linux-stable.git/kernel/time/timekeeping.c, branch v4.1.41

timekeeping_Force_unsigned_clocksource_to_nanoseconds_conversion

2017-01-13T01:56:56+00:00

[ Upstream commit 9c1645727b8fa90d07256fdfcc45bf831242a3ab ]

The clocksource delta to nanoseconds conversion is using signed math, but
the delta is unsigned. This makes the conversion space smaller than
necessary and in case of a multiplication overflow the conversion can
become negative. The conversion is done with scaled math:

    s64 nsec_delta = ((s64)clkdelta * clk->mult) >> clk->shift;

Shifting a signed integer right obvioulsy preserves the sign, which has
interesting consequences:

 - Time jumps backwards

 - __iter_div_u64_rem() which is used in one of the calling code pathes
   will take forever to piecewise calculate the seconds/nanoseconds part.

This has been reported by several people with different scenarios:

David observed that when stopping a VM with a debugger:

 "It was essentially the stopped by debugger case.  I forget exactly why,
  but the guest was being explicitly stopped from outside, it wasn't just
  scheduling lag.  I think it was something in the vicinity of 10 minutes
  stopped."

 When lifting the stop the machine went dead.

The stopped by debugger case is not really interesting, but nevertheless it
would be a good thing not to die completely.

But this was also observed on a live system by Liav:

 "When the OS is too overloaded, delta will get a high enough value for the
  msb of the sum delta * tkr->mult + tkr->xtime_nsec to be set, and so
  after the shift the nsec variable will gain a value similar to
  0xffffffffff000000."

Unfortunately this has been reintroduced recently with commit 6bd58f09e1d8
("time: Add cycles to nanoseconds translation"). It had been fixed a year
ago already in commit 35a4933a8959 ("time: Avoid signed overflow in
timekeeping_get_ns()").

Though it's not surprising that the issue has been reintroduced because the
function itself and the whole call chain uses s64 for the result and the
propagation of it. The change in this recent commit is subtle:

   s64 nsec;

-  nsec = (d * m + n) >> s:
+  nsec = d * m + n;
+  nsec >>= s;

d being type of cycle_t adds another level of obfuscation.

This wouldn't have happened if the previous change to unsigned computation
would have made the 'nsec' variable u64 right away and a follow up patch
had cleaned up the whole call chain.

There have been patches submitted which basically did a revert of the above
patch leaving everything else unchanged as signed. Back to square one. This
spawned a admittedly pointless discussion about potential users which rely
on the unsigned behaviour until someone pointed out that it had been fixed
before. The changelogs of said patches added further confusion as they made
finally false claims about the consequences for eventual users which expect
signed results.

Despite delta being cycle_t, aka. u64, it's very well possible to hand in
a signed negative value and the signed computation will happily return the
correct result. But nobody actually sat down and analyzed the code which
was added as user after the propably unintended signed conversion.

Though in sensitive code like this it's better to analyze it proper and
make sure that nothing relies on this than hunting the subtle wreckage half
a year later. After analyzing all call chains it stands that no caller can
hand in a negative value (which actually would work due to the s64 cast)
and rely on the signed math to do the right thing.

Change the conversion function to unsigned math. The conversion of all call
chains is done in a follow up patch.

This solves the starvation issue, which was caused by the negative result,
but it does not solve the underlying problem. It merily procrastinates
it. When the timekeeper update is deferred long enough that the unsigned
multiplication overflows, then time going backwards is observable again.

It does neither solve the issue of clocksources with a small counter width
which will wrap around possibly several times and cause random time stamps
to be generated. But those are usually not found on systems used for
virtualization, so this is likely a non issue.

I took the liberty to claim authorship for this simply because
analyzing all callsites and writing the changelog took substantially
more time than just making the simple s/s64/u64/ change and ignore the
rest.

Fixes: 6bd58f09e1d8 ("time: Add cycles to nanoseconds translation")
Reported-by: David Gibson 
Reported-by: Liav Rehana 
Signed-off-by: Thomas Gleixner 
Reviewed-by: David Gibson 
Acked-by: Peter Zijlstra (Intel) 
Cc: Parit Bhargava 
Cc: Laurent Vivier 
Cc: "Christopher S. Hall" 
Cc: Chris Metcalf 
Cc: Richard Cochran 
Cc: John Stultz 
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/20161208204228.688545601@linutronix.de
Signed-off-by: Thomas Gleixner 
Signed-off-by: Sasha Levin

timekeeping: Fix __ktime_get_fast_ns() regression

2016-10-23T23:37:51+00:00

[ Upstream commit 58bfea9532552d422bde7afa207e1a0f08dffa7d ]

In commit 27727df240c7 ("Avoid taking lock in NMI path with
CONFIG_DEBUG_TIMEKEEPING"), I changed the logic to open-code
the timekeeping_get_ns() function, but I forgot to include
the unit conversion from cycles to nanoseconds, breaking the
function's output, which impacts users like perf.

