The system reset interrupt is used for crash/debug situations, so it is
desirable to have as little impact on the normal state of the system as
possible.

Currently it uses the current kernel stack to process the exception.
This stores into the stack which may be involved with the crash. The
stack pointer may be corrupted, or it may have overflowed.

Avoid or minimise these problems by creating a dedicated NMI stack for
the system reset interrupt to use.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

In preparation for using a dedicated stack for system reset interrupts,
prevent a nested system reset from recovering, in order to simplify
code that is called in crash/debug path. This allows a system reset
interrupt to just use the base stack pointer.

Keep an in_nmi nesting counter similarly to the in_mce counter. Consider
the interrrupt non-recoverable if it is taken inside another system
reset.

Interrupt nesting could be allowed similarly to MCE, but system reset
is a special case that's not for normal operation, so simplicity wins
until there is requirement for nested system reset interrupts.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The system reset interrupt can occur when MSR_EE=0, and it currently
uses the PACA_EXGEN save area.

Some PACA_EXGEN interrupts have a window where MSR_RI=1 and MSR_EE=0
when the save area is still in use. A system reset interrupt in this
window can lead to undetected corruption when the save area gets
overwritten.

This patch introduces PACA_EXNMI save area for system reset exceptions,
which closes this corruption window. It's also helpful to retain the
EXGEN state for debugging situations, even if not considering the
recoverability aspect.

This patch also moves the PACA_EXMC area down to a less frequently used
part of the paca with the new save area.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This code is common to a few exceptions, and another user will be added.
This causes a trivial change to generated code:

-     604: std     r9,416(r1)
-     608: mfspr   r11,314
-     60c: std     r11,368(r1)
-     610: mfspr   r12,315
+     604: mfspr   r11,314
+     608: mfspr   r12,315
+     60c: std     r9,416(r1)
+     610: std     r11,368(r1)

machine_check_powernv_early could also use this, but that requires non
trivial changes to generated code, so that's for another patch.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Subsequent patches will add more non-RI variant exceptions, so
create a macro for it rather than open-code it.

This does not change generated instructions.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Cell will wake from low power state at the system reset interrupt,
with the event encoded in SRR1, rather than waking at the interrupt
vector that corresponds to that event.

The system reset handler for this platform decodes SRR1 event reason
and calls the interrupt handler to process it directly from the system
reset handlre.

A subsequent change will treat the system reset interrupt as a Linux NMI
with its own per-CPU stack, and this will no longer work. Remove the
external and decrementer handlers from the system reset handler.

- The external exception remains raised and will fire again at the
  EE interrupt vector when system reset returns.

- The decrementer is set to 1 so it will be raised again and fire when
  the system reset returns.

It is possible to branch to an idle handler from the system reset
interrupt (like POWER does), then restore a normal stack and restore
this optimisation. But simplicity wins for now.
Tested-by: NMichael Ellerman <mpe@ellerman.id.au>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

PA Semi will wake from low power state at the system reset interrupt,
with the event encoded in SRR1, rather than waking at the interrupt
vector that corresponds to that event.

The system reset handler for this platform decodes SRR1 event reason
and calls the interrupt handler to process it directly from the system
reset handlre.

A subsequent change will treat the system reset interrupt as a Linux NMI
with its own per-CPU stack, and this will no longer work. Remove the
external and decrementer handlers from the system reset handler.

- The external exception remains raised and will fire again at the
  EE interrupt vector when system reset returns.

- The decrementer is set to 1 so it will be raised again and fire when
  the system reset returns.

It is possible to branch to an idle handler from the system reset
interrupt (like POWER does), then restore a normal stack and restore
this optimisation. But simplicity wins for now.
Tested-by: NChristian Zigotzky <chzigotzky@xenosoft.de>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Merge the topic branch we were sharing with kvm-ppc, Paul has also
merged it.

Split ftrace_64.S further retaining the core ftrace 64-bit aspects
in ftrace_64.S and moving ftrace_caller() and ftrace_graph_caller() into
separate files based on -mprofile-kernel. The livepatch routines are all
now contained within the mprofile file.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

entry_*.S now includes a lot more than just kernel entry/exit code. As a
first step at cleaning this up, let's split out the ftrace bits into
separate files. Also move all related tracing code into a new trace/
subdirectory.

