use APC virtual machines

APC virtual machines arent used on POWER-9 chips and are already
disabled in on-chip CAPP. They also need to be disabled on the PSL via
'PSL Data Send Control Register' by setting bit(47). This forces the
PSL to send commands to CAPP with queue.id == 0.

Fixes: 5632874311db ("cxl: Add support for POWER9 DD2")
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Reviewed-by: Alastair D'Silva <alastair@d-silva.org>
Reviewed-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agocxl: Disable prefault_mode in Radix mode
Vaibhav Jain [Fri, 18 May 2018 09:42:23 +0000 (15:12 +0530)]
cxl: Disable prefault_mode in Radix mode

Currently we see a kernel-oops reported on Power-9 while attaching a
context to an AFU, with radix-mode and sysfs attr 'prefault_mode' set
to anything other than 'none'. The backtrace of the oops is of this
form:

  Unable to handle kernel paging request for data at address 0x00000080
  Faulting instruction address: 0xc00800000bcf3b20
  cpu 0x1: Vector: 300 (Data Access) at [c00000037f003800]
      pc: c00800000bcf3b20: cxl_load_segment+0x178/0x290 [cxl]
      lr: c00800000bcf39f0: cxl_load_segment+0x48/0x290 [cxl]
      sp: c00000037f003a80
     msr: 9000000000009033
     dar: 80
   dsisr: 40000000
    current = 0xc00000037f280000
    paca    = 0xc0000003ffffe600   softe: 3        irq_happened: 0x01
      pid   = 3529, comm = afp_no_int
  <snip>
  cxl_prefault+0xfc/0x248 [cxl]
  process_element_entry_psl9+0xd8/0x1a0 [cxl]
  cxl_attach_dedicated_process_psl9+0x44/0x130 [cxl]
  native_attach_process+0xc0/0x130 [cxl]
  afu_ioctl+0x3f4/0x5e0 [cxl]
  do_vfs_ioctl+0xdc/0x890
  ksys_ioctl+0x68/0xf0
  sys_ioctl+0x40/0xa0
  system_call+0x58/0x6c

The issue is caused as on Power-8 the AFU attr 'prefault_mode' was
used to improve initial storage fault performance by prefaulting
process segments. However on Power-9 with radix mode we don't have
Storage-Segments that we can prefault. Also prefaulting process Pages
will be too costly and fine-grained.

Hence, since the prefaulting mechanism doesn't makes sense of
radix-mode, this patch updates prefault_mode_store() to not allow any
other value apart from CXL_PREFAULT_NONE when radix mode is enabled.

Fixes: f24be42aab37 ("cxl: Add psl9 specific code")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Acked-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/kbuild: Use flags variables rather than overriding LD/CC/AS
Nicholas Piggin [Wed, 30 May 2018 12:19:21 +0000 (22:19 +1000)]
powerpc/kbuild: Use flags variables rather than overriding LD/CC/AS

The powerpc toolchain can compile combinations of 32/64 bit and
big/little endian, so it's convenient to consider, e.g.,

  `CC -m64 -mbig-endian`

To be the C compiler for the purpose of invoking it to build target
artifacts. So overriding the CC variable to include these flags works
for this purpose.

Unfortunately that is not compatible with the way the proposed new
Kconfig macro language will work.

After previous patches in this series, these flags can be carefully
passed in using flags instead.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/kbuild: Remove CROSS32 defines from top level powerpc Makefile
Nicholas Piggin [Wed, 30 May 2018 12:19:20 +0000 (22:19 +1000)]
powerpc/kbuild: Remove CROSS32 defines from top level powerpc Makefile

Switch VDSO32 build over to use CROSS32_COMPILE directly, and have
it pass in -m32 after the standard c_flags. This allows endianness
overrides to be removed and the endian and bitness flags moved into
standard flags variables.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/kbuild: Set default generic machine type for 32-bit compile
Nicholas Piggin [Wed, 30 May 2018 12:19:19 +0000 (22:19 +1000)]
powerpc/kbuild: Set default generic machine type for 32-bit compile

Some 64-bit toolchains uses the wrong ISA variant for compiling 32-bit
kernels, even with -m32. Debian's powerpc64le is one such case, and
that is because it is built with --with-cpu=power8.

So when cross compiling a 32-bit kernel with a 64-bit toolchain, set
-mcpu=powerpc initially, which is the generic 32-bit powerpc machine
type and scheduling model. CPU and platform code can override this
with subsequent -mcpu flags if necessary.

This is not done for 32-bit toolchains otherwise it would override
their defaults, which are presumably set appropriately for the
environment (moreso than a 64-bit cross compiler).

This fixes a lot of build failures due to incompatible assembly when
compiling 32-bit kernel with the Debian powerpc64le 64-bit toolchain.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/livepatch: Fix build error with kprobes disabled.
Aneesh Kumar K.V [Tue, 22 May 2018 09:08:20 +0000 (14:38 +0530)]
powerpc/livepatch: Fix build error with kprobes disabled.

arch/powerpc/kernel/stacktrace.c: In function ‘save_stack_trace_tsk_reliable’:
arch/powerpc/kernel/stacktrace.c:176:28: error: ‘kretprobe_trampoline’ undeclared
   if (ip == (unsigned long)kretprobe_trampoline)
                            ^~~~~~~~~~~~~~~~~~~~

Fixes: df78d3f61480 ("powerpc/livepatch: Implement reliable stack tracing for the consistency model")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoselftests/powerpc: Add core file test for Protection Key registers
Thiago Jung Bauermann [Fri, 25 May 2018 02:11:45 +0000 (23:11 -0300)]
selftests/powerpc: Add core file test for Protection Key registers

This test verifies that the AMR, IAMR and UAMOR are being written to a
process' core file.

Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
[mpe: Simplify make rule]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agoselftests/powerpc: Add ptrace tests for Protection Key registers
Thiago Jung Bauermann [Fri, 25 May 2018 02:11:44 +0000 (23:11 -0300)]
selftests/powerpc: Add ptrace tests for Protection Key registers

This test exercises read and write access to the AMR, IAMR and UAMOR.

Signed-off-by: Thiago Jung Bauermann <bauerman@linux.ibm.com>
[mpe: Simplify make rule]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/modules: remove unused mod_arch_specific.toc field
Josh Poimboeuf [Fri, 25 May 2018 03:48:34 +0000 (22:48 -0500)]
powerpc/modules: remove unused mod_arch_specific.toc field

The toc field in the mod_arch_specific struct isn't actually used
anywhere, so remove it.

Also the ftrace-specific fields are now common between 32-bit and
64-bit, so simplify the struct definition a bit by moving them out of
the __powerpc64__ #ifdef.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
5 years agopowerpc/powernv/cpuidle: Init all present cpus for deep states
Akshay Adiga [Wed, 16 May 2018 12:02:14 +0000 (17:32 +0530)]
powerpc/powernv/cpuidle: Init all present cpus for deep states

Init all present cpus for deep states instead of "all possible" cpus.
Init fails if a possible cpu is guarded. Resulting in making only
non-deep states available for cpuidle/hotplug.

Stewart says, this means that for single threaded workloads, if you
guard out a CPU core you'll not get WoF (Workload Optimised
Frequency), which means that performance goes down when you wouldn't
expect it to.

Fixes: 77b54e9f213f ("powernv/powerpc: Add winkle support for offline cpus")
Cc: stable@vger.kernel.org # v3.19+
Signed-off-by: Akshay Adiga <akshay.adiga@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/perf: Update raw-event code encoding comment for power8
Madhavan Srinivasan [Tue, 8 May 2018 05:00:24 +0000 (10:30 +0530)]
powerpc/perf: Update raw-event code encoding comment for power8

Comment explanning the raw event code encoding for Power8 was
moved to isa207_common.h file when re-factoring the code to
support power9. But then Power9 pmu branched out due to changes
specific to power9. So move the encoding comment back to power8-pmu.c
Just comment movement and no logic change.

Fixes: 4d3576b20716 ('powerpc/perf: factor out power8 pmu macros and defines')
Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Add a missing include header
Mathieu Malaterre [Thu, 22 Mar 2018 20:20:04 +0000 (21:20 +0100)]
powerpc: Add a missing include header

The header file <asm/switch_to.h> was missing from the includes. Fix the
following warning, treated as error with W=1:

  arch/powerpc/kernel/vecemu.c:260:5: error: no previous prototype for ‘emulate_altivec’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/32: Add a missing include header
Mathieu Malaterre [Thu, 22 Mar 2018 20:20:03 +0000 (21:20 +0100)]
powerpc/32: Add a missing include header

The header file <linux/syscalls.h> was missing from the includes. Fix the
following warning, treated as error with W=1:

  arch/powerpc/kernel/pci_32.c:286:6: error: no previous prototype for ‘sys_pciconfig_iobase’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/chrp/time: Make some functions static, add missing header include
Mathieu Malaterre [Thu, 22 Mar 2018 20:19:56 +0000 (21:19 +0100)]
powerpc/chrp/time: Make some functions static, add missing header include

Add a missing include <platforms/chrp/chrp.h>.

These functions can all be static, make it so. Fix warnings treated as
errors with W=1:

  arch/powerpc/platforms/chrp/time.c:41:13: error: no previous prototype for ‘chrp_time_init’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/time.c:66:5: error: no previous prototype for ‘chrp_cmos_clock_read’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/time.c:74:6: error: no previous prototype for ‘chrp_cmos_clock_write’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/time.c:86:5: error: no previous prototype for ‘chrp_set_rtc_time’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/time.c:130:6: error: no previous prototype for ‘chrp_get_rtc_time’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/tau: Make some function static
Mathieu Malaterre [Thu, 22 Mar 2018 20:19:54 +0000 (21:19 +0100)]
powerpc/tau: Make some function static

These functions can all be static, make it so. Fix warnings treated as
errors with W=1:

  arch/powerpc/kernel/tau_6xx.c:53:6: error: no previous prototype for ‘set_thresholds’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:73:6: error: no previous prototype for ‘TAUupdate’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:208:13: error: no previous prototype for ‘TAU_init_smp’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:220:12: error: no previous prototype for ‘TAU_init’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:126:6: error: no previous prototype for ‘TAUException’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Make function btext_initialize static
Mathieu Malaterre [Thu, 22 Mar 2018 20:19:52 +0000 (21:19 +0100)]
powerpc: Make function btext_initialize static

This function can be static, make it so, this fix a warning treated as
error with W=1:

  arch/powerpc/kernel/btext.c:173:5: error: no previous prototype for ‘btext_initialize’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/tau: Synchronize function prototypes and body
Mathieu Malaterre [Wed, 4 Apr 2018 20:10:28 +0000 (22:10 +0200)]
powerpc/tau: Synchronize function prototypes and body

