]> git.karo-electronics.de Git - karo-tx-linux.git/log
karo-tx-linux.git
9 years agotimekeeping: Rework frequency adjustments to work better w/ nohz
John Stultz [Sat, 7 Dec 2013 01:25:21 +0000 (17:25 -0800)]
timekeeping: Rework frequency adjustments to work better w/ nohz

The existing timekeeping_adjust logic has always been complicated
to understand. Further, since it was developed prior to NOHZ becoming
common, its not surprising it performs poorly when NOHZ is enabled.

Since Miroslav pointed out the problematic nature of the existing code
in the NOHZ case, I've tried to refactor the code to perform better.

The problem with the previous approach was that it tried to adjust
for the total cumulative error using a scaled dampening factor. This
resulted in large errors to be corrected slowly, while small errors
were corrected quickly. With NOHZ the timekeeping code doesn't know
how far out the next tick will be, so this results in bad
over-correction to small errors, and insufficient correction to large
errors.

Inspired by Miroslav's patch, I've refactored the code to try to
address the correction in two steps.

1) Check the future freq error for the next tick, and if the frequency
error is large, try to make sure we correct it so it doesn't cause
much accumulated error.

2) Then make a small single unit adjustment to correct any cumulative
error that has collected over time.

This method performs fairly well in the simulator Miroslav created.

Major credit to Miroslav for pointing out the issue, providing the
original patch to resolve this, a simulator for testing, as well as
helping debug and resolve issues in my implementation so that it
performed closer to his original implementation.

Cc: Miroslav Lichvar <mlichvar@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Reported-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Minor fixup for timespec64->timespec assignment
John Stultz [Wed, 23 Jul 2014 21:35:39 +0000 (14:35 -0700)]
timekeeping: Minor fixup for timespec64->timespec assignment

In the GENERIC_TIME_VSYSCALL_OLD update_vsyscall implementation,
we take the tk_xtime() value, which returns a timespec64, and
store it in a timespec.

This luckily is ok, since the only architectures that use
GENERIC_TIME_VSYSCALL_OLD are ia64 and ppc64, which are both
64 bit systems where timespec64 is the same as a timespec.

Even so, for cleanliness reasons, use the conversion function
to assign the proper type.

Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoftrace: Provide trace clocks monotonic
Thomas Gleixner [Wed, 16 Jul 2014 21:05:25 +0000 (21:05 +0000)]
ftrace: Provide trace clocks monotonic

Expose the new NMI safe accessor to clock monotonic to the tracer.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide fast and NMI safe access to CLOCK_MONOTONIC
Thomas Gleixner [Wed, 16 Jul 2014 21:05:23 +0000 (21:05 +0000)]
timekeeping: Provide fast and NMI safe access to CLOCK_MONOTONIC

Tracers want a correlated time between the kernel instrumentation and
user space. We really do not want to export sched_clock() to user
space, so we need to provide something sensible for this.

Using separate data structures with an non blocking sequence count
based update mechanism allows us to do that. The data structure
required for the readout has a sequence counter and two copies of the
timekeeping data.

On the update side:

  smp_wmb();
  tkf->seq++;
  smp_wmb();
  update(tkf->base[0], tk);
  smp_wmb();
  tkf->seq++;
  smp_wmb();
  update(tkf->base[1], tk);

On the reader side:

  do {
     seq = tkf->seq;
     smp_rmb();
     idx = seq & 0x01;
     now = now(tkf->base[idx]);
     smp_rmb();
  } while (seq != tkf->seq)

So if a NMI hits the update of base[0] it will use base[1] which is
still consistent, but this timestamp is not guaranteed to be monotonic
across an update.

The timestamp is calculated by:

now = base_mono + clock_delta * slope

So if the update lowers the slope, readers who are forced to the
not yet updated second array are still using the old steeper slope.

 tmono
 ^
 |    o  n
 |   o n
 |  u
 | o
 |o
 |12345678---> reader order

 o = old slope
 u = update
 n = new slope

So reader 6 will observe time going backwards versus reader 5.

While other CPUs are likely to be able observe that, the only way
for a CPU local observation is when an NMI hits in the middle of
the update. Timestamps taken from that NMI context might be ahead
of the following timestamps. Callers need to be aware of that and
deal with it.

V2: Got rid of clock monotonic raw and reorganized the data
    structures. Folded in the barrier fix from Mathieu.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoseqcount: Add raw_write_seqcount_latch()
Mathieu Desnoyers [Wed, 16 Jul 2014 21:05:21 +0000 (21:05 +0000)]
seqcount: Add raw_write_seqcount_latch()

For NMI safe access to clock monotonic we use the seqcount LSB as
index of a timekeeper array. The update sequence looks like this:

      smp_wmb();      <- prior stores to a[1]
      seq++;
      smp_wmb();      <- seq increment before update of a[0]
      update(a[0]);
      smp_wmb();      <- update of a[0]
      seq++;
      smp_wmb();      <- seq increment before update of a[1]
      update(a[1]);

To avoid open coded barriers, provide a helper function.

[ tglx: Split out of a combo patch against the first implementation of
   the NMI safe accessor ]

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoseqcount: Provide raw_read_seqcount()
Thomas Gleixner [Wed, 16 Jul 2014 21:05:20 +0000 (21:05 +0000)]
seqcount: Provide raw_read_seqcount()

raw_read_seqcount opens a read critical section of the given seqcount
without any lockdep checking and without checking or masking the
LSB. Calling code is responsible for handling that.