No functional changes.
Suggested-by: NMichael Ellerman <mpe@ellerman.id.au>
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Page table dump debugfs file is named 'kernel_page_tables' on
all other architectures implementing it, while is is named
'kernel_pagetables' on powerpc. This patch renames it.
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

On some targets, _PAGE_RW is 0 and this is _PAGE_RO which is used.
There is also _PAGE_SHARED that is missing.
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

On PPC32 (eg. mpc885_ads_defconfig), page table dump compilation fails as
follows. This is because the memory layout is slightly different on PPC32. This
patch adapts it.

  arch/powerpc/mm/dump_linuxpagetables.c: In function 'walk_pagetables':
  arch/powerpc/mm/dump_linuxpagetables.c:369:10: error: 'KERN_VIRT_START' undeclared (first use in this function)
  ...

Fixes: 8eb07b18 ("powerpc/mm: Dump linux pagetables")
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

_tlbiel_pid() is called with a ric (Radix Invalidation Control) argument of
either RIC_FLUSH_TLB or RIC_FLUSH_ALL.

RIC_FLUSH_ALL says to invalidate the entire TLB and the Page Walk Cache (PWC).

To flush the whole TLB, we have to iterate over each set (congruence class) of
the TLB. Currently we do that and pass RIC_FLUSH_ALL each time. That is not
incorrect but it means we flush the PWC 128 times, when once would suffice.

Fix it by doing the first flush with the ric value we're passed, and then if it
was RIC_FLUSH_ALL, we downgrade it to RIC_FLUSH_TLB, because we know we have
just flushed the PWC and don't need to do it again.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Split out of combined patch, tweak logic, rewrite change log]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently we implement flushing of the page walk cache (PWC) by calling
_tlbiel_pid() with a RIC (Radix Invalidation Control) value of 1 which says to
only flush the PWC.

But _tlbiel_pid() loops over each set (congruence class) of the TLB, which is
not necessary when we're just flushing the PWC.

In fact the set argument is ignored for a PWC flush, so essentially we're just
flushing the PWC 127 extra times for no benefit.

Fix it by adding tlbiel_pwc() which just does a single flush of the PWC.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Split out of combined patch, drop _ in name, rewrite change log]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Recently we merged the native xive support for Power9, and then separately some
reworks for doorbell IPI support. In isolation both series were OK, but the
merged result had a bug in one case.

On P9 DD1 we use pnv_p9_dd1_cause_ipi() which tries to use doorbells, and then
falls back to the interrupt controller. However the fallback is implemented by
calling icp_ops->cause_ipi. But now that xive support is merged we might be
using xive, in which case icp_ops is not initialised, it's a xics specific
structure. This leads to an oops such as:

  Unable to handle kernel paging request for data at address 0x00000028
  Oops: Kernel access of bad area, sig: 11 [#1]
  NIP pnv_p9_dd1_cause_ipi+0x74/0xe0
  LR smp_muxed_ipi_message_pass+0x54/0x70

To fix it, rather than using icp_ops which might be NULL, have both xics and
xive set smp_ops->cause_ipi, and then in the powernv code we save that as
ic_cause_ipi before overriding smp_ops->cause_ipi. For paranoia add a WARN_ON()
to check if somehow smp_ops->cause_ipi is NULL.

Fixes: b866cc21 ("powerpc: Change the doorbell IPI calling convention")
Tested-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

In opal_export_attrs() we dynamically allocate some bin_attributes. They're
allocated with kmalloc() and although we initialise most of the fields, we don't
initialise write() or mmap(), and in particular we don't initialise the lockdep
related fields in the embedded struct attribute.

This leads to a lockdep warning at boot:

  BUG: key c0000000f11906d8 not in .data!
  WARNING: CPU: 0 PID: 1 at ../kernel/locking/lockdep.c:3136 lockdep_init_map+0x28c/0x2a0
  ...
  Call Trace:
    lockdep_init_map+0x288/0x2a0 (unreliable)
    __kernfs_create_file+0x8c/0x170
    sysfs_add_file_mode_ns+0xc8/0x240
    __machine_initcall_powernv_opal_init+0x60c/0x684
    do_one_initcall+0x60/0x1c0
    kernel_init_freeable+0x2f4/0x3d4
    kernel_init+0x24/0x160
    ret_from_kernel_thread+0x5c/0xb0

Fix it by kzalloc'ing the attr, which fixes the uninitialised write() and
mmap(), and calling sysfs_bin_attr_init() on it to initialise the lockdep
fields.

