Commit Graph

28 Commits

Author SHA1 Message Date
Ingo Molnar
fa623d1b02 Merge branches 'x86/apic', 'x86/cleanups', 'x86/cpufeature', 'x86/crashdump', 'x86/debug', 'x86/defconfig', 'x86/detect-hyper', 'x86/doc', 'x86/dumpstack', 'x86/early-printk', 'x86/fpu', 'x86/idle', 'x86/io', 'x86/memory-corruption-check', 'x86/microcode', 'x86/mm', 'x86/mtrr', 'x86/nmi-watchdog', 'x86/pat2', 'x86/pci-ioapic-boot-irq-quirks', 'x86/ptrace', 'x86/quirks', 'x86/reboot', 'x86/setup-memory', 'x86/signal', 'x86/sparse-fixes', 'x86/time', 'x86/uv' and 'x86/xen' into x86/core 2008-12-23 16:27:23 +01:00
Ingo Molnar
7cbaef9c83 sched: optimize sched_clock() a bit
sched_clock() uses cycles_2_ns() needlessly - which is an irq-disabling
variant of __cycles_2_ns().

Most of the time sched_clock() is called with irqs disabled already.
The few places that call it with irqs enabled need to be updated.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-11-08 17:05:38 +01:00
Alok Kataria
70de9a9704 x86: don't use tsc_khz to calculate lpj if notsc is passed
Impact: fix udelay when "notsc" boot parameter is passed

With notsc passed on commandline, tsc may not be used for
udelays, make sure that we do not use tsc_khz to calculate
the lpj value in such cases.

Reported-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Signed-off-by: Alok N Kataria <akataria@vmware.com>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-11-04 09:55:26 +01:00
Alok Kataria
395628ef4e x86: Skip verification by the watchdog for TSC clocksource.
Impact: Changes timekeeping on Vmware (or with tsc=reliable).

This is achieved by resetting the CLOCKSOURCE_MUST_VERIFY flag.

We add a tsc=reliable commandline option to enable this.
This enables legacy hardware without HPET, LAPIC, or ACPI timers
to enter high-resolution timer mode.

Along with that have extended this to be used in virtualization environement
too. Now we also set this flag if the X86_FEATURE_TSC_RELIABLE bit is set.

This is important since there is a wrap-around problem with the acpi_pm timer.
The acpi_pm counter is just 24bits and this can overflow in ~4 seconds. With
the NO_HZ kernels in virtualized environment, there can be situations when
the guest is descheduled for longer duration, as a result we may miss the wrap
of the acpi counter. When TSC is used as a clocksource and acpi_pm timer is
being used as the watchdog clocksource this error in acpi_pm results in TSC
being marked as unstable, and essentially results in time dropping in chunks
of 4 seconds whenever this wrap is missed. Since the virtualized TSC is
reliable on VMware, we should always use the TSCs clocksource on VMware, so
we skip the verfication at runtime, by checking for the feature bit.

Since we reset the flag for mgeode systems too, i have combined
the mgeode case with the feature bit check.

Signed-off-by: Jeff Hansen <jhansen@cardaccess-inc.com>
Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2008-11-01 18:59:03 -07:00
Alok Kataria
88b094fb8d x86: Hypervisor detection and get tsc_freq from hypervisor
Impact: Changes timebase calibration on Vmware.

v3->v2 : Abstract the hypervisor detection and feature (tsc_freq) request
	 behind a hypervisor.c file
v2->v1 : Add a x86_hyper_vendor field to the cpuinfo_x86 structure.
	 This avoids multiple calls to the hypervisor detection function.

This patch adds function to detect if we are running under VMware.
The current way to check if we are on VMware is following,
#  check if "hypervisor present bit" is set, if so read the 0x40000000
   cpuid leaf and check for "VMwareVMware" signature.
#  if the above fails, check the DMI vendors name for "VMware" string
   if we find one we query the VMware hypervisor port to check if we are
   under VMware.

The DMI + "VMware hypervisor port check" is needed for older VMware products,
which don't implement the hypervisor signature cpuid leaf.
Also note that since we are checking for the DMI signature the hypervisor
port should never be accessed on native hardware.

This patch also adds a hypervisor_get_tsc_freq function, instead of
calibrating the frequency which can be error prone in virtualized
environment, we ask the hypervisor for it. We get the frequency from
the hypervisor by accessing the hypervisor port if we are running on VMware.
Other hypervisors too can add code to the generic routine to get frequency on
their platform.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2008-11-01 18:57:08 -07:00
James Bottomley
017d9d20d8 x86: use CONFIG_X86_SMP instead of CONFIG_SMP
Impact: fix x86/Voyager boot

CONFIG_SMP is used for features which work on *all* x86 boxes.
CONFIG_X86_SMP is used for standard PC like x86 boxes (for things like
multi core and apics)

Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-10-30 22:53:10 +01:00
Ingo Molnar
e496e3d645 Merge branches 'x86/alternatives', 'x86/cleanups', 'x86/commandline', 'x86/crashdump', 'x86/debug', 'x86/defconfig', 'x86/doc', 'x86/exports', 'x86/fpu', 'x86/gart', 'x86/idle', 'x86/mm', 'x86/mtrr', 'x86/nmi-watchdog', 'x86/oprofile', 'x86/paravirt', 'x86/reboot', 'x86/sparse-fixes', 'x86/tsc', 'x86/urgent' and 'x86/vmalloc' into x86-v28-for-linus-phase1 2008-10-06 18:17:07 +02:00
Ingo Molnar
5df4551551 x86, tsc calibration: fix
my brown paperbag day ...

Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-06 23:55:40 +02:00
Ingo Molnar
4156e9a8ef x86: quick TSC calibration, improve
- make sure the final TSC timestamp is reliable too

Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-04 23:21:57 +02:00
Linus Torvalds
6ac40ed041 x86: quick TSC calibration
Introduce a fast TSC-calibration method on sane hardware.

It only uses 17920 PIT timer ticks to calibrate the TSC, plus 256 ticks on
each side to make sure the TSC values were very close to the tick, so the
whole calibration takes 15ms. Yet, despite only takign 15ms,
we can actually give pretty stringent guarantees of accuracy:

 - the code requires that we hit each 256-counter block at least 50 times,
   so the TSC error is basically at *MOST* just a few PIT cycles off in
   any direction. In practice, it's going to be about one microseconds
   off (which is how long it takes to read the counter)

 - so over 17920 PIT cycles, we can pretty much guarantee that the
   calibration error is less than one half of a percent.

My testing bears this out: on my machine, the quick-calibration reports
2934.085kHz, while the slow one reports 2933.415.

Yes, the slower calibration is still more precise. For me, the slow
calibration is stable to within about one hundreth of a percent, so it's
(at a guess) roughly an order-and-a-half of magnitude more precise. The
longer you wait, the more precise you can be.

However, the nice thing about the fast TSC PIT synchronization is that
it's pretty much _guaranteed_ to give that 0.5% precision, and fail
gracefully (and very quickly) if it doesn't get it. And it really is
fairly simple (even if there's a lot of _details_ there, and I didn't get
all of those right ont he first try or even the second ;)

The patch says "110 insertions", but 63 of those new lines are actually
comments.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 arch/x86/kernel/tsc.c |  111 ++++++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 110 insertions(+), 1 deletions(-)
2008-09-04 22:54:50 +02:00
Thomas Gleixner
a977c40095 x86: TSC make the calibration loop smarter
The last changes made the calibration loop 250ms long which is far
too much. Try to do that more clever.

Experiments have shown that using a 10ms delay for the PIT based calibration
gives us a good enough value. If we have a reference (HPET/PMTIMER) and the
result of the PIT and the reference is close enough, then we can break out of
the calibration loop on a match right away and use the reference value.

Otherwise we just loop 3 times and decide then, which value to take.

One caveat is that for virtualized environments the PIT calibration often does
not work at all and I found out that 10us is a bit too short as well for the
reference to give a sane result. The solution here is to make the last loop
longer when the first two PIT calibrations failed.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-04 17:35:35 +02:00
Thomas Gleixner
827014be05 x86: TSC: use one set of reference variables
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-04 17:35:34 +02:00
Thomas Gleixner
d683ef7afe x86: TSC: separate hpet/pmtimer calculation out
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-04 17:35:33 +02:00
Thomas Gleixner
cce3e05724 x86: TSC: define the PIT latch value separate
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-04 17:35:33 +02:00
Alok N Kataria
de014d6176 x86: Change warning message in TSC calibration.
When calibration against PIT fails, the warning that we print is misleading.
In a virtualized environment the VM may get descheduled while calibration
or, the check in PIT calibration may fail due to other virtualization
overheads.

The warning message explicitly assumes that calibration failed due to SMI's
which may not be the case. Change that to something proper.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-03 20:10:37 -07:00
Linus Torvalds
ec0c15afb4 Split up PIT part of TSC calibration from native_calibrate_tsc
The TSC calibration function is still very complicated, but this makes
it at least a little bit less so by moving the PIT part out into a
helper function of its own.

Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-03 07:30:13 -07:00
Thomas Gleixner
fbb16e2438 [x86] Fix TSC calibration issues
Larry Finger reported at http://lkml.org/lkml/2008/9/1/90:
An ancient laptop of mine started throwing errors from b43legacy when
I started using 2.6.27 on it. This has been bisected to commit bfc0f59
"x86: merge tsc calibration".

The unification of the TSC code adopted mostly the 64bit code, which
prefers PMTIMER/HPET over the PIT calibration.

Larrys system has an AMD K6 CPU. Such systems are known to have
PMTIMER incarnations which run at double speed. This results in a
miscalibration of the TSC by factor 0.5. So the resulting calibrated
CPU/TSC speed is half of the real CPU speed, which means that the TSC
based delay loop will run half the time it should run. That might
explain why the b43legacy driver went berserk.

On the other hand we know about systems, where the PIT based
calibration results in random crap due to heavy SMI/SMM
disturbance. On those systems the PMTIMER/HPET based calibration logic
with SMI detection shows better results.

According to Alok also virtualized systems suffer from the PIT
calibration method.