Preparatory patch to provide a NMI safe clock monotonic accessor
function.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use tk_read_base as argument for timekeeping_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:05:18 +0000 (21:05 +0000)]
timekeeping: Use tk_read_base as argument for timekeeping_get_ns()

All the function needs is in the tk_read_base struct. No functional
change for the current code, just a preparatory patch for the NMI safe
accessor to clock monotonic which will use struct tk_read_base as well.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Create struct tk_read_base and use it in struct timekeeper
Thomas Gleixner [Wed, 16 Jul 2014 21:05:16 +0000 (21:05 +0000)]
timekeeping: Create struct tk_read_base and use it in struct timekeeper

The members of the new struct are the required ones for the new NMI
safe accessor to clcok monotonic. In order to reuse the existing
timekeeping code and to make the update of the fast NMI safe
timekeepers a simple memcpy use the struct for the timekeeper as well
and convert all users.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Restructure the timekeeper some more
Thomas Gleixner [Wed, 16 Jul 2014 21:05:15 +0000 (21:05 +0000)]
timekeeping: Restructure the timekeeper some more

Access to time requires to touch two cachelines at minimum

   1) The timekeeper data structure

   2) The clocksource data structure

The access to the clocksource data structure can be avoided as almost
all clocksource implementations ignore the argument to the read
callback, which is a pointer to the clocksource.

But the core needs to touch it to access the members @read and @mask.

So we are better off by copying the @read function pointer and the
@mask from the clocksource to the core data structure itself.

For the most used ktime_get() access all required data including the
@read and @mask copies fits together with the sequence counter into a
single 64 byte cacheline.

For the other time access functions we touch in the current code three
cache lines in the worst case. But with the clocksource data copies we
can reduce that to two adjacent cachelines, which is more efficient
than disjunct cache lines.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoclocksource: Get rid of cycle_last
Thomas Gleixner [Wed, 16 Jul 2014 21:05:13 +0000 (21:05 +0000)]
clocksource: Get rid of cycle_last

cycle_last was added to the clocksource to support the TSC
validation. We moved that to the core code, so we can get rid of the
extra copy.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoclocksource: Move cycle_last validation to core code
Thomas Gleixner [Wed, 16 Jul 2014 21:05:12 +0000 (21:05 +0000)]
clocksource: Move cycle_last validation to core code

The only user of the cycle_last validation is the x86 TSC. In order to
provide NMI safe accessor functions for clock monotonic and
monotonic_raw we need to do that in the core.

We can't do the TSC specific

    if (now < cycle_last)
            now = cycle_last;

for the other wrapping around clocksources, but TSC has
CLOCKSOURCE_MASK(64) which actually does not mask out anything so if
now is less than cycle_last the subtraction will give a negative
result. So we can check for that in clocksource_delta() and return 0
for that case.

Implement and enable it for x86

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoclocksource: Make delta calculation a function
Thomas Gleixner [Wed, 16 Jul 2014 21:05:10 +0000 (21:05 +0000)]
clocksource: Make delta calculation a function

We want to move the TSC sanity check into core code to make NMI safe
accessors to clock monotonic[_raw] possible. For this we need to
sanity check the delta calculation. Create a helper function and
convert all sites to use it.

[ Build fix from jstultz ]

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agowireless: ath9k: Get rid of timespec conversions
Thomas Gleixner [Wed, 16 Jul 2014 21:05:09 +0000 (21:05 +0000)]
wireless: ath9k: Get rid of timespec conversions

We have interfaces. Remove the open coded cruft. Reduces text size
along with the code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: QCA ath9k Development <ath9k-devel@qca.qualcomm.com>
Cc: John W. Linville <linville@tuxdriver.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agodrm: vmwgfx: Use nsec based interfaces
Thomas Gleixner [Wed, 16 Jul 2014 21:05:07 +0000 (21:05 +0000)]
drm: vmwgfx: Use nsec based interfaces

No point in converting timespecs back and forth.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agodrm: i915: Use nsec based interfaces
Thomas Gleixner [Wed, 16 Jul 2014 21:05:06 +0000 (21:05 +0000)]
drm: i915: Use nsec based interfaces

Use ktime_get_raw_ns() and get rid of the back and forth timespec
conversions.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide ktime_get_raw()
Thomas Gleixner [Wed, 16 Jul 2014 21:05:04 +0000 (21:05 +0000)]
timekeeping: Provide ktime_get_raw()

Provide a ktime_t based interface for raw monotonic time.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agohangcheck-timer: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:05:03 +0000 (21:05 +0000)]
hangcheck-timer: Use ktime_get_ns()

There is no point in having a S390 private implementation and there is
no point in using the raw monotonic time. The NTP freqeuency
adjustment of CLOCK_MONOTONIC is really not doing any harm for the
hang check timer.

Use ktime_get_ns() for everything and get rid of the timespec
conversions.

V2: Drop the raw monotonic and the S390 special case

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Simplify timekeeping_clocktai()
Thomas Gleixner [Wed, 16 Jul 2014 21:05:01 +0000 (21:05 +0000)]
timekeeping: Simplify timekeeping_clocktai()

timekeeping_clocktai() is not used in fast pathes, so the extra
timespec conversion is not problematic.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Remove timekeeper.total_sleep_time
Thomas Gleixner [Wed, 16 Jul 2014 21:05:00 +0000 (21:05 +0000)]
timekeeping: Remove timekeeper.total_sleep_time

No more users. Remove it

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Simplify getboottime()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:58 +0000 (21:04 +0000)]
timekeeping: Simplify getboottime()

Subtracting plain nsec values and converting to timespec is simpler
than the whole timespec math. Not really fastpath code, so the
division is not an issue.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use ktime_get_boottime() for get_monotonic_boottime()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:57 +0000 (21:04 +0000)]
timekeeping: Use ktime_get_boottime() for get_monotonic_boottime()

get_monotonic_boottime() is not used in fast pathes, so the extra
timespec conversion is not problematic.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Remove monotonic_to_bootbased
Thomas Gleixner [Wed, 16 Jul 2014 21:04:55 +0000 (21:04 +0000)]
timekeeping: Remove monotonic_to_bootbased

No more users.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agox86: kvm: Make kvm_get_time_and_clockread() nanoseconds based
Thomas Gleixner [Wed, 16 Jul 2014 21:04:54 +0000 (21:04 +0000)]
x86: kvm: Make kvm_get_time_and_clockread() nanoseconds based

Convert the relevant base data right away to nanoseconds instead of
doing the conversion on every readout. Reduces text size by 160 bytes.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: kvm@vger.kernel.org
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agox86: kvm: Use ktime_get_boot_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:52 +0000 (21:04 +0000)]
x86: kvm: Use ktime_get_boot_ns()

