src/patches/suse-2.6.27.25/patches.arch/x86-vmware-tsc-05-skip-tsc-clocksource

   1 From: Alok Kataria <akataria@vmware.com>
   2 Subject: x86: Skip verification by the watchdog for TSC clocksource.
   3 Patch-mainline:
   4 References: bnc#441338
   5
   6 Impact: Changes timekeeping on Vmware (or with tsc=reliable).
   7
   8 This is achieved by resetting the CLOCKSOURCE_MUST_VERIFY flag.
   9
  10 We add a tsc=reliable commandline option to enable this.
  11 This enables legacy hardware without HPET, LAPIC, or ACPI timers
  12 to enter high-resolution timer mode.
  13
  14 Along with that have extended this to be used in virtualization environement
  15 too. Now we also set this flag if the X86_FEATURE_TSC_RELIABLE bit is set.
  16 This is important since there is a wrap-around problem with the acpi_pm timer.
  17 The acpi_pm counter is just 24bits and this can overflow in ~4 seconds. With
  18 the NO_HZ kernels in virtualized environment, there can be situations when
  19 the guest is descheduled for longer duration, as a result we may miss the wrap
  20 of the acpi counter. When TSC is used as a clocksource and acpi_pm timer is
  21 being used as the watchdog clocksource this error in acpi_pm results in TSC
  22 being marked as unstable, and essentially results in time dropping in chunks
  23 of 4 seconds whenever this wrap is missed. Since the virtualized TSC is
  24 reliable on VMware, we should always use the TSCs clocksource on VMware, so
  25 we skip the verfication at runtime, by checking for the feature bit.
  26
  27 Since we reset the flag for mgeode systems too, i have combined
  28 the mgeode case with the feature bit check.
  29
  30 Signed-off-by: Jeff Hansen <jhansen@cardaccess-inc.com>
  31 Signed-off-by: Alok N Kataria <akataria@vmware.com>
  32 Signed-off-by: Dan Hecht <dhecht@vmware.com>
  33 Signed-off-by: H. Peter Anvin <hpa@zytor.com>
  34 Signed-off-by: Takashi Iwai <tiwai@suse.de>
  35
  36 ---
  37
  38  Documentation/kernel-parameters.txt |    7 +++++++
  39  arch/x86/kernel/tsc.c               |   33 +++++++++++++++++++++------------
  40  2 files changed, 28 insertions(+), 12 deletions(-)
  41
  42
  43 --- a/arch/x86/kernel/tsc.c
  44 +++ b/arch/x86/kernel/tsc.c
  45 @@ -32,6 +32,7 @@ static int tsc_unstable;
  46     erroneous rdtsc usage on !cpu_has_tsc processors */
  47  static int tsc_disabled = -1;
  48
  49 +static int tsc_clocksource_reliable;
  50  /*
  51   * Scheduler clock - returns current time in nanosec units.
  52   */
  53 @@ -99,6 +100,15 @@ int __init notsc_setup(char *str)
  54
  55  __setup("notsc", notsc_setup);
  56
  57 +static int __init tsc_setup(char *str)
  58 +{
  59 +       if (!strcmp(str, "reliable"))
  60 +               tsc_clocksource_reliable = 1;
  61 +       return 1;
  62 +}
  63 +
  64 +__setup("tsc=", tsc_setup);
  65 +
  66  #define MAX_RETRIES     5
  67  #define SMI_TRESHOLD    50000
  68
  69 @@ -564,24 +574,21 @@ static struct dmi_system_id __initdata b
  70         {}
  71  };
  72
  73 -/*
  74 - * Geode_LX - the OLPC CPU has a possibly a very reliable TSC
  75 - */
  76 +static void __init check_system_tsc_reliable(void)
  77 +{
  78  #ifdef CONFIG_MGEODE_LX
  79 -/* RTSC counts during suspend */
  80 +       /* RTSC counts during suspend */
  81  #define RTSC_SUSP 0x100
  82 -
  83 -static void __init check_geode_tsc_reliable(void)
  84 -{
  85         unsigned long res_low, res_high;
  86
  87         rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
  88 +       /* Geode_LX - the OLPC CPU has a possibly a very reliable TSC */
  89         if (res_low & RTSC_SUSP)
  90 -               clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
  91 -}
  92 -#else
  93 -static inline void check_geode_tsc_reliable(void) { }
  94 +               tsc_clocksource_reliable = 1;
  95  #endif
  96 +       if (boot_cpu_has(X86_FEATURE_TSC_RELIABLE))
  97 +               tsc_clocksource_reliable = 1;
  98 +}
  99
 100  /*
 101   * Make an educated guess if the TSC is trustworthy and synchronized
 102 @@ -616,6 +623,8 @@ static void __init init_tsc_clocksource(
 103  {
 104         clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
 105                         clocksource_tsc.shift);
 106 +       if (tsc_clocksource_reliable)
 107 +               clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
 108         /* lower the rating if we already know its unstable: */
 109         if (check_tsc_unstable()) {
 110                 clocksource_tsc.rating = 0;
 111 @@ -676,7 +685,7 @@ void __init tsc_init(void)
 112         if (unsynchronized_tsc())
 113                 mark_tsc_unstable("TSCs unsynchronized");
 114
 115 -       check_geode_tsc_reliable();
 116 +       check_system_tsc_reliable();
 117         init_tsc_clocksource();
 118  }
 119
 120 --- a/Documentation/kernel-parameters.txt
 121 +++ b/Documentation/kernel-parameters.txt
 122 @@ -2200,6 +2200,13 @@ and is between 256 and 4096 characters.
 123                         Format:
 124                         <io>,<irq>,<dma>,<dma2>,<sb_io>,<sb_irq>,<sb_dma>,<mpu_io>,<mpu_irq>
 125
 126 +       tsc=            Disable clocksource-must-verify flag for TSC.
 127 +                       Format: <string>
 128 +                       [x86] reliable: mark tsc clocksource as reliable, this
 129 +                       disables clocksource verification at runtime.
 130 +                       Used to enable high-resolution timer mode on older
 131 +                       hardware, and in virtualized environment.
 132 +
 133         turbografx.map[2|3]=    [HW,JOY]
 134                         TurboGraFX parallel port interface
 135                         Format: