Cpufreq governors like the ondemand governor calculate the load on the CPU
periodically by employing deferrable timers. A deferrable timer won't fire
if the CPU is completely idle (and there are no other timers to be run), in
order to avoid unnecessary wakeups and thus save CPU power.

However, the load calculation logic is agnostic to all this, and this can
lead to the problem described below.

Time (ms)               CPU 1

100                Task-A running

110                Governor's timer fires, finds load as 100% in the last
                   10ms interval and increases the CPU frequency.

110.5              Task-A running

120		   Governor's timer fires, finds load as 100% in the last
		   10ms interval and increases the CPU frequency.

125		   Task-A went to sleep. With nothing else to do, CPU 1
		   went completely idle.

200		   Task-A woke up and started running again.

200.5		   Governor's deferred timer (which was originally programmed
		   to fire at time 130) fires now. It calculates load for the
		   time period 120 to 200.5, and finds the load is almost zero.
		   Hence it decreases the CPU frequency to the minimum.

210		   Governor's timer fires, finds load as 100% in the last
		   10ms interval and increases the CPU frequency.

So, after the workload woke up and started running, the frequency was suddenly
dropped to absolute minimum, and after that, there was an unnecessary delay of
10ms (sampling period) to increase the CPU frequency back to a reasonable value.
And this pattern repeats for every wake-up-from-cpu-idle for that workload.
This can be quite undesirable for latency- or response-time sensitive bursty
workloads. So we need to fix the governor's logic to detect such wake-up-from-
cpu-idle scenarios and start the workload at a reasonably high CPU frequency.

One extreme solution would be to fake a load of 100% in such scenarios. But
that might lead to undesirable side-effects such as frequency spikes (which
might also need voltage changes) especially if the previous frequency happened
to be very low.

We just want to avoid the stupidity of dropping down the frequency to a minimum
and then enduring a needless (and long) delay before ramping it up back again.
So, let us simply carry forward the previous load - that is, let us just pretend
that the 'load' for the current time-window is the same as the load for the
previous window. That way, the frequency and voltage will continue to be set
to whatever values they were set at previously. This means that bursty workloads
will get a chance to influence the CPU frequency at which they wake up from
cpu-idle, based on their past execution history. Thus, they might be able to
avoid suffering from slow wakeups and long response-times.

However, we should take care not to over-do this. For example, such a "copy
previous load" logic will benefit cases like this: (where # represents busy
and . represents idle)

##########.........#########.........###########...........##########........

but it will be detrimental in cases like the one shown below, because it will
retain the high frequency (copied from the previous interval) even in a mostly
idle system:

##########.........#.................#.....................#...............

(i.e., the workload finished and the remaining tasks are such that their busy
periods are smaller than the sampling interval, which causes the timer to
always get deferred. So, this will make the copy-previous-load logic copy
the initial high load to subsequent idle periods over and over again, thus
keeping the frequency high unnecessarily).

So, we modify this copy-previous-load logic such that it is used only once
upon every wakeup-from-idle. Thus if we have 2 consecutive idle periods, the
previous load won't get blindly copied over; cpufreq will freshly evaluate the
load in the second idle interval, thus ensuring that the system comes back to
its normal state.

[ The right way to solve this whole problem is to teach the CPU frequency
governors to also track load on a per-task basis, not just a per-CPU basis,
and then use both the data sources intelligently to set the appropriate
frequency on the CPUs. But that involves redesigning the cpufreq subsystem,
so this patch should make the situation bearable until then. ]

Experimental results:
+-------------------+

I ran a modified version of ebizzy (called 'sleeping-ebizzy') that sleeps in
between its execution such that its total utilization can be a user-defined
value, say 10% or 20% (higher the utilization specified, lesser the amount of
sleeps injected). This ebizzy was run with a single-thread, tied to CPU 8.