Fixes: 11fe909d ("powerpc/powernv: Add OPAL exports attributes to sysfs")
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The recent patch to add runtime configuration of the ASLR limits added a bug in
arch_mmap_rnd() where we may shift an integer (32-bits) by up to 33 bits,
leading to undefined behaviour.

In practice it exhibits as every process seg faulting instantly, presumably
because the rnd value hasn't been restricited by the modulus at all. We didn't
notice because it only happens under certain kernel configurations and if the
number of bits is actually set to a large value.

Fix it by switching to unsigned long.

Fixes: 9fea59bd ("powerpc/mm: Add support for runtime configuration of ASLR limits")
Reported-by: NBalbir Singh <bsingharora@gmail.com>
Reviewed-by: NKees Cook <keescook@chromium.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

powerpc expects IRQs to already be (soft) disabled when switch_mm() is
called, as made clear in the commit message of 9c1e1052 ("powerpc: Allow
perf_counters to access user memory at interrupt time").

Aside from any race conditions that might exist between switch_mm() and an IRQ,
there is also an unconditional hard_irq_disable() in switch_slb(). If that isn't
followed at some point by an IRQ enable then interrupts will remain disabled
until we return to userspace.

It is true that when switch_mm() is called from the scheduler IRQs are off, but
not when it's called by use_mm(). Looking closer we see that last year in commit
f98db601 ("sched/core: Add switch_mm_irqs_off() and use it in the scheduler")
this was made more explicit by the addition of switch_mm_irqs_off() which is now
called by the scheduler, vs switch_mm() which is used by use_mm().

Arguably it is a bug in use_mm() to call switch_mm() in a different context than
it expects, but fixing that will take time.

This was discovered recently when vhost started throwing warnings such as:

  BUG: sleeping function called from invalid context at kernel/mutex.c:578
  in_atomic(): 0, irqs_disabled(): 1, pid: 10768, name: vhost-10760
  no locks held by vhost-10760/10768.
  irq event stamp: 10
  hardirqs last  enabled at (9):  _raw_spin_unlock_irq+0x40/0x80
  hardirqs last disabled at (10): switch_slb+0x2e4/0x490
  softirqs last  enabled at (0):  copy_process+0x5e8/0x1260
  softirqs last disabled at (0):  (null)
  Call Trace:
    show_stack+0x88/0x390 (unreliable)
    dump_stack+0x30/0x44
    __might_sleep+0x1c4/0x2d0
    mutex_lock_nested+0x74/0x5c0
    cgroup_attach_task_all+0x5c/0x180
    vhost_attach_cgroups_work+0x58/0x80 [vhost]
    vhost_worker+0x24c/0x3d0 [vhost]
    kthread+0xec/0x100
    ret_from_kernel_thread+0x5c/0xd4

Prior to commit 04b96e55 ("vhost: lockless enqueuing") (Aug 2016) the
vhost_worker() would do a spin_unlock_irq() not long after calling use_mm(),
which had the effect of reenabling IRQs. Since that commit removed the locking
in vhost_worker() the body of the vhost_worker() loop now runs with interrupts
off causing the warnings.

This patch addresses the problem by making the powerpc code mirror the x86 code,
ie. we disable interrupts in switch_mm(), and optimise the scheduler case by
defining switch_mm_irqs_off().

Cc: stable@vger.kernel.org # v4.7+
Signed-off-by: NDavid Gibson <david@gibson.dropbear.id.au>
[mpe: Flesh out/rewrite change log, add stable]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

For CPUs present at boot each logical CPU acquires a reference to the
associated device node of the core. This happens in register_cpu() which
is called by topology_init(). The result of this is that we end up with
a reference held by each thread of the core. However, these references
are never freed if the CPU core is DLPAR removed.

This patch fixes the reference leaks by acquiring and releasing the references
in the CPU hotplug callbacks un/register_cpu_online(). With this patch symmetric
reference counting is observed with both CPUs present at boot, and those DLPAR
added after boot.

Fixes: f86e4718 ("driver/core: cpu: initialize of_node in cpu's device struture")
Cc: stable@vger.kernel.org # v3.12+
Signed-off-by: NTyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Historically struct device_node references were tracked using a kref embedded as
a struct field. Commit 75b57ecf ("of: Make device nodes kobjects so they
show up in sysfs") (Mar 2014) refactored device_nodes to be kobjects such that
the device tree could by more simply exposed to userspace using sysfs.

