arm_pmu interrupts are maked as PERCPU even when these are not local
physical interrupts to a single CPU. When using non-local interrupts,
interrupts marked as PERCPU will not get freed not disabled properly
by the PMU driver.

Check if interrupts are local to a single CPU with PERCPU_DEVID since
this is what the PMU driver really needs to know.
Acked-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NJulien Thierry <julien.thierry@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Rather than continue adding CPU-specific event maps, instead look up by
default in the PMUv3 event map and only fallback to the CPU-specific maps
if either the event isn't described by PMUv3, or it is described but
the PMCEID registers say that it is unsupported by the current CPU.
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Since the PMU register interface is banked per CPU, CPU PMU interrrupts
cannot be handled by a CPU other than the one with the PMU asserting the
interrupt. This means that migrating PMU SPIs, as we do during a CPU
hotplug operation doesn't make any sense and can lead to the IRQ being
disabled entirely if we route a spurious IRQ to the new affinity target.

This has been observed in practice on AMD Seattle, where CPUs on the
non-boot cluster appear to take a spurious PMU IRQ when coming online,
which is routed to CPU0 where it cannot be handled.

This patch passes IRQF_PERCPU for PMU SPIs and forcefully sets their
affinity prior to requesting them, ensuring that they cannot
be migrated during hotplug events. This interacts badly with the DB8500
erratum workaround that ping-pongs the interrupt affinity from the handler,
so we avoid passing IRQF_PERCPU in that case by allowing the IRQ flags
to be overridden in the platdata.

Fixes: 3cf7ee98 ("drivers/perf: arm_pmu: move irq request/free into probe")
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

This patch adds framework code to handle parsing PMU data out of the
MADT, sanity checking this, and managing the association of CPUs (and
their interrupts) with appropriate logical PMUs.

For the time being, we expect that only one PMU driver (PMUv3) will make
use of this, and we simply pass in a single probe function.

This is based on an earlier patch from Jeremy Linton.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Now that we've split the pdev and DT probing logic from the runtime
management, let's move the former into its own file. We gain a few lines
due to the copyright header and includes, but this should keep the logic
clearly separated, and paves the way for adding ACPI support in a
similar fashion.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
[will: rename nr_irqs to avoid conflict with global variable]
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Currently we request (and potentially free) all IRQs for a given PMU in
cpu_pmu_init(). This works for platform/DT probing today, but it doesn't
fit ACPI well as we don't have all our affinity data up-front.

In preparation for ACPI support, fold the IRQ request/free into
arm_pmu_device_probe(), which will remain specific to platform/DT
probing.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Currently we have functions to request/free all IRQs for a given PMU.
While this works today, this won't work for ACPI, where we don't know
the full set of IRQs up front, and need to request them separately.

To enable supporting ACPI, this patch splits out the cpu-local
request/free into new functions, allowing us to request/free individual
IRQs.

As this makes it possible/necessary to request a PPI once per cpu, an
additional check is added to detect mismatched PPIs. This shouldn't
matter for the DT / platform case, as we check this when parsing.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

For historical reasons, portions of the arm_pmu code use a cpu_pmu_
prefix rather than an armpmu_ prefix. While a minor annoyance, this
hasn't been a problem thusfar.

However, to enable ACPI support, we'll need to expose a few things in
header files, and we should aim to keep those consistently namespaced.
In preparation for exporting our IRQ request/free functions, rename
these to have an armpmu_ prefix. For consistency, the 'cpu_pmu'
parameter is also renamed to 'armpmu'.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

In armpmu_dispatch_irq() we look at arm_pmu::plat_device to acquire
platdata, so that we can defer to platform-specific IRQ handling,
required on some 32-bit parts. With the advent of ACPI we won't always
have a platform_device, and so we must avoid trying to dereference
fields from it.

This patch fixes up armpmu_dispatch_irq() to avoid doing so, introducing
a new armpmu_get_platdata() helper.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

The ARM PMU framework code always uses armpmu_dispatch_irq as its common
IRQ handler. Passing this down from cpu_pmu_init() is somewhat
pointless, and gets in the way of refactoring.

This patch makes cpu_pmu_request_irqs() always use armpmu_dispatch_irq
as the handler when requesting IRQs, and removes the handler parameter
from its prototype.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Currently arm_pmu_device_probe contains probing logic specific to the
platform_device infrastructure, and some logic required to safely
register the PMU with various systems.