Use the new nanoseconds based interface and get rid of the timespec
conversion dance.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: kvm@vger.kernel.org
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoarm: bL_switcher:k Use ktime_get_real_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:50 +0000 (21:04 +0000)]
arm: bL_switcher:k Use ktime_get_real_ns()

Use the nanoseconds based interface instead of converting from a
timespec.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoiio: Use ktime_get_real_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:49 +0000 (21:04 +0000)]
iio: Use ktime_get_real_ns()

No idea why iio needs wall clock based time stamps, but we can avoid
the timespec conversion dance by using the new interfaces.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Jonathan Cameron <jic23@kernel.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agohwmon: ibmaem: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:47 +0000 (21:04 +0000)]
hwmon: ibmaem: Use ktime_get_ns()

Using the wall clock time for delta time calculations is wrong to
begin with because wall clock time can be set from userspace and NTP.
Such data wants to be based on clock monotonic.

The calculations also are done on a nanosecond basis. Use the
nanoseconds based interface right away.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jean Delvare <jdelvare@suse.de>
Acked-by: Jean Delvare <jdelvare@suse.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agofs: lockd: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:46 +0000 (21:04 +0000)]
fs: lockd: Use ktime_get_ns()

Replace the ever recurring:
        ts = ktime_get_ts();
        ns = timespec_to_ns(&ts);
with
        ns = ktime_get_ns();

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agonet: mlx5: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:44 +0000 (21:04 +0000)]
net: mlx5: Use ktime_get_ns()

This code is beyond silly:

     struct timespec ts = ktime_get_ts();
     ktime_t ktime = timespec_to_ktime(ts);

Further down the code builds the delta of two ktime_t values and
converts the result to nanoseconds.

Use ktime_get_ns() and replace all the nonsense.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Eli Cohen <eli@mellanox.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agomisc: ioc4: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:43 +0000 (21:04 +0000)]
misc: ioc4: Use ktime_get_ns()

Replace the ever recurring:
        ts = ktime_get_ts();
        ns = timespec_to_ns(&ts);
with
        ns = ktime_get_ns();

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agomfd: cros_ec_spi: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:41 +0000 (21:04 +0000)]
mfd: cros_ec_spi: Use ktime_get_ns()

Replace the ever recurring:
ts = ktime_get_ts();
ns = timespec_to_ns(&ts);
with
ns = ktime_get_ns();

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoconnector: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:40 +0000 (21:04 +0000)]
connector: Use ktime_get_ns()

Replace the ever recurring:
        ts = ktime_get_ts();
        ns = timespec_to_ns(&ts);
with
        ns = ktime_get_ns();

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agopowerpc: cell: Use ktime_get_ns()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:38 +0000 (21:04 +0000)]
powerpc: cell: Use ktime_get_ns()

Replace the ever recurring:
ts = ktime_get_ts();
ns = timespec_to_ns(&ts);
with
ns = ktime_get_ns();

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agodelayacct: Remove braindamaged type conversions
Thomas Gleixner [Wed, 16 Jul 2014 21:04:37 +0000 (21:04 +0000)]
delayacct: Remove braindamaged type conversions

Converting cputime to timespec and timespec to nanoseconds makes no
sense. Use cputime_to_ns() and be done with it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agodelayacct: Make accounting nanosecond based
Thomas Gleixner [Wed, 16 Jul 2014 21:04:35 +0000 (21:04 +0000)]
delayacct: Make accounting nanosecond based

Kill the timespec juggling and calculate with plain nanoseconds.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agosched: Make task->start_time nanoseconds based
Thomas Gleixner [Wed, 16 Jul 2014 21:04:34 +0000 (21:04 +0000)]
sched: Make task->start_time nanoseconds based

Simplify the timespec to nsec/usec conversions.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agosched: Make task->real_start_time nanoseconds based
Thomas Gleixner [Wed, 16 Jul 2014 21:04:32 +0000 (21:04 +0000)]
sched: Make task->real_start_time nanoseconds based

Simplify the only user of this data by removing the timespec
conversion.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotime: Export nsecs_to_jiffies()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:31 +0000 (21:04 +0000)]
time: Export nsecs_to_jiffies()

Required for moving drivers to the nanosecond based interfaces.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide ktime_get[*]_ns() helpers
Thomas Gleixner [Wed, 16 Jul 2014 21:04:29 +0000 (21:04 +0000)]
timekeeping: Provide ktime_get[*]_ns() helpers

A lot of code converts either timespecs or ktime_t to
nanoseconds. Provide helper functions.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Remove ktime_get_monotonic_offset()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:28 +0000 (21:04 +0000)]
timekeeping: Remove ktime_get_monotonic_offset()

No more users.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agodrm: Use ktime_mono_to_real()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:26 +0000 (21:04 +0000)]
drm: Use ktime_mono_to_real()

Convert the monotonic timestamp with ktime_mono_to_real() in
drm_calc_vbltimestamp_from_scanoutpos().

In get_drm_timestamp we can call either ktime_get() or
ktime_get_real() depending on drm_timestamp_monotonic. No point in
having two calls into the core for CLOCK_REALTIME.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoinput: evdev: Use ktime_mono_to_real()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:25 +0000 (21:04 +0000)]
input: evdev: Use ktime_mono_to_real()

Convert the monotonic timestamp with ktime_mono_to_real() in
evdev_events().

In evdev_queue_syn_dropped() we can call either ktime_get() or
ktime_get_real() depending on the clkid. No point in having two calls
for CLOCK_REALTIME.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimerfd: Use ktime_mono_to_real()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:23 +0000 (21:04 +0000)]
timerfd: Use ktime_mono_to_real()

We have a few other use cases of ktime_get_monotonic_offset() which
can be optimized with ktime_mono_to_real(). The timerfd code uses the
offset only for comparison, so we can use ktime_mono_to_real(0) for
this as well.

Funny enough text size shrinks with that on ARM and x8664 !?