Commit 0829f6d1 ("of: device_node kobject lifecycle fixes") (Mar 2014)
followed up these changes to better control the kobject lifecycle and in
particular the referecne counting via of_node_get(), of_node_put(), and
of_node_init().

A result of this second commit was that it introduced an of_node_put() call when
a dynamic node is detached, in of_node_remove(), that removes the initial kobj
reference created by of_node_init().

Traditionally as the original dynamic device node user the pseries code had
assumed responsibilty for releasing this final reference in its platform
specific DLPAR detach code.

This patch fixes a refcount underflow introduced by commit 0829f6d1, and
recently exposed by the upstreaming of the recount API.

Messages like the following are no longer seen in the kernel log with this
patch following DLPAR remove operations of cpus and pci devices.

  rpadlpar_io: slot PHB 72 removed
  refcount_t: underflow; use-after-free.
  ------------[ cut here ]------------
  WARNING: CPU: 5 PID: 3335 at lib/refcount.c:128 refcount_sub_and_test+0xf4/0x110

Fixes: 0829f6d1 ("of: device_node kobject lifecycle fixes")
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: NTyrel Datwyler <tyreld@linux.vnet.ibm.com>
[mpe: Make change log commit references more verbose]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently the code that dumps SLB entries uses a double-nested if. This
means the actual dumping logic is a bit squashed. Deindent it by using
continue.
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>
Reviewed-by: NRashmica Gupta <rashmica.g@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Although most of these kprobes patches are powerpc specific, there's a couple
that touch generic code (with Acks). At the moment there's one conflict with
acme's tree, but it's not too bad. Still just in case some other conflicts show
up, we've put these in a topic branch so another tree could merge some or all of
it if necessary.

KPROBES_ON_FTRACE avoids much of the overhead of regular kprobes as it
eliminates the need for a trap, as well as the need to emulate or single-step
instructions.

Though OPTPROBES provides us with similar performance, we have limited
optprobes trampoline slots. As such, when asked to probe at a function
entry, default to using the ftrace infrastructure.

With:
  # cd /sys/kernel/debug/tracing
  # echo 'p _do_fork' > kprobe_events

before patch:
  # cat ../kprobes/list
  c0000000000daf08  k  _do_fork+0x8    [DISABLED]
  c000000000044fc0  k  kretprobe_trampoline+0x0    [OPTIMIZED]

and after patch:
  # cat ../kprobes/list
  c0000000000d074c  k  _do_fork+0xc    [DISABLED][FTRACE]
  c0000000000412b0  k  kretprobe_trampoline+0x0    [OPTIMIZED]
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

kprobe_lookup_name() is specific to the kprobe subsystem and may not always
return the function entry point (in a subsequent patch for KPROBES_ON_FTRACE).
For looking up function entry points, introduce a separate helper and use it
in optprobes.c
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Allow kprobes to be placed on ftrace _mcount() call sites. This optimization
avoids the use of a trap, by riding on ftrace infrastructure.

This depends on HAVE_DYNAMIC_FTRACE_WITH_REGS which depends on MPROFILE_KERNEL,
which is only currently enabled on powerpc64le with newer toolchains.

Based on the x86 code by Masami.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Pass the real LR to the ftrace handler. This is needed for KPROBES_ON_FTRACE for
the pre handlers.

Also, with KPROBES_ON_FTRACE, the link register may be updated by the pre
handlers or by a registed kretprobe. Honor updated LR by restoring it from
pt_regs, rather than from the stack save area.

Live patch and function graph continue to work fine with this change.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Blacklist all the exception common/OOL handlers as the kernel stack is not yet
setup, which means we can't take a trap at this point.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Introduce __head_end to mark end of the early fixed sections and use it to
blacklist all exception handlers from kprobes.

mpe: We do not need to do anything special for relocatable kernels, where the
exception vectors are split from the main kernel, as the split vectors are
already excluded by the check for kernel_text_address().
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Move __head_end outside #ifdef 64-bit to unbreak the 32-bit build]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Along similar lines as commit 9326638c ("kprobes, x86: Use NOKPROBE_SYMBOL()
instead of __kprobes annotation"), convert __kprobes annotation to either
NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case
the caller needs to be added to NOKPROBE_SYMBOL().