This patch factors out the logic relating to the registration of the
PMU. This makes arm_pmu_device_probe a little easier to read, and will
make it easier to reuse the logic for an ACPI-specific probing
mechanism.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Given we always want to initialise common fields on an allocated PMU,
this patch folds this common initialisation into armpmu_alloc(). This
will make it simpler to reuse this code for an ACPI-specific probe path.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

We expect an ARM PMU's init function to have a particular prototype,
which we open-code in a few places. This is less than ideal, considering
that we cast a void value to this type in one location, and a mismatch
could easily be missed.

Add a typedef so that we can ensure this is consistent.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

We currently disable the PMU temporarily in armpmu_add(). We may have
required this historically, but the perf core always disables an event's
PMU when calling event::pmu::add(), so this is not necessary.

We don't do similarly in armpmu_del(), or elsewhere, so this is
unnecessary and inconsistent, and only serves to confuse the reader.

Remove the pointless disable, simplifying armpmu_add() in the process.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NJeremy Linton <jeremy.linton@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

For historical reasons, we lazily request and free interrupts in the
arm pmu driver. This requires us to refcount use of the pmu (by way of
counting the active events) in order to request/free interrupts at the
correct times, which complicates the driver somewhat.

The existing logic is flawed, as it only considers currently online CPUs
when requesting, freeing, or managing the affinity of interrupts.
Intervening hotplug events can result in erroneous IRQ affinity, online
CPUs for which interrupts have not been requested, or offline CPUs whose
interrupts are still requested.

To fix this, this patch splits the requesting of interrupts from any
per-cpu management (i.e. per-cpu enable/disable, and configuration of
cpu affinity). We now request all interrupts up-front at probe time (and
never free them, since we never unregister PMUs).

The management of affinity, and per-cpu enable/disable now happens in
our cpu hotplug callback, ensuring it occurs consistently. This means
that we must now invoke the CPU hotplug callback at boot time in order
to configure IRQs, and since the callback also resets the PMU hardware,
we can remove the duplicate reset in the probe path.

This rework renders our event refcounting unnecessary, so this is
removed.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
[will: make armpmu_get_cpu_irq static]
Signed-off-by: NWill Deacon <will.deacon@arm.com>

When requesting or freeing interrupts, we use platform_get_irq() to find
relevant irqs, backing this up with additional information in an
optional irq_affinity table.

This means that our irq request and free paths are tied to a
platform_device, and our request path must jump through a number of
hoops in order to determine the required affinity of each interrupt.

Given that the affinity must be static, we can compute the affinity once
up-front at probe time, simplifying the irq request and free paths. By
recording interrupts in a per-cpu data structure, we simplify a few
paths, and permit a subsequent rework of the request and free paths.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
[will: rename local nr_irqs variable to avoid conflict with global]
Signed-off-by: NWill Deacon <will.deacon@arm.com>

For historical reasons, we allocate per-cpu data associated with a PMU
rather late, in cpu_pmu_init, after we've parsed whatever hardware
information we were provided with.

In order to allow use to store some per-cpu data early in the probe
path, we need to allocate (and initialise) the per-cpu data earlier.
This patch reworks the way we allocate the pmu and associated per-cpu
data in order to make that possible.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
[will: make armpmu_{alloc,free} static
Signed-off-by: NWill Deacon <will.deacon@arm.com>

We are going to split <linux/sched/clock.h> out of <linux/sched.h>, which
will have to be picked up from other headers and .c files.

Create a trivial placeholder <linux/sched/clock.h> file that just
maps to <linux/sched.h> to make this patch obviously correct and
bisectable.

Include the new header in the files that are going to need it.
Acked-by: NLinus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: NIngo Molnar <mingo@kernel.org>

When the state names got added a script was used to add the extra argument
to the calls. The script basically converted the state constant to a
string, but the cleanup to convert these strings into meaningful ones did
not happen.

Replace all the useless strings with 'subsys/xxx/yyy:state' strings which
are used in all the other places already.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Link: http://lkml.kernel.org/r/20161221192112.085444152@linutronix.deSigned-off-by: NThomas Gleixner <tglx@linutronix.de>

In preparation for ACPI support, add a pmu_probe_info table to
the arm_pmu_device_probe() call. This table gets used when
probing in the absence of a devicetree node for PMU.
Signed-off-by: NMark Salter <msalter@redhat.com>
Signed-off-by: NJeremy Linton <jeremy.linton@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

In systems with heterogeneous CPUs, there are multiple logical CPU PMUs,
each of which covers a subset of CPUs in the system. In some cases
userspace needs to know which CPUs a given logical PMU covers, so we'd
like to expose a cpumask under sysfs, similar to what is done for uncore
PMUs.