Some function prototypes and body for Thermal Assist Units were not in
sync. Update the function definition to match the existing function
declaration found in `setup-common.c`, changing an `int` return type to a
`u32` return type. Move the prototypes to a header file. Fix the following
warnings, treated as error with W=1:

  arch/powerpc/kernel/tau_6xx.c:257:5: error: no previous prototype for ‘cpu_temp_both’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:262:5: error: no previous prototype for ‘cpu_temp’ [-Werror=missing-prototypes]
  arch/powerpc/kernel/tau_6xx.c:267:5: error: no previous prototype for ‘tau_interrupts’ [-Werror=missing-prototypes]

Compile tested with CONFIG_TAU_INT.

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Add missing prototype
Mathieu Malaterre [Wed, 4 Apr 2018 20:11:42 +0000 (22:11 +0200)]
powerpc: Add missing prototype

Add one missing prototype for function rh_dump_blk. Fix warning treated as
error in W=1:

  arch/powerpc/lib/rheap.c:740:6: error: no previous prototype for ‘rh_dump_blk’ [-Werror=missing-prototypes]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/52xx: Add missing functions prototypes
Mathieu Malaterre [Wed, 4 Apr 2018 20:12:30 +0000 (22:12 +0200)]
powerpc/52xx: Add missing functions prototypes

The function prototypes were declared within a `#ifdef CONFIG_PPC_LITE5200`
block which would prevent them from being visible when compiling
`mpc52xx_pm.c`. Move the prototypes outside of the `#ifdef` block to fix
the following warnings treated as errors with W=1:

  arch/powerpc/platforms/52xx/mpc52xx_pm.c:58:5: error: no previous prototype for ‘mpc52xx_pm_prepare’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/52xx/mpc52xx_pm.c:113:5: error: no previous prototype for ‘mpc52xx_pm_enter’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/52xx/mpc52xx_pm.c:181:6: error: no previous prototype for ‘mpc52xx_pm_finish’ [-Werror=missing-prototypes]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Add missing prototype for note_bootable_part()
Mathieu Malaterre [Wed, 4 Apr 2018 20:13:05 +0000 (22:13 +0200)]
powerpc/powermac: Add missing prototype for note_bootable_part()

Add a missing prototype for function `note_bootable_part` to silence a
warning treated as error with W=1:

  arch/powerpc/platforms/powermac/setup.c:361:12: error: no previous prototype for ‘note_bootable_part’ [-Werror=missing-prototypes]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Move pmac_pfunc_base_install prototype to header file
Mathieu Malaterre [Wed, 4 Apr 2018 20:13:55 +0000 (22:13 +0200)]
powerpc/powermac: Move pmac_pfunc_base_install prototype to header file

The pmac_pfunc_base_install prototype was declared in powermac/smp.c since
function was used there, move it to pmac_pfunc.h header to be visible in
pfunc_base.c. Fix a warning treated as error with W=1:

  arch/powerpc/platforms/powermac/pfunc_base.c:330:12: error: no previous prototype for ‘pmac_pfunc_base_install’ [-Werror=missing-prototypes]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/chrp/pci: Make some functions static
Mathieu Malaterre [Wed, 4 Apr 2018 20:15:03 +0000 (22:15 +0200)]
powerpc/chrp/pci: Make some functions static

These functions can all be static, make it so. Fix warnings treated as
errors with W=1:

  arch/powerpc/platforms/chrp/pci.c:34:5: error: no previous prototype for ‘gg2_read_config’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/pci.c:61:5: error: no previous prototype for ‘gg2_write_config’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/pci.c:97:5: error: no previous prototype for ‘rtas_read_config’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/pci.c:112:5: error: no previous prototype for ‘rtas_write_config’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powermac: Mark variable x as unused
Mathieu Malaterre [Wed, 4 Apr 2018 20:07:46 +0000 (22:07 +0200)]
powerpc/powermac: Mark variable x as unused

Since the value of x is never intended to be read, declare it with gcc
attribute as unused. Fix warning treated as error with W=1:

  arch/powerpc/platforms/powermac/bootx_init.c:471:21: error: variable ‘x’ set but not used [-Werror=unused-but-set-variable]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/chrp/setup: Remove idu_size variable and make some functions static
Mathieu Malaterre [Wed, 4 Apr 2018 20:09:11 +0000 (22:09 +0200)]
powerpc/chrp/setup: Remove idu_size variable and make some functions static

Remove variable declaration idu_size and associated code since not used.