This results in bogus perf timestamps like:
 swapper     0 [000]   253.427536:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.426573:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.426687:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.426800:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.426905:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.427022:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.427127:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.427239:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.427346:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   254.427463:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]   255.426572:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])

Instead of more reasonable expected timestamps like:
 swapper     0 [000]    39.953768:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.064839:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.175956:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.287103:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.398217:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.509324:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.620437:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.731546:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.842654:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    40.953772:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])
 swapper     0 [000]    41.064881:  111111111 cpu-clock:  ffffffff810a0de6 native_safe_halt+0x6 ([kernel.kallsyms])

Add the proper use of timekeeping_delta_to_ns() to convert
the cycle delta to nanoseconds as needed.

Thanks to Brendan and Alexei for finding this quickly after
the v4.8 release. Unfortunately the problematic commit has
landed in some -stable trees so they'll need this fix as
well.

Many apologies for this mistake. I'll be looking to add a
perf-clock sanity test to the kselftest timers tests soon.

Fixes: 27727df240c7 "timekeeping: Avoid taking lock in NMI path with CONFIG_DEBUG_TIMEKEEPING"
Reported-by: Brendan Gregg 
Reported-by: Alexei Starovoitov 
Tested-and-reviewed-by: Mathieu Desnoyers 
Signed-off-by: John Stultz 
Cc: Peter Zijlstra 
Cc: stable 
Cc: Steven Rostedt 
Link: http://lkml.kernel.org/r/1475636148-26539-1-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner 
Signed-off-by: Sasha Levin

time: Add cycles to nanoseconds translation

2016-10-23T23:37:45+00:00

[ Upstream commit 6bd58f09e1d8cc6c50a824c00bf0d617919986a1 ]

The timekeeping code does not currently provide a way to translate
externally provided clocksource cycles to system time. The cycle count
is always provided by the result clocksource read() method internal to
the timekeeping code. The added function timekeeping_cycles_to_ns()
calculated a nanosecond value from a cycle count that can be added to
tk_read_base.base value yielding the current system time. This allows
clocksource cycle values external to the timekeeping code to provide a
cycle count that can be transformed to system time.

Cc: Prarit Bhargava 
Cc: Richard Cochran 
Cc: Thomas Gleixner 
Cc: Ingo Molnar 
Cc: Andy Lutomirski 
Cc: kevin.b.stanton@intel.com
Cc: kevin.j.clarke@intel.com
Cc: hpa@zytor.com
Cc: jeffrey.t.kirsher@intel.com
Cc: netdev@vger.kernel.org
Reviewed-by: Thomas Gleixner 
Signed-off-by: Christopher S. Hall 
Signed-off-by: John Stultz 
Signed-off-by: Sasha Levin

timekeeping: Avoid taking lock in NMI path with CONFIG_DEBUG_TIMEKEEPING

2016-08-31T23:21:09+00:00

[ Upstream commit 27727df240c7cc84f2ba6047c6f18d5addfd25ef ]

When I added some extra sanity checking in timekeeping_get_ns() under
CONFIG_DEBUG_TIMEKEEPING, I missed that the NMI safe __ktime_get_fast_ns()
method was using timekeeping_get_ns().

Thus the locking added to the debug checks broke the NMI-safety of
__ktime_get_fast_ns().

This patch open-codes the timekeeping_get_ns() logic for
__ktime_get_fast_ns(), so can avoid any deadlocks in NMI.

Fixes: 4ca22c2648f9 "timekeeping: Add warnings when overflows or underflows are observed"
Reported-by: Steven Rostedt 
Reported-by: Peter Zijlstra 
Signed-off-by: John Stultz 
Cc: stable 
Link: http://lkml.kernel.org/r/1471993702-29148-2-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner 
Signed-off-by: Sasha Levin

time: Avoid signed overflow in timekeeping_get_ns()

2016-02-01T16:37:05+00:00

[ Upstream commit 35a4933a895927990772ae96fdcfd2f806929ee2 ]

1e75fa8 "time: Condense timekeeper.xtime into xtime_sec" replaced a call to
clocksource_cyc2ns() from timekeeping_get_ns() with an open-coded version
of the same logic to avoid keeping a semi-redundant struct timespec
in struct timekeeper.

However, the commit also introduced a subtle semantic change - where
clocksource_cyc2ns() uses purely unsigned math, the new version introduces
a signed temporary, meaning that if (delta * tk->mult) has a 63-bit
overflow the following shift will still give a negative result.  The
choice of 'maxsec' in __clocksource_updatefreq_scale() means this will
generally happen if there's a ~10 minute pause in examining the
clocksource.

This can be triggered on a powerpc KVM guest by stopping it from qemu for
a bit over 10 minutes.  After resuming time has jumped backwards several
minutes causing numerous problems (jiffies does not advance, msleep()s can
be extended by minutes..).  It doesn't happen on x86 KVM guests, because
the guest TSC is effectively frozen while the guest is stopped, which is
not the case for the powerpc timebase.

