linux-2.6-microblaze.git
7 years agocpufreq: intel_pstate: Do not reinit performance limits in ->setpolicy
Rafael J. Wysocki [Thu, 2 Mar 2017 22:29:12 +0000 (23:29 +0100)]
cpufreq: intel_pstate: Do not reinit performance limits in ->setpolicy

If the current P-state selection algorithm is set to "performance"
in intel_pstate_set_policy(), the limits may be initialized from
scratch, but only if no_turbo is not set and the maximum frequency
allowed for the given CPU (i.e. the policy object representing it)
is at least equal to the max frequency supported by the CPU.  In all
of the other cases, the limits will not be updated.

For example, the following can happen:

 # cat intel_pstate/status
 active
 # echo performance > cpufreq/policy0/scaling_governor
 # cat intel_pstate/min_perf_pct
 100
 # echo 94 > intel_pstate/min_perf_pct
 # cat intel_pstate/min_perf_pct
 100
 # cat cpufreq/policy0/scaling_max_freq
 3100000
 echo 3000000 > cpufreq/policy0/scaling_max_freq
 # cat intel_pstate/min_perf_pct
 94
 # echo 95 > intel_pstate/min_perf_pct
 # cat intel_pstate/min_perf_pct
 95

That is confusing for two reasons.  First, the initial attempt to
change min_perf_pct to 94 seems to have no effect, even though
setting the global limits should always work.  Second, after
changing scaling_max_freq for policy0 the global min_perf_pct
attribute shows 94, even though it should have not been affected
by that operation in principle.

Moreover, the final attempt to change min_perf_pct to 95 worked
as expected, because scaling_max_freq for the only policy with
scaling_governor equal to "performance" was different from the
maximum at that time.

To make all that confusion go away, modify intel_pstate_set_policy()
so that it doesn't reinitialize the limits at all.

At the same time, change intel_pstate_set_performance_limits() to
set min_sysfs_pct to 100 in the "performance" limits set so that
switching the P-state selection algorithm to "performance" causes
intel_pstate/min_perf_pct in sysfs to go to 100 (or whatever value
min_sysfs_pct in the "performance" limits is set to later).

That requires per-CPU limits to be initialized explicitly rather
than by copying the global limits to avoid setting min_sysfs_pct
in the per-CPU limits to 100.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Fix intel_pstate_verify_policy()
Rafael J. Wysocki [Tue, 28 Feb 2017 23:11:05 +0000 (00:11 +0100)]
cpufreq: intel_pstate: Fix intel_pstate_verify_policy()

The code added to intel_pstate_verify_policy() by commit 1443ebbacfd7
(cpufreq: intel_pstate: Fix sysfs limits enforcement for performance
policy) should use perf_limits instead of limits, because otherwise
setting global limits via sysfs may affect policies inconsistently.

For example, in the sequence of shell commands below, the
scaling_min_freq attribute for policy1 and policy2 should be
affected in the same way, because scaling_governor is set in
the same way for both of them:

 # cat cpufreq/policy1/scaling_governor
 powersave
 # cat cpufreq/policy2/scaling_governor
 powersave
 # echo performance > cpufreq/policy0/scaling_governor
 # echo 94 > intel_pstate/min_perf_pct
 # cat cpufreq/policy0/scaling_min_freq
 2914000
 # cat cpufreq/policy1/scaling_min_freq
 2914000
 # cat cpufreq/policy2/scaling_min_freq
 800000

The are affected differently, because intel_pstate_verify_policy()
is invoked with limits set to &performance_limits (left behind by
policy0) for policy1 and with limits set to &powersave_limits (left
behind by policy1) for policy2.  Since perf_limits is set to the
set of limits matching the policy being updated, using it instead
of limits fixes the inconsistency.

Fixes: 1443ebbacfd7 (cpufreq: intel_pstate: Fix sysfs limits enforcement for performance policy)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Fix global settings in active mode
Rafael J. Wysocki [Tue, 28 Feb 2017 23:07:36 +0000 (00:07 +0100)]
cpufreq: intel_pstate: Fix global settings in active mode

Commit 111b8b3fe4fa (cpufreq: intel_pstate: Always keep all
limits settings in sync) changed intel_pstate to invoke
cpufreq_update_policy() for every registered CPU on global sysfs
attributes updates, but that led to undesirable effects in the
active mode if the "performance" P-state selection algorithm is
configufred for one CPU and the "powersave" one is chosen for
all of the other CPUs.

Namely, in that case, the following is possible:

 # cd /sys/devices/system/cpu/
 # cat intel_pstate/max_perf_pct
 100
 # cat intel_pstate/min_perf_pct
 26
 # echo performance > cpufreq/policy0/scaling_governor
 # cat intel_pstate/max_perf_pct
 100
 # cat intel_pstate/min_perf_pct
 100
 # echo 94 > intel_pstate/min_perf_pct
 # cat intel_pstate/min_perf_pct
 26

The reason why this happens is because intel_pstate attempts to
maintain two sets of global limits in the active mode, one for
the "performance" P-state selection algorithm and one for the
"powersave"  P-state selection algorithm, but the P-state selection
algorithms are set per policy, so the global limits cannot reflect
all of them at the same time if they are different for different
policies.

In the particular situation above, the attempt to change
min_perf_pct to 94 caused cpufreq_update_policy() to be run
for a CPU with the "powersave"  P-state selection algorithm
and intel_pstate_set_policy() called by it silently switched the
global limits to the "powersave" set which finally was reflected
by the sysfs interface.

To prevent that from happening, modify intel_pstate_update_policies()
to always switch back to the set of limits that was used right before
it has been invoked.

Fixes: 111b8b3fe4fa (cpufreq: intel_pstate: Always keep all limits settings in sync)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Add the "cpufreq.off=1" cmdline option
Len Brown [Tue, 28 Feb 2017 21:44:16 +0000 (16:44 -0500)]
cpufreq: Add the "cpufreq.off=1" cmdline option

Add the "cpufreq.off=1" cmdline option.

At boot-time, this allows a user to request CONFIG_CPU_FREQ=n
behavior from a kernel built with CONFIG_CPU_FREQ=y.

This is analogous to the existing "cpuidle.off=1" option
and CONFIG_CPU_IDLE=y

This capability is valuable when we need to debug end-user
issues in the BIOS or in Linux.  It is also convenient
for enabling comparisons, which may otherwise require a new kernel,
or help from BIOS SETUP, which may be buggy or unavailable.

Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Avoid triggering cpu_frequency tracepoint unnecessarily
Rafael J. Wysocki [Fri, 3 Mar 2017 22:51:31 +0000 (23:51 +0100)]
cpufreq: intel_pstate: Avoid triggering cpu_frequency tracepoint unnecessarily

In the passive mode the cpu_frequency trace event is already
triggered by the cpufreq core or by scaling governors, so
intel_pstate should not trigger it once again for the same
P-state updates.

In addition to that, the frequency returned by
intel_cpufreq_fast_switch() and passed via freqs.new from
intel_cpufreq_target() to cpufreq_freq_transition_end() should
reflect the P-state actually set, so make that happen.

Fixes: 001c76f05b01 (cpufreq: intel_pstate: Generic governors support)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Fix intel_cpufreq_verify_policy()
Rafael J. Wysocki [Thu, 2 Mar 2017 22:24:36 +0000 (23:24 +0100)]
cpufreq: intel_pstate: Fix intel_cpufreq_verify_policy()

The intel_pstate_update_perf_limits() called from
intel_cpufreq_verify_policy() may cause global P-state limits
to change which is generally confusing and unnecessary.

In the passive mode the global limits are only applied to the
frequency selected by the scaling governor (they are not taken
into account by governors when making decisions anyway), so making
them follow the per-policy limits serves no purpose and may go
against user expectations (as it generally causes the global
attributes in sysfs to change even though they have not been
written to in some cases).

Fix that by dropping the intel_pstate_update_perf_limits()
invocation from intel_cpufreq_verify_policy() (which also
reduces the code size by a few lines).

This change does not affect the per-CPU limits case, because those
limits allow any P-state to be set by default in the passive mode
and it removes the only piece of code updating them in that mode,
so the per-policy settings will be the only ones taken into account
in that case as expected.

Fixes: 001c76f05b01 (cpufreq: intel_pstate: Generic governors support)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Do not use performance_limits in passive mode
Rafael J. Wysocki [Thu, 2 Mar 2017 22:22:29 +0000 (23:22 +0100)]
cpufreq: intel_pstate: Do not use performance_limits in passive mode

Using performance_limits in the passive mode doesn't make
sense, because in that mode the global limits are applied to the
frequency selected by the scaling governor.

The maximum and minimum P-state limits in performance_limits are both
set to 100 percent which will put all CPUs into the turbo range
regardless of what governor is used and what frequencies are
selected by it (that is particularly undesirable on CPUs with the
generic powersave governor attached).

For this reason, make intel_pstate_register_driver() always point
limits to powersave_limits in the passive mode.

Fixes: 001c76f05b01 (cpufreq: intel_pstate: Generic governors support)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Fix limits issue with operation mode switching
Rafael J. Wysocki [Mon, 27 Feb 2017 23:05:01 +0000 (00:05 +0100)]
cpufreq: intel_pstate: Fix limits issue with operation mode switching

There is a problem with intel_pstate operation mode switching
introduced by commit fb1fe1041c04 (cpufreq: intel_pstate: Operation
mode control from sysfs), because the global sysfs limits are
preserved across operation modes while per-policy limits are
reinitialized from scratch on a mode switch and both sets of limits
may get out of sync this way.