Behavior observed with tracing (sample taken from 40% utilization runs):
------------------------------------------------------------------------

Without patch:
~~~~~~~~~~~~~~
kworker/8:2-12137  416.335742: cpu_frequency: state=2061000 cpu_id=8
kworker/8:2-12137  416.335744: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40753  416.345741: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137  416.345744: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-12137  416.345746: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40753  416.355738: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
<snip>  ---------------------------------------------------------------------  <snip>
      <...>-40753  416.402202: sched_switch: prev_comm=ebizzy ==> next_comm=swapper/8
     <idle>-0      416.502130: sched_switch: prev_comm=swapper/8 ==> next_comm=ebizzy
      <...>-40753  416.505738: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137  416.505739: cpu_frequency: state=2061000 cpu_id=8
kworker/8:2-12137  416.505741: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40753  416.515739: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137  416.515742: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-12137  416.515744: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy

Observation: Ebizzy went idle at 416.402202, and started running again at
416.502130. But cpufreq noticed the long idle period, and dropped the frequency
at 416.505739, only to increase it back again at 416.515742, realizing that the
workload is in-fact CPU bound. Thus ebizzy needlessly ran at the lowest frequency
for almost 13 milliseconds (almost 1 full sample period), and this pattern
repeats on every sleep-wakeup. This could hurt latency-sensitive workloads quite
a lot.

With patch:
~~~~~~~~~~~

kworker/8:2-29802  464.832535: cpu_frequency: state=2061000 cpu_id=8
<snip>  ---------------------------------------------------------------------  <snip>
kworker/8:2-29802  464.962538: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40738  464.972533: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802  464.972536: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-29802  464.972538: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40738  464.982531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
<snip>  ---------------------------------------------------------------------  <snip>
kworker/8:2-29802  465.022533: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40738  465.032531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802  465.032532: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40738  465.035797: sched_switch: prev_comm=ebizzy ==> next_comm=swapper/8
     <idle>-0      465.240178: sched_switch: prev_comm=swapper/8 ==> next_comm=ebizzy
      <...>-40738  465.242533: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802  465.242535: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
      <...>-40738  465.252531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2

Observation: Ebizzy went idle at 465.035797, and started running again at
465.240178. Since ebizzy was the only real workload running on this CPU,
cpufreq retained the frequency at 4.1Ghz throughout the run of ebizzy, no
matter how many times ebizzy slept and woke-up in-between. Thus, ebizzy
got the 10ms worth of 4.1 Ghz benefit during every sleep-wakeup (as compared
to the run without the patch) and this boost gave a modest improvement in total
throughput, as shown below.

Sleeping-ebizzy records-per-second:
-----------------------------------

Utilization  Without patch  With patch  Difference (Absolute and % values)
    10%         274767        277046        +  2279 (+0.829%)
    20%         543429        553484        + 10055 (+1.850%)
    40%        1090744       1107959        + 17215 (+1.578%)
    60%        1634908       1662018        + 27110 (+1.658%)

A rudimentary and somewhat approximately latency-sensitive workload such as
sleeping-ebizzy itself showed a consistent, noticeable performance improvement
with this patch. Hence, workloads that are truly latency-sensitive will benefit
quite a bit from this change. Moreover, this is an overall win-win since this
patch does not hurt power-savings at all (because, this patch does not reduce
the idle time or idle residency; and the high frequency of the CPU when it goes
to cpu-idle does not affect/hurt the power-savings of deep idle states).
Signed-off-by: NSrivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Reviewed-by: NGautham R. Shenoy <ego@linux.vnet.ibm.com>
Acked-by: NViresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Commit 6712d293 (cpufreq: ppc-corenet-cpufreq: Fix __udivdi3 modpost
error) used the remainder from do_div instead of the quotient.  Fix that
and add one to ensure minimum is met.

Fixes: 6712d293 (cpufreq: ppc-corenet-cpufreq: Fix __udivdi3 modpost error)
Signed-off-by: NEd Swarthout <Ed.Swarthout@freescale.com>
Cc: 3.15+ <stable@vger.kernel.org> # 3.15+
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

This reverts commit 4920ab84 (cpufreq: Enable big.LITTLE cpufreq
driver on arm64) that breaks build on arm64.