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide ktime_mono_to_any()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:22 +0000 (21:04 +0000)]
timekeeping: Provide ktime_mono_to_any()

ktime based conversion function to map a monotonic time stamp to a
different CLOCK.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping; Use ktime based data for ktime_get_update_offsets_tick()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:20 +0000 (21:04 +0000)]
timekeeping; Use ktime based data for ktime_get_update_offsets_tick()

No need to juggle with timespecs.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use ktime_t data for ktime_get_update_offsets_now()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:19 +0000 (21:04 +0000)]
timekeeping: Use ktime_t data for ktime_get_update_offsets_now()

No need to juggle with timespecs.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use ktime_t based data for ktime_get_clocktai()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:17 +0000 (21:04 +0000)]
timekeeping: Use ktime_t based data for ktime_get_clocktai()

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping; Use ktime_t based data for ktime_get_boottime()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:16 +0000 (21:04 +0000)]
timekeeping; Use ktime_t based data for ktime_get_boottime()

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use ktime_t based data for ktime_get_real()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:14 +0000 (21:04 +0000)]
timekeeping: Use ktime_t based data for ktime_get_real()

Speed up the readout.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide ktime_get_with_offset()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:13 +0000 (21:04 +0000)]
timekeeping: Provide ktime_get_with_offset()

Provide a helper function which lets us implement ktime_t based
interfaces for real, boot and tai clocks.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use ktime_t based data for ktime_get()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:12 +0000 (21:04 +0000)]
timekeeping: Use ktime_t based data for ktime_get()

Speed up ktime_get() by using ktime_t based data. Text size shrinks by
64 bytes on x8664.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide internal ktime_t based data
Thomas Gleixner [Wed, 16 Jul 2014 21:04:10 +0000 (21:04 +0000)]
timekeeping: Provide internal ktime_t based data

The ktime_t based interfaces are used a lot in performance critical
code pathes. Add ktime_t based data so the interfaces don't have to
convert from the xtime/timespec based data.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Use timekeeping_update() instead of memcpy()
Thomas Gleixner [Wed, 16 Jul 2014 21:04:09 +0000 (21:04 +0000)]
timekeeping: Use timekeeping_update() instead of memcpy()

We already have a function which does the right thing, that also makes
sure that the coming ktime_t based cached values are getting updated.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Cache optimize struct timekeeper
Thomas Gleixner [Wed, 16 Jul 2014 21:04:07 +0000 (21:04 +0000)]
timekeeping: Cache optimize struct timekeeper

struct timekeeper is quite badly sorted for the hot readout path. Most
time access functions need to load two cache lines.

Rearrange it so ktime_get() and getnstimeofday() are happy with a
single cache line.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeper: Move tk_xtime to core code
Thomas Gleixner [Wed, 16 Jul 2014 21:04:05 +0000 (21:04 +0000)]
timekeeper: Move tk_xtime to core code

No users outside of the core.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Provide timespec64 based interfaces
Thomas Gleixner [Wed, 16 Jul 2014 21:04:04 +0000 (21:04 +0000)]
timekeeping: Provide timespec64 based interfaces

To convert callers of the core code to timespec64 we need to provide
the proper interfaces.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotime: Consolidate the time accessor prototypes
Thomas Gleixner [Wed, 16 Jul 2014 21:04:02 +0000 (21:04 +0000)]
time: Consolidate the time accessor prototypes

Right now we have time related prototypes in 3 different header
files. Move it to a single timekeeping header file and move the core
internal stuff into a core private header.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Convert timekeeping core to use timespec64s
John Stultz [Wed, 16 Jul 2014 21:04:01 +0000 (21:04 +0000)]
timekeeping: Convert timekeeping core to use timespec64s

Convert the core timekeeping logic to use timespec64s. This moves the
2038 issues out of the core logic and into all of the accessor
functions.

Future changes will need to push the timespec64s out to all
timekeeping users, but that can be done interface by interface.

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotime: More core infrastructure for timespec64
John Stultz [Wed, 16 Jul 2014 21:03:59 +0000 (21:03 +0000)]
time: More core infrastructure for timespec64

Helper and conversion functions for timespec64.

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotime64: Add time64.h header and define struct timespec64
John Stultz [Wed, 16 Jul 2014 21:03:58 +0000 (21:03 +0000)]
time64: Add time64.h header and define struct timespec64

Define the timespec64 structure and standard helper functions.

[ tglx: Make it 32bit only. 64bit really can map timespec to timespec64 ]

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoktime: Change ktime_set() to take 64bit seconds value
John Stultz [Wed, 16 Jul 2014 21:03:56 +0000 (21:03 +0000)]
ktime: Change ktime_set() to take 64bit seconds value

In order to support dates past 2038 on 32bit systems, ktime_set()
needs to handle 64bit second values.

[ tglx: Removed the BITS_PER_LONG check ]

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoktime: Sanitize ktime_to_us/ms conversion
Thomas Gleixner [Wed, 16 Jul 2014 21:03:55 +0000 (21:03 +0000)]
ktime: Sanitize ktime_to_us/ms conversion

With the plain nanoseconds based ktime_t we can simply use
ktime_divns() instead of going through loops and hoops of
timespec/timeval conversion.

Reported-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoktime: Kill non-scalar ktime_t implementation for 2038
John Stultz [Wed, 16 Jul 2014 21:03:53 +0000 (21:03 +0000)]
ktime: Kill non-scalar ktime_t implementation for 2038

The non-scalar ktime_t implementation is basically a timespec
which has to be changed to support dates past 2038 on 32bit
systems.

This patch removes the non-scalar ktime_t implementation, forcing
the scalar s64 nanosecond version on all architectures.

This may have additional performance overhead on some 32bit
systems when converting between ktime_t and timespec structures,
however the majority of 32bit systems (arm and i386) were already
using scalar ktime_t, so no performance regressions will be seen
on those platforms.

On affected platforms, I'm open to finding optimizations, including
avoiding converting to timespecs where possible.