Fix that by always reinitializing the global limits upon the
registration of the driver.

Fixes: fb1fe1041c04 (cpufreq: intel_pstate: Operation mode control from sysfs)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
7 years agocpufreq: qoriq: clean up unused code
Tang Yuantian [Tue, 21 Feb 2017 06:51:24 +0000 (14:51 +0800)]
cpufreq: qoriq: clean up unused code

This snip code is not needed anymore since its user
get_hard_smp_processor_id() has been removed.

Signed-off-by: Tang Yuantian <yuantian.tang@nxp.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMAINTAINERS: cpufreq: add bmips-cpufreq.c
Markus Mayer [Fri, 17 Feb 2017 20:27:03 +0000 (12:27 -0800)]
MAINTAINERS: cpufreq: add bmips-cpufreq.c

Add maintainer information for bmips-cpufreq.c.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: CPPC: add ACPI_PROCESSOR dependency
Arnd Bergmann [Tue, 14 Feb 2017 21:19:18 +0000 (22:19 +0100)]
cpufreq: CPPC: add ACPI_PROCESSOR dependency

Without the Kconfig dependency, we can get this warning:

warning: ACPI_CPPC_CPUFREQ selects ACPI_CPPC_LIB which has unmet direct dependencies (ACPI && ACPI_PROCESSOR)

Fixes: 5477fb3bd1e8 (ACPI / CPPC: Add a CPUFreq driver for use with CPPC)
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: make ti-cpufreq explicitly non-modular
Paul Gortmaker [Tue, 14 Feb 2017 00:45:02 +0000 (19:45 -0500)]
cpufreq: make ti-cpufreq explicitly non-modular

The Kconfig currently controlling compilation of this code is:

drivers/cpufreq/Kconfig.arm:config ARM_TI_CPUFREQ
drivers/cpufreq/Kconfig.arm:    bool "Texas Instruments CPUFreq support"

...meaning that it currently is not being built as a module by anyone.

Lets remove the couple traces of modular infrastructure use, so that
when reading the driver there is no doubt it is builtin-only.

Since module_init translates to device_initcall in the non-modular
case, the init ordering remains unchanged with this commit.

We also delete the MODULE_LICENSE tag etc. since all that information
is already contained at the top of the file in the comments.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Do not clear real_cpus mask on policy init
Rafael J. Wysocki [Mon, 13 Feb 2017 12:44:57 +0000 (13:44 +0100)]
cpufreq: Do not clear real_cpus mask on policy init

If new_policy is set in cpufreq_online(), the policy object has just
been created and its real_cpus mask has been zeroed on allocation,
and the driver's ->init() callback should not touch it.

It doesn't need to be cleared again, so don't do that.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
7 years agocpufreq: dt: Don't use generic platdev driver for ti-cpufreq platforms
Dave Gerlach [Fri, 3 Feb 2017 17:29:29 +0000 (11:29 -0600)]
cpufreq: dt: Don't use generic platdev driver for ti-cpufreq platforms

Some TI platforms, specifically those in the am33xx, am43xx, dra7xx, and
am57xx families of SoCs can make use of the ti-cpufreq driver to
selectively enable OPPs based on the exact configuration in use. The
ti-cpufreq is given the responsibility of creating the cpufreq-dt
platform device when the driver is in use so drop am33xx and dra7xx
from the cpufreq-dt-platdev driver so it is not created twice.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: ti: Add cpufreq driver to determine available OPPs at runtime
Dave Gerlach [Fri, 3 Feb 2017 17:29:28 +0000 (11:29 -0600)]
cpufreq: ti: Add cpufreq driver to determine available OPPs at runtime

Some TI SoCs, like those in the AM335x, AM437x, DRA7x, and AM57x families,
have different OPPs available for the MPU depending on which specific
variant of the SoC is in use. This can be determined through use of the
revision and an eFuse register present in the silicon. Introduce a
ti-cpufreq driver that can read the aformentioned values and provide
them as version matching data to the opp framework. Through this the
opp-supported-hw dt binding that is part of the operating-points-v2
table can be used to indicate availability of OPPs for each device.

This driver also creates the "cpufreq-dt" platform_device after passing
the version matching data to the OPP framework so that the cpufreq-dt
handles the actual cpufreq implementation. Even without the necessary
data to pass the version matching data the driver will still create this
device to maintain backwards compatibility with operating-points v1
tables.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoDocumentation: dt: add bindings for ti-cpufreq
Dave Gerlach [Thu, 9 Feb 2017 15:35:49 +0000 (09:35 -0600)]
Documentation: dt: add bindings for ti-cpufreq

Add the device tree bindings document for the TI CPUFreq/OPP driver
on AM33xx, AM43xx, DRA7xx, and AM57xx SoCs. The operating-points-v2
binding allows us to provide an opp-supported-hw property for each OPP
to define when it is available. This driver is responsible for reading
and parsing registers to determine which OPPs can be selectively enabled
based on the specific SoC in use by matching against the opp-supported-hw
data.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMerge OPP material for v4.11 to satisfy dependencies.
Rafael J. Wysocki [Thu, 9 Feb 2017 21:52:35 +0000 (22:52 +0100)]
Merge OPP material for v4.11 to satisfy dependencies.

7 years agoPM / OPP: Expose _of_get_opp_desc_node as dev_pm_opp API
Dave Gerlach [Fri, 3 Feb 2017 17:29:26 +0000 (11:29 -0600)]
PM / OPP: Expose _of_get_opp_desc_node as dev_pm_opp API

Rename _of_get_opp_desc_node to dev_pm_opp_of_get_opp_desc_node and add it
to include/linux/pm_opp.h to allow other drivers, such as platform OPP
and cpufreq drivers, to make use of it.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: qoriq: Don't look at clock implementation details
Tang Yuantian [Thu, 9 Feb 2017 02:33:02 +0000 (10:33 +0800)]
cpufreq: qoriq: Don't look at clock implementation details

Get the CPU clock's potential parent clocks from the clock interface
itself, rather than manually parsing the clocks property to find a
phandle, looking at the clock-names property of that, and assuming that
those are valid parent clocks for the cpu clock.

This is necessary now that the clocks are generated based on the clock
driver's knowledge of the chip rather than a fragile device-tree
description of the mux options.

We can now rely on the clock driver to ensure that the mux only exposes
options that are valid.  The cpufreq driver was currently being overly
conservative in some cases -- for example, the "min_cpufreq =
get_bus_freq()" restriction only applies to chips with erratum
A-004510, and whether the freq_mask used on p5020 is needed depends on
the actual frequencies of the PLLs (FWIW, p5040 has a similar
limitation but its .freq_mask was zero) -- and the frequency mask
mechanism made assumptions about particular parent clock indices that
are no longer valid.

Signed-off-by: Scott Wood <scottwood@nxp.com>
Signed-off-by: Tang Yuantian <yuantian.tang@nxp.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: qoriq: add ARM64 SoCs support
Tang Yuantian [Thu, 9 Feb 2017 02:33:01 +0000 (10:33 +0800)]
cpufreq: qoriq: add ARM64 SoCs support

Add ARM64 config to Kconfig to enable CPU frequency feature on
NXP ARM64 SoCs.

Signed-off-by: Tang Yuantian <yuantian.tang@nxp.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: brcmstb-avs-cpufreq: remove unnecessary platform_set_drvdata()
Wei Yongjun [Tue, 7 Feb 2017 15:56:45 +0000 (15:56 +0000)]
cpufreq: brcmstb-avs-cpufreq: remove unnecessary platform_set_drvdata()

The driver core clears the driver data to NULL after device_release
or on probe failure. Thus, it is not needed to manually clear the
device driver data to NULL.

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: s3c2416: double free on driver init error path
Dan Carpenter [Tue, 7 Feb 2017 13:19:06 +0000 (16:19 +0300)]
cpufreq: s3c2416: double free on driver init error path

The "goto err_armclk;" error path already does a clk_put(s3c_freq->hclk);
so this is a double free.

Fixes: 34ee55075265 ([CPUFREQ] Add S3C2416/S3C2450 cpufreq driver)
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMIPS: BMIPS: enable CPUfreq
Markus Mayer [Tue, 7 Feb 2017 21:58:56 +0000 (13:58 -0800)]
MIPS: BMIPS: enable CPUfreq

Enable all applicable CPUfreq options.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Reviewed-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: bmips-cpufreq: CPUfreq driver for Broadcom's BMIPS SoCs
Markus Mayer [Tue, 7 Feb 2017 21:58:55 +0000 (13:58 -0800)]
cpufreq: bmips-cpufreq: CPUfreq driver for Broadcom's BMIPS SoCs

Add the MIPS CPUfreq driver. This driver currently supports CPUfreq on
BMIPS5xxx-based SoCs.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoBMIPS: Enable prerequisites for CPUfreq in MIPS Kconfig.
Markus Mayer [Tue, 7 Feb 2017 21:58:54 +0000 (13:58 -0800)]
BMIPS: Enable prerequisites for CPUfreq in MIPS Kconfig.