Unfortunately, prior to commit 00e727bb ("perf stat: Balance
opening and reading events"), perf stat only correctly handled a cpumask
holding a single CPU, and only when profiling in system-wide mode. In
other cases, the presence of a cpumask file could cause perf stat to
behave erratically.

Thus, exposing a cpumask file would break older perf binaries in cases
where they would otherwise work.

To avoid this issue while still providing userspace with the information
it needs, this patch exposes a differently-named file (cpus) under
sysfs. New tools can look for this and operate correctly, while older
tools will not be adversely affected by its presence.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Now that the 32-bit and 64-bit perf backends use the common groups
directly, remove the fallback and no longer allow the groups array to be
overridden.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

In preparation for adding common attribute groups, add an array of
attribute group pointers to arm_pmu, which will be used if the
backend hasn't already set pmu::attr_groups.

Subsequent patches will move backends over to using these, before adding
common fields.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

As declared by the chief penguin, and enforced by the NO_IRQ brigade,
IRQ0 doesn't exist, and is considered as an error (no irq).

Unfortunately, the arm_pmu driver still considers it as valid in
a large number of cases. Let's fix this.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Signed-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: rt@linutronix.de
Link: http://lkml.kernel.org/r/20160817171420.sdwk2qivxunzryz4@linutronix.deSigned-off-by: NThomas Gleixner <tglx@linutronix.de>

Patch 7f1d642f ("drivers/perf: arm-pmu: Fix handling of SPI lacking
interrupt-affinity property") unintended also fixes perf_event support
for bcm2835 which doesn't have PMU interrupts. Unfortunately this change
introduce a NULL pointer dereference on bcm2835, because irq_is_percpu
always expected to be called with a valid IRQ. So fix this regression
by validating the IRQ before.
Tested-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NStefan Wahren <stefan.wahren@i2se.com>
Fixes: 7f1d642f ("drivers/perf: arm-pmu: Fix handling of SPI lacking "interrupt-affinity" property")
Signed-off-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>

In case of a IRQ type mismatch in of_pmu_irq_cfg() the
device node for interrupt affinity isn't freed. So fix this
issue by calling of_node_put().
Signed-off-by: NStefan Wahren <stefan.wahren@i2se.com>
Fixes: fa8ad788 ("arm: perf: factor arm_pmu core out to drivers")
Signed-off-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>

Patch 19a469a5 ("drivers/perf: arm-pmu: Handle per-interrupt
affinity mask") added support for partitionned PPI setups, but
inadvertently broke setups using SPIs without the "interrupt-affinity"
property (which is the case for UP platforms).

This patch restore the broken functionnality by testing whether the
interrupt is percpu or not instead of relying on the using_spi flag
that really means "SPI *and* interrupt-affinity property".
Acked-by: NMark Rutland <mark.rutland@arm.com>
Reported-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Tested-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Fixes: 19a469a5 ("drivers/perf: arm-pmu: Handle per-interrupt affinity mask")
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

arm_pmu_mutex is never held long and we don't want to sleep while the
lock is being held as it's executed in the context of hotplug notifiers.
So it can be converted to a simple spinlock instead.

Without this patch we get the following warning:

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:620
in_atomic(): 1, irqs_disabled(): 128, pid: 0, name: swapper/2
no locks held by swapper/2/0.
irq event stamp: 381314
hardirqs last  enabled at (381313): _raw_spin_unlock_irqrestore+0x7c/0x88
hardirqs last disabled at (381314): cpu_die+0x28/0x48
softirqs last  enabled at (381294): _local_bh_enable+0x28/0x50
softirqs last disabled at (381293): irq_enter+0x58/0x78
CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.7.0 #12
Call trace:
 dump_backtrace+0x0/0x220
 show_stack+0x24/0x30
 dump_stack+0xb4/0xf0
 ___might_sleep+0x1d8/0x1f0
 __might_sleep+0x5c/0x98
 mutex_lock_nested+0x54/0x400
 arm_perf_starting_cpu+0x34/0xb0
 cpuhp_invoke_callback+0x88/0x3d8
 notify_cpu_starting+0x78/0x98
 secondary_start_kernel+0x108/0x1a8

This patch converts the mutex to spinlock to eliminate the above
warnings. This constraints pmu->reset to be non-blocking call which is
the case with all the ARM PMU backends.