[ tglx: We can now cleanup the ktime_t.tv64 mess, but thats a
  different issue and we can throw a coccinelle script at it ]

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agohrtimer: Cleanup hrtimer accessors to the timekepeing state
John Stultz [Wed, 16 Jul 2014 21:03:52 +0000 (21:03 +0000)]
hrtimer: Cleanup hrtimer accessors to the timekepeing state

Rather then having two similar but totally different implementations
that provide timekeeping state to the hrtimer code, try to unify the
two implementations to be more simliar.

Thus this clarifies ktime_get_update_offsets to
ktime_get_update_offsets_now and changes get_xtime...  to
ktime_get_update_offsets_tick.

Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotimekeeping: Simplify arch_gettimeoffset()
Thomas Gleixner [Wed, 16 Jul 2014 21:03:50 +0000 (21:03 +0000)]
timekeeping: Simplify arch_gettimeoffset()

Provide a default stub function instead of having the extra
conditional. Cuts binary size on a m68k build by ~100 bytes.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotile: Convert VDSO timekeeping to the precise mechanism
Thomas Gleixner [Wed, 16 Jul 2014 21:03:49 +0000 (21:03 +0000)]
tile: Convert VDSO timekeeping to the precise mechanism

The code was only halfarsed converted to the new VSDO update mechanism
and still uses the inaccurate base value which lacks the fractional
part of xtime_nsec. Fix it up.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agotools: add script to test udelay
David Riley [Mon, 16 Jun 2014 21:58:33 +0000 (14:58 -0700)]
tools: add script to test udelay

This script makes use of the udelay_test module to exercise udelay()
and ensure that it is delaying long enough (as compared to ktime).

Signed-off-by: David Riley <davidriley@chromium.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agokernel: time: Add udelay_test module to validate udelay
David Riley [Mon, 16 Jun 2014 21:58:32 +0000 (14:58 -0700)]
kernel: time: Add udelay_test module to validate udelay

Create a module that allows udelay() to be executed to ensure that
it is delaying at least as long as requested (with a little bit of
error allowed).

There are some configurations which don't have reliably udelay
due to using a loop delay with cpufreq changes which should use
a counter time based delay instead.  This test aims to identify
those configurations where timing is unreliable.

Signed-off-by: David Riley <davidriley@chromium.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
9 years agoMerge branch 'clockevents/3.17' of git://git.linaro.org/people/daniel.lezcano/linux...
Thomas Gleixner [Wed, 23 Jul 2014 11:22:59 +0000 (13:22 +0200)]
Merge branch 'clockevents/3.17' of git://git.linaro.org/people/daniel.lezcano/linux into timers/core

Pull clockevents from Danel Lezcano:
 * New timer driver for the Cirrus Logic CLPS711X SoC
 * New driver for the Mediatek SoC which includes:
 * A new function for of, acked by Rob Herring
 * Move the PXA driver to drivers/clocksource, add DT support
 * Optimization of the exynos_mct driver
 * DT support for the renesas timers family.
 * Some Kconfig and driver fixlets

9 years agoclocksource: exynos_mct: Only use 32-bits where possible
Doug Anderson [Fri, 4 Jul 2014 21:43:26 +0000 (06:43 +0900)]
clocksource: exynos_mct: Only use 32-bits where possible

The MCT has a nice 64-bit counter.  That means that we _can_ register
as a 64-bit clocksource and sched_clock.  ...but that doesn't mean we
should.

The 64-bit counter is read by reading two 32-bit registers.  That
means reading needs to be something like:
- Read upper half
- Read lower half
- Read upper half and confirm that it hasn't changed.

That wouldn't be terrible, but:
- THe MCT isn't very fast to access (hundreds of nanoseconds).
- The clocksource is queried _all the time_.

In total system profiles of real workloads on ChromeOS, we've seen
exynos_frc_read() taking 2% or more of CPU time even after optimizing
the 3 reads above to 2 (see below).

The MCT is clocked at ~24MHz on all known systems.  That means that
the 32-bit half of the counter rolls over every ~178 seconds.  This
inspired an optimization in ChromeOS to cache the upper half between
calls, moving 3 reads to 2.  ...but we can do better!  Having a 32-bit
timer that flips every 178 seconds is more than sufficient for Linux.
Let's just use the lower half of the MCT.

Times on 5420 to do 1000000 gettimeofday() calls from userspace:
* Original code:                      1323852 us
* ChromeOS cache upper half:          1173084 us
* ChromeOS + ldmia to optimize:       1045674 us
* Use lower 32-bit only (this code):  1014429 us

As you can see, the time used doesn't increase linearly with the
number of reads and we can make 64-bit work almost as fast as 32-bit
with a bit of assembly code.  But since there's no real gain for
64-bit, let's go with the simplest and fastest implementation.

Note: with this change roughly half the time for gettimeofday() is
spent in exynos_frc_read().  The rest is timer / system call overhead.

Also note: this patch disables the use of the MCT on ARM64 systems
until we've sorted out how to make "cycles_t" always 32-bit.  Really
ARM64 systems should be using arch timers anyway.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: exynos_mct: Use readl_relaxed/writel_relaxed
Doug Anderson [Fri, 4 Jul 2014 21:43:20 +0000 (06:43 +0900)]
clocksource: exynos_mct: Use readl_relaxed/writel_relaxed

Using the __raw functions is discouraged.  Update the file to
consistently use the proper functions.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Signed-off-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoARM: pxa: Add non device-tree timer link to clocksource
Robert Jarzmik [Mon, 14 Jul 2014 16:52:04 +0000 (18:52 +0200)]
ARM: pxa: Add non device-tree timer link to clocksource

As clocksource pxa_timer was moved to clocksource framework, the
pxa_timer initialization needs to be a bit amended, to pass the
necessary informations to clocksource, ie :
 - the timer interrupt (mach specific)
 - the timer registers base (ditto)
 - the timer clockrate

Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoARM: pxa: Add CLKSRC_OF dependency
Robert Jarzmik [Mon, 14 Jul 2014 16:52:03 +0000 (18:52 +0200)]
ARM: pxa: Add CLKSRC_OF dependency

Select CLKSRC_OF for PXA architectures.

Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: pxa: Add device-tree support for PXA timer
Robert Jarzmik [Mon, 14 Jul 2014 16:52:02 +0000 (18:52 +0200)]
clocksource: pxa: Add device-tree support for PXA timer

Add device-tree support to PXA platforms.
The driver still needs to maintain backward non device-tree
compatibility as well, which implies :
 - a non device-tree init function
 - a static registers base address in the driver

Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: pxa: Move PXA timer to clocksource framework
Robert Jarzmik [Mon, 14 Jul 2014 16:52:01 +0000 (18:52 +0200)]
clocksource: pxa: Move PXA timer to clocksource framework

Move time.c from arch/arm/mach-pxa/time.c to
drivers/clocksource/pxa_timer.c.

Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: clps711x: Add DT bindings documentation
Alexander Shiyan [Sun, 13 Jul 2014 04:45:36 +0000 (08:45 +0400)]
clocksource: clps711x: Add DT bindings documentation

This patch adds DT binding documentation for the Cirrus Logic
CLPS711X-based CPUs clocksource subsystem.

Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: clps711x: Add CLPS711X clocksource driver
Alexander Shiyan [Sun, 13 Jul 2014 04:45:13 +0000 (08:45 +0400)]
clocksource: clps711x: Add CLPS711X clocksource driver

This adds the clocksource driver for Cirrus Logic CLPS711X series SoCs.
Designed primarily for migration CLPS711X subarch for multiplatform & DT,
for this as the "OF" and "non-OF" calls implemented.

Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: sirf: Fix incorrect clock enable counter for timer
Zhiwu Song [Thu, 3 Jul 2014 12:52:51 +0000 (20:52 +0800)]
clocksource: sirf: Fix incorrect clock enable counter for timer

In the clocksource driver, we didn't explicitly enable the clock. it makes the
clk reference counter wrong. We didn't encounter any hang issue because the
tick's clock input has been open and is shared by some other hardware
components, but if we don't enable those components in kernel, in the stage of
disabling unused clk in kernel boot, Linux tick hangs.

This patch fixes it. it does an explicit prepare and enable to the clock input,
and increases the usage counter of the clk.

Signed-off-by: Zhiwu Song <Zhiwu.Song@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: Kconfig: Let EM_TIMER_STI depend on HAS_IOMEM
Chen Gang [Tue, 8 Jul 2014 12:39:40 +0000 (20:39 +0800)]
clocksource: Kconfig: Let EM_TIMER_STI depend on HAS_IOMEM

In 'em_sti.c', it will call devm_ioremap_resource() which need
HAS_IOMEM. So need let EM_TIMER_STI depend on HAS_IOMEM, too.

The related error (with allmodconfig under score):

  LD      init/built-in.o
em_sti.c:(.text.em_sti_probe+0x84): undefined reference to `devm_ioremap_resource'
make: *** [vmlinux] Error 1

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agovendor-prefixes: Add prefix for Mediatek Inc.
Matthias Brugger [Fri, 18 Jul 2014 09:36:49 +0000 (11:36 +0200)]
vendor-prefixes: Add prefix for Mediatek Inc.

Signed-off-by: Matthias Brugger <matthias.bgg@gmail.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agodt-bindings: Add mtk-timer bindings
Matthias Brugger [Fri, 18 Jul 2014 09:36:46 +0000 (11:36 +0200)]
dt-bindings: Add mtk-timer bindings

Add binding documentation for the General Purpose Timer driver of
the Mediatek SoCs.

Signed-off-by: Matthias Brugger <matthias.bgg@gmail.com>
Acked-by: Sören Brinkmann <soren.brinkmann@xilinx.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: Add support for the Mediatek SoCs
Matthias Brugger [Fri, 18 Jul 2014 09:36:43 +0000 (11:36 +0200)]
clocksource: Add support for the Mediatek SoCs

This patch adds a clock source and clock event for the timer found
on the Mediatek SoCs.

The Mediatek General Purpose Timer block provides five 32 bit timers and
one 64 bit timer.

Two 32 bit timers are used by this driver:
TIMER1: clock events supporting periodic and oneshot events
TIMER2: clock source configured as a free running counter

The General Purpose Timer block can be run with two clocks. A 13 MHz system
clock and the RTC clock running at 32 KHz. This implementation uses the system
clock with no clock source divider.

The interrupts are shared between the different timers and have to be read back
from a register. We just enable one interrupt for the clock event. The clock
event timer is used by all cores.

Signed-off-by: Matthias Brugger <matthias.bgg@gmail.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoof: Provide a function to request and map memory
Matthias Brugger [Fri, 18 Jul 2014 09:36:39 +0000 (11:36 +0200)]
of: Provide a function to request and map memory

A call to of_iomap does not request the memory region. This patch adds the
function of_io_request_and_map which requests the memory region before
mapping it.

Signed-off-by: Matthias Brugger <matthias.bgg@gmail.com>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Suggested-by: Rob Herring <robh@kernel.org>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoclocksource: sh_mtu2: Tidy up Kconfig typo for MTU2
Kuninori Morimoto [Fri, 18 Jul 2014 09:36:36 +0000 (11:36 +0200)]
clocksource: sh_mtu2: Tidy up Kconfig typo for MTU2

It should be "MTU2" instead of "TMU2"

Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Acked-by: Wolfram Sang <wsa@sang-engineering.com>
Acked-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
9 years agoMerge branch 'clockevents/renesas-timers-dt' into clockevents/3.17
Daniel Lezcano [Wed, 23 Jul 2014 10:00:00 +0000 (12:00 +0200)]
Merge branch 'clockevents/renesas-timers-dt' into clockevents/3.17

9 years agotimerfd: Implement timerfd_ioctl method to restore timerfd_ctx::ticks, v3
Cyrill Gorcunov [Tue, 15 Jul 2014 21:54:54 +0000 (01:54 +0400)]
timerfd: Implement timerfd_ioctl method to restore timerfd_ctx::ticks, v3

The read() of timerfd files allows to fetch the number of timer ticks
while there is no way to set it back from userspace.

To restore the timer's state as it was at checkpoint moment we need
a path to bring @ticks back. Initially I thought about writing ticks
back via write() interface but it seems such API is somehow obscure.