These functions can all be static, make it so. Fix warnings treated as
errors with W=1:

  arch/powerpc/platforms/chrp/setup.c:97:6: error: no previous prototype for ‘chrp_show_cpuinfo’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/setup.c:302:13: error: no previous prototype for ‘chrp_setup_arch’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/setup.c:385:16: error: variable ‘idu_size’ set but not used [-Werror=unused-but-set-variable]
  arch/powerpc/platforms/chrp/setup.c:526:13: error: no previous prototype for ‘chrp_init_IRQ’ [-Werror=missing-prototypes]
  arch/powerpc/platforms/chrp/setup.c:559:1: error: no previous prototype for ‘chrp_init2’ [-Werror=missing-prototypes]

Suggested-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/wii: Make hlwd_pic_init function static
Mathieu Malaterre [Mon, 23 Apr 2018 19:45:32 +0000 (21:45 +0200)]
powerpc/wii: Make hlwd_pic_init function static

The function hlwd_pic_init can be made static, so do it. Fix the following
warning treated as error (W=1):

../arch/powerpc/platforms/embedded6xx/hlwd-pic.c:158:20: error: no previous prototype for ‘hlwd_pic_init’ [-Werror=missing-prototypes]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm/radix: Use do/while(0) trick for single statement block
Mathieu Malaterre [Mon, 23 Apr 2018 19:36:38 +0000 (21:36 +0200)]
powerpc/mm/radix: Use do/while(0) trick for single statement block

In commit 7a22d6321c3d ("powerpc/mm/radix: Update command line parsing for
disable_radix") an `if` statement was added for a possible empty body
(prom_debug).

Fix the following warning, treated as error with W=1:

  arch/powerpc/kernel/prom_init.c:656:46: error: suggest braces around empty body in an ‘if’ statement [-Werror=empty-body]

Suggested-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/sparse: Fix plain integer as NULL pointer warning
Mathieu Malaterre [Fri, 13 Apr 2018 18:41:43 +0000 (20:41 +0200)]
powerpc/sparse: Fix plain integer as NULL pointer warning

Trivial fix to remove the following sparse warnings:

  arch/powerpc/kernel/module_32.c:112:74: warning: Using plain integer as NULL pointer
  arch/powerpc/kernel/module_32.c:117:74: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:1155:28: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:1230:20: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:1385:36: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:1752:23: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:2084:19: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:2110:32: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:2167:19: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:2183:19: warning: Using plain integer as NULL pointer
  drivers/macintosh/via-pmu.c:277:20: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/setup.c:155:67: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/setup.c:247:27: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/setup.c:249:27: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/setup.c:252:37: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/tlb_hash32.c:127:21: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/tlb_hash32.c:148:21: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/tlb_hash32.c:44:21: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/tlb_hash32.c:57:21: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/tlb_hash32.c:87:21: warning: Using plain integer as NULL pointer
  arch/powerpc/kernel/btext.c:160:31: warning: Using plain integer as NULL pointer
  arch/powerpc/kernel/btext.c:167:22: warning: Using plain integer as NULL pointer
  arch/powerpc/kernel/btext.c:274:21: warning: Using plain integer as NULL pointer
  arch/powerpc/kernel/btext.c:285:31: warning: Using plain integer as NULL pointer
  arch/powerpc/include/asm/hugetlb.h:204:16: warning: Using plain integer as NULL pointer
  arch/powerpc/mm/ppc_mmu_32.c:170:21: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/pci.c:1227:23: warning: Using plain integer as NULL pointer
  arch/powerpc/platforms/powermac/pci.c:65:24: warning: Using plain integer as NULL pointer

Also use `--fix` command line option from `script/checkpatch --strict` to
remove the following:

  CHECK: Comparison to NULL could be written "!dispDeviceBase"
  #72: FILE: arch/powerpc/kernel/btext.c:160:
  + if (dispDeviceBase == NULL)

  CHECK: Comparison to NULL could be written "!vbase"
  #80: FILE: arch/powerpc/kernel/btext.c:167:
  + if (vbase == NULL)

  CHECK: Comparison to NULL could be written "!base"
  #89: FILE: arch/powerpc/kernel/btext.c:274:
  + if (base == NULL)

  CHECK: Comparison to NULL could be written "!dispDeviceBase"
  #98: FILE: arch/powerpc/kernel/btext.c:285:
  + if (dispDeviceBase == NULL)

  CHECK: Comparison to NULL could be written "strstr"
  #117: FILE: arch/powerpc/kernel/module_32.c:117:
  + if (strstr(secstrings + sechdrs[i].sh_name, ".debug") != NULL)

  CHECK: Comparison to NULL could be written "!Hash"
  #130: FILE: arch/powerpc/mm/ppc_mmu_32.c:170:
  + if (Hash == NULL)

  CHECK: Comparison to NULL could be written "Hash"
  #143: FILE: arch/powerpc/mm/tlb_hash32.c:44:
  + if (Hash != NULL) {

  CHECK: Comparison to NULL could be written "!Hash"
  #152: FILE: arch/powerpc/mm/tlb_hash32.c:57:
  + if (Hash == NULL) {

  CHECK: Comparison to NULL could be written "!Hash"
  #161: FILE: arch/powerpc/mm/tlb_hash32.c:87:
  + if (Hash == NULL) {

  CHECK: Comparison to NULL could be written "!Hash"
  #170: FILE: arch/powerpc/mm/tlb_hash32.c:127:
  + if (Hash == NULL) {

  CHECK: Comparison to NULL could be written "!Hash"
  #179: FILE: arch/powerpc/mm/tlb_hash32.c:148:
  + if (Hash == NULL) {

  ERROR: space required after that ';' (ctx:VxV)
  #192: FILE: arch/powerpc/platforms/powermac/pci.c:65:
  + for (; node != NULL;node = node->sibling) {

  CHECK: Comparison to NULL could be written "node"
  #192: FILE: arch/powerpc/platforms/powermac/pci.c:65:
  + for (; node != NULL;node = node->sibling) {

  CHECK: Comparison to NULL could be written "!region"
  #201: FILE: arch/powerpc/platforms/powermac/pci.c:1227:
  + if (region == NULL)