Fixes: 4920ab84 (cpufreq: Enable big.LITTLE cpufreq driver on arm64)
Reported-by: NCatalin Marinas <catalin.marinas@arm.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Tegra has been switching to intermediate frequency (pll_p_clk) forever.
CPUFreq core has better support for handling notifications for these
frequencies and so we can adapt Tegra's driver to it.

Also do a WARN() if clk_set_parent() fails while moving back to pll_x
as we should have atleast restored to earlier frequency on error.
Tested-by: NStephen Warren <swarren@nvidia.com>
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: NDoug Anderson <dianders@chromium.org>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Douglas Anderson, recently pointed out an interesting problem due to which
udelay() was expiring earlier than it should.

While transitioning between frequencies few platforms may temporarily switch to
a stable frequency, waiting for the main PLL to stabilize.

For example: When we transition between very low frequencies on exynos, like
between 200MHz and 300MHz, we may temporarily switch to a PLL running at 800MHz.
No CPUFREQ notification is sent for that. That means there's a period of time
when we're running at 800MHz but loops_per_jiffy is calibrated at between 200MHz
and 300MHz. And so udelay behaves badly.

To get this fixed in a generic way, introduce another set of callbacks
get_intermediate() and target_intermediate(), only for drivers with
target_index() and CPUFREQ_ASYNC_NOTIFICATION unset.

get_intermediate() should return a stable intermediate frequency platform wants
to switch to, and target_intermediate() should set CPU to that frequency,
before jumping to the frequency corresponding to 'index'. Core will take care of
sending notifications and driver doesn't have to handle them in
target_intermediate() or target_index().

NOTE: ->target_index() should restore to policy->restore_freq in case of
failures as core would send notifications for that.
Tested-by: NStephen Warren <swarren@nvidia.com>
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: NDoug Anderson <dianders@chromium.org>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Conflicts:
	arch/mips/loongson/lemote-2f/clock.c
	drivers/cpufreq/intel_pstate.c

This change makes the busy calculation using 64 bit math which prevents
overflow for large values of aperf/mperf.

Cc: 3.14+ <stable@vger.kernel.org> # 3.14+
Signed-off-by: NDoug Smythies <dsmythies@telus.net>
Signed-off-by: NDirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

The PID assumes that samples are of equal time, which for a deferable
timers this is not true when the system goes idle.  This causes the
PID to take a long time to converge to the min P state and depending
on the pattern of the idle load can make the P state appear stuck.

The hold-off value of three sample times before using the scaling is
to give a grace period for applications that have high performance
requirements and spend a lot of time idle,  The poster child for this
behavior is the ffmpeg benchmark in the Phoronix test suite.

Cc: 3.14+ <stable@vger.kernel.org> # 3.14+
Signed-off-by: NDirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Changing to fixed point math throughout the busy calculation in
commit e66c1768 (Change busy calculation to use fixed point
math.) Introduced some inaccuracies by rounding the busy value at two
points in the calculation.  This change removes roundings and moves
the rounding to the output of the PID where the calculations are
complete and the value returned as an integer.

Fixes: e66c1768 (intel_pstate: Change busy calculation to use fixed point math.)
Reported-by: NDoug Smythies <dsmythies@telus.net>
Cc: 3.14+ <stable@vger.kernel.org> # 3.14+
Signed-off-by: NDirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Commit fcb6a15c (intel_pstate: Take core C0 time into account for core
busy calculation) introduced a regression referenced below.  The issue
with "lockup" after suspend that this commit was addressing is now dealt
with in the suspend path.

Fixes: fcb6a15c (intel_pstate: Take core C0 time into account for core busy calculation)
Link: https://bugzilla.kernel.org/show_bug.cgi?id=66581
Link: https://bugzilla.kernel.org/show_bug.cgi?id=75121Reported-by: NDoug Smythies <dsmythies@telus.net>
Cc: 3.14+ <stable@vger.kernel.org> # 3.14+
Signed-off-by: NDirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Pull powerpc fix from Ben Herrenschmidt:
 "Here's just one trivial patch to wire up sys_renameat2 which I seem to
  have completely missed so far.