Also:
 - blacklist arch_deref_entry_point(), and
 - convert a few regular inlines to nokprobe_inline in lib/sstep.c

A key benefit is the ability to detect such symbols as being
blacklisted. Before this patch:

  $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem
  $ perf probe read_mem
  Failed to write event: Invalid argument
    Error: Failed to add events.
  $ dmesg | tail -1
  [ 3736.112815] Could not insert probe at _text+10014968: -22

After patch:
  $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem
  0xc000000000072b50-0xc000000000072d20	read_mem
  $ perf probe read_mem
  read_mem is blacklisted function, skip it.
  Added new events:
    (null):(null)        (on read_mem)
    probe:read_mem       (on read_mem)

  You can now use it in all perf tools, such as:

	  perf record -e probe:read_mem -aR sleep 1

  $ grep " read_mem" /proc/kallsyms
  c000000000072b50 t read_mem
  c0000000005f3b40 t read_mem
  $ cat /sys/kernel/debug/kprobes/list
  c0000000005f3b48  k  read_mem+0x8    [DISABLED]
Acked-by: NMasami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Minor change log formatting, fix up some conflicts]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Move the stack setup and teardown code into ftrace_graph_caller(). This way, we
don't incur the cost of setting it up unless function graph is enabled for this
function.

Also, remove the extraneous LR restore code after the function graph stub. LR
has previously been restored and neither livepatch_handler() nor
ftrace_graph_caller() return back here.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Drop bad change to non-mprofile-kernel version of ftrace_graph_caller]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

set_current_kprobe() already saves regs->msr into kprobe_saved_msr. Remove the
redundant save.
Signed-off-by: NNaveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Reviewed-by: NMasami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The idle workaround does not need to load PACATOC, and it does not
need to be called within a nested function that requires LR to be
saved.

Load the PACATOC at entry to the idle wakeup. It does not matter which
PACA this comes from, so it's okay to call before the workaround. Then
apply the workaround to get the right PACA.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

If not all threads were in winkle, full state loss recovery is not
necessary and can be avoided. A previous patch removed this optimisation
due to some complexity with the implementation. Re-implement it by
counting the number of threads in winkle with the per-core idle state.
Only restore full state loss if all threads were in winkle.

This has a small window of false positives right before threads execute
winkle and just after they wake up, when the winkle count does not
reflect the true number of threads in winkle. This is not a significant
problem in comparison with even the minimum winkle duration. For
correctness, a false positive is not a problem (only false negatives
would be).
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When taking the core idle state lock, grab it immediately like a regular
lock, rather than adding more tests in there. Holding the lock keeps it
stable, so there is no need to do it whole holding the reservation.
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

In preparation for adding more bits to the core idle state word, move
the lock bit up, and unlock by flipping the lock bit rather than masking
off all but the thread bits.

Add branch hints for atomic operations while we're here.
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The ISA specifies power save wakeup due to a machine check exception can
cause a machine check interrupt (rather than the usual system reset
interrupt).

The machine check handler copes with this by doing low level machine
check recovery without restoring full state from idle, then queues up a
machine check event for logging, then directly executes the same idle
instruction it woke from. This minimises the work done before recovery
is performed.

The problem is that it requires machine specific instructions and
knowledge of the book3s idle code. Currently it only has code to handle
POWER8 idle, so POWER9 crashes when trying to execute the P8 idle
instructions which don't exist in ISAv3.0B.

cpu 0x0: Vector: e40 (Emulation Assist) at [c0000000008f3810]
    pc: c000000000008380: machine_check_handle_early+0x130/0x2f0
    lr: c00000000053a098: stop_loop+0x68/0xd0
    sp: c0000000008f3a90
   msr: 9000000000081001
  current = 0xc0000000008a1080
  paca    = 0xc00000000ffd0000   softe: 0        irq_happened: 0x01
    pid   = 0, comm = swapper/0

Instead of going to sleep after recovery, do the usual idle wakeup and
state restoration by calling into the normal idle wakeup path. This
reuses the normal idle wakeup paths.
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Reviewed-by: NMahesh J Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This reduces the number of nops for POWER8.
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The POWER8 idle code has a neat trick of programming the power on engine
to restore a low bit into HSPRG0, so idle wakeup code can test and see
if it has been programmed this way and therefore lost all state. Restore
time can be reduced if winkle has not been reached.

However this messes with our r13 PACA pointer, and requires HSPRG0 to be
written to. It also optimizes the slowest and most uncommon case at the
expense of another SPR write in the common nap state wakeup.

Remove this complexity and assume winkle sleeps always require a state
restore. This speedup could be made entirely contained within the winkle
idle code by counting per-core winkles and setting a thread bitmap when
all have gone to winkle.
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>