Turn on CPU_SUPPORTS_CPUFREQ and MIPS_EXTERNAL_TIMER for BMIPS.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMIPS: BMIPS: Update defconfig
Markus Mayer [Tue, 7 Feb 2017 21:58:53 +0000 (13:58 -0800)]
MIPS: BMIPS: Update defconfig

Ran "make savedefconfig" to bring bmips_stb_defconfig up to date.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Reviewed-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMerge back earlier cpufreq changes for v4.11.
Rafael J. Wysocki [Thu, 9 Feb 2017 00:18:14 +0000 (01:18 +0100)]
Merge back earlier cpufreq changes for v4.11.

7 years agoPM / OPP: Make _find_opp_table_unlocked() static
Wei Yongjun [Mon, 6 Feb 2017 14:29:55 +0000 (14:29 +0000)]
PM / OPP: Make _find_opp_table_unlocked() static

Fixes the following sparse warning:

drivers/base/power/opp/core.c:49:18: warning:
 symbol '_find_opp_table_unlocked' was not declared. Should it be static?

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Fix typos in comments
Viresh Kumar [Fri, 3 Feb 2017 09:56:25 +0000 (15:26 +0530)]
cpufreq: Fix typos in comments

- s/freqnency/frequency/
- s/accomodating/accommodating/

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Disable energy efficiency optimization
Srinivas Pandruvada [Fri, 3 Feb 2017 22:18:39 +0000 (14:18 -0800)]
cpufreq: intel_pstate: Disable energy efficiency optimization

Some Kabylake desktop processors may not reach max turbo when running in
HWP mode, even if running under sustained 100% utilization.

This occurs when the HWP.EPP (Energy Performance Preference) is set to
"balance_power" (0x80) -- the default on most systems.

It occurs because the platform BIOS may erroneously enable an
energy-efficiency setting -- MSR_IA32_POWER_CTL BIT-EE, which is not
recommended to be enabled on this SKU.

On the failing systems, this BIOS issue was not discovered when the
desktop motherboard was tested with Windows, because the BIOS also
neglects to provide the ACPI/CPPC table, that Windows requires to enable
HWP, and so Windows runs in legacy P-state mode, where this setting has
no effect.

Linux' intel_pstate driver does not require ACPI/CPPC to enable HWP, and
so it runs in HWP mode, exposing this incorrect BIOS configuration.

There are several ways to address this problem.

First, Linux can also run in legacy P-state mode on this system.
As intel_pstate is how Linux enables HWP, booting with
"intel_pstate=disable"
will run in acpi-cpufreq/ondemand legacy p-state mode.

Or second, the "performance" governor can be used with intel_pstate,
which will modify HWP.EPP to 0.

Or third, starting in 4.10, the
/sys/devices/system/cpu/cpufreq/policy*/energy_performance_preference
attribute in can be updated from "balance_power" to "performance".

Or fourth, apply this patch, which fixes the erroneous setting of
MSR_IA32_POWER_CTL BIT_EE on this model, allowing the default
configuration to function as designed.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Reviewed-by: Len Brown <len.brown@intel.com>
Cc: 4.6+ <stable@vger.kernel.org> # 4.6+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Calculate guaranteed performance for HWP
Srinivas Pandruvada [Thu, 19 Jan 2017 23:03:14 +0000 (15:03 -0800)]
cpufreq: intel_pstate: Calculate guaranteed performance for HWP

When HWP is active, turbo activation ratio is not used to calculate max
non turbo ratio. But on these systems the max non turbo ratio is decided
by config TDP settings.

This change removes usage of MSR_TURBO_ACTIVATION_RATIO for HWP systems,
instead directly use TDP ratios, when more than one TDPs are available.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Make HWP limits compatible with legacy
Srinivas Pandruvada [Wed, 18 Jan 2017 18:48:24 +0000 (10:48 -0800)]
cpufreq: intel_pstate: Make HWP limits compatible with legacy

Under HWP the performance limits are calculated using max_perf_pct
and min_perf_pct using possible performance, not available performance.
The available performance can be reduced by no_turbo setting. To make
compatible with legacy mode, use max/min performance percentage with
respect to available performance.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Lower frequency than expected under no_turbo
Srinivas Pandruvada [Wed, 18 Jan 2017 18:48:23 +0000 (10:48 -0800)]
cpufreq: intel_pstate: Lower frequency than expected under no_turbo

When turbo is not disabled by BIOS, but user disabled from intel P-State
sysfs and changes max/min using cpufreq sysfs, the resultant frequency
is lower than what user requested.

The reason for this, when the perf limits are calculated in set_policy()
callback, they are with reference to max cpu frequency (turbo frequency
), but when enforced in the intel_pstate_get_min_max() they are with
reference to max available performance as documented in the intel_pstate
documentation (in this case max non turbo P-State).

This needs similar change as done in intel_cpufreq_verify_policy() for
passive mode. Set policy->cpuinfo.max_freq based on the turbo status.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Operation mode control from sysfs
Rafael J. Wysocki [Thu, 5 Jan 2017 01:53:12 +0000 (02:53 +0100)]
cpufreq: intel_pstate: Operation mode control from sysfs

Make it possible to change the operation mode of intel_pstate with
the help of a new sysfs attribute called "status".

There are three possible configurations that can be selected using
this attribute:

 "off"     - The driver is not in use at this time.
 "active"  - The driver works as a P-state governor (default).
 "passive" - The driver works as a regular cpufreq one and collaborates
             with the generic cpufreq governors (it sets P-states as
             requested by those governors).  [This is the same mode
             the driver can be started in by passing intel_pstate=passive
             in the kernel command line.]

The current setting is returned by reads from this attribute.  Writing
one of the above strings to it changes the operation mode as indicated
by that string, if possible.

If HW-managed P-states (HWP) feature is enabled, it is not possible
to change the driver's operation mode and attempts to write to this
attribute will fail.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: intel_pstate: Expose global sysfs attributes upfront
Rafael J. Wysocki [Wed, 11 Jan 2017 03:12:16 +0000 (04:12 +0100)]
cpufreq: intel_pstate: Expose global sysfs attributes upfront

Expose the intel_pstate's global sysfs attributes before registering
the driver to prepare for the addition of an attribute that also will
have to work if the driver is not registered.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Remove CPUFREQ_START notifier event
Viresh Kumar [Thu, 5 Jan 2017 06:04:31 +0000 (11:34 +0530)]
cpufreq: Remove CPUFREQ_START notifier event

Its not used anymore, remove it.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoACPI: processor_perflib: Simplify code and stop using CPUFREQ_START
Viresh Kumar [Mon, 30 Jan 2017 04:29:57 +0000 (09:59 +0530)]
ACPI: processor_perflib: Simplify code and stop using CPUFREQ_START

acpi_processor_ppc_notifier() can live without using CPUFREQ_START
(which is gonna be removed soon), as it is only used while setting
ignore_ppc to 0. This can be done with the help of "ignore_ppc < 0"
check alone. The notifier function anyway ignores all events except
CPUFREQ_ADJUST and dropping CPUFREQ_START wouldn't harm at all.

Once CPUFREQ_START event is removed from the cpufreq core,
acpi_processor_ppc_notifier() will get called only for CPUFREQ_NOTIFY or
CPUFREQ_ADJUST event. Drop the return statement from the first if block
to make sure we don't ignore any such events.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: powernv: Add boost files to export ultra-turbo frequencies
Shilpasri G Bhat [Tue, 3 Jan 2017 11:06:00 +0000 (16:36 +0530)]
cpufreq: powernv: Add boost files to export ultra-turbo frequencies

In P8+, Workload Optimized Frequency(WOF) provides the capability to
boost the cpu frequency based on the utilization of the other cpus
running in the chip. The On-Chip-Controller(OCC) firmware will control
the achievability of these frequencies depending on the power headroom
available in the chip. Currently the ultra-turbo frequencies provided
by this feature are exported along with the turbo and sub-turbo
frequencies as scaling_available_frequencies. This patch will export
the ultra-turbo frequencies separately as scaling_boost_frequencies in
WOF enabled systems. This patch will add the boost sysfs file which
can be used to disable/enable ultra-turbo frequencies.

Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Documentation: Updates based on current code
Viresh Kumar [Fri, 6 Jan 2017 05:38:05 +0000 (11:08 +0530)]
cpufreq: Documentation: Updates based on current code

The cpufreq core has gone though lots of updates in recent times, but on
many occasions the documentation wasn't updated along with the code.
This patch tries to catchup the documentation with the code.

Also add Rafael and Viresh as the contributors to the documentation.

Based on a patch from Claudio Scordino.

Signed-off-by: Claudio Scordino <claudio@evidence.eu.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Documentation: Minor reformatting
Viresh Kumar [Fri, 6 Jan 2017 05:38:04 +0000 (11:08 +0530)]
cpufreq: Documentation: Minor reformatting

This patch doesn't change the content of the documentation, but rather
reformat it to make it more readable.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Remove CONFIG_CPU_FREQ_STAT_DETAILS config option
Viresh Kumar [Thu, 5 Jan 2017 08:27:41 +0000 (13:57 +0530)]
cpufreq: Remove CONFIG_CPU_FREQ_STAT_DETAILS config option

This doesn't have any benefit apart from saving a small amount of memory
when it is disabled. The ifdef hackery in the code makes it dirty
unnecessarily.