  (My test build scripts fwd me warnings but miss the ones generated for
  missing syscalls)"

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
  powerpc: Wire renameat2() syscall

Pull MIPS fixes from Ralf Baechle:
 "A fair number of fixes across the field.  Nothing terribly
  complicated; the one liners in below changelog should be fairly
  descriptive.

  Noteworthy is the SB1 change which the result of changes to binutils
  resulting in one big gas warning for most files being assembled as
  well as the asid_cache and branch emulation fixes which fix corruption
  or possible uninteded behaviour of kernel or application code.  The
  remainder of fixes are more platforms or subsystem specific"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: R46000: Fix Micro-assembler field overflow for R4600 V2
  MIPS: ptrace: Avoid smp_processor_id() in preemptible code
  MIPS: Lemote 2F: cs5536: mfgpt: use raw locks
  MIPS: SB1: Fix excessive kernel warnings.
  MIPS: RC32434: fix broken PCI resource initialization
  MIPS: malta: memory.c: Initialize the 'memsize' variable
  MIPS: Fix typo when reporting cache and ftlb errors for ImgTec cores
  MIPS: Fix inconsistancy of __NR_Linux_syscalls value
  MIPS: Fix branch emulation of branch likely instructions.
  MIPS: Fix a typo error in AUDIT_ARCH definition
  MIPS: Change type of asid_cache to unsigned long

Pull scheduler fixes from Ingo Molnar:
 "Various fixlets, mostly related to the (root-only) SCHED_DEADLINE
  policy, but also a hotplug bug fix and a fix for a NR_CPUS related
  overallocation bug causing a suspend/resume regression"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: Fix hotplug vs. set_cpus_allowed_ptr()
  sched/cpupri: Replace NR_CPUS arrays
  sched/deadline: Replace NR_CPUS arrays
  sched/deadline: Restrict user params max value to 2^63 ns
  sched/deadline: Change sched_getparam() behaviour vs SCHED_DEADLINE
  sched: Disallow sched_attr::sched_policy < 0
  sched: Make sched_setattr() correctly return -EFBIG

Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

Pull core futex/rtmutex fixes from Thomas Gleixner:
 "Three fixlets for long standing issues in the futex/rtmutex code
  unearthed by Dave Jones syscall fuzzer:

   - Add missing early deadlock detection checks in the futex code
   - Prevent user space from attaching a futex to kernel threads
   - Make the deadlock detector of rtmutex work again

  Looks large, but is more comments than code change"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  rtmutex: Fix deadlock detector for real
  futex: Prevent attaching to kernel threads
  futex: Add another early deadlock detection check

Pull drm fixes from Dave Airlie:
 "Mostly quiet now:

  i915:
    fixing userspace visiblie issues, all stable marked

  radeon:
    one more pll fix, two crashers, one suspend/resume regression"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/radeon: Resume fbcon last
  drm/radeon: only allocate necessary size for vm bo list
  drm/radeon: don't allow RADEON_GEM_DOMAIN_CPU for command submission
  drm/radeon: avoid crash if VM command submission isn't available
  drm/radeon: lower the ref * post PLL maximum once more
  drm/i915: Prevent negative relocation deltas from wrapping
  drm/i915: Only copy back the modified fields to userspace from execbuffer
  drm/i915: Fix dynamic allocation of physical handles

lock_parent() very much on purpose does nested locking of dentries, and
is careful to maintain the right order (lock parent first).  But because
it didn't annotate the nested locking order, lockdep thought it might be
a deadlock on d_lock, and complained.

Add the proper annotation for the inner locking of the child dentry to
make lockdep happy.

Introduced by commit 046b961b ("shrink_dentry_list(): take parent's
->d_lock earlier").
Reported-and-tested-by: NJosh Boyer <jwboyer@fedoraproject.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

So a few people complained that

commit 177cf92d
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Apr 1 22:14:59 2014 +0200

    drm/crtc-helpers: fix dpms on logic

which was merged into 3.15-rc1, broke resume on radeons. Strangely git
bisect lead everyone to

commit 25f397a4
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Fri Jul 19 18:57:11 2013 +0200

    drm/crtc-helper: explicit DPMS on after modeset

which was merged long ago and actually part of 3.14.