  CHECK: Comparison to NULL could be written "of_get_property"
  #214: FILE: arch/powerpc/platforms/powermac/setup.c:155:
  + if (of_get_property(np, "cache-unified", NULL) != NULL && dc) {

  CHECK: Comparison to NULL could be written "!np"
  #223: FILE: arch/powerpc/platforms/powermac/setup.c:247:
  + if (np == NULL)

  CHECK: Comparison to NULL could be written "np"
  #226: FILE: arch/powerpc/platforms/powermac/setup.c:249:
  + if (np != NULL) {

  CHECK: Comparison to NULL could be written "l2cr"
  #230: FILE: arch/powerpc/platforms/powermac/setup.c:252:
  + if (l2cr != NULL) {

  CHECK: Comparison to NULL could be written "via"
  #243: FILE: drivers/macintosh/via-pmu.c:277:
  + if (via != NULL)

  CHECK: Comparison to NULL could be written "current_req"
  #252: FILE: drivers/macintosh/via-pmu.c:1155:
  + if (current_req != NULL) {

  CHECK: Comparison to NULL could be written "!req"
  #261: FILE: drivers/macintosh/via-pmu.c:1230:
  + if (req == NULL || pmu_state != idle

  CHECK: Comparison to NULL could be written "!req"
  #270: FILE: drivers/macintosh/via-pmu.c:1385:
  + if (req == NULL) {

  CHECK: Comparison to NULL could be written "!pp"
  #288: FILE: drivers/macintosh/via-pmu.c:2084:
  + if (pp == NULL)

  CHECK: Comparison to NULL could be written "!pp"
  #297: FILE: drivers/macintosh/via-pmu.c:2110:
  + if (count < 1 || pp == NULL)

  CHECK: Comparison to NULL could be written "!pp"
  #306: FILE: drivers/macintosh/via-pmu.c:2167:
  + if (pp == NULL)

  CHECK: Comparison to NULL could be written "pp"
  #315: FILE: drivers/macintosh/via-pmu.c:2183:
  + if (pp != NULL) {

Link: https://github.com/linuxppc/linux/issues/37
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/altivec: Add missing prototypes for altivec
Mathieu Malaterre [Wed, 28 Mar 2018 18:55:25 +0000 (20:55 +0200)]
powerpc/altivec: Add missing prototypes for altivec

Some functions prototypes were missing for the non-altivec code. Add the
missing prototypes in a new header file, fix warnings treated as errors
with W=1:

  arch/powerpc/lib/xor_vmx_glue.c:18:6: error: no previous prototype for ‘xor_altivec_2’ [-Werror=missing-prototypes]
  arch/powerpc/lib/xor_vmx_glue.c:29:6: error: no previous prototype for ‘xor_altivec_3’ [-Werror=missing-prototypes]
  arch/powerpc/lib/xor_vmx_glue.c:40:6: error: no previous prototype for ‘xor_altivec_4’ [-Werror=missing-prototypes]
  arch/powerpc/lib/xor_vmx_glue.c:52:6: error: no previous prototype for ‘xor_altivec_5’ [-Werror=missing-prototypes]

The prototypes were already present in <asm/xor.h> but this header file is
meant to be included after <include/linux/raid/xor.h>. Trying to re-use
<asm/xor.h> directly would lead to warnings such as:

  arch/powerpc/include/asm/xor.h:39:15: error: variable ‘xor_block_altivec’ has initializer but incomplete type

Trying to re-use <asm/xor.h> after <include/linux/raid/xor.h> in
xor_vmx_glue.c would in turn trigger the following warnings:

  include/asm-generic/xor.h:688:34: error: ‘xor_block_32regs’ defined but not used [-Werror=unused-variable]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Add __printf verification to prom_printf
Mathieu Malaterre [Fri, 6 Apr 2018 20:12:19 +0000 (22:12 +0200)]
powerpc: Add __printf verification to prom_printf

__printf is useful to verify format and arguments. Fix arg mismatch
reported by gcc, remove the following warnings (with W=1):

  arch/powerpc/kernel/prom_init.c:1467:31: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1471:31: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1504:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1505:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1506:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1507:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1508:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1509:33: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:1975:39: error: format ‘%lu’ expects argument of type ‘long unsigned int’, but argument 2 has type ‘unsigned int’
  arch/powerpc/kernel/prom_init.c:1986:27: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:2567:38: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:2567:46: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 3 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:2569:38: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
  arch/powerpc/kernel/prom_init.c:2569:46: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 3 has type ‘long unsigned int’

The patch also include arg mismatch fix for case with #define DEBUG_PROM
(warning not listed here).