Clean it up by removing the Kconfig option completely. Few defconfigs
are also updated and CONFIG_CPU_FREQ_STAT_DETAILS is replaced with
CONFIG_CPU_FREQ_STAT now in them, as users wanted stats to be enabled.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Gregory CLEMENT <gregory.clement@free-electrons.com>
Reviewed-by: Chanwoo Choi <cw00.choi@samsung.com>
Acked-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: Remove policy create/remove notifiers
Viresh Kumar [Thu, 5 Jan 2017 04:47:27 +0000 (10:17 +0530)]
cpufreq: Remove policy create/remove notifiers

Those were added by:

commit fcd7af917abb ("cpufreq: stats: handle cpufreq_unregister_driver()
and suspend/resume properly")

but aren't used anymore since:

commit 1aefc75b2449 ("cpufreq: stats: Make the stats code non-modular").

Remove them. Also remove the redundant parameter to the respective
routines.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Update Documentation to remove RCU specific bits
Viresh Kumar [Mon, 23 Jan 2017 04:41:52 +0000 (10:11 +0530)]
PM / OPP: Update Documentation to remove RCU specific bits

Update OPP documentation to remove the RCU specific bits.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Simplify dev_pm_opp_get_max_volt_latency()
Viresh Kumar [Mon, 23 Jan 2017 04:41:51 +0000 (10:11 +0530)]
PM / OPP: Simplify dev_pm_opp_get_max_volt_latency()

dev_pm_opp_get_max_volt_latency() calls _find_opp_table() two times
effectively.

Merge _get_regulator_count() into dev_pm_opp_get_max_volt_latency() to
avoid that.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Simplify _opp_set_availability()
Viresh Kumar [Mon, 23 Jan 2017 04:41:50 +0000 (10:11 +0530)]
PM / OPP: Simplify _opp_set_availability()

As we don't use RCU locking anymore, there is no need to replace an
earlier OPP node with a new one. Just update the existing one.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Move away from RCU locking
Viresh Kumar [Mon, 23 Jan 2017 04:41:49 +0000 (10:11 +0530)]
PM / OPP: Move away from RCU locking

The RCU locking isn't well suited for the OPP core. The RCU locking fits
better for reader heavy stuff, while the OPP core have at max one or two
readers only at a time.

Over that, it was getting very confusing the way RCU locking was used
with the OPP core. The individual OPPs are mostly well handled, i.e. for
an update a new structure was created and then that replaced the older
one. But the OPP tables were updated directly all the time from various
parts of the core. Though they were mostly used from within RCU locked
region, they didn't had much to do with RCU and were governed by the
mutex instead.

And that mixed with the 'opp_table_lock' has made the core even more
confusing.

Now that we are already managing the OPPs and the OPP tables with kernel
reference infrastructure, we can get rid of RCU locking completely and
simplify the code a lot.

Remove all RCU references from code and comments.

Acquire opp_table->lock while parsing the list of OPPs though.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Take kref from _find_opp_table()
Viresh Kumar [Mon, 23 Jan 2017 04:41:48 +0000 (10:11 +0530)]
PM / OPP: Take kref from _find_opp_table()

Take reference of the OPP table from within _find_opp_table(). Also
update the callers of _find_opp_table() to call
dev_pm_opp_put_opp_table() after they have used the OPP table.

Note that _find_opp_table() increments the reference under the
opp_table_lock.

Now that the OPP table wouldn't get freed until the callers of
_find_opp_table() call dev_pm_opp_put_opp_table(), there is no need to
take the opp_table_lock or rcu_read_lock() around it. Drop them.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Update OPP users to put reference
Viresh Kumar [Mon, 23 Jan 2017 04:41:47 +0000 (10:11 +0530)]
PM / OPP: Update OPP users to put reference

This patch updates dev_pm_opp_find_freq_*() routines to get a reference
to the OPPs returned by them.

Also updates the users of dev_pm_opp_find_freq_*() routines to call
dev_pm_opp_put() after they are done using the OPPs.

As it is guaranteed the that OPPs wouldn't get freed while being used,
the RCU read side locking present with the users isn't required anymore.
Drop it as well.

This patch also updates all users of devfreq_recommended_opp() which was
returning an OPP received from the OPP core.

Note that some of the OPP core routines have gained
rcu_read_{lock|unlock}() calls, as those still use RCU specific APIs
within them.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Chanwoo Choi <cw00.choi@samsung.com> [Devfreq]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Add 'struct kref' to struct dev_pm_opp
Viresh Kumar [Mon, 23 Jan 2017 04:41:46 +0000 (10:11 +0530)]
PM / OPP: Add 'struct kref' to struct dev_pm_opp

Add kref to struct dev_pm_opp for easier accounting of the OPPs.

Note that the OPPs are freed under the opp_table->lock mutex only.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Use dev_pm_opp_get_opp_table() instead of _add_opp_table()
Viresh Kumar [Mon, 23 Jan 2017 04:41:45 +0000 (10:11 +0530)]
PM / OPP: Use dev_pm_opp_get_opp_table() instead of _add_opp_table()

Migrate all users of _add_opp_table() to use dev_pm_opp_get_opp_table()
to guarantee that the OPP table doesn't get freed while being used.

Also update _managed_opp() to get the reference to the OPP table.

Now that the OPP table wouldn't get freed while these routines are
executing after dev_pm_opp_get_opp_table() is called, there is no need
to take opp_table_lock. Drop them as well.

Now that _add_opp_table(), _remove_opp_table() and the unlocked release
routines aren't used anymore, remove them.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Take reference of the OPP table while adding/removing OPPs
Viresh Kumar [Mon, 23 Jan 2017 04:41:44 +0000 (10:11 +0530)]
PM / OPP: Take reference of the OPP table while adding/removing OPPs

Take reference of the OPP table while adding and removing OPPs, that
helps us remove special checks in _remove_opp_table().

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Return opp_table from dev_pm_opp_set_*() routines
Viresh Kumar [Mon, 23 Jan 2017 04:41:43 +0000 (10:11 +0530)]
PM / OPP: Return opp_table from dev_pm_opp_set_*() routines

Now that we have proper kernel reference infrastructure in place for OPP
tables, use it to guarantee that the OPP table isn't freed while being
used by the callers of dev_pm_opp_set_*() APIs.

Make them all return the pointer to the OPP table after taking its
reference and put the reference back with dev_pm_opp_put_*() APIs.

Now that the OPP table wouldn't get freed while these routines are
executing after dev_pm_opp_get_opp_table() is called, there is no need
to take opp_table_lock. Drop them as well.

Remove the rcu specific comments from these routines as they aren't
relevant anymore.

Note that prototypes of dev_pm_opp_{set|put}_regulators() were already
updated by another patch.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Add 'struct kref' to OPP table
Viresh Kumar [Mon, 23 Jan 2017 04:41:42 +0000 (10:11 +0530)]
PM / OPP: Add 'struct kref' to OPP table

Add kref to struct opp_table for easier accounting of the OPP table.

Note that the new routine dev_pm_opp_get_opp_table() takes the reference
from under the opp_table_lock, which guarantees that the OPP table
doesn't get freed unless dev_pm_opp_put_opp_table() is called for the
OPP table.

Two separate release mechanisms are added: locked and unlocked. In
unlocked version the routines aren't required to take/drop
opp_table_lock as the callers have already done that. This is required
to avoid breaking git bisect, otherwise we may get lockdeps between
commits. Once all the users of OPP table are updated the unlocked
version shall be removed.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Add per OPP table mutex
Viresh Kumar [Mon, 23 Jan 2017 04:41:41 +0000 (10:11 +0530)]
PM / OPP: Add per OPP table mutex

Add per OPP table lock to protect opp_table->opp_list.

Note that at few places opp_list is used under the rcu_read_lock() and
so a mutex can't be added there for now. This will be fixed by a later
patch.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoLinux 4.10-rc6
Linus Torvalds [Sun, 29 Jan 2017 22:25:17 +0000 (14:25 -0800)]
Linux 4.10-rc6

7 years agodrm/i915: Check for NULL i915_vma in intel_unpin_fb_obj()
Linus Torvalds [Sun, 29 Jan 2017 21:50:06 +0000 (13:50 -0800)]
drm/i915: Check for NULL i915_vma in intel_unpin_fb_obj()

I've seen this trigger twice now, where the i915_gem_object_to_ggtt()
call in intel_unpin_fb_obj() returns NULL, resulting in an oops
immediately afterwards as the (inlined) call to i915_vma_unpin_fence()
tries to dereference it.