Digging deeper I've noticed (again) that the call to
drm_helper_resume_force_mode in the radeon resume handlers was a no-op
previously because everything gets shut down on suspend. radeon does
this with explicit calls to drm_helper_connector_dpms with DPMS_OFF.
But with 177c we now force the dpms state to ON, so suddenly
resume_force_mode actually forced the crtcs back on.

This is the intention of the change after all, the problem is that
radeon resumes the fbdev console layer _before_ restoring the display,
through calling fb_set_suspend. And fbcon does an immediate ->set_par,
which in turn causes the same forced mode restore to happen.

Two concurrent modeset operations didn't lead to happiness. Fix this
by delaying the fbcon resume until the end of the readeon resum
functions.

v2: Fix up a bit of the spelling fail.

References: https://lkml.org/lkml/2014/5/29/1043
References: https://lkml.org/lkml/2014/5/2/388
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=74751Tested-by: NKen Moffat <zarniwhoop@ntlworld.com>
Cc: Alex Deucher <alexdeucher@gmail.com>
Cc: Ken Moffat <zarniwhoop@ntlworld.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@gmail.com>

this is the next pull request for stashed up radeon fixes for 3.15. This is finally calming down with only four patches in this pull request.

* 'drm-fixes-3.15' of git://people.freedesktop.org/~deathsimple/linux:
  drm/radeon: only allocate necessary size for vm bo list
  drm/radeon: don't allow RADEON_GEM_DOMAIN_CPU for command submission
  drm/radeon: avoid crash if VM command submission isn't available
  drm/radeon: lower the ref * post PLL maximum once more

Pull input subsystem fixes from Dmitry Torokhov:
 "A couple of driver/build fixups and also redone quirk for Synaptics
  touchpads on Lenovo boxes (now using PNP IDs instead of DMI data to
  limit number of quirks)"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: synaptics - change min/max quirk table to pnp-id matching
  Input: synaptics - add a matches_pnp_id helper function
  Input: synaptics - T540p - unify with other LEN0034 models
  Input: synaptics - add min/max quirk for the ThinkPad W540
  Input: ambakmi - request a shared interrupt for AMBA KMI devices
  Input: pxa27x-keypad - fix generating scancode
  Input: atmel-wm97xx - only build for AVR32
  Input: fix ps2/serio module dependency

Pull firewire fix from Stefan Richter:
 "A regression fix for the IEEE 1394 subsystem: re-enable IRQ-based
  asynchronous request reception at addresses below 128 TB"

* tag 'firewire-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: revert to 4 GB RDMA, fix protocols using Memory Space

Pull device-mapper fixes from Mike Snitzer:
 "A dm-cache stable fix to split discards on cache block boundaries
  because dm-cache cannot yet handle discards that span cache blocks.

  Really fix a dm-mpath LOCKDEP warning that was introduced in -rc1.

  Add a 'no_space_timeout' control to dm-thinp to restore the ability to
  queue IO indefinitely when no data space is available.  This fixes a
  change in behavior that was introduced in -rc6 where the timeout
  couldn't be disabled"

* tag 'dm-3.15-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm mpath: really fix lockdep warning
  dm cache: always split discards on cache block boundaries
  dm thin: add 'no_space_timeout' dm-thin-pool module param

While I play inhouse patches with much memory pressure on qemu-kvm,
3.14 kernel was randomly crashed. The reason was kernel stack overflow.

When I investigated the problem, the callstack was a little bit deeper
by involve with reclaim functions but not direct reclaim path.

I tried to diet stack size of some functions related with alloc/reclaim
so did a hundred of byte but overflow was't disappeard so that I encounter
overflow by another deeper callstack on reclaim/allocator path.