Cc: Stephen Boyd <sboyd@codeaurora.org>
Fixes: 37b502f1 ("arm/perf: Fix hotplug state machine conversion")
Acked-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NSudeep Holla <sudeep.holla@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Mark Rutland pointed out that this commit is incomplete:

  7d88eb69 ("arm/perf: Convert to hotplug state machine")

The problem is that:

 > We may have multiple PMUs (e.g. two in big.LITTLE systems), and
 > __oprofile_cpu_pmu only contains one of these. So this conversion is not
 > correct.
 >
 > We were relying on the notifier list implicitly containing a list of
 > those PMUs. It seems like we need an explicit list here.
 >
 > We keep __oprofile_cpu_pmu around for legacy 32-bit users of OProfile
 > (on non-hetereogeneous systems), and that's all that the variable should
 > be used for.

Introduce arm_pmu_list to correctly handle multiple PMUs in the system.
Signed-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: NMark Rutland <mark.rutland@arm.com>
Cc: Anna-Maria Gleixner <anna-maria@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-tip-commits@vger.kernel.org
Cc: rt@linutronix.de
Link: http://lkml.kernel.org/r/20160719111733.GA22911@linutronix.deSigned-off-by: NIngo Molnar <mingo@kernel.org>

Straight forward conversion w/o bells and whistles.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Signed-off-by: NAnna-Maria Gleixner <anna-maria@linutronix.de>
Reviewed-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: rt@linutronix.de
Link: http://lkml.kernel.org/r/20160713153335.794097159@linutronix.deSigned-off-by: NIngo Molnar <mingo@kernel.org>

On a big-little system, PMUs can be wired to CPUs using per CPU
interrups (PPI). In this case, it is important to make sure that
the enable/disable do happen on the right set of CPUs.

So instead of relying on the interrupt-affinity property, we can
use the actual percpu affinity that DT exposes as part of the
interrupt specifier. The DT binding is also updated to reflect
the fact that the interrupt-affinity property shouldn't be used
in that case.
Acked-by: NRob Herring <robh@kernel.org>
Tested-by: NCaesar Wang <wxt@rock-chips.com>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>

There is a problem in the non-devicetree PMU probing where some
probe functions may get the number of supported events through
smp_call_function_any() using the arm_pmu supported_cpus mask.
But at the time the probe function is called, the supported_cpus
mask is empty so the call fails. This patch makes sure the mask
is set before calling the init function rather than after.
Signed-off-by: NMark Salter <msalter@redhat.com>
Signed-off-by: NJeremy Linton <jeremy.linton@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

pmu->irq_affinity will not be freed if an error occurred within
arm_pmu_device_probe after of_pmu_irq_cfg has been called.

Note that in the case of_pmu_irq_cfg is returning an error,
pmu->irq_affinity will not be set, but it should be NULL as pmu was
kzalloc'd. Therefore the result kfree(NULL) is benign.
Signed-off-by: NJulien Grall <julien.grall@arm.com>
Acked-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

The global variable __oprofile_cpu_pmu is set before the PMU is fully
initialized. If an error occurs before the end of the initialization,
the PMU will be freed and the variable will contain an invalid pointer.

This will result in a kernel crash when perf will be used.

Fix it by moving the setting of __oprofile_cpu_pmu when the PMU is fully
initialized (i.e when it is no longer possible to fail).

Cc: <stable@vger.kernel.org>
Signed-off-by: NJulien Grall <julien.grall@arm.com>
Acked-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

The only function called by of_pmu_irq_cfg that will increment the
reference count on dn is of_parse_phandle.

Each time we successfully parse a possible CPU from an
interrupt-affinity property, we increment the refcount of that CPU node
once via of_parse_handle. After validating the CPU is possible, we
decrement the refcount once. Subsequently, we decrement the refcount
again, either as part of an early break if we don't have a matching SPI,
or as part of the end of the loop body.

This will lead to decrementing twice the refcounnt.
Remove the second pairs of call to of_node_put as nobody is using dn
between the first and second call to of_node_put.
Signed-off-by: NJulien Grall <julien.grall@arm.com>
Acked-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Commit:

  26657848 ("perf/core: Verify we have a single perf_hw_context PMU")

forcefully prevents multiple PMUs from sharing perf_hw_context, as this
generally doesn't make sense. It is a common bug for uncore PMUs to
use perf_hw_context rather than perf_invalid_context, which this detects.

However, systems exist with heterogeneous CPUs (and hence heterogeneous
HW PMUs), for which sharing perf_hw_context is necessary, and possible
in some limited cases.

To make this work we have to perform some gymnastics, as we did in these
commits:

  66eb579e ("perf: allow for PMU-specific event filtering")
  c904e32a ("arm: perf: filter unschedulable events")

To allow those systems to work, we must allow PMUs for heterogeneous
CPUs to share perf_hw_context, though we must still disallow sharing
otherwise to detect the common misuse of perf_hw_context.