It seems to be some race condition where the object is going away at
shutdown time, since both times happened when shutting down the X
server.  The call chains were different:

 - VT ioctl(KDSETMODE, KD_TEXT):

    intel_cleanup_plane_fb+0x5b/0xa0 [i915]
    drm_atomic_helper_cleanup_planes+0x6f/0x90 [drm_kms_helper]
    intel_atomic_commit_tail+0x749/0xfe0 [i915]
    intel_atomic_commit+0x3cb/0x4f0 [i915]
    drm_atomic_commit+0x4b/0x50 [drm]
    restore_fbdev_mode+0x14c/0x2a0 [drm_kms_helper]
    drm_fb_helper_restore_fbdev_mode_unlocked+0x34/0x80 [drm_kms_helper]
    drm_fb_helper_set_par+0x2d/0x60 [drm_kms_helper]
    intel_fbdev_set_par+0x18/0x70 [i915]
    fb_set_var+0x236/0x460
    fbcon_blank+0x30f/0x350
    do_unblank_screen+0xd2/0x1a0
    vt_ioctl+0x507/0x12a0
    tty_ioctl+0x355/0xc30
    do_vfs_ioctl+0xa3/0x5e0
    SyS_ioctl+0x79/0x90
    entry_SYSCALL_64_fastpath+0x13/0x94

 - i915 unpin_work workqueue:

    intel_unpin_work_fn+0x58/0x140 [i915]
    process_one_work+0x1f1/0x480
    worker_thread+0x48/0x4d0
    kthread+0x101/0x140

and this patch purely papers over the issue by adding a NULL pointer
check and a WARN_ON_ONCE() to avoid the oops that would then generally
make the machine unresponsive.

Other callers of i915_gem_object_to_ggtt() seem to also check for the
returned pointer being NULL and warn about it, so this clearly has
happened before in other places.

[ Reported it originally to the i915 developers on Jan 8, applying the
  ugly workaround on my own now after triggering the problem for the
  second time with no feedback.

  This is likely to be the same bug reported as

     https://bugs.freedesktop.org/show_bug.cgi?id=98829
     https://bugs.freedesktop.org/show_bug.cgi?id=99134

  which has a patch for the underlying problem, but it hasn't gotten to
  me, so I'm applying the workaround. ]

Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge branch 'parisc-4.10-3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sun, 29 Jan 2017 18:56:56 +0000 (10:56 -0800)]
Merge branch 'parisc-4.10-3' of git://git./linux/kernel/git/deller/parisc-linux

Pull two parisc fixes from Helge Deller:
 "One fix to avoid usage of BITS_PER_LONG in user-space exported swab.h
  header which breaks compiling qemu, and one trivial fix for printk
  continuation in the parisc parport driver"

* 'parisc-4.10-3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Don't use BITS_PER_LONG in userspace-exported swab.h header
  parisc, parport_gsc: Fixes for printk continuation lines

7 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sat, 28 Jan 2017 23:09:23 +0000 (15:09 -0800)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Two I2C driver bugfixes.

  The 'VLLS mode support' patch should have been entitled 'reconfigure
  pinctrl after suspend' to make the bugfix more clear. Sorry, I missed
  that, yet didn't want to rebase"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: imx-lpi2c: add VLLS mode support
  i2c: i2c-cadence: Initialize configuration before probing devices

7 years agoparisc: Don't use BITS_PER_LONG in userspace-exported swab.h header
Helge Deller [Sat, 28 Jan 2017 10:52:02 +0000 (11:52 +0100)]
parisc: Don't use BITS_PER_LONG in userspace-exported swab.h header

In swab.h the "#if BITS_PER_LONG > 32" breaks compiling userspace programs if
BITS_PER_LONG is #defined by userspace with the sizeof() compiler builtin.

Solve this problem by using __BITS_PER_LONG instead.  Since we now
#include asm/bitsperlong.h avoid further potential userspace pollution
by moving the #define of SHIFT_PER_LONG to bitops.h which is not
exported to userspace.

This patch unbreaks compiling qemu on hppa/parisc.

Signed-off-by: Helge Deller <deller@gmx.de>
Cc: <stable@vger.kernel.org>
7 years agoparisc, parport_gsc: Fixes for printk continuation lines
Helge Deller [Tue, 3 Jan 2017 21:55:50 +0000 (22:55 +0100)]
parisc, parport_gsc: Fixes for printk continuation lines

Signed-off-by: Helge Deller <deller@gmx.de>
7 years agoMerge tag 'nfs-for-4.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Sat, 28 Jan 2017 19:50:17 +0000 (11:50 -0800)]
Merge tag 'nfs-for-4.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:
 "Stable patches:
   - NFSv4.1: Fix a deadlock in layoutget
   - NFSv4 must not bump sequence ids on NFS4ERR_MOVED errors
   - NFSv4 Fix a regression with OPEN EXCLUSIVE4 mode
   - Fix a memory leak when removing the SUNRPC module

  Bugfixes:
   - Fix a reference leak in _pnfs_return_layout"

* tag 'nfs-for-4.10-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  pNFS: Fix a reference leak in _pnfs_return_layout
  nfs: Fix "Don't increment lock sequence ID after NFS4ERR_MOVED"
  SUNRPC: cleanup ida information when removing sunrpc module
  NFSv4.0: always send mode in SETATTR after EXCLUSIVE4
  nfs: Don't increment lock sequence ID after NFS4ERR_MOVED
  NFSv4.1: Fix a deadlock in layoutget

7 years agoMerge tag 'md/4.10-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md
Linus Torvalds [Sat, 28 Jan 2017 19:09:04 +0000 (11:09 -0800)]
Merge tag 'md/4.10-rc6' of git://git./linux/kernel/git/shli/md

Pull MD fixes from Shaohua Li:
 "This fixes several corner cases for raid5 cache, which is merged into
  this cycle"

* tag 'md/4.10-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md:
  md/r5cache: disable write back for degraded array
  md/r5cache: shift complex rmw from read path to write path
  md/r5cache: flush data only stripes in r5l_recovery_log()
  md/raid5: move comment of fetch_block to right location
  md/r5cache: read data into orig_page for prexor of cached data
  md/raid5-cache: delete meaningless code

7 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Sat, 28 Jan 2017 19:06:42 +0000 (11:06 -0800)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fix from Catalin Marinas:
 "Fix kernel panic on ACPI-based systems where CPU capacity description
  is not currently handled"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: skip register_cpufreq_notifier on ACPI-based systems

7 years agoMerge tag 'arc-4.10-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
Linus Torvalds [Sat, 28 Jan 2017 19:00:08 +0000 (11:00 -0800)]
Merge tag 'arc-4.10-rc6' of git://git./linux/kernel/git/vgupta/arc

Pull ARC fixes from Vineet Gupta:
 "Hopefully last set of changes for ARC for 4.10:

   - fix for unaligned access emulation corner case

   - fix for udelay loop inline asm regression

   - fix irq affinity finally for AXS103 board [Yuriy]

   - final fixes for setting IO-coherency sanely in SMP"

* tag 'arc-4.10-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: [arcompact] handle unaligned access delay slot corner case
  ARCv2: smp-boot: wake_flag polling by non-Masters needs to be uncached
  ARC: smp-boot: Decouple Non masters waiting API from jump to entry point
  ARCv2: MCIP: update the BCR per current changes
  ARC: udelay: fix inline assembler by adding LP_COUNT to clobber list
  ARCv2: MCIP: Deprecate setting of affinity in Device Tree

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 27 Jan 2017 20:54:16 +0000 (12:54 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) GTP fixes from Andreas Schultz (missing genl module alias, clear IP
    DF on transmit).

 2) Netfilter needs to reflect the fwmark when sending resets, from Pau
    Espin Pedrol.

 3) nftable dump OOPS fix from Liping Zhang.

 4) Fix erroneous setting of VIRTIO_NET_HDR_F_DATA_VALID on transmit,
    from Rolf Neugebauer.

 5) Fix build error of ipt_CLUSTERIP when procfs is disabled, from Arnd
    Bergmann.

 6) Fix regression in handling of NETIF_F_SG in harmonize_features(),
    from Eric Dumazet.

 7) Fix RTNL deadlock wrt. lwtunnel module loading, from David Ahern.

 8) tcp_fastopen_create_child() needs to setup tp->max_window, from
    Alexey Kodanev.

 9) Missing kmemdup() failure check in ipv6 segment routing code, from
    Eric Dumazet.

10) Don't execute unix_bind() under the bindlock, otherwise we deadlock
    with splice. From WANG Cong.

11) ip6_tnl_parse_tlv_enc_lim() potentially reallocates the skb buffer,
    therefore callers must reload cached header pointers into that skb.
    Fix from Eric Dumazet.

12) Fix various bugs in legacy IRQ fallback handling in alx driver, from
    Tobias Regnery.

13) Do not allow lwtunnel drivers to be unloaded while they are
    referenced by active instances, from Robert Shearman.

14) Fix truncated PHY LED trigger names, from Geert Uytterhoeven.

15) Fix a few regressions from virtio_net XDP support, from John
    Fastabend and Jakub Kicinski.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (102 commits)
  ISDN: eicon: silence misleading array-bounds warning
  net: phy: micrel: add support for KSZ8795
  gtp: fix cross netns recv on gtp socket
  gtp: clear DF bit on GTP packet tx
  gtp: add genl family modules alias
  tcp: don't annotate mark on control socket from tcp_v6_send_response()
  ravb: unmap descriptors when freeing rings
  virtio_net: reject XDP programs using header adjustment
  virtio_net: use dev_kfree_skb for small buffer XDP receive
  r8152: check rx after napi is enabled
  r8152: re-schedule napi for tx
  r8152: avoid start_xmit to schedule napi when napi is disabled
  r8152: avoid start_xmit to call napi_schedule during autosuspend
  net: dsa: Bring back device detaching in dsa_slave_suspend()
  net: phy: leds: Fix truncated LED trigger names
  net: phy: leds: Break dependency of phy.h on phy_led_triggers.h
  net: phy: leds: Clear phy_num_led_triggers on failure to avoid crash
  net-next: ethernet: mediatek: change the compatible string
  Documentation: devicetree: change the mediatek ethernet compatible string
  bnxt_en: Fix RTNL lock usage on bnxt_get_port_module_status().
  ...