Of course, we might sweep every sites we have found for reducing
stack usage but I'm not sure how long it saves the world(surely,
lots of developer start to add nice features which will use stack
agains) and if we consider another more complex feature in I/O layer
and/or reclaim path, it might be better to increase stack size(
meanwhile, stack usage on 64bit machine was doubled compared to 32bit
while it have sticked to 8K. Hmm, it's not a fair to me and arm64
already expaned to 16K. )

So, my stupid idea is just let's expand stack size and keep an eye
toward stack consumption on each kernel functions via stacktrace of ftrace.
For example, we can have a bar like that each funcion shouldn't exceed 200K
and emit the warning when some function consumes more in runtime.
Of course, it could make false positive but at least, it could make a
chance to think over it.

I guess this topic was discussed several time so there might be
strong reason not to increase kernel stack size on x86_64, for me not
knowing so Ccing x86_64 maintainers, other MM guys and virtio
maintainers.

Here's an example call trace using up the kernel stack:

         Depth    Size   Location    (51 entries)
         -----    ----   --------
   0)     7696      16   lookup_address
   1)     7680      16   _lookup_address_cpa.isra.3
   2)     7664      24   __change_page_attr_set_clr
   3)     7640     392   kernel_map_pages
   4)     7248     256   get_page_from_freelist
   5)     6992     352   __alloc_pages_nodemask
   6)     6640       8   alloc_pages_current
   7)     6632     168   new_slab
   8)     6464       8   __slab_alloc
   9)     6456      80   __kmalloc
  10)     6376     376   vring_add_indirect
  11)     6000     144   virtqueue_add_sgs
  12)     5856     288   __virtblk_add_req
  13)     5568      96   virtio_queue_rq
  14)     5472     128   __blk_mq_run_hw_queue
  15)     5344      16   blk_mq_run_hw_queue
  16)     5328      96   blk_mq_insert_requests
  17)     5232     112   blk_mq_flush_plug_list
  18)     5120     112   blk_flush_plug_list
  19)     5008      64   io_schedule_timeout
  20)     4944     128   mempool_alloc
  21)     4816      96   bio_alloc_bioset
  22)     4720      48   get_swap_bio
  23)     4672     160   __swap_writepage
  24)     4512      32   swap_writepage
  25)     4480     320   shrink_page_list
  26)     4160     208   shrink_inactive_list
  27)     3952     304   shrink_lruvec
  28)     3648      80   shrink_zone
  29)     3568     128   do_try_to_free_pages
  30)     3440     208   try_to_free_pages
  31)     3232     352   __alloc_pages_nodemask
  32)     2880       8   alloc_pages_current
  33)     2872     200   __page_cache_alloc
  34)     2672      80   find_or_create_page
  35)     2592      80   ext4_mb_load_buddy
  36)     2512     176   ext4_mb_regular_allocator
  37)     2336     128   ext4_mb_new_blocks
  38)     2208     256   ext4_ext_map_blocks
  39)     1952     160   ext4_map_blocks
  40)     1792     384   ext4_writepages
  41)     1408      16   do_writepages
  42)     1392      96   __writeback_single_inode
  43)     1296     176   writeback_sb_inodes
  44)     1120      80   __writeback_inodes_wb
  45)     1040     160   wb_writeback
  46)      880     208   bdi_writeback_workfn
  47)      672     144   process_one_work
  48)      528     112   worker_thread
  49)      416     240   kthread
  50)      176     176   ret_from_fork

[ Note: the problem is exacerbated by certain gcc versions that seem to
  generate much bigger stack frames due to apparently bad coalescing of
  temporaries and generating too many spills.  Rusty saw gcc-4.6.4 using
  35% more stack on the virtio path than 4.8.2 does, for example.

  Minchan not only uses such a bad gcc version (4.6.3 in his case), but
  some of the stack use is due to debugging (CONFIG_DEBUG_PAGEALLOC is
  what causes that kernel_map_pages() frame, for example). But we're
  clearly getting too close.

  The VM code also seems to have excessive stack frames partly for the
  same compiler reason, triggered by excessive inlining and lots of
  function arguments.