Obviously an unsigned (64 bit) overflow will only take twice as long as a
signed, 63-bit overflow.  I don't know the time code well enough to know
if that will still cause incorrect calculations, or if a 64-bit overflow
is avoided elsewhere.

Still, an incorrect forwards clock adjustment will cause less trouble than
time going backwards.  So, this patch removes the potential for
intermediate signed overflow.

Cc: stable@vger.kernel.org  (3.7+)
Suggested-by: Laurent Vivier 
Tested-by: Laurent Vivier 
Signed-off-by: David Gibson 
Signed-off-by: John Stultz 
Signed-off-by: Sasha Levin

time: Fix timekeeping_freqadjust()'s incorrect use of abs() instead of abs64()

2015-10-22T21:43:11+00:00

commit 2619d7e9c92d524cb155ec89fd72875321512e5b upstream.

The internal clocksteering done for fine-grained error
correction uses a logarithmic approximation, so any time
adjtimex() adjusts the clock steering, timekeeping_freqadjust()
quickly approximates the correct clock frequency over a series
of ticks.

Unfortunately, the logic in timekeeping_freqadjust(), introduced
in commit:

  dc491596f639 ("timekeeping: Rework frequency adjustments to work better w/ nohz")

used the abs() function with a s64 error value to calculate the
size of the approximated adjustment to be made.

Per include/linux/kernel.h:

  "abs() should not be used for 64-bit types (s64, u64, long long) - use abs64()".

Thus on 32-bit platforms, this resulted in the clocksteering to
take a quite dampended random walk trying to converge on the
proper frequency, which caused the adjustments to be made much
slower then intended (most easily observed when large
adjustments are made).

This patch fixes the issue by using abs64() instead.

Reported-by: Nuno Gonçalves 
Tested-by: Nuno Goncalves 
Signed-off-by: John Stultz 
Cc: Linus Torvalds 
Cc: Miroslav Lichvar 
Cc: Peter Zijlstra 
Cc: Prarit Bhargava 
Cc: Richard Cochran 
Cc: Thomas Gleixner 
Link: http://lkml.kernel.org/r/1441840051-20244-1-git-send-email-john.stultz@linaro.org
Signed-off-by: Ingo Molnar 
Signed-off-by: Greg Kroah-Hartman

timekeeping: Get rid of stale comment

2015-04-03T06:44:37+00:00

Arch specific management of xtime/jiffies/wall_to_monotonic is
gone for quite a while. Zap the stale comment.

Signed-off-by: Thomas Gleixner 
Signed-off-by: Rafael J. Wysocki 
Acked-by: John Stultz 
Cc: Peter Zijlstra 
Cc: John Stultz 
Link: http://lkml.kernel.org/r/2422730.dmO29q661S@vostro.rjw.lan
Signed-off-by: Ingo Molnar

time, drivers/rtc: Don't bother with rtc_resume() for the nonstop clocksource

2015-04-03T06:18:34+00:00

If a system does not provide a persistent_clock(), the time
will be updated on resume by rtc_resume(). With the addition
of the non-stop clocksources for suspend timing, those systems
set the time on resume in timekeeping_resume(), but may not
provide a valid persistent_clock().

This results in the rtc_resume() logic thinking no one has set
the time and it then will over-write the suspend time again,
which is not necessary and only increases clock error.

So, fix this for rtc_resume().

This patch also improves the name of persistent_clock_exist to
make it more grammatical.

Signed-off-by: Xunlei Pang 
Signed-off-by: John Stultz 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Link: http://lkml.kernel.org/r/1427945681-29972-19-git-send-email-john.stultz@linaro.org
Signed-off-by: Ingo Molnar

time: Fix a bug in timekeeping_suspend() with no persistent clock

2015-04-03T06:18:33+00:00

When there's no persistent clock, normally
timekeeping_suspend_time should always be zero, but this can
break in timekeeping_suspend().

At T1, there was a system suspend, so old_delta was assigned T1.
After some time, one time adjustment happened, and xtime got the
value of T1-dt(0s
Signed-off-by: John Stultz 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Link: http://lkml.kernel.org/r/1427945681-29972-18-git-send-email-john.stultz@linaro.org
Signed-off-by: Ingo Molnar

time: Don't build timekeeping_inject_sleeptime64() if no one uses it

2015-04-03T06:18:31+00:00

timekeeping_inject_sleeptime64() is only used by RTC
suspend/resume, so add build dependencies on the necessary RTC
related macros.

Signed-off-by: Xunlei Pang 
[ Improve commit message clarity. ]
Signed-off-by: John Stultz 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Link: http://lkml.kernel.org/r/1427945681-29972-16-git-send-email-john.stultz@linaro.org
Signed-off-by: Ingo Molnar