Instead implement timerfd_ioctl() method with TFD_IOC_SET_TICKS
command which allows to adjust @ticks into non-zero value waking
up the waiters.

I wrapped code with CONFIG_CHECKPOINT_RESTORE which can be
dropped off if there users except c/r camp appear.

v2 (by akpm@):
 - Use define timerfd_ioctl NULL for non c/r config

v3:
 - Use copy_from_user for @ticks fetching since
   not all arch support get_user for 8 byte argument

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christopher Covington <cov@codeaurora.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Link: http://lkml.kernel.org/r/20140715215703.285617923@openvz.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agodocs: Procfs -- Document timerfd output
Cyrill Gorcunov [Tue, 15 Jul 2014 21:54:53 +0000 (01:54 +0400)]
docs: Procfs -- Document timerfd output

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Link: http://lkml.kernel.org/r/20140715215703.199905126@openvz.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agotimerfd: Implement show_fdinfo method
Cyrill Gorcunov [Tue, 15 Jul 2014 21:54:52 +0000 (01:54 +0400)]
timerfd: Implement show_fdinfo method

For checkpoint/restore of timerfd files we need to know how exactly
the timer were armed, to be able to recreate it on restore stage.
Thus implement show_fdinfo method which provides enough information
for that.

One of significant changes I think is the addition of @settime_flags
member. Currently there are two flags TFD_TIMER_ABSTIME and
TFD_TIMER_CANCEL_ON_SET, and the second can be found from
@might_cancel variable but in case if the flags will be extended
in future we most probably will have to somehow remember them
explicitly anyway so I guss doing that right now won't hurt.

To not bloat the timerfd_ctx structure I've converted @expired
to short integer and defined @settime_flags as short too.

v2 (by avagin@, vdavydov@ and tglx@):

 - Add it_value/it_interval fields
 - Save flags being used in timerfd_setup in context

v3 (by tglx@):
 - don't forget to use CONFIG_PROC_FS

v4 (by akpm@):
 -Use define timerfd_show NULL for non c/r config

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Link: http://lkml.kernel.org/r/20140715215703.114365649@openvz.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoMerge tag 'v3.16-rc5' into timers/core
Thomas Gleixner [Wed, 16 Jul 2014 19:57:38 +0000 (21:57 +0200)]
Merge tag 'v3.16-rc5' into timers/core

Reason: Bring in upstream modifications, so the pending changes which
depend on them can be queued.

9 years agoLinux 3.16-rc5 v3.16-rc5
Linus Torvalds [Sun, 13 Jul 2014 21:04:33 +0000 (14:04 -0700)]
Linux 3.16-rc5

9 years agoMerge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 13 Jul 2014 20:14:55 +0000 (13:14 -0700)]
Merge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4

Pull ext4 bugfixes from Ted Ts'o:
 "More bug fixes for ext4 -- most importantly, a fix for a bug
  introduced in 3.15 that can end up triggering a file system corruption
  error after a journal replay.

  It shouldn't lead to any actual data corruption, but it is scary and
  can force file systems to be remounted read-only, etc"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: fix potential null pointer dereference in ext4_free_inode
  ext4: fix a potential deadlock in __ext4_es_shrink()
  ext4: revert commit which was causing fs corruption after journal replays
  ext4: disable synchronous transaction batching if max_batch_time==0
  ext4: clarify ext4_error message in ext4_mb_generate_buddy_error()
  ext4: clarify error count warning messages
  ext4: fix unjournalled bg descriptor while initializing inode bitmap

9 years agoMerge tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mike.turquette/linux
Linus Torvalds [Sun, 13 Jul 2014 19:21:04 +0000 (12:21 -0700)]
Merge tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mike.turquette/linux

Pull clock driver fixes from Mike Turquette:
 "This batch of fixes is for a handful of clock drivers from Allwinner,
  Samsung, ST & TI.  Most of them are of the "this hardware won't work
  without this fix" variety, including patches that fix platforms that
  did not boot under certain configurations.  Other fixes are the result
  of changes to the clock core introduced in 3.15 that had subtle
  impacts on the clock drivers.

  There are no fixes to the clock framework core in this pull request"

* tag 'clk-fixes-for-linus' of git://git.linaro.org/people/mike.turquette/linux:
  clk: spear3xx: Set proper clock parent of uart1/2
  clk: spear3xx: Use proper control register offset
  clk: qcom: HDMI source sel is 3 not 2
  clk: sunxi: fix devm_ioremap_resource error detection code
  clk: s2mps11: Fix double free corruption during driver unbind
  clk: ti: am43x: Fix boot with CONFIG_SOC_AM33XX disabled
  clk: exynos5420: Remove aclk66_peric from the clock tree description
  clk/exynos5250: fix bit number for tv sysmmu clock
  clk: s3c64xx: Hookup SPI clocks correctly
  clk: samsung: exynos4: Remove SRC_MASK_ISP gates
  clk: samsung: add more aliases for s3c24xx
  clk: samsung: fix several typos to fix boot on s3c2410
  clk: ti: set CLK_SET_RATE_NO_REPARENT for ti,mux-clock
  clk: ti: am43x: Fix boot with CONFIG_SOC_AM33XX disabled
  clk: ti: dra7: return error code in failure case
  clk: ti: apll: not allocating enough data

9 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Sun, 13 Jul 2014 19:10:18 +0000 (12:10 -0700)]
Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "This week's arm-soc fixes:

   - Another set of OMAP fixes
     * Clock fixes
     * Restart handling
     * PHY regulators
     * SATA hwmod data for DRA7
     + Some trivial fixes and removal of a bit of dead code
   - Exynos fixes
     * A bunch of clock fixes
     * Some SMP fixes
     * Exynos multi-core timer: register as clocksource and fix ftrace.
     + a few other minor fixes

  There's also a couple more patches, and at91 fix for USB caused by
  common clock conversion, and more MAINTAINERS entries for shmobile.