  We need to improve on our stack use, but in the meantime let's do this
  simple stack increase too.  Unlike most earlier reports, there is
  nothing simple that stands out as being really horribly wrong here,
  apart from the fact that the stack frames are just bigger than they
  should need to be.        - Linus ]
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Michael S Tsirkin <mst@redhat.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: PJ Waskiewicz <pjwaskiewicz@gmail.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Pull vfs dcache livelock fix from Al Viro:
 "Fixes for livelocks in shrink_dentry_list() introduced by fixes to
  shrink list corruption; the root cause was that trylock of parent's
  ->d_lock could be disrupted by d_walk() happening on other CPUs,
  resulting in shrink_dentry_list() making no progress *and* the same
  d_walk() being called again and again for as long as
  shrink_dentry_list() doesn't get past that mess.

  The solution is to have shrink_dentry_list() treat that trylock
  failure not as 'try to do the same thing again', but 'lock them in the
  right order'"

* 'for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  dentry_kill() doesn't need the second argument now
  dealing with the rest of shrink_dentry_list() livelock
  shrink_dentry_list(): take parent's ->d_lock earlier
  expand dentry_kill(dentry, 0) in shrink_dentry_list()
  split dentry_kill()
  lift the "already marked killed" case into shrink_dentry_list()

it's 1 in the only remaining caller.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

We have the same problem with ->d_lock order in the inner loop, where
we are dropping references to ancestors.  Same solution, basically -
instead of using dentry_kill() we use lock_parent() (introduced in the
previous commit) to get that lock in a safe way, recheck ->d_count
(in case if lock_parent() has ended up dropping and retaking ->d_lock
and somebody managed to grab a reference during that window), trylock
the inode->i_lock and use __dentry_kill() to do the rest.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The cause of livelocks there is that we are taking ->d_lock on
dentry and its parent in the wrong order, forcing us to use
trylock on the parent's one.  d_walk() takes them in the right
order, and unfortunately it's not hard to create a situation
when shrink_dentry_list() can't make progress since trylock
keeps failing, and shrink_dcache_parent() or check_submounts_and_drop()
keeps calling d_walk() disrupting the very shrink_dentry_list() it's
waiting for.

Solution is straightforward - if that trylock fails, let's unlock
the dentry itself and take locks in the right order.  We need to
stabilize ->d_parent without holding ->d_lock, but that's doable
using RCU.  And we'd better do that in the very beginning of the
loop in shrink_dentry_list(), since the checks on refcount, etc.
would need to be redone anyway.

That deals with a half of the problem - killing dentries on the
shrink list itself.  Another one (dropping their parents) is
in the next commit.

locking parent is interesting - it would be easy to do rcu_read_lock(),
lock whatever we think is a parent, lock dentry itself and check
if the parent is still the right one.  Except that we need to check
that *before* locking the dentry, or we are risking taking ->d_lock
out of order.  Fortunately, once the D1 is locked, we can check if
D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent
can start or stop pointing to D1 only under D1->d_lock, so taking
D1->d_lock is enough.  In other words, the right solution is
rcu_read_lock/lock what looks like parent right now/check if it's
still our parent/rcu_read_unlock/lock the child.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

No need to always allocate the theoretical maximum here.
Signed-off-by: NChristian König <christian.koenig@amd.com>

It hangs the hardware.
Signed-off-by: NMarek Olšák <marek.olsak@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org

Signed-off-by: NChristian König <christian.koenig@amd.com>
CC: stable@vger.kernel.org

Let's be conservative and use 100 here until we find something better.

Bugs: https://bugzilla.kernel.org/show_bug.cgi?id=75241Signed-off-by: NChristian König <christian.koenig@amd.com>

Pull ARM fixes from Russell King:
 "The usual random collection of relatively small ARM fixes"

* 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  ARM: 8063/1: bL_switcher: fix individual online status reporting of removed CPUs
  ARM: 8064/1: fix v7-M signal return
  ARM: 8057/1: amba: Add Qualcomm vendor ID.
  ARM: 8052/1: unwind: Fix handling of "Pop r4-r[4+nnn],r14" opcode
  ARM: 8051/1: put_user: fix possible data corruption in put_user
  ARM: 8048/1: fix v7-M setup stack location

Pull arm64 fix from Will Deacon:
 "Fix CoW regression for transparent hugepages by routing set_pmd_at to
  set_pte_at, which correctly handles PTE_WRITE and will mark the
  resulting table entry as read-only where appropriate"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: mm: fix pmd_write CoW brokenness

Pull ACPI and power management fixes from Rafael Wysocki:
 "These are three stable-candidate fixes, one for the ACPI thermal
  driver and two for cpufreq drivers.