This patch fix also the following warnings revealed by checkpatch:

  WARNING: Prefer using '"%s...", __func__' to using 'alloc_up', this function's name, in a string
  #101: FILE: arch/powerpc/kernel/prom_init.c:1235:
  + prom_debug("alloc_up(%lx, %lx)\n", size, align);

and

  WARNING: Prefer using '"%s...", __func__' to using 'alloc_down', this function's name, in a string
  #138: FILE: arch/powerpc/kernel/prom_init.c:1278:
  + prom_debug("alloc_down(%lx, %lx, %s)\n", size, align,

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/xmon: Update paca fields dumped in xmon
Michael Ellerman [Mon, 21 May 2018 09:47:20 +0000 (19:47 +1000)]
powerpc/xmon: Update paca fields dumped in xmon

The set of paca fields we dump in xmon has gotten somewhat out of
date. Update to add some recently added fields.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/xmon: Realign paca dump fields
Michael Ellerman [Mon, 21 May 2018 11:28:34 +0000 (21:28 +1000)]
powerpc/xmon: Realign paca dump fields

We've added some fields with longer names since we originally wrote
this, so the fields are no longer lined up. Adjust the widths to make
it all look nice again, eg:

  0:mon> dp
  paca for cpu 0x0 @ c000000001fa0000:
   possible                  = yes
   ...
   slb_shadow            [0] = 0xc000000008000000 0x400ea1b217000500
   slb_shadow            [1] = 0xd000000008000001 0x400d43642f000510
   ...
   rfi_flush_fallback_area   = c0000000fff80000   (0xcc8)
   ...
   accounting.starttime_user = 0x51582f07         (0xae8)

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/xmon: Add __printf annotation to xmon_printf()
Mathieu Malaterre [Sun, 25 Mar 2018 09:06:47 +0000 (11:06 +0200)]
powerpc/xmon: Add __printf annotation to xmon_printf()

This allows the compiler to verify the format strings vs the types of
the arguments.

Update the other prototype declarations in asm/xmon.h.

Silence warnings (triggered at W=1) by adding relevant __printf
attribute. Move #define at bottom of the file to prevent conflict with
gcc attribute.

Solves the original warning:

  arch/powerpc/xmon/nonstdio.c:178:2: error: function might be
  possible candidate for ‘gnu_printf’ format attribute

In turn this uncovered many formatting errors in xmon.c, all fixed in
this patch.

Signed-off-by: Mathieu Malaterre <malat@debian.org>
[mpe: Always use px not p, fixup the 44x specific code, tweak change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/xmon: Specify the full format in DUMP() macro
Michael Ellerman [Mon, 21 May 2018 11:06:19 +0000 (21:06 +1000)]
powerpc/xmon: Specify the full format in DUMP() macro

In dump_one_paca() the DUMP macro unconditionally prepends '#' to the
printf format specifier. In most cases we're using either 'x' or 'lx'
etc. and that is OK. But for 'p' and other formats using '#' is
actually undefined, and once we enable printf() checking for
xmon_printf() we will get warnings from the compiler.

So just have each usage specify the full format, that way we can omit
'#' when it's inappropriate.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Mathieu Malaterre <malat@debian.org>
6 years agopowerpc/xmon: Also setup debugger hooks when single-stepping
Michal Suchanek [Wed, 23 May 2018 18:00:54 +0000 (20:00 +0200)]
powerpc/xmon: Also setup debugger hooks when single-stepping

When single-stepping kernel code from xmon without a debug hook
enabled the kernel crashes. This can happen when kernel starts with
xmon on crash disabled but xmon is entered using sysrq.

Call force_enable_xmon when single-stepping in xmon to install the
xmon debug hooks.

Fixes: e1368d0c9edb ("powerpc/xmon: Setup debugger hooks when first break-point is set")
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Reviewed-by: Vaibhav Jain <vaibhav@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/8xx: fix invalid register expression in head_8xx.S
Christophe Leroy [Thu, 24 May 2018 11:02:06 +0000 (11:02 +0000)]
powerpc/8xx: fix invalid register expression in head_8xx.S

New binutils generate the following warning

  AS      arch/powerpc/kernel/head_8xx.o
arch/powerpc/kernel/head_8xx.S: Assembler messages:
arch/powerpc/kernel/head_8xx.S:916: Warning: invalid register expression

This patch fixes it.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc: Add ptrace hw breakpoint test
Michael Neuling [Tue, 22 May 2018 06:14:27 +0000 (16:14 +1000)]
selftests/powerpc: Add ptrace hw breakpoint test

This test the ptrace hw breakpoints via PTRACE_SET_DEBUGREG and
PPC_PTRACE_SETHWDEBUG.  This test was use to find the bugs fixed by
these recent commits:

  4f7c06e26e powerpc/ptrace: Fix setting 512B aligned breakpoints with PTRACE_SET_DEBUGREG
  cd6ef7eebf powerpc/ptrace: Fix enforcement of DAWR constraints

Signed-off-by: Michael Neuling <mikey@neuling.org>
[mpe: Add SPDX tag, clang format it]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc: Add missing .gitignores
Michael Neuling [Tue, 22 May 2018 06:13:59 +0000 (16:13 +1000)]
selftests/powerpc: Add missing .gitignores

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm: Only read faulting instruction when necessary in do_page_fault()
Christophe Leroy [Wed, 23 May 2018 08:53:22 +0000 (10:53 +0200)]
powerpc/mm: Only read faulting instruction when necessary in do_page_fault()

Commit a7a9dcd882a67 ("powerpc: Avoid taking a data miss on every
userspace instruction miss") has shown that limiting the read of
faulting instruction to likely cases improves performance.

This patch goes further into this direction by limiting the read
of the faulting instruction to the only cases where it is likely
needed.