  We're definitely switching to only regression fixes from here on out,
  we've been a little less strict than usual up until now"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (26 commits)
  ARM: at91: at91sam9x5: add clocks for usb device
  ARM: EXYNOS: Register cpuidle device only on exynos4210 and 5250
  ARM: dts: Add clock property for mfc_pd in exynos5420
  clk: exynos5420: Add IDs for clocks used in PD mfc
  ARM: EXYNOS: Add support for clock handling in power domain
  ARM: OMAP2+: Remove non working OMAP HDMI audio initialization
  ARM: imx: fix shared gate clock
  ARM: dts: Update the parent for Audss clocks in Exynos5420
  ARM: EXYNOS: Update secondary boot addr for secure mode
  ARM: dts: Fix TI CPSW Phy mode selection on IGEP COM AQUILA.
  ARM: dts: am335x-evmsk: Enable the McASP FIFO for audio
  ARM: dts: am335x-evm: Enable the McASP FIFO for audio
  ARM: OMAP2+: Make GPMC skip disabled devices
  ARM: OMAP2+: create dsp device only on OMAP3 SoCs
  ARM: dts: dra7-evm: Make VDDA_1V8_PHY supply always on
  ARM: DRA7/AM43XX: fix header definition for omap44xx_restart
  ARM: OMAP2+: clock/dpll: fix _dpll_test_fint arithmetics overflow
  ARM: DRA7: hwmod: Add SYSCONFIG for usb_otg_ss
  ARM: DRA7: hwmod: Fixup SATA hwmod
  ARM: OMAP3: PRM/CM: Add back macros used by TI DSP/Bridge driver
  ...

9 years agoMerge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 13 Jul 2014 19:09:18 +0000 (12:09 -0700)]
Merge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 "Another round of fixes for ARM:
   - a set of kprobes fixes from Jon Medhurst
   - fix the revision checking for the L2 cache which wasn't noticed to
     have been broken"

* 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  ARM: l2c: fix revision checking
  ARM: kprobes: Fix test code compilation errors for ARMv4 targets
  ARM: kprobes: Disallow instructions with PC and register specified shift
  ARM: kprobes: Prevent known test failures stopping other tests running

9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux...
Linus Torvalds [Sun, 13 Jul 2014 19:04:06 +0000 (12:04 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k

Pull m68k fixes from Geert Uytterhoeven:
 "Summary:
  - Fix for a boot regression introduced in v3.16-rc1,
  - Fix for a build issue in -next"

Christoph Hellwig questioned why mach_random_get_entropy should be
exported to modules, and Geert explains that random_get_entropy() is
called by at least the crypto layer and ends up using it on m68k.  On
most other architectures it just uses get_cycles() (which is typically
inlined and doesn't need exporting),

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k:
  m68k: Export mach_random_get_entropy to modules
  m68k: Fix boot regression on machines with RAM at non-zero

9 years agoMerge branch 'parisc-3.16-5' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sun, 13 Jul 2014 19:02:05 +0000 (12:02 -0700)]
Merge branch 'parisc-3.16-5' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "The major patch in here is one which fixes the fanotify_mark() syscall
  in the compat layer of the 64bit parisc kernel.  It went unnoticed so
  long, because the calling syntax when using a 64bit parameter in a
  32bit syscall is quite complex and even worse, it may be even
  different if you call syscall() or the glibc wrapper.  This patch
  makes the kernel accept the calling convention when called by the
  glibc wrapper.

  The other two patches are trivial and remove unused headers, #includes
  and adds the serial ports of the fastest C8000 workstation to the
  parisc-kernel internal hardware database"

* 'parisc-3.16-5' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: drop unused defines and header includes
  parisc: fix fanotify_mark() syscall on 32bit compat kernel
  parisc: add serial ports of C8000/1GHz machine to hardware database

9 years agoclk: spear3xx: Set proper clock parent of uart1/2
Thomas Gleixner [Thu, 19 Jun 2014 21:52:24 +0000 (21:52 +0000)]
clk: spear3xx: Set proper clock parent of uart1/2

The uarts only work when the parent is ras_ahb_clk. The stale 3.5
based ST tree does this in the board file.

Add it to the clk init function. Not pretty, but the mess there is
amazing anyway.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Mike Turquette <mturquette@linaro.org>
9 years agoclk: spear3xx: Use proper control register offset
Thomas Gleixner [Thu, 19 Jun 2014 21:52:23 +0000 (21:52 +0000)]
clk: spear3xx: Use proper control register offset

The control register is at offset 0x10, not 0x0. This is wreckaged
since commit 5df33a62c (SPEAr: Switch to common clock framework).

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Mike Turquette <mturquette@linaro.org>
9 years agoparisc: drop unused defines and header includes
Helge Deller [Thu, 10 Jul 2014 16:08:11 +0000 (18:08 +0200)]
parisc: drop unused defines and header includes

Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # 3.13+
9 years agoparisc: fix fanotify_mark() syscall on 32bit compat kernel
Helge Deller [Thu, 10 Jul 2014 16:07:17 +0000 (18:07 +0200)]
parisc: fix fanotify_mark() syscall on 32bit compat kernel

On parisc we can not use the existing compat implementation for fanotify_mark()
because for the 64bit mask parameter the higher and lower 32bits are ordered
differently than what the compat function expects from big endian
architectures.

Specifically:
It finally turned out, that on hppa we end up with different assignments
of parameters to kernel arguments depending on if we call the glibc
wrapper function
 int fanotify_mark (int __fanotify_fd, unsigned int __flags,
                    uint64_t __mask, int __dfd, const char *__pathname);
or directly calling the syscall manually
 syscall(__NR_fanotify_mark, ...)

Reason is, that the syscall() function is implemented as C-function and
because we now have the sysno as first parameter in front of the other
parameters the compiler will unexpectedly add an empty paramenter in
front of the u64 value to ensure the correct calling alignment for 64bit
values.
This means, on hppa you can't simply use syscall() to call the kernel
fanotify_mark() function directly, but you have to use the glibc
function instead.

This patch fixes the kernel in the hppa-arch specifc coding to adjust
the parameters in a way as if userspace calls the glibc wrapper function
fanotify_mark().

Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # 3.13+