  Specifics:

   - A workqueue is destroyed too early during the ACPI thermal driver
     module unload which leads to a NULL pointer dereference in the
     driver's remove callback.  Fix from Aaron Lu.

   - A wrong argument is passed to devm_regulator_get_optional() in the
     probe routine of the cpu0 cpufreq driver which leads to resource
     leaks if the driver is unbound from the cpufreq platform device.
     Fix from Lucas Stach.

   - A lock is missing in cpufreq_governor_dbs() which leads to memory
     corruption and NULL pointer dereferences during system
     suspend/resume, for example.  Fix from Bibek Basu"

* tag 'pm+acpi-3.15-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI / thermal: fix workqueue destroy order
  cpufreq: cpu0: drop wrong devm usage
  cpufreq: remove race while accessing cur_policy

Pull clock fixes from Mike Turquette:
 "Small number of user-visible regression fixes for clock drivers.

  There is a memory leak fix for an ST platform, an infinite Loop Of
  Doom fix for the recent changes to the basic clock divider (hopefully
  the last fix for those recent changes) and some Tegra PLL changes
  which keep PCI from being hosed on that platform"

* tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mike.turquette/linux:
  clk: st: Fix memory leak
  clk: divider: Fix table round up function
  clk: tegra: Fix enabling of PLLE
  clk: tegra: Introduce divider mask and shift helpers
  clk: tegra: Fix PLLE programming

Undo a feature introduced in v3.14 by commit fcd46b34
"firewire: Enable remote DMA above 4 GB".  That change raised the
minimum address at which protocol drivers and user programs can register
for request reception from 0x0001'0000'0000 to 0x8000'0000'0000.
It turned out that at least one vendor-specific protocol exists which
uses lower addresses:  https://bugzilla.kernel.org/show_bug.cgi?id=76921

For the time being, revert most of commit fcd46b34 so that affected
protocols work like with kernel v3.13 and before.  Just keep the valid
documentation parts from the regressing commit, and the ability to
identify controllers which could be programmed to accept >32 bit
physical DMA addresses.  The rest of fcd46b34 should probably be
brought back as an optional instead of default feature.
Reported-by: NFabien Spindler <fabien.spindler@inria.fr>
Cc: <stable@vger.kernel.org> # 3.14+
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Result will be massaged to saner shape in the next commits.  It is
ugly, no questions - the point of that one is to be a provably
equivalent transformation (and it might be worth splitting a bit
more).
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

... into trylocks and everything else.  The latter (actual killing)
is __dentry_kill().
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Commit 9c7e535f ("arm64: mm: Route pmd thp functions through pte
equivalents") changed the pmd manipulator and accessor functions to
convert the target pmd to a pte, process it with the pte functions, then
convert it back. Along the way, we gained support for PTE_WRITE, however
this is completely ignored by set_pmd_at, and so we fail to set the
PMD_SECT_RDONLY for PMDs, resulting in all sorts of lovely failures (like
CoW not working).

Partially reverting the offending commit (by making use of
PMD_SECT_RDONLY explicitly for pmd_{write,wrprotect,mkwrite} functions)
leads to further issues because pmd_write can then return potentially
incorrect values for page table entries marked as RDONLY, leading to
BUG_ON(pmd_write(entry)) tripping under some THP workloads.

This patch fixes the issue by routing set_pmd_at through set_pte_at,
which correctly takes the PTE_WRITE flag into account. Given that
THP mappings are always anonymous, the additional cache-flushing code
in __sync_icache_dcache won't impose any significant overhead as the
flush will be skipped.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: NSteve Capper <steve.capper@arm.com>
Tested-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>