7 years agoMerge tag 'xfs-for-linus-4.10-rc6-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Linus Torvalds [Fri, 27 Jan 2017 20:44:32 +0000 (12:44 -0800)]
Merge tag 'xfs-for-linus-4.10-rc6-5' of git://git./fs/xfs/xfs-linux

Pull xfs uodates from Darrick Wong:
 "I have some more fixes this week: better input validation, corruption
  avoidance, build fixes, memory leak fixes, and a couple from Christoph
  to avoid an ENOSPC failure.

  Summary:
   - Fix race conditions in the CoW code
   - Fix some incorrect input validation checks
   - Avoid crashing fs by running out of space when freeing inodes
   - Fix toctou race wrt whether or not an inode has an attr
   - Fix build error on arm
   - Fix page refcount corruption when readahead fails
   - Don't corrupt userspace in the bmap ioctl"

* tag 'xfs-for-linus-4.10-rc6-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  xfs: prevent quotacheck from overloading inode lru
  xfs: fix bmv_count confusion w/ shared extents
  xfs: clear _XBF_PAGES from buffers when readahead page
  xfs: extsize hints are not unlikely in xfs_bmap_btalloc
  xfs: remove racy hasattr check from attr ops
  xfs: use per-AG reservations for the finobt
  xfs: only update mount/resv fields on success in __xfs_ag_resv_init
  xfs: verify dirblocklog correctly
  xfs: fix COW writeback race

7 years agoMerge branch 'for-linus-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
Linus Torvalds [Fri, 27 Jan 2017 20:41:46 +0000 (12:41 -0800)]
Merge branch 'for-linus-4.10' of git://git./linux/kernel/git/mason/linux-btrfs

Pull btrfs updates from Chris Mason:
 "Some fixes that we've collected from the list.

  We still have one more pending to nail down a regression in lzo
  compression, but I wanted to get this batch out the door"

* 'for-linus-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  Btrfs: remove ->{get, set}_acl() from btrfs_dir_ro_inode_operations
  Btrfs: disable xattr operations on subvolume directories
  Btrfs: remove old tree_root case in btrfs_read_locked_inode()
  Btrfs: fix truncate down when no_holes feature is enabled
  Btrfs: Fix deadlock between direct IO and fast fsync
  btrfs: fix false enospc error when truncating heavily reflinked file

7 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 27 Jan 2017 20:36:39 +0000 (12:36 -0800)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A set of fixes for this series. This contains:

   - Set of fixes for the nvme target code

   - A revert of patch from this merge window, causing a regression with
     WRITE_SAME on iSCSI targets at least.

   - A fix for a use-after-free in the new O_DIRECT bdev code.

   - Two fixes for the xen-blkfront driver"

* 'for-linus' of git://git.kernel.dk/linux-block:
  Revert "sd: remove __data_len hack for WRITE SAME"
  nvme-fc: use blk_rq_nr_phys_segments
  nvmet-rdma: Fix missing dma sync to nvme data structures
  nvmet: Call fatal_error from keep-alive timout expiration
  nvmet: cancel fatal error and flush async work before free controller
  nvmet: delete controllers deletion upon subsystem release
  nvmet_fc: correct logic in disconnect queue LS handling
  block: fix use after free in __blkdev_direct_IO
  xen-blkfront: correct maximum segment accounting
  xen-blkfront: feature flags handling adjustments

7 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Linus Torvalds [Fri, 27 Jan 2017 20:29:30 +0000 (12:29 -0800)]
Merge tag 'for-linus' of git://git./linux/kernel/git/dledford/rdma

Pull rdma fixes from Doug Ledford:
 "Second round of -rc fixes for 4.10.

  This -rc cycle has been slow for the rdma subsystem. I had already
  sent you the first batch before the Holiday break. After that, we kept
  only getting a few here or there. Up until this week, when I got a
  drop of 13 to one driver (qedr). So, here's the -rc patches I have. I
  currently have none held in reserve, so unless something new comes in,
  this is it until the next merge window opens.

  Summary:

   - series of iw_cxgb4 fixes to make it work with the drain cq API

   - one or two patches each to: srp, iser, cxgb3, vmw_pvrdma, umem,
     rxe, and ipoib

   - one big series (13 patches) for the new qedr driver"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (27 commits)
  RDMA/cma: Fix unknown symbol when CONFIG_IPV6 is not enabled
  IB/rxe: Prevent from completer to operate on non valid QP
  IB/rxe: Fix rxe dev insertion to rxe_dev_list
  IB/umem: Release pid in error and ODP flow
  RDMA/qedr: Dispatch port active event from qedr_add
  RDMA/qedr: Fix and simplify memory leak in PD alloc
  RDMA/qedr: Fix RDMA CM loopback
  RDMA/qedr: Fix formatting
  RDMA/qedr: Mark three functions as static
  RDMA/qedr: Don't reset QP when queues aren't flushed
  RDMA/qedr: Don't spam dmesg if QP is in error state
  RDMA/qedr: Remove CQ spinlock from CM completion handlers
  RDMA/qedr: Return max inline data in QP query result
  RDMA/qedr: Return success when not changing QP state
  RDMA/qedr: Add uapi header qedr-abi.h
  RDMA/qedr: Fix MTU returned from QP query
  RDMA/core: Add the function ib_mtu_int_to_enum
  IB/vmw_pvrdma: Fix incorrect cleanup on pvrdma_pci_probe error path
  IB/vmw_pvrdma: Don't leak info from alloc_ucontext
  IB/cxgb3: fix misspelling in header guard
  ...

7 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Fri, 27 Jan 2017 20:25:26 +0000 (12:25 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Martin Schwidefsky:
 "Another two bug fixes:

   - ptrace partial write information leak

   - a guest page hinting regression introduced with v4.6"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/mm: Fix cmma unused transfer from pgste into pte
  s390/ptrace: Preserve previous registers for short regset write

7 years agoMerge branch 'stable/for-linus-4.10' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 27 Jan 2017 20:17:07 +0000 (12:17 -0800)]
Merge branch 'stable/for-linus-4.10' of git://git./linux/kernel/git/konrad/swiotlb

Pull swiotlb fix from Konrad Rzeszutek Wilk:
 "An ARM fix in the Xen SWIOTLB - mainly the translation of physical to
  bus addresses was done just a tad too late"

* 'stable/for-linus-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/swiotlb:
  swiotlb-xen: update dev_addr after swapping pages

7 years agoMerge tag 'vfio-v4.10-rc6' of git://github.com/awilliam/linux-vfio
Linus Torvalds [Fri, 27 Jan 2017 20:10:58 +0000 (12:10 -0800)]
Merge tag 'vfio-v4.10-rc6' of git://github.com/awilliam/linux-vfio

Pull VFIO fix from Alex Williamson:
 "mdev IOMMU groups are not yet compatible with the powerpc SPAPR IOMMU
  backend, detect and fail group attach (Greg Kurz)"

* tag 'vfio-v4.10-rc6' of git://github.com/awilliam/linux-vfio:
  vfio/spapr: fail tce_iommu_attach_group() when iommu_data is null

7 years agoRDMA/cma: Fix unknown symbol when CONFIG_IPV6 is not enabled
Jack Morgenstein [Sun, 15 Jan 2017 18:15:00 +0000 (20:15 +0200)]
RDMA/cma: Fix unknown symbol when CONFIG_IPV6 is not enabled

If IPV6 has not been enabled in the underlying kernel, we must avoid
calling IPV6 procedures in rdma_cm.ko.

This requires using "IS_ENABLED(CONFIG_IPV6)" in "if" statements
surrounding any code which calls external IPV6 procedures.

In the instance fixed here, procedure cma_bind_addr() called
ipv6_addr_type() -- which resulted in calling external procedure
__ipv6_addr_type().

Fixes: 6c26a77124ff ("RDMA/cma: fix IPv6 address resolution")
Cc: <stable@vger.kernel.org> # v4.2+
Cc: Spencer Baugh <sbaugh@catern.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
7 years agoMerge branch 'stable/for-jens-4.10' of git://git.kernel.org/pub/scm/linux/kernel...
Jens Axboe [Fri, 27 Jan 2017 18:56:06 +0000 (11:56 -0700)]
Merge branch 'stable/for-jens-4.10' of git://git./linux/kernel/git/konrad/xen into for-linus

Konrad writes:

Please pull in your 'for-linus' branch two little fixes for Xen
block front:

One fix is for handling the XEN_PAGE_SIZE != PAGE_SIZE (4KB vs 64KB
on ARM for example) mishandling while the other is fixing
the accounting for the configuration changes.

7 years agoARC: [arcompact] handle unaligned access delay slot corner case
Vineet Gupta [Fri, 27 Jan 2017 18:45:27 +0000 (10:45 -0800)]
ARC: [arcompact] handle unaligned access delay slot corner case

After emulating an unaligned access in delay slot of a branch, we
pretend as the delay slot never happened - so return back to actual
branch target (or next PC if branch was not taken).