On an MPC885, with the same benchmark app as in the commit referred
above, we see a reduction of about 3900 dTLB misses (approx 3%):

Before the patch:
 Performance counter stats for './fault 500' (10 runs):

         683033312      cpu-cycles                                                    ( +-  0.03% )
            134538      dTLB-load-misses                                              ( +-  0.03% )
             46099      iTLB-load-misses                                              ( +-  0.02% )
             19681      faults                                                        ( +-  0.02% )

       5.389747878 seconds time elapsed                                          ( +-  0.06% )

With the patch:

 Performance counter stats for './fault 500' (10 runs):

         682112862      cpu-cycles                                                    ( +-  0.03% )
            130619      dTLB-load-misses                                              ( +-  0.03% )
             46073      iTLB-load-misses                                              ( +-  0.05% )
             19681      faults                                                        ( +-  0.01% )

       5.381342641 seconds time elapsed                                          ( +-  0.07% )

The proper work of the huge stack expansion was tested with the
following app:

int main(int argc, char **argv)
{
char buf[1024 * 1025];

sprintf(buf, "Hello world !\n");
printf(buf);

exit(0);
}

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Add include of pagemap.h to fix build errors]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/mm: Use instruction symbolic names in store_updates_sp()
Christophe Leroy [Wed, 23 May 2018 07:04:04 +0000 (09:04 +0200)]
powerpc/mm: Use instruction symbolic names in store_updates_sp()

Use symbolic names defined in asm/ppc-opcode.h
instead of hardcoded values.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Export tm_enable()/tm_disable/tm_abort() APIs
Simon Guo [Wed, 23 May 2018 07:01:46 +0000 (15:01 +0800)]
powerpc: Export tm_enable()/tm_disable/tm_abort() APIs

This patch exports tm_enable()/tm_disable/tm_abort() APIs, which
will be used for PR KVM transactional memory logic.

Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Reviewed-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/reg: Add TEXASR related macros
Simon Guo [Wed, 23 May 2018 07:01:45 +0000 (15:01 +0800)]
powerpc/reg: Add TEXASR related macros

This patches add some macros for CR0/TEXASR bits so that PR KVM TM
logic (tbegin./treclaim./tabort.) can make use of them later.

Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Reviewed-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc: Export msr_check_and_set() to modules
Simon Guo [Wed, 23 May 2018 07:01:44 +0000 (15:01 +0800)]
powerpc: Export msr_check_and_set() to modules

PR KVM will need to reuse msr_check_and_set().
This patch exports this API for reuse.

Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Reviewed-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agohwmon: (ibmpowernv) Add energy sensors
Shilpasri G Bhat [Mon, 7 May 2018 10:25:38 +0000 (15:55 +0530)]
hwmon: (ibmpowernv) Add energy sensors

This patch exports the accumulated power numbers of each power
sensor maintained by OCC.

Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agohwmon: (ibmpowernv): Add support to read 64 bit sensors
Shilpasri G Bhat [Mon, 7 May 2018 10:25:37 +0000 (15:55 +0530)]
hwmon: (ibmpowernv): Add support to read 64 bit sensors

The firmware has supported for reading sensor values of size u32.
This patch adds support to use newer firmware functions which allows
to read the sensors of size u64.

Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowernv: opal-sensor: Add support to read 64bit sensor values
Shilpasri G Bhat [Mon, 7 May 2018 10:25:36 +0000 (15:55 +0530)]
powernv: opal-sensor: Add support to read 64bit sensor values

This patch adds support to read 64-bit sensor values. This method is
used to read energy sensors and counters which are of type u64.

Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agoselftests/powerpc: Remove redundant cp_abort test
Michael Neuling [Thu, 5 Oct 2017 23:48:57 +0000 (10:48 +1100)]
selftests/powerpc: Remove redundant cp_abort test

Paste on POWER9 only works on accelerators and no longer on real
memory. Hence this test is broken so remove it.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/fsl/dts: fix the i2c-mux compatible for t104xqds
Peter Rosin [Thu, 3 Aug 2017 12:59:34 +0000 (14:59 +0200)]
powerpc/fsl/dts: fix the i2c-mux compatible for t104xqds

The sanctioned compatible is "nxp,pca9547".

Signed-off-by: Peter Rosin <peda@axentia.se>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/ptrace: Fix setting 512B aligned breakpoints with PTRACE_SET_DEBUGREG
Michael Neuling [Thu, 17 May 2018 05:37:15 +0000 (15:37 +1000)]
powerpc/ptrace: Fix setting 512B aligned breakpoints with PTRACE_SET_DEBUGREG

In commit e2a800beaca1 ("powerpc/hw_brk: Fix off by one error when
validating DAWR region end") we fixed setting the DAWR end point to
its max value via PPC_PTRACE_SETHWDEBUG. Unfortunately we broke
PTRACE_SET_DEBUGREG when setting a 512 byte aligned breakpoint.

PTRACE_SET_DEBUGREG currently sets the length of the breakpoint to
zero (memset() in hw_breakpoint_init()). This worked with
arch_validate_hwbkpt_settings() before the above patch was applied but
is now broken if the breakpoint is 512byte aligned.

This sets the length of the breakpoint to 8 bytes when using
PTRACE_SET_DEBUGREG.

Fixes: e2a800beaca1 ("powerpc/hw_brk: Fix off by one error when validating DAWR region end")
Cc: stable@vger.kernel.org # v3.11+
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/ptrace: Fix enforcement of DAWR constraints
Michael Neuling [Thu, 17 May 2018 05:37:14 +0000 (15:37 +1000)]
powerpc/ptrace: Fix enforcement of DAWR constraints

Back when we first introduced the DAWR, in commit 4ae7ebe9522a
("powerpc: Change hardware breakpoint to allow longer ranges"), we
screwed up the constraint making it a 1024 byte boundary rather than a
512. This makes the check overly permissive. Fortunately GDB is the
only real user and it always did they right thing, so we never
noticed.