This patch adds a new PERF_PMU_CAP_HETEROGENEOUS_CPUS for this, updates
the core logic to account for this, and makes use of it in the arm_pmu
code that is used for systems with heterogeneous CPUs. Comments are
added to make the rationale clear and hopefully avoid accidental abuse.
Signed-off-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/20160426103346.GA20836@leverpostejSigned-off-by: NIngo Molnar <mingo@kernel.org>

Commit da4e4f18 ("drivers/perf: arm_pmu: implement CPU_PM notifier")
added code in the arm perf infrastructure that allows the kernel to
save/restore perf counters whenever the CPU enters a low-power
state. The kernel saves/restores the counters for each active event
through the armpmu_{stop/start} ARM pmu API, so that the low-power state
enter/exit cycle is emulated through pmu start/stop operations for each
event in use.

However, calling armpmu_start() for each active event on power up
executes code that requires RCU locking (perf_event_update_userpage())
to be functional, so, given that the core may call the CPU_PM notifiers
while running the idle thread in an quiescent RCU state this is not
allowed as detected through the following splat when kernel is run with
CONFIG_PROVE_LOCKING enabled:

[   49.293286]
[   49.294761] ===============================
[   49.298895] [ INFO: suspicious RCU usage. ]
[   49.303031] 4.6.0-rc3+ #421 Not tainted
[   49.306821] -------------------------------
[   49.310956] include/linux/rcupdate.h:872 rcu_read_lock() used
illegally while idle!
[   49.318530]
[   49.318530] other info that might help us debug this:
[   49.318530]
[   49.326451]
[   49.326451] RCU used illegally from idle CPU!
[   49.326451] rcu_scheduler_active = 1, debug_locks = 0
[   49.337209] RCU used illegally from extended quiescent state!
[   49.342892] 2 locks held by swapper/2/0:
[   49.346768]  #0:  (cpu_pm_notifier_lock){......}, at:
[<ffffff8008163c28>] cpu_pm_exit+0x18/0x80
[   49.355492]  #1:  (rcu_read_lock){......}, at: [<ffffff800816dc38>]
perf_event_update_userpage+0x0/0x260

This patch wraps the armpmu_start() call (that indirectly calls
perf_event_update_userpage()) on CPU_PM notifier power state exit (or
failed entry) within the RCU_NONIDLE() macro so that the RCU subsystem
is made aware the calling cpu is not idle from an RCU perspective for
the armpmu_start() call duration, therefore fixing the issue.

Fixes: da4e4f18 ("drivers/perf: arm_pmu: implement CPU_PM notifier")
Signed-off-by: NLorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Reported-by: NJames Morse <james.morse@arm.com>
Suggested-by: NKevin Hilman <khilman@baylibre.com>
Cc: Ashwin Chaugule <ashwin.chaugule@linaro.org>
Cc: Kevin Hilman <khilman@baylibre.com>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Acked-by: NMark Rutland <mark.rutland@arm.com>
Acked-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>

Commit c6b90653 ("drivers/perf: arm_pmu: make info messages more
verbose") breaks booting on systems where the PMU is probed without
devicetree (e.g by inspecting the MIDR of the current CPU). In this case,
pdev->dev.of_node is NULL and we shouldn't try to access its ->fullname
field when printing probe error messages.

This patch fixes the probing code to use of_node_full_name, which safely
handles NULL nodes and removes the "Error %i" part of the string, since
it's not terribly useful.
Reported-by: NGuenter Roeck <private@roeck-us.net>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

When a CPU is suspended (either through suspend-to-RAM or CPUidle),
its PMU registers content can be lost, which means that counters
registers values that were initialized on power down entry have to be
reprogrammed on power-up to make sure the counters set-up is preserved
(ie on power-up registers take the reset values on Cold or Warm reset,
which can be architecturally UNKNOWN).

To guarantee seamless profiling conditions across a core power down
this patch adds a CPU PM notifier to ARM pmus, that upon CPU PM
entry/exit from low-power states saves/restores the pmu registers
set-up (by using the ARM perf API), so that the power-down/up cycle does
not affect the perf behaviour (apart from a black-out period between
power-up/down CPU PM notifications that is unavoidable).

Cc: Will Deacon <will.deacon@arm.com>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Acked-by: NAshwin Chaugule <ashwin.chaugule@linaro.org>
Acked-by: NKevin Hilman <khilman@baylibre.com>
Signed-off-by: NLorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>