Curently we did this by handling STATUS32.DE, we also need to clear the
BTA.T bit, which is disregarded when returning from original misaligned
exception, but could cause weirdness if it took the interrupt return
path (in case interrupt was acive too)

One ARC700 customer ran into this when enabling unaligned access fixup
for kernel mode accesses as well

Cc: stable@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
7 years agoMerge tag 'media/v4.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Fri, 27 Jan 2017 18:29:33 +0000 (10:29 -0800)]
Merge tag 'media/v4.10-2' of git://git./linux/kernel/git/mchehab/linux-media

Pull media fixes from Mauro Carvalho Chehab:

 - fix a regression on tvp5150 causing failures at input selection and
   image glitches

 - CEC was moved out of staging for v4.10. Fix some bugs on it while not
   too late

 - fix a regression on pctv452e caused by VM stack changes

 - fix suspend issued with smiapp

 - fix a regression on cobalt driver

 - fix some warnings and Kconfig issues with some random configs.

* tag 'media/v4.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
  [media] s5k4ecgx: select CRC32 helper
  [media] dvb: avoid warning in dvb_net
  [media] v4l: tvp5150: Don't override output pinmuxing at stream on/off time
  [media] v4l: tvp5150: Fix comment regarding output pin muxing
  [media] v4l: tvp5150: Reset device at probe time, not in get/set format handlers
  [media] pctv452e: move buffer to heap, no mutex
  [media] media/cobalt: use pci_irq_allocate_vectors
  [media] cec: fix race between configuring and unconfiguring
  [media] cec: move cec_report_phys_addr into cec_config_thread_func
  [media] cec: replace cec_report_features by cec_fill_msg_report_features
  [media] cec: update log_addr[] before finishing configuration
  [media] cec: CEC_MSG_GIVE_FEATURES should abort for CEC version < 2
  [media] cec: when canceling a message, don't overwrite old status info
  [media] cec: fix report_current_latency
  [media] smiapp: Make suspend and resume functions __maybe_unused
  [media] smiapp: Implement power-on and power-off sequences without runtime PM

7 years agoMerge tag 'mmc-v4.10-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 27 Jan 2017 18:25:31 +0000 (10:25 -0800)]
Merge tag 'mmc-v4.10-rc5' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fix from Ulf Hansson:
 "MMC host: fix runtime PM resume path in dw_mmc"

* tag 'mmc-v4.10-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: dw_mmc: force setup bus if active slots exist

7 years agoMerge branch 'for-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
Linus Torvalds [Fri, 27 Jan 2017 18:22:00 +0000 (10:22 -0800)]
Merge branch 'for-rc' of git://git./linux/kernel/git/rzhang/linux

Pull thermal management fix from Zhang Rui:
 "A single revert from a recently introduced problem.

  Specifics:

  Commit 7611fb68062f ("thermal: thermal_hwmon: Convert to
  hwmon_device_register_with_info()"), which was introduced in 4.10-rc5,
  uses new hwmon API. But this breaks some soc thermal driver because
  the new hwmon API has a strict rule for the hwmon device name. Revert
  the offending commit as a quick solution for 4.10"

* 'for-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux:
  Revert "thermal: thermal_hwmon: Convert to hwmon_device_register_with_info()"

7 years agoxfs: prevent quotacheck from overloading inode lru
Brian Foster [Thu, 26 Jan 2017 21:18:09 +0000 (13:18 -0800)]
xfs: prevent quotacheck from overloading inode lru

Quotacheck runs at mount time in situations where quota accounting must
be recalculated. In doing so, it uses bulkstat to visit every inode in
the filesystem. Historically, every inode processed during quotacheck
was released and immediately tagged for reclaim because quotacheck runs
before the superblock is marked active by the VFS. In other words,
the final iput() lead to an immediate ->destroy_inode() call, which
allowed the XFS background reclaim worker to start reclaiming inodes.

Commit 17c12bcd3 ("xfs: when replaying bmap operations, don't let
unlinked inodes get reaped") marks the XFS superblock active sooner as
part of the mount process to support caching inodes processed during log
recovery. This occurs before quotacheck and thus means all inodes
processed by quotacheck are inserted to the LRU on release.  The
s_umount lock is held until the mount has completed and thus prevents
the shrinkers from operating on the sb. This means that quotacheck can
excessively populate the inode LRU and lead to OOM conditions on systems
without sufficient RAM.

Update the quotacheck bulkstat handler to set XFS_IGET_DONTCACHE on
inodes processed by quotacheck. This causes ->drop_inode() to return 1
and in turn causes iput_final() to evict the inode. This preserves the
original quotacheck behavior and prevents it from overloading the LRU
and running out of memory.

CC: stable@vger.kernel.org # v4.9
Reported-by: Martin Svec <martin.svec@zoner.cz>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
7 years agoISDN: eicon: silence misleading array-bounds warning
Arnd Bergmann [Fri, 27 Jan 2017 12:32:14 +0000 (13:32 +0100)]
ISDN: eicon: silence misleading array-bounds warning

With some gcc versions, we get a warning about the eicon driver,
and that currently shows up as the only remaining warning in one
of the build bots:

In file included from ../drivers/isdn/hardware/eicon/message.c:30:0:
eicon/message.c: In function 'mixer_notify_update':
eicon/platform.h:333:18: warning: array subscript is above array bounds [-Warray-bounds]

The code is easily changed to open-code the unusual PUT_WORD() line
causing this to avoid the warning.

Cc: stable@vger.kernel.org
Link: http://arm-soc.lixom.net/buildlogs/stable-rc/v4.4.45/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: micrel: add support for KSZ8795
Sean Nyekjaer [Fri, 27 Jan 2017 07:46:23 +0000 (08:46 +0100)]
net: phy: micrel: add support for KSZ8795

This is adds support for the PHYs in the KSZ8795 5port managed switch.

It will allow to detect the link between the switch and the soc
and uses the same read_status functions as the KSZ8873MLL switch.

Signed-off-by: Sean Nyekjaer <sean.nyekjaer@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'gtp-fixes'
David S. Miller [Fri, 27 Jan 2017 15:39:10 +0000 (10:39 -0500)]
Merge branch 'gtp-fixes'

Andreas Schultz says:

====================
various gtp fixes

I'm sorry for the compile error mess up in the last version.
It's no excuse for not test compiling, but the hunks got lost in
a rebase.

This is the part of the previous "simple gtp improvements" series
that Pablo indicated should go into net.

The addition of the module alias fixes genl family autoloading,
clearing the DF bit fixes a protocol violation in regard to the
specification and the netns comparison fixes a corner case of
cross netns recv.

v2->v3: fix compiler error introduced in rebase
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogtp: fix cross netns recv on gtp socket
Andreas Schultz [Fri, 27 Jan 2017 09:40:58 +0000 (10:40 +0100)]
gtp: fix cross netns recv on gtp socket

The use of the passed through netlink src_net to check for a
cross netns operation was wrong. Using the GTP socket and the
GTP netdevice is always correct (even if the netdev has been
moved to new netns after link creation).

Remove the now obsolete net field from gtp_dev.

Signed-off-by: Andreas Schultz <aschultz@tpip.net>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogtp: clear DF bit on GTP packet tx
Andreas Schultz [Fri, 27 Jan 2017 09:40:57 +0000 (10:40 +0100)]
gtp: clear DF bit on GTP packet tx

3GPP TS 29.281 and 3GPP TS 29.060 imply that GTP-U packets should be
sent with the DF bit cleared. For example 3GPP TS 29.060, Release 8,
Section 13.2.2:

> Backbone router: Any router in the backbone may fragment the GTP
> packet if needed, according to IPv4.

Signed-off-by: Andreas Schultz <aschultz@tpip.net>
Acked-by: Harald Welte <laforge@netfilter.org>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogtp: add genl family modules alias
Andreas Schultz [Fri, 27 Jan 2017 09:40:56 +0000 (10:40 +0100)]
gtp: add genl family modules alias

Auto-load the module when userspace asks for the gtp netlink
family.

Signed-off-by: Andreas Schultz <aschultz@tpip.net>
Acked-by: Harald Welte <laforge@netfilter.org>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: don't annotate mark on control socket from tcp_v6_send_response()
Pablo Neira [Thu, 26 Jan 2017 21:56:21 +0000 (22:56 +0100)]
tcp: don't annotate mark on control socket from tcp_v6_send_response()

Unlike ipv4, this control socket is shared by all cpus so we cannot use
it as scratchpad area to annotate the mark that we pass to ip6_xmit().

Add a new parameter to ip6_xmit() to indicate the mark. The SCTP socket
family caches the flowi6 structure in the sctp_transport structure, so
we cannot use to carry the mark unless we later on reset it back, which
I discarded since it looks ugly to me.

Fixes: bf99b4ded5f8 ("tcp: fix mark propagation with fwmark_reflect enabled")
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoarm64: skip register_cpufreq_notifier on ACPI-based systems
Prashanth Prakash [Thu, 26 Jan 2017 18:08:32 +0000 (11:08 -0700)]
arm64: skip register_cpufreq_notifier on ACPI-based systems

On ACPI based systems where the topology is setup using the API
store_cpu_topology, at the moment we do not have necessary code
to parse cpu capacity and handle cpufreq notifier, thus
resulting in a kernel panic.