The solution is to use a more wreckage aware aproach than the current
either/or decision.

1) reimplement the retry loop which was dropped from the 32bit code
during the merge. It repeats the calibration and selects the lowest
frequency value as this is probably the closest estimate to the real
frequency

2) Monitor the delta of the TSC values in the delay loop which waits
for the PIT counter to reach zero. If the maximum value is
significantly different from the minimum, then we have a pretty safe
indicator that the loop was disturbed by an SMI.

3) keep the pmtimer/hpet reference as a backup solution for systems
where the SMI disturbance is a permanent point of failure for PIT
based calibration

4) do the loop iteration for both methods, record the lowest value and
decide after all iterations finished.

5) Set a clear preference to PIT based calibration when the result
makes sense.

The implementation does the reference calibration based on
HPET/PMTIMER around the delay, which is necessary for the PIT anyway,
but keeps separate TSC values to ensure the "independency" of the
resulting calibration values.

Tested on various 32bit/64bit machines including Geode 266Mhz, AMD K6
(affected machine with a double speed pmtimer which I grabbed out of
the dump), Pentium class machines and AMD/Intel 64 bit boxen.

Bisected-by:  Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-02 20:35:56 -07:00
Peter Zijlstra
52a8968ce9 x86: fix cpufreq + sched_clock() regression
I noticed that my sched_clock() was slow on a number of machine, so I
started looking at cpufreq.

The below seems to fix the problem for me.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-08-25 14:39:19 +02:00
Linus Torvalds
060700b571 x86: do not enable TSC notifier if we don't need it
Impact: crash on non-TSC-equipped CPUs

Don't enable the TSC notifier if we *either*:

1. don't have a CPU, or
2. have a CPU with constant TSC.

In either of those cases, the notifier is either damaging (1) or useless(2).

From: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2008-08-24 17:16:28 -07:00
Marcin Slusarz
d554d9a429 x86, tsc: fix section mismatch warning
WARNING: vmlinux.o(.text+0x7950): Section mismatch in reference from the function native_calibrate_tsc() to the function .init.text:tsc_read_refs()
The function native_calibrate_tsc() references
the function __init tsc_read_refs().
This is often because native_calibrate_tsc lacks a __init
annotation or the annotation of tsc_read_refs is wrong.

tsc_read_refs is called from native_calibrate_tsc which is not __init
and native_calibrate_tsc cannot be marked __init

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-08-18 07:48:07 +02:00
Thomas Gleixner
431ceb83f7 x86: fix TSC build error on 32bit
Dave Hansen reported a build error on 32bit which went unnoticed
as newer gcc versions seem to optimize unused static functions
away before compiling them.

Make vread_tsc() depend on CONFIG_X86_64

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-07-15 22:46:47 +02:00
Glauber Costa
e54afe3863 x86: remove duplicate call to use_tsc_delay
Integration generated a duplicate call to use_tsc_delay.
Particularly, the one that is done before we check for general
tsc usability seems wrong.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-10 19:47:55 +02:00
Glauber Costa
0a4d8a472f x86: provide delay loop for x86_64.
This is for consistency with i386. We call use_tsc_delay()
at tsc initialization for x86_64, so we'll be always using it.

Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 08:51:41 +02:00
Alok Kataria
e93ef949fd x86: rename paravirtualized TSC functions
Rename the paravirtualized calculate_cpu_khz to calibrate_tsc.
In all cases, we actually calibrate_tsc and use that as the cpu_khz value.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Cc: Dan Hecht <dhecht@vmware.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 07:43:28 +02:00
Alok Kataria
8fbbc4b45c x86: merge tsc_init and clocksource code
Unify the clocksource code.
Unify the tsc_init code.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Cc: Dan Hecht <dhecht@vmware.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 07:43:27 +02:00
Alok Kataria
2dbe06faf3 x86: merge the TSC cpu-freq code
Unify the TSC cpufreq code.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Cc: Dan Hecht <dhecht@vmware.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 07:43:26 +02:00
Alok Kataria
bfc0f5947a x86: merge tsc calibration
Merge the tsc calibration code for the 32bit and 64bit kernel.
The paravirtualized calculate_cpu_khz for 64bit now points to the correct
tsc_calibrate code as in 32bit.
Original native_calculate_cpu_khz for 64 bit is now called as calibrate_cpu.

Also moved the recalibrate_cpu_khz function in the common file.
Note that this function is called only from powernow K7 cpu freq driver.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Cc: Dan Hecht <dhecht@vmware.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 07:43:25 +02:00
Alok Kataria
0ef9553332 x86: merge sched_clock handling
Move the basic global variable definitions and sched_clock handling in the
common "tsc.c" file.

 - Unify notsc kernel command line handling for 32 bit and 64bit.
 - Functional changes for 64bit.
        - "tsc_disabled" is updated if "notsc" is passed at boottime.
        - Fallback to jiffies for sched_clock, incase notsc is passed on
	  commandline.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: Dan Hecht <dhecht@vmware.com>
Cc: Dan Hecht <dhecht@vmware.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-09 07:43:25 +02:00