This fixes the constraint to 512 bytes.

Fixes: 4ae7ebe9522a ("powerpc: Change hardware breakpoint to allow longer ranges")
Cc: stable@vger.kernel.org # v3.9+
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/powernv: Use __raw_[rm_]writeq_be() in npu-dma.c
Michael Ellerman [Mon, 14 May 2018 12:50:33 +0000 (22:50 +1000)]
powerpc/powernv: Use __raw_[rm_]writeq_be() in npu-dma.c

This allows us to squash some sparse warnings and also avoids having
to do explicity endian conversions in the code.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
6 years agopowerpc/powernv: Use __raw_[rm_]writeq_be() in pci-ioda.c
Michael Ellerman [Mon, 14 May 2018 12:50:32 +0000 (22:50 +1000)]
powerpc/powernv: Use __raw_[rm_]writeq_be() in pci-ioda.c

This allows us to squash some sparse warnings and also avoids having
to do explicity endian conversions in the code.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
6 years agopowerpc/io: Add __raw_writeq_be() __raw_rm_writeq_be()
Michael Ellerman [Mon, 14 May 2018 12:50:31 +0000 (22:50 +1000)]
powerpc/io: Add __raw_writeq_be() __raw_rm_writeq_be()

Add byte-swapping versions of __raw_writeq() and __raw_rm_writeq().

This allows us to avoid sparse warnings caused by passing __be64 to
__raw_writeq(), which takes unsigned long:

  arch/powerpc/platforms/powernv/pci-ioda.c:1981:38:
  warning: incorrect type in argument 1 (different base types)
      expected unsigned long [unsigned] v
      got restricted __be64 [usertype] <noident>

It's also generally preferable to use a byte-swapping accessor rather
than doing it by hand in the code, which is more bug prone.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
6 years agopowerpc/perf: Fix memory allocation for core-imc based on num_possible_cpus()
Anju T Sudhakar [Wed, 16 May 2018 06:35:18 +0000 (12:05 +0530)]
powerpc/perf: Fix memory allocation for core-imc based on num_possible_cpus()

Currently memory is allocated for core-imc based on cpu_present_mask,
which has bit 'cpu' set iff cpu is populated. We use (cpu number / threads
per core) as the array index to access the memory.

Under some circumstances firmware marks a CPU as GUARDed CPU and boot the
system, until cleared of errors, these CPU's are unavailable for all
subsequent boots. GUARDed CPUs are possible but not present from linux
view, so it blows a hole when we assume the max length of our allocation
is driven by our max present cpus, where as one of the cpus might be online
and be beyond the max present cpus, due to the hole.
So (cpu number / threads per core) value bounds the array index and leads
to memory overflow.

Call trace observed during a guard test:

Faulting instruction address: 0xc000000000149f1c
cpu 0x69: Vector: 380 (Data Access Out of Range) at [c000003fea303420]
    pc:c000000000149f1c: prefetch_freepointer+0x14/0x30
    lr:c00000000014e0f8: __kmalloc+0x1a8/0x1ac
    sp:c000003fea3036a0
   msr:9000000000009033
   dar:c9c54b2c91dbf6b7
  current = 0xc000003fea2c0000
  paca    = 0xc00000000fddd880  softe: 3  irq_happened: 0x01
    pid   = 1, comm = swapper/104
Linux version 4.16.7-openpower1 (smc@smc-desktop) (gcc version 6.4.0
(Buildroot 2018.02.1-00006-ga8d1126)) #2 SMP Fri May 4 16:44:54 PDT 2018
enter ? for help
call trace:
 __kmalloc+0x1a8/0x1ac
 (unreliable)
 init_imc_pmu+0x7f4/0xbf0
 opal_imc_counters_probe+0x3fc/0x43c
 platform_drv_probe+0x48/0x80
 driver_probe_device+0x22c/0x308
 __driver_attach+0xa0/0xd8
 bus_for_each_dev+0x88/0xb4
 driver_attach+0x2c/0x40
 bus_add_driver+0x1e8/0x228
 driver_register+0xd0/0x114
 __platform_driver_register+0x50/0x64
 opal_imc_driver_init+0x24/0x38
 do_one_initcall+0x150/0x15c
 kernel_init_freeable+0x250/0x254
 kernel_init+0x1c/0x150
 ret_from_kernel_thread+0x5c/0xc8

Allocating memory for core-imc based on cpu_possible_mask, which has
bit 'cpu' set iff cpu is populatable, will fix this issue.

Reported-by: Pridhiviraj Paidipeddi <ppaidipe@linux.vnet.ibm.com>
Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Tested-by: Pridhiviraj Paidipeddi <ppaidipe@linux.vnet.ibm.com>
Fixes: 39a846db1d57 ("powerpc/perf: Add core IMC PMU support")
Cc: stable@vger.kernel.org # v4.14+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agopowerpc/rtas: Fix spelling mistake "Discharching" -> "Discharging"
Colin Ian King [Fri, 18 May 2018 09:31:17 +0000 (10:31 +0100)]
powerpc/rtas: Fix spelling mistake "Discharching" -> "Discharging"

Trivial fix to spelling mistake in battery_charging array.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>