Stack:
        init_cpu_capacity_callback+0xb4/0x1c8
        notifier_call_chain+0x5c/0xa0
        __blocking_notifier_call_chain+0x58/0xa0
        blocking_notifier_call_chain+0x3c/0x50
        cpufreq_set_policy+0xe4/0x328
        cpufreq_init_policy+0x80/0x100
        cpufreq_online+0x418/0x710
        cpufreq_add_dev+0x118/0x180
        subsys_interface_register+0xa4/0xf8
        cpufreq_register_driver+0x1c0/0x298
        cppc_cpufreq_init+0xdc/0x1000 [cppc_cpufreq]
        do_one_initcall+0x5c/0x168
        do_init_module+0x64/0x1e4
        load_module+0x130c/0x14d0
        SyS_finit_module+0x108/0x120
        el0_svc_naked+0x24/0x28

Fixes: 7202bde8b7ae ("arm64: parse cpu capacity-dmips-mhz from DT")
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
7 years agoPM / OPP: Split out part of _add_opp_table() and _remove_opp_table()
Viresh Kumar [Mon, 2 Jan 2017 09:11:04 +0000 (14:41 +0530)]
PM / OPP: Split out part of _add_opp_table() and _remove_opp_table()

Split out parts of _add_opp_table() and _remove_opp_table() into
separate routines. This improves readability as well.

Should result in no functional changes.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Don't expose srcu_head to register notifiers
Viresh Kumar [Mon, 2 Jan 2017 09:11:03 +0000 (14:41 +0530)]
PM / OPP: Don't expose srcu_head to register notifiers

Let the OPP core provide helpers to register notifiers for any device,
instead of exposing srcu_head outside of the core.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Rename dev_pm_opp_get_suspend_opp() and return OPP rate
Viresh Kumar [Mon, 2 Jan 2017 09:11:02 +0000 (14:41 +0530)]
PM / OPP: Rename dev_pm_opp_get_suspend_opp() and return OPP rate

There is only one user of dev_pm_opp_get_suspend_opp() and that uses it
to get the OPP rate for the suspend_opp.

Rename dev_pm_opp_get_suspend_opp() as dev_pm_opp_get_suspend_opp_freq()
and return the rate directly from it.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Don't allocate OPP table from _opp_allocate()
Viresh Kumar [Mon, 2 Jan 2017 09:11:01 +0000 (14:41 +0530)]
PM / OPP: Don't allocate OPP table from _opp_allocate()

There is no point in trying to find/allocate the table for every OPP
that is added for a device. It would be far more efficient to allocate
the table only once and pass its pointer to the routines that add the
OPP entry.

Locking is removed from _opp_add_static_v2() and _opp_add_v1() now as
the callers call them with that lock already held.

Call to _remove_opp_table() routine is also removed from _opp_free()
now, as opp_table isn't allocated from within _opp_allocate(). This is
handled by the routines which created the OPP table in the first place.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Rename and split _dev_pm_opp_remove_table()
Viresh Kumar [Mon, 2 Jan 2017 09:11:00 +0000 (14:41 +0530)]
PM / OPP: Rename and split _dev_pm_opp_remove_table()

Later patches would want to remove OPP table (and its OPPs) using the
opp_table pointer instead of 'dev'.

In order to prepare for that, rename _dev_pm_opp_remove_table() as
_dev_pm_opp_find_and_remove_table() split out part of it.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Add light weight _opp_free() routine
Viresh Kumar [Mon, 2 Jan 2017 09:10:59 +0000 (14:40 +0530)]
PM / OPP: Add light weight _opp_free() routine

The OPPs which are never successfully added using _opp_add() are not
required to be freed with the _opp_remove() routine, as a simple kfree()
is enough for them.

Introduce a new light weight routine _opp_free(), which will do that.

That also helps us removing the 'notify' parameter to _opp_remove(),
which isn't required anymore.

Note that _opp_free() contains a call to _remove_opp_table() as the OPP
table might have been added for this very OPP only. The
_remove_opp_table() routine returns quickly if there are more OPPs in
the table. This will be simplified in later patches though.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Error out on failing to add static OPPs for v1 bindings
Viresh Kumar [Mon, 2 Jan 2017 09:10:58 +0000 (14:40 +0530)]
PM / OPP: Error out on failing to add static OPPs for v1 bindings

The code adding static OPPs for V2 bindings already does so. Make the V1
bindings specific code behave the same.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Rename _allocate_opp() to _opp_allocate()
Viresh Kumar [Mon, 2 Jan 2017 09:10:57 +0000 (14:40 +0530)]
PM / OPP: Rename _allocate_opp() to _opp_allocate()

Make the naming consistent with how other routines are named.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Remove useless TODO
Viresh Kumar [Mon, 2 Jan 2017 09:10:56 +0000 (14:40 +0530)]
PM / OPP: Remove useless TODO

This TODO doesn't make sense anymore as we have all the information in a
single OPP table. Remove it.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoPM / OPP: Fix memory leak while adding duplicate OPPs
Viresh Kumar [Mon, 2 Jan 2017 09:10:55 +0000 (14:40 +0530)]
PM / OPP: Fix memory leak while adding duplicate OPPs

There are two types of duplicate OPPs that get different behavior from
the core:
A) An earlier OPP is marked 'available' and has same freq/voltages as
   the new one.
B) An earlier OPP with same frequency, but is marked 'unavailable' OR
   doesn't have same voltages as the new one.

The OPP core returns 0 for the first one, but -EEXIST for the second.

While the OPP core returns 0 for the first case, its callers don't free
the newly allocated OPP structure which isn't used anymore. Fix that by
returning -EBUSY instead of 0, but make the callers return 0 eventually.

As this isn't a critical fix, its not getting marked for stable kernel.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: brcmstb-avs-cpufreq: properly retrieve P-state upon suspend
Markus Mayer [Mon, 19 Dec 2016 20:10:28 +0000 (12:10 -0800)]
cpufreq: brcmstb-avs-cpufreq: properly retrieve P-state upon suspend

The AVS GET_PMAP command does return a P-state along with the P-map
information. However, that P-state is the initial P-state when the
P-map was first downloaded to AVS. It is *not* the current P-state.

Therefore, we explicitly retrieve the P-state using the GET_PSTATE
command.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agocpufreq: brcmstb-avs-cpufreq: extend sysfs entry brcm_avs_pmap
Markus Mayer [Mon, 19 Dec 2016 20:10:27 +0000 (12:10 -0800)]
cpufreq: brcmstb-avs-cpufreq: extend sysfs entry brcm_avs_pmap

We extend the brcm_avs_pmap sysfs entry (which issues the GET_PMAP
command to AVS) to include all fields from struct pmap. This means
adding mode (AVS, DVS, DVFS) and state (the P-state) to the output.

Signed-off-by: Markus Mayer <mmayer@broadcom.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
7 years agoMerge tag 'drm-fixes-for-v4.10-rc6-part-two' of git://people.freedesktop.org/~airlied...
Linus Torvalds [Fri, 27 Jan 2017 02:04:56 +0000 (18:04 -0800)]
Merge tag 'drm-fixes-for-v4.10-rc6-part-two' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "This is the main request for rc6, since really the one earlier was the
  rc5 one :-)

  The main thing are the nouveau specific race fixes for the connector
  locking bug we fixed in -next and reverted here as it has quite large
  prereqs. These two fixes should solve the problem at that level and we
  can fix it properly in 4.11

  Otherwise i915 has a bunch of changes, one ABI change for GVT related
  stuff, some VC4 leak fixes, one core fence fix and some AMD changes,
  oh and one ast hang avoidance fix.

  Hoping it calms down around now"

* tag 'drm-fixes-for-v4.10-rc6-part-two' of git://people.freedesktop.org/~airlied/linux: (25 commits)
  drm/nouveau: Handle fbcon suspend/resume in seperate worker
  drm/nouveau: Don't enabling polling twice on runtime resume
  drm/ast: Fixed system hanged if disable P2A
  Revert "drm/radeon: always apply pci shutdown callbacks"
  drm/i915: reinstate call to trace_i915_vma_bind
  drm/i915: Move atomic state free from out of fence release
  drm/i915: Check for NULL atomic state in intel_crtc_disable_noatomic()
  drm/i915: Fix calculation of rotated x and y offsets for planar formats
  drm/i915: Don't init hpd polling for vlv and chv from runtime_suspend()
  drm/i915: Don't leak edid in intel_crt_detect_ddc()
  drm/i915: Release temporary load-detect state upon switching
  drm/i915: prevent crash with .disable_display parameter
  drm/i915: Avoid drm_atomic_state_put(NULL) in intel_display_resume
  MAINTAINERS: update new mail list for intel gvt driver
  drm/i915/gvt: Fix kmem_cache_create() name
  drm/i915/gvt/kvmgt: mdev ABI is available_instances, not available_instance
  drm/amdgpu: fix unload driver issue for virtual display
  drm/amdgpu: check ring being ready before using
  drm/vc4: Return -EINVAL on the overflow checks failing.
  drm/vc4: Fix an integer overflow in temporary allocation